xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision 4e1a33b1)
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 #include <linux/firmware.h>
24 #include "drmP.h"
25 #include "amdgpu.h"
26 #include "amdgpu_ih.h"
27 #include "amdgpu_gfx.h"
28 #include "cikd.h"
29 #include "cik.h"
30 #include "atom.h"
31 #include "amdgpu_ucode.h"
32 #include "clearstate_ci.h"
33 
34 #include "dce/dce_8_0_d.h"
35 #include "dce/dce_8_0_sh_mask.h"
36 
37 #include "bif/bif_4_1_d.h"
38 #include "bif/bif_4_1_sh_mask.h"
39 
40 #include "gca/gfx_7_0_d.h"
41 #include "gca/gfx_7_2_enum.h"
42 #include "gca/gfx_7_2_sh_mask.h"
43 
44 #include "gmc/gmc_7_0_d.h"
45 #include "gmc/gmc_7_0_sh_mask.h"
46 
47 #include "oss/oss_2_0_d.h"
48 #include "oss/oss_2_0_sh_mask.h"
49 
50 #define GFX7_NUM_GFX_RINGS     1
51 #define GFX7_NUM_COMPUTE_RINGS 8
52 
53 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev);
54 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev);
55 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev);
56 
57 MODULE_FIRMWARE("radeon/bonaire_pfp.bin");
58 MODULE_FIRMWARE("radeon/bonaire_me.bin");
59 MODULE_FIRMWARE("radeon/bonaire_ce.bin");
60 MODULE_FIRMWARE("radeon/bonaire_rlc.bin");
61 MODULE_FIRMWARE("radeon/bonaire_mec.bin");
62 
63 MODULE_FIRMWARE("radeon/hawaii_pfp.bin");
64 MODULE_FIRMWARE("radeon/hawaii_me.bin");
65 MODULE_FIRMWARE("radeon/hawaii_ce.bin");
66 MODULE_FIRMWARE("radeon/hawaii_rlc.bin");
67 MODULE_FIRMWARE("radeon/hawaii_mec.bin");
68 
69 MODULE_FIRMWARE("radeon/kaveri_pfp.bin");
70 MODULE_FIRMWARE("radeon/kaveri_me.bin");
71 MODULE_FIRMWARE("radeon/kaveri_ce.bin");
72 MODULE_FIRMWARE("radeon/kaveri_rlc.bin");
73 MODULE_FIRMWARE("radeon/kaveri_mec.bin");
74 MODULE_FIRMWARE("radeon/kaveri_mec2.bin");
75 
76 MODULE_FIRMWARE("radeon/kabini_pfp.bin");
77 MODULE_FIRMWARE("radeon/kabini_me.bin");
78 MODULE_FIRMWARE("radeon/kabini_ce.bin");
79 MODULE_FIRMWARE("radeon/kabini_rlc.bin");
80 MODULE_FIRMWARE("radeon/kabini_mec.bin");
81 
82 MODULE_FIRMWARE("radeon/mullins_pfp.bin");
83 MODULE_FIRMWARE("radeon/mullins_me.bin");
84 MODULE_FIRMWARE("radeon/mullins_ce.bin");
85 MODULE_FIRMWARE("radeon/mullins_rlc.bin");
86 MODULE_FIRMWARE("radeon/mullins_mec.bin");
87 
88 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
89 {
90 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
91 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
92 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
93 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
94 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
95 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
96 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
97 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
98 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
99 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
100 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
101 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
102 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
103 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
104 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
105 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
106 };
107 
108 static const u32 spectre_rlc_save_restore_register_list[] =
109 {
110 	(0x0e00 << 16) | (0xc12c >> 2),
111 	0x00000000,
112 	(0x0e00 << 16) | (0xc140 >> 2),
113 	0x00000000,
114 	(0x0e00 << 16) | (0xc150 >> 2),
115 	0x00000000,
116 	(0x0e00 << 16) | (0xc15c >> 2),
117 	0x00000000,
118 	(0x0e00 << 16) | (0xc168 >> 2),
119 	0x00000000,
120 	(0x0e00 << 16) | (0xc170 >> 2),
121 	0x00000000,
122 	(0x0e00 << 16) | (0xc178 >> 2),
123 	0x00000000,
124 	(0x0e00 << 16) | (0xc204 >> 2),
125 	0x00000000,
126 	(0x0e00 << 16) | (0xc2b4 >> 2),
127 	0x00000000,
128 	(0x0e00 << 16) | (0xc2b8 >> 2),
129 	0x00000000,
130 	(0x0e00 << 16) | (0xc2bc >> 2),
131 	0x00000000,
132 	(0x0e00 << 16) | (0xc2c0 >> 2),
133 	0x00000000,
134 	(0x0e00 << 16) | (0x8228 >> 2),
135 	0x00000000,
136 	(0x0e00 << 16) | (0x829c >> 2),
137 	0x00000000,
138 	(0x0e00 << 16) | (0x869c >> 2),
139 	0x00000000,
140 	(0x0600 << 16) | (0x98f4 >> 2),
141 	0x00000000,
142 	(0x0e00 << 16) | (0x98f8 >> 2),
143 	0x00000000,
144 	(0x0e00 << 16) | (0x9900 >> 2),
145 	0x00000000,
146 	(0x0e00 << 16) | (0xc260 >> 2),
147 	0x00000000,
148 	(0x0e00 << 16) | (0x90e8 >> 2),
149 	0x00000000,
150 	(0x0e00 << 16) | (0x3c000 >> 2),
151 	0x00000000,
152 	(0x0e00 << 16) | (0x3c00c >> 2),
153 	0x00000000,
154 	(0x0e00 << 16) | (0x8c1c >> 2),
155 	0x00000000,
156 	(0x0e00 << 16) | (0x9700 >> 2),
157 	0x00000000,
158 	(0x0e00 << 16) | (0xcd20 >> 2),
159 	0x00000000,
160 	(0x4e00 << 16) | (0xcd20 >> 2),
161 	0x00000000,
162 	(0x5e00 << 16) | (0xcd20 >> 2),
163 	0x00000000,
164 	(0x6e00 << 16) | (0xcd20 >> 2),
165 	0x00000000,
166 	(0x7e00 << 16) | (0xcd20 >> 2),
167 	0x00000000,
168 	(0x8e00 << 16) | (0xcd20 >> 2),
169 	0x00000000,
170 	(0x9e00 << 16) | (0xcd20 >> 2),
171 	0x00000000,
172 	(0xae00 << 16) | (0xcd20 >> 2),
173 	0x00000000,
174 	(0xbe00 << 16) | (0xcd20 >> 2),
175 	0x00000000,
176 	(0x0e00 << 16) | (0x89bc >> 2),
177 	0x00000000,
178 	(0x0e00 << 16) | (0x8900 >> 2),
179 	0x00000000,
180 	0x3,
181 	(0x0e00 << 16) | (0xc130 >> 2),
182 	0x00000000,
183 	(0x0e00 << 16) | (0xc134 >> 2),
184 	0x00000000,
185 	(0x0e00 << 16) | (0xc1fc >> 2),
186 	0x00000000,
187 	(0x0e00 << 16) | (0xc208 >> 2),
188 	0x00000000,
189 	(0x0e00 << 16) | (0xc264 >> 2),
190 	0x00000000,
191 	(0x0e00 << 16) | (0xc268 >> 2),
192 	0x00000000,
193 	(0x0e00 << 16) | (0xc26c >> 2),
194 	0x00000000,
195 	(0x0e00 << 16) | (0xc270 >> 2),
196 	0x00000000,
197 	(0x0e00 << 16) | (0xc274 >> 2),
198 	0x00000000,
199 	(0x0e00 << 16) | (0xc278 >> 2),
200 	0x00000000,
201 	(0x0e00 << 16) | (0xc27c >> 2),
202 	0x00000000,
203 	(0x0e00 << 16) | (0xc280 >> 2),
204 	0x00000000,
205 	(0x0e00 << 16) | (0xc284 >> 2),
206 	0x00000000,
207 	(0x0e00 << 16) | (0xc288 >> 2),
208 	0x00000000,
209 	(0x0e00 << 16) | (0xc28c >> 2),
210 	0x00000000,
211 	(0x0e00 << 16) | (0xc290 >> 2),
212 	0x00000000,
213 	(0x0e00 << 16) | (0xc294 >> 2),
214 	0x00000000,
215 	(0x0e00 << 16) | (0xc298 >> 2),
216 	0x00000000,
217 	(0x0e00 << 16) | (0xc29c >> 2),
218 	0x00000000,
219 	(0x0e00 << 16) | (0xc2a0 >> 2),
220 	0x00000000,
221 	(0x0e00 << 16) | (0xc2a4 >> 2),
222 	0x00000000,
223 	(0x0e00 << 16) | (0xc2a8 >> 2),
224 	0x00000000,
225 	(0x0e00 << 16) | (0xc2ac  >> 2),
226 	0x00000000,
227 	(0x0e00 << 16) | (0xc2b0 >> 2),
228 	0x00000000,
229 	(0x0e00 << 16) | (0x301d0 >> 2),
230 	0x00000000,
231 	(0x0e00 << 16) | (0x30238 >> 2),
232 	0x00000000,
233 	(0x0e00 << 16) | (0x30250 >> 2),
234 	0x00000000,
235 	(0x0e00 << 16) | (0x30254 >> 2),
236 	0x00000000,
237 	(0x0e00 << 16) | (0x30258 >> 2),
238 	0x00000000,
239 	(0x0e00 << 16) | (0x3025c >> 2),
240 	0x00000000,
241 	(0x4e00 << 16) | (0xc900 >> 2),
242 	0x00000000,
243 	(0x5e00 << 16) | (0xc900 >> 2),
244 	0x00000000,
245 	(0x6e00 << 16) | (0xc900 >> 2),
246 	0x00000000,
247 	(0x7e00 << 16) | (0xc900 >> 2),
248 	0x00000000,
249 	(0x8e00 << 16) | (0xc900 >> 2),
250 	0x00000000,
251 	(0x9e00 << 16) | (0xc900 >> 2),
252 	0x00000000,
253 	(0xae00 << 16) | (0xc900 >> 2),
254 	0x00000000,
255 	(0xbe00 << 16) | (0xc900 >> 2),
256 	0x00000000,
257 	(0x4e00 << 16) | (0xc904 >> 2),
258 	0x00000000,
259 	(0x5e00 << 16) | (0xc904 >> 2),
260 	0x00000000,
261 	(0x6e00 << 16) | (0xc904 >> 2),
262 	0x00000000,
263 	(0x7e00 << 16) | (0xc904 >> 2),
264 	0x00000000,
265 	(0x8e00 << 16) | (0xc904 >> 2),
266 	0x00000000,
267 	(0x9e00 << 16) | (0xc904 >> 2),
268 	0x00000000,
269 	(0xae00 << 16) | (0xc904 >> 2),
270 	0x00000000,
271 	(0xbe00 << 16) | (0xc904 >> 2),
272 	0x00000000,
273 	(0x4e00 << 16) | (0xc908 >> 2),
274 	0x00000000,
275 	(0x5e00 << 16) | (0xc908 >> 2),
276 	0x00000000,
277 	(0x6e00 << 16) | (0xc908 >> 2),
278 	0x00000000,
279 	(0x7e00 << 16) | (0xc908 >> 2),
280 	0x00000000,
281 	(0x8e00 << 16) | (0xc908 >> 2),
282 	0x00000000,
283 	(0x9e00 << 16) | (0xc908 >> 2),
284 	0x00000000,
285 	(0xae00 << 16) | (0xc908 >> 2),
286 	0x00000000,
287 	(0xbe00 << 16) | (0xc908 >> 2),
288 	0x00000000,
289 	(0x4e00 << 16) | (0xc90c >> 2),
290 	0x00000000,
291 	(0x5e00 << 16) | (0xc90c >> 2),
292 	0x00000000,
293 	(0x6e00 << 16) | (0xc90c >> 2),
294 	0x00000000,
295 	(0x7e00 << 16) | (0xc90c >> 2),
296 	0x00000000,
297 	(0x8e00 << 16) | (0xc90c >> 2),
298 	0x00000000,
299 	(0x9e00 << 16) | (0xc90c >> 2),
300 	0x00000000,
301 	(0xae00 << 16) | (0xc90c >> 2),
302 	0x00000000,
303 	(0xbe00 << 16) | (0xc90c >> 2),
304 	0x00000000,
305 	(0x4e00 << 16) | (0xc910 >> 2),
306 	0x00000000,
307 	(0x5e00 << 16) | (0xc910 >> 2),
308 	0x00000000,
309 	(0x6e00 << 16) | (0xc910 >> 2),
310 	0x00000000,
311 	(0x7e00 << 16) | (0xc910 >> 2),
312 	0x00000000,
313 	(0x8e00 << 16) | (0xc910 >> 2),
314 	0x00000000,
315 	(0x9e00 << 16) | (0xc910 >> 2),
316 	0x00000000,
317 	(0xae00 << 16) | (0xc910 >> 2),
318 	0x00000000,
319 	(0xbe00 << 16) | (0xc910 >> 2),
320 	0x00000000,
321 	(0x0e00 << 16) | (0xc99c >> 2),
322 	0x00000000,
323 	(0x0e00 << 16) | (0x9834 >> 2),
324 	0x00000000,
325 	(0x0000 << 16) | (0x30f00 >> 2),
326 	0x00000000,
327 	(0x0001 << 16) | (0x30f00 >> 2),
328 	0x00000000,
329 	(0x0000 << 16) | (0x30f04 >> 2),
330 	0x00000000,
331 	(0x0001 << 16) | (0x30f04 >> 2),
332 	0x00000000,
333 	(0x0000 << 16) | (0x30f08 >> 2),
334 	0x00000000,
335 	(0x0001 << 16) | (0x30f08 >> 2),
336 	0x00000000,
337 	(0x0000 << 16) | (0x30f0c >> 2),
338 	0x00000000,
339 	(0x0001 << 16) | (0x30f0c >> 2),
340 	0x00000000,
341 	(0x0600 << 16) | (0x9b7c >> 2),
342 	0x00000000,
343 	(0x0e00 << 16) | (0x8a14 >> 2),
344 	0x00000000,
345 	(0x0e00 << 16) | (0x8a18 >> 2),
346 	0x00000000,
347 	(0x0600 << 16) | (0x30a00 >> 2),
348 	0x00000000,
349 	(0x0e00 << 16) | (0x8bf0 >> 2),
350 	0x00000000,
351 	(0x0e00 << 16) | (0x8bcc >> 2),
352 	0x00000000,
353 	(0x0e00 << 16) | (0x8b24 >> 2),
354 	0x00000000,
355 	(0x0e00 << 16) | (0x30a04 >> 2),
356 	0x00000000,
357 	(0x0600 << 16) | (0x30a10 >> 2),
358 	0x00000000,
359 	(0x0600 << 16) | (0x30a14 >> 2),
360 	0x00000000,
361 	(0x0600 << 16) | (0x30a18 >> 2),
362 	0x00000000,
363 	(0x0600 << 16) | (0x30a2c >> 2),
364 	0x00000000,
365 	(0x0e00 << 16) | (0xc700 >> 2),
366 	0x00000000,
367 	(0x0e00 << 16) | (0xc704 >> 2),
368 	0x00000000,
369 	(0x0e00 << 16) | (0xc708 >> 2),
370 	0x00000000,
371 	(0x0e00 << 16) | (0xc768 >> 2),
372 	0x00000000,
373 	(0x0400 << 16) | (0xc770 >> 2),
374 	0x00000000,
375 	(0x0400 << 16) | (0xc774 >> 2),
376 	0x00000000,
377 	(0x0400 << 16) | (0xc778 >> 2),
378 	0x00000000,
379 	(0x0400 << 16) | (0xc77c >> 2),
380 	0x00000000,
381 	(0x0400 << 16) | (0xc780 >> 2),
382 	0x00000000,
383 	(0x0400 << 16) | (0xc784 >> 2),
384 	0x00000000,
385 	(0x0400 << 16) | (0xc788 >> 2),
386 	0x00000000,
387 	(0x0400 << 16) | (0xc78c >> 2),
388 	0x00000000,
389 	(0x0400 << 16) | (0xc798 >> 2),
390 	0x00000000,
391 	(0x0400 << 16) | (0xc79c >> 2),
392 	0x00000000,
393 	(0x0400 << 16) | (0xc7a0 >> 2),
394 	0x00000000,
395 	(0x0400 << 16) | (0xc7a4 >> 2),
396 	0x00000000,
397 	(0x0400 << 16) | (0xc7a8 >> 2),
398 	0x00000000,
399 	(0x0400 << 16) | (0xc7ac >> 2),
400 	0x00000000,
401 	(0x0400 << 16) | (0xc7b0 >> 2),
402 	0x00000000,
403 	(0x0400 << 16) | (0xc7b4 >> 2),
404 	0x00000000,
405 	(0x0e00 << 16) | (0x9100 >> 2),
406 	0x00000000,
407 	(0x0e00 << 16) | (0x3c010 >> 2),
408 	0x00000000,
409 	(0x0e00 << 16) | (0x92a8 >> 2),
410 	0x00000000,
411 	(0x0e00 << 16) | (0x92ac >> 2),
412 	0x00000000,
413 	(0x0e00 << 16) | (0x92b4 >> 2),
414 	0x00000000,
415 	(0x0e00 << 16) | (0x92b8 >> 2),
416 	0x00000000,
417 	(0x0e00 << 16) | (0x92bc >> 2),
418 	0x00000000,
419 	(0x0e00 << 16) | (0x92c0 >> 2),
420 	0x00000000,
421 	(0x0e00 << 16) | (0x92c4 >> 2),
422 	0x00000000,
423 	(0x0e00 << 16) | (0x92c8 >> 2),
424 	0x00000000,
425 	(0x0e00 << 16) | (0x92cc >> 2),
426 	0x00000000,
427 	(0x0e00 << 16) | (0x92d0 >> 2),
428 	0x00000000,
429 	(0x0e00 << 16) | (0x8c00 >> 2),
430 	0x00000000,
431 	(0x0e00 << 16) | (0x8c04 >> 2),
432 	0x00000000,
433 	(0x0e00 << 16) | (0x8c20 >> 2),
434 	0x00000000,
435 	(0x0e00 << 16) | (0x8c38 >> 2),
436 	0x00000000,
437 	(0x0e00 << 16) | (0x8c3c >> 2),
438 	0x00000000,
439 	(0x0e00 << 16) | (0xae00 >> 2),
440 	0x00000000,
441 	(0x0e00 << 16) | (0x9604 >> 2),
442 	0x00000000,
443 	(0x0e00 << 16) | (0xac08 >> 2),
444 	0x00000000,
445 	(0x0e00 << 16) | (0xac0c >> 2),
446 	0x00000000,
447 	(0x0e00 << 16) | (0xac10 >> 2),
448 	0x00000000,
449 	(0x0e00 << 16) | (0xac14 >> 2),
450 	0x00000000,
451 	(0x0e00 << 16) | (0xac58 >> 2),
452 	0x00000000,
453 	(0x0e00 << 16) | (0xac68 >> 2),
454 	0x00000000,
455 	(0x0e00 << 16) | (0xac6c >> 2),
456 	0x00000000,
457 	(0x0e00 << 16) | (0xac70 >> 2),
458 	0x00000000,
459 	(0x0e00 << 16) | (0xac74 >> 2),
460 	0x00000000,
461 	(0x0e00 << 16) | (0xac78 >> 2),
462 	0x00000000,
463 	(0x0e00 << 16) | (0xac7c >> 2),
464 	0x00000000,
465 	(0x0e00 << 16) | (0xac80 >> 2),
466 	0x00000000,
467 	(0x0e00 << 16) | (0xac84 >> 2),
468 	0x00000000,
469 	(0x0e00 << 16) | (0xac88 >> 2),
470 	0x00000000,
471 	(0x0e00 << 16) | (0xac8c >> 2),
472 	0x00000000,
473 	(0x0e00 << 16) | (0x970c >> 2),
474 	0x00000000,
475 	(0x0e00 << 16) | (0x9714 >> 2),
476 	0x00000000,
477 	(0x0e00 << 16) | (0x9718 >> 2),
478 	0x00000000,
479 	(0x0e00 << 16) | (0x971c >> 2),
480 	0x00000000,
481 	(0x0e00 << 16) | (0x31068 >> 2),
482 	0x00000000,
483 	(0x4e00 << 16) | (0x31068 >> 2),
484 	0x00000000,
485 	(0x5e00 << 16) | (0x31068 >> 2),
486 	0x00000000,
487 	(0x6e00 << 16) | (0x31068 >> 2),
488 	0x00000000,
489 	(0x7e00 << 16) | (0x31068 >> 2),
490 	0x00000000,
491 	(0x8e00 << 16) | (0x31068 >> 2),
492 	0x00000000,
493 	(0x9e00 << 16) | (0x31068 >> 2),
494 	0x00000000,
495 	(0xae00 << 16) | (0x31068 >> 2),
496 	0x00000000,
497 	(0xbe00 << 16) | (0x31068 >> 2),
498 	0x00000000,
499 	(0x0e00 << 16) | (0xcd10 >> 2),
500 	0x00000000,
501 	(0x0e00 << 16) | (0xcd14 >> 2),
502 	0x00000000,
503 	(0x0e00 << 16) | (0x88b0 >> 2),
504 	0x00000000,
505 	(0x0e00 << 16) | (0x88b4 >> 2),
506 	0x00000000,
507 	(0x0e00 << 16) | (0x88b8 >> 2),
508 	0x00000000,
509 	(0x0e00 << 16) | (0x88bc >> 2),
510 	0x00000000,
511 	(0x0400 << 16) | (0x89c0 >> 2),
512 	0x00000000,
513 	(0x0e00 << 16) | (0x88c4 >> 2),
514 	0x00000000,
515 	(0x0e00 << 16) | (0x88c8 >> 2),
516 	0x00000000,
517 	(0x0e00 << 16) | (0x88d0 >> 2),
518 	0x00000000,
519 	(0x0e00 << 16) | (0x88d4 >> 2),
520 	0x00000000,
521 	(0x0e00 << 16) | (0x88d8 >> 2),
522 	0x00000000,
523 	(0x0e00 << 16) | (0x8980 >> 2),
524 	0x00000000,
525 	(0x0e00 << 16) | (0x30938 >> 2),
526 	0x00000000,
527 	(0x0e00 << 16) | (0x3093c >> 2),
528 	0x00000000,
529 	(0x0e00 << 16) | (0x30940 >> 2),
530 	0x00000000,
531 	(0x0e00 << 16) | (0x89a0 >> 2),
532 	0x00000000,
533 	(0x0e00 << 16) | (0x30900 >> 2),
534 	0x00000000,
535 	(0x0e00 << 16) | (0x30904 >> 2),
536 	0x00000000,
537 	(0x0e00 << 16) | (0x89b4 >> 2),
538 	0x00000000,
539 	(0x0e00 << 16) | (0x3c210 >> 2),
540 	0x00000000,
541 	(0x0e00 << 16) | (0x3c214 >> 2),
542 	0x00000000,
543 	(0x0e00 << 16) | (0x3c218 >> 2),
544 	0x00000000,
545 	(0x0e00 << 16) | (0x8904 >> 2),
546 	0x00000000,
547 	0x5,
548 	(0x0e00 << 16) | (0x8c28 >> 2),
549 	(0x0e00 << 16) | (0x8c2c >> 2),
550 	(0x0e00 << 16) | (0x8c30 >> 2),
551 	(0x0e00 << 16) | (0x8c34 >> 2),
552 	(0x0e00 << 16) | (0x9600 >> 2),
553 };
554 
555 static const u32 kalindi_rlc_save_restore_register_list[] =
556 {
557 	(0x0e00 << 16) | (0xc12c >> 2),
558 	0x00000000,
559 	(0x0e00 << 16) | (0xc140 >> 2),
560 	0x00000000,
561 	(0x0e00 << 16) | (0xc150 >> 2),
562 	0x00000000,
563 	(0x0e00 << 16) | (0xc15c >> 2),
564 	0x00000000,
565 	(0x0e00 << 16) | (0xc168 >> 2),
566 	0x00000000,
567 	(0x0e00 << 16) | (0xc170 >> 2),
568 	0x00000000,
569 	(0x0e00 << 16) | (0xc204 >> 2),
570 	0x00000000,
571 	(0x0e00 << 16) | (0xc2b4 >> 2),
572 	0x00000000,
573 	(0x0e00 << 16) | (0xc2b8 >> 2),
574 	0x00000000,
575 	(0x0e00 << 16) | (0xc2bc >> 2),
576 	0x00000000,
577 	(0x0e00 << 16) | (0xc2c0 >> 2),
578 	0x00000000,
579 	(0x0e00 << 16) | (0x8228 >> 2),
580 	0x00000000,
581 	(0x0e00 << 16) | (0x829c >> 2),
582 	0x00000000,
583 	(0x0e00 << 16) | (0x869c >> 2),
584 	0x00000000,
585 	(0x0600 << 16) | (0x98f4 >> 2),
586 	0x00000000,
587 	(0x0e00 << 16) | (0x98f8 >> 2),
588 	0x00000000,
589 	(0x0e00 << 16) | (0x9900 >> 2),
590 	0x00000000,
591 	(0x0e00 << 16) | (0xc260 >> 2),
592 	0x00000000,
593 	(0x0e00 << 16) | (0x90e8 >> 2),
594 	0x00000000,
595 	(0x0e00 << 16) | (0x3c000 >> 2),
596 	0x00000000,
597 	(0x0e00 << 16) | (0x3c00c >> 2),
598 	0x00000000,
599 	(0x0e00 << 16) | (0x8c1c >> 2),
600 	0x00000000,
601 	(0x0e00 << 16) | (0x9700 >> 2),
602 	0x00000000,
603 	(0x0e00 << 16) | (0xcd20 >> 2),
604 	0x00000000,
605 	(0x4e00 << 16) | (0xcd20 >> 2),
606 	0x00000000,
607 	(0x5e00 << 16) | (0xcd20 >> 2),
608 	0x00000000,
609 	(0x6e00 << 16) | (0xcd20 >> 2),
610 	0x00000000,
611 	(0x7e00 << 16) | (0xcd20 >> 2),
612 	0x00000000,
613 	(0x0e00 << 16) | (0x89bc >> 2),
614 	0x00000000,
615 	(0x0e00 << 16) | (0x8900 >> 2),
616 	0x00000000,
617 	0x3,
618 	(0x0e00 << 16) | (0xc130 >> 2),
619 	0x00000000,
620 	(0x0e00 << 16) | (0xc134 >> 2),
621 	0x00000000,
622 	(0x0e00 << 16) | (0xc1fc >> 2),
623 	0x00000000,
624 	(0x0e00 << 16) | (0xc208 >> 2),
625 	0x00000000,
626 	(0x0e00 << 16) | (0xc264 >> 2),
627 	0x00000000,
628 	(0x0e00 << 16) | (0xc268 >> 2),
629 	0x00000000,
630 	(0x0e00 << 16) | (0xc26c >> 2),
631 	0x00000000,
632 	(0x0e00 << 16) | (0xc270 >> 2),
633 	0x00000000,
634 	(0x0e00 << 16) | (0xc274 >> 2),
635 	0x00000000,
636 	(0x0e00 << 16) | (0xc28c >> 2),
637 	0x00000000,
638 	(0x0e00 << 16) | (0xc290 >> 2),
639 	0x00000000,
640 	(0x0e00 << 16) | (0xc294 >> 2),
641 	0x00000000,
642 	(0x0e00 << 16) | (0xc298 >> 2),
643 	0x00000000,
644 	(0x0e00 << 16) | (0xc2a0 >> 2),
645 	0x00000000,
646 	(0x0e00 << 16) | (0xc2a4 >> 2),
647 	0x00000000,
648 	(0x0e00 << 16) | (0xc2a8 >> 2),
649 	0x00000000,
650 	(0x0e00 << 16) | (0xc2ac >> 2),
651 	0x00000000,
652 	(0x0e00 << 16) | (0x301d0 >> 2),
653 	0x00000000,
654 	(0x0e00 << 16) | (0x30238 >> 2),
655 	0x00000000,
656 	(0x0e00 << 16) | (0x30250 >> 2),
657 	0x00000000,
658 	(0x0e00 << 16) | (0x30254 >> 2),
659 	0x00000000,
660 	(0x0e00 << 16) | (0x30258 >> 2),
661 	0x00000000,
662 	(0x0e00 << 16) | (0x3025c >> 2),
663 	0x00000000,
664 	(0x4e00 << 16) | (0xc900 >> 2),
665 	0x00000000,
666 	(0x5e00 << 16) | (0xc900 >> 2),
667 	0x00000000,
668 	(0x6e00 << 16) | (0xc900 >> 2),
669 	0x00000000,
670 	(0x7e00 << 16) | (0xc900 >> 2),
671 	0x00000000,
672 	(0x4e00 << 16) | (0xc904 >> 2),
673 	0x00000000,
674 	(0x5e00 << 16) | (0xc904 >> 2),
675 	0x00000000,
676 	(0x6e00 << 16) | (0xc904 >> 2),
677 	0x00000000,
678 	(0x7e00 << 16) | (0xc904 >> 2),
679 	0x00000000,
680 	(0x4e00 << 16) | (0xc908 >> 2),
681 	0x00000000,
682 	(0x5e00 << 16) | (0xc908 >> 2),
683 	0x00000000,
684 	(0x6e00 << 16) | (0xc908 >> 2),
685 	0x00000000,
686 	(0x7e00 << 16) | (0xc908 >> 2),
687 	0x00000000,
688 	(0x4e00 << 16) | (0xc90c >> 2),
689 	0x00000000,
690 	(0x5e00 << 16) | (0xc90c >> 2),
691 	0x00000000,
692 	(0x6e00 << 16) | (0xc90c >> 2),
693 	0x00000000,
694 	(0x7e00 << 16) | (0xc90c >> 2),
695 	0x00000000,
696 	(0x4e00 << 16) | (0xc910 >> 2),
697 	0x00000000,
698 	(0x5e00 << 16) | (0xc910 >> 2),
699 	0x00000000,
700 	(0x6e00 << 16) | (0xc910 >> 2),
701 	0x00000000,
702 	(0x7e00 << 16) | (0xc910 >> 2),
703 	0x00000000,
704 	(0x0e00 << 16) | (0xc99c >> 2),
705 	0x00000000,
706 	(0x0e00 << 16) | (0x9834 >> 2),
707 	0x00000000,
708 	(0x0000 << 16) | (0x30f00 >> 2),
709 	0x00000000,
710 	(0x0000 << 16) | (0x30f04 >> 2),
711 	0x00000000,
712 	(0x0000 << 16) | (0x30f08 >> 2),
713 	0x00000000,
714 	(0x0000 << 16) | (0x30f0c >> 2),
715 	0x00000000,
716 	(0x0600 << 16) | (0x9b7c >> 2),
717 	0x00000000,
718 	(0x0e00 << 16) | (0x8a14 >> 2),
719 	0x00000000,
720 	(0x0e00 << 16) | (0x8a18 >> 2),
721 	0x00000000,
722 	(0x0600 << 16) | (0x30a00 >> 2),
723 	0x00000000,
724 	(0x0e00 << 16) | (0x8bf0 >> 2),
725 	0x00000000,
726 	(0x0e00 << 16) | (0x8bcc >> 2),
727 	0x00000000,
728 	(0x0e00 << 16) | (0x8b24 >> 2),
729 	0x00000000,
730 	(0x0e00 << 16) | (0x30a04 >> 2),
731 	0x00000000,
732 	(0x0600 << 16) | (0x30a10 >> 2),
733 	0x00000000,
734 	(0x0600 << 16) | (0x30a14 >> 2),
735 	0x00000000,
736 	(0x0600 << 16) | (0x30a18 >> 2),
737 	0x00000000,
738 	(0x0600 << 16) | (0x30a2c >> 2),
739 	0x00000000,
740 	(0x0e00 << 16) | (0xc700 >> 2),
741 	0x00000000,
742 	(0x0e00 << 16) | (0xc704 >> 2),
743 	0x00000000,
744 	(0x0e00 << 16) | (0xc708 >> 2),
745 	0x00000000,
746 	(0x0e00 << 16) | (0xc768 >> 2),
747 	0x00000000,
748 	(0x0400 << 16) | (0xc770 >> 2),
749 	0x00000000,
750 	(0x0400 << 16) | (0xc774 >> 2),
751 	0x00000000,
752 	(0x0400 << 16) | (0xc798 >> 2),
753 	0x00000000,
754 	(0x0400 << 16) | (0xc79c >> 2),
755 	0x00000000,
756 	(0x0e00 << 16) | (0x9100 >> 2),
757 	0x00000000,
758 	(0x0e00 << 16) | (0x3c010 >> 2),
759 	0x00000000,
760 	(0x0e00 << 16) | (0x8c00 >> 2),
761 	0x00000000,
762 	(0x0e00 << 16) | (0x8c04 >> 2),
763 	0x00000000,
764 	(0x0e00 << 16) | (0x8c20 >> 2),
765 	0x00000000,
766 	(0x0e00 << 16) | (0x8c38 >> 2),
767 	0x00000000,
768 	(0x0e00 << 16) | (0x8c3c >> 2),
769 	0x00000000,
770 	(0x0e00 << 16) | (0xae00 >> 2),
771 	0x00000000,
772 	(0x0e00 << 16) | (0x9604 >> 2),
773 	0x00000000,
774 	(0x0e00 << 16) | (0xac08 >> 2),
775 	0x00000000,
776 	(0x0e00 << 16) | (0xac0c >> 2),
777 	0x00000000,
778 	(0x0e00 << 16) | (0xac10 >> 2),
779 	0x00000000,
780 	(0x0e00 << 16) | (0xac14 >> 2),
781 	0x00000000,
782 	(0x0e00 << 16) | (0xac58 >> 2),
783 	0x00000000,
784 	(0x0e00 << 16) | (0xac68 >> 2),
785 	0x00000000,
786 	(0x0e00 << 16) | (0xac6c >> 2),
787 	0x00000000,
788 	(0x0e00 << 16) | (0xac70 >> 2),
789 	0x00000000,
790 	(0x0e00 << 16) | (0xac74 >> 2),
791 	0x00000000,
792 	(0x0e00 << 16) | (0xac78 >> 2),
793 	0x00000000,
794 	(0x0e00 << 16) | (0xac7c >> 2),
795 	0x00000000,
796 	(0x0e00 << 16) | (0xac80 >> 2),
797 	0x00000000,
798 	(0x0e00 << 16) | (0xac84 >> 2),
799 	0x00000000,
800 	(0x0e00 << 16) | (0xac88 >> 2),
801 	0x00000000,
802 	(0x0e00 << 16) | (0xac8c >> 2),
803 	0x00000000,
804 	(0x0e00 << 16) | (0x970c >> 2),
805 	0x00000000,
806 	(0x0e00 << 16) | (0x9714 >> 2),
807 	0x00000000,
808 	(0x0e00 << 16) | (0x9718 >> 2),
809 	0x00000000,
810 	(0x0e00 << 16) | (0x971c >> 2),
811 	0x00000000,
812 	(0x0e00 << 16) | (0x31068 >> 2),
813 	0x00000000,
814 	(0x4e00 << 16) | (0x31068 >> 2),
815 	0x00000000,
816 	(0x5e00 << 16) | (0x31068 >> 2),
817 	0x00000000,
818 	(0x6e00 << 16) | (0x31068 >> 2),
819 	0x00000000,
820 	(0x7e00 << 16) | (0x31068 >> 2),
821 	0x00000000,
822 	(0x0e00 << 16) | (0xcd10 >> 2),
823 	0x00000000,
824 	(0x0e00 << 16) | (0xcd14 >> 2),
825 	0x00000000,
826 	(0x0e00 << 16) | (0x88b0 >> 2),
827 	0x00000000,
828 	(0x0e00 << 16) | (0x88b4 >> 2),
829 	0x00000000,
830 	(0x0e00 << 16) | (0x88b8 >> 2),
831 	0x00000000,
832 	(0x0e00 << 16) | (0x88bc >> 2),
833 	0x00000000,
834 	(0x0400 << 16) | (0x89c0 >> 2),
835 	0x00000000,
836 	(0x0e00 << 16) | (0x88c4 >> 2),
837 	0x00000000,
838 	(0x0e00 << 16) | (0x88c8 >> 2),
839 	0x00000000,
840 	(0x0e00 << 16) | (0x88d0 >> 2),
841 	0x00000000,
842 	(0x0e00 << 16) | (0x88d4 >> 2),
843 	0x00000000,
844 	(0x0e00 << 16) | (0x88d8 >> 2),
845 	0x00000000,
846 	(0x0e00 << 16) | (0x8980 >> 2),
847 	0x00000000,
848 	(0x0e00 << 16) | (0x30938 >> 2),
849 	0x00000000,
850 	(0x0e00 << 16) | (0x3093c >> 2),
851 	0x00000000,
852 	(0x0e00 << 16) | (0x30940 >> 2),
853 	0x00000000,
854 	(0x0e00 << 16) | (0x89a0 >> 2),
855 	0x00000000,
856 	(0x0e00 << 16) | (0x30900 >> 2),
857 	0x00000000,
858 	(0x0e00 << 16) | (0x30904 >> 2),
859 	0x00000000,
860 	(0x0e00 << 16) | (0x89b4 >> 2),
861 	0x00000000,
862 	(0x0e00 << 16) | (0x3e1fc >> 2),
863 	0x00000000,
864 	(0x0e00 << 16) | (0x3c210 >> 2),
865 	0x00000000,
866 	(0x0e00 << 16) | (0x3c214 >> 2),
867 	0x00000000,
868 	(0x0e00 << 16) | (0x3c218 >> 2),
869 	0x00000000,
870 	(0x0e00 << 16) | (0x8904 >> 2),
871 	0x00000000,
872 	0x5,
873 	(0x0e00 << 16) | (0x8c28 >> 2),
874 	(0x0e00 << 16) | (0x8c2c >> 2),
875 	(0x0e00 << 16) | (0x8c30 >> 2),
876 	(0x0e00 << 16) | (0x8c34 >> 2),
877 	(0x0e00 << 16) | (0x9600 >> 2),
878 };
879 
880 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev);
881 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer);
882 static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev);
883 static void gfx_v7_0_init_pg(struct amdgpu_device *adev);
884 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev);
885 
886 /*
887  * Core functions
888  */
889 /**
890  * gfx_v7_0_init_microcode - load ucode images from disk
891  *
892  * @adev: amdgpu_device pointer
893  *
894  * Use the firmware interface to load the ucode images into
895  * the driver (not loaded into hw).
896  * Returns 0 on success, error on failure.
897  */
898 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev)
899 {
900 	const char *chip_name;
901 	char fw_name[30];
902 	int err;
903 
904 	DRM_DEBUG("\n");
905 
906 	switch (adev->asic_type) {
907 	case CHIP_BONAIRE:
908 		chip_name = "bonaire";
909 		break;
910 	case CHIP_HAWAII:
911 		chip_name = "hawaii";
912 		break;
913 	case CHIP_KAVERI:
914 		chip_name = "kaveri";
915 		break;
916 	case CHIP_KABINI:
917 		chip_name = "kabini";
918 		break;
919 	case CHIP_MULLINS:
920 		chip_name = "mullins";
921 		break;
922 	default: BUG();
923 	}
924 
925 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
926 	err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
927 	if (err)
928 		goto out;
929 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
930 	if (err)
931 		goto out;
932 
933 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
934 	err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
935 	if (err)
936 		goto out;
937 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
938 	if (err)
939 		goto out;
940 
941 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
942 	err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
943 	if (err)
944 		goto out;
945 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
946 	if (err)
947 		goto out;
948 
949 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec.bin", chip_name);
950 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
951 	if (err)
952 		goto out;
953 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
954 	if (err)
955 		goto out;
956 
957 	if (adev->asic_type == CHIP_KAVERI) {
958 		snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec2.bin", chip_name);
959 		err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
960 		if (err)
961 			goto out;
962 		err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
963 		if (err)
964 			goto out;
965 	}
966 
967 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name);
968 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
969 	if (err)
970 		goto out;
971 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
972 
973 out:
974 	if (err) {
975 		printk(KERN_ERR
976 		       "gfx7: Failed to load firmware \"%s\"\n",
977 		       fw_name);
978 		release_firmware(adev->gfx.pfp_fw);
979 		adev->gfx.pfp_fw = NULL;
980 		release_firmware(adev->gfx.me_fw);
981 		adev->gfx.me_fw = NULL;
982 		release_firmware(adev->gfx.ce_fw);
983 		adev->gfx.ce_fw = NULL;
984 		release_firmware(adev->gfx.mec_fw);
985 		adev->gfx.mec_fw = NULL;
986 		release_firmware(adev->gfx.mec2_fw);
987 		adev->gfx.mec2_fw = NULL;
988 		release_firmware(adev->gfx.rlc_fw);
989 		adev->gfx.rlc_fw = NULL;
990 	}
991 	return err;
992 }
993 
994 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev)
995 {
996 	release_firmware(adev->gfx.pfp_fw);
997 	adev->gfx.pfp_fw = NULL;
998 	release_firmware(adev->gfx.me_fw);
999 	adev->gfx.me_fw = NULL;
1000 	release_firmware(adev->gfx.ce_fw);
1001 	adev->gfx.ce_fw = NULL;
1002 	release_firmware(adev->gfx.mec_fw);
1003 	adev->gfx.mec_fw = NULL;
1004 	release_firmware(adev->gfx.mec2_fw);
1005 	adev->gfx.mec2_fw = NULL;
1006 	release_firmware(adev->gfx.rlc_fw);
1007 	adev->gfx.rlc_fw = NULL;
1008 }
1009 
1010 /**
1011  * gfx_v7_0_tiling_mode_table_init - init the hw tiling table
1012  *
1013  * @adev: amdgpu_device pointer
1014  *
1015  * Starting with SI, the tiling setup is done globally in a
1016  * set of 32 tiling modes.  Rather than selecting each set of
1017  * parameters per surface as on older asics, we just select
1018  * which index in the tiling table we want to use, and the
1019  * surface uses those parameters (CIK).
1020  */
1021 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev)
1022 {
1023 	const u32 num_tile_mode_states =
1024 			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
1025 	const u32 num_secondary_tile_mode_states =
1026 			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
1027 	u32 reg_offset, split_equal_to_row_size;
1028 	uint32_t *tile, *macrotile;
1029 
1030 	tile = adev->gfx.config.tile_mode_array;
1031 	macrotile = adev->gfx.config.macrotile_mode_array;
1032 
1033 	switch (adev->gfx.config.mem_row_size_in_kb) {
1034 	case 1:
1035 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1036 		break;
1037 	case 2:
1038 	default:
1039 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1040 		break;
1041 	case 4:
1042 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1043 		break;
1044 	}
1045 
1046 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1047 		tile[reg_offset] = 0;
1048 	for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1049 		macrotile[reg_offset] = 0;
1050 
1051 	switch (adev->asic_type) {
1052 	case CHIP_BONAIRE:
1053 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1054 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1055 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1056 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1057 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1058 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1059 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1060 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1061 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1062 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1063 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1064 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1065 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1066 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1067 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1068 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1069 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1070 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1071 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1072 			   TILE_SPLIT(split_equal_to_row_size));
1073 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1074 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1075 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1076 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1077 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1078 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1079 			   TILE_SPLIT(split_equal_to_row_size));
1080 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1081 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1082 			   PIPE_CONFIG(ADDR_SURF_P4_16x16));
1083 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1084 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1085 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1086 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1087 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1088 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1089 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1090 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1091 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1092 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1093 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1094 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1095 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1096 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1097 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1098 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1099 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1100 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1101 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1102 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1103 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1104 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1105 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1106 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1107 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1108 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1109 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1110 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1111 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1112 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1113 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1114 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1115 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1116 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1117 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1118 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1119 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1120 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1121 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1122 		tile[21] =  (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1123 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1124 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1125 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1126 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1127 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1128 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1129 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1130 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1131 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1132 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1133 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1134 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1135 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1136 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1137 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1138 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1139 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1140 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1141 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1142 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1143 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1144 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1145 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1146 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1147 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1148 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1149 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1150 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1151 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1152 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1153 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1154 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1155 
1156 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1157 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1158 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1159 				NUM_BANKS(ADDR_SURF_16_BANK));
1160 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1161 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1162 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1163 				NUM_BANKS(ADDR_SURF_16_BANK));
1164 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1165 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1166 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1167 				NUM_BANKS(ADDR_SURF_16_BANK));
1168 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1169 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1170 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1171 				NUM_BANKS(ADDR_SURF_16_BANK));
1172 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1173 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1174 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1175 				NUM_BANKS(ADDR_SURF_16_BANK));
1176 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1177 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1178 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1179 				NUM_BANKS(ADDR_SURF_8_BANK));
1180 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1181 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1182 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1183 				NUM_BANKS(ADDR_SURF_4_BANK));
1184 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1185 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1186 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1187 				NUM_BANKS(ADDR_SURF_16_BANK));
1188 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1189 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1190 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1191 				NUM_BANKS(ADDR_SURF_16_BANK));
1192 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1193 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1194 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1195 				NUM_BANKS(ADDR_SURF_16_BANK));
1196 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1197 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1198 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1199 				NUM_BANKS(ADDR_SURF_16_BANK));
1200 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1201 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1202 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1203 				NUM_BANKS(ADDR_SURF_16_BANK));
1204 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1205 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1206 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1207 				NUM_BANKS(ADDR_SURF_8_BANK));
1208 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1209 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1210 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1211 				NUM_BANKS(ADDR_SURF_4_BANK));
1212 
1213 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1214 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1215 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1216 			if (reg_offset != 7)
1217 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1218 		break;
1219 	case CHIP_HAWAII:
1220 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1221 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1222 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1223 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1224 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1225 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1226 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1227 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1228 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1229 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1230 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1231 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1232 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1233 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1234 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1235 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1236 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1237 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1238 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1239 			   TILE_SPLIT(split_equal_to_row_size));
1240 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1241 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1242 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1243 			   TILE_SPLIT(split_equal_to_row_size));
1244 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1245 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1246 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1247 			   TILE_SPLIT(split_equal_to_row_size));
1248 		tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1249 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1250 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1251 			   TILE_SPLIT(split_equal_to_row_size));
1252 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1253 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
1254 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1255 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1256 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1257 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1258 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1259 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1260 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1261 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1262 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1263 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1264 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1265 		tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1266 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1267 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1268 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1269 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1270 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1271 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1272 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1273 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1274 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1275 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1276 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1277 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1278 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1279 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1280 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1281 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1282 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1283 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1284 		tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1285 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1286 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1287 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1288 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1289 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1290 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1291 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1292 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1293 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1294 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1295 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1296 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1297 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1298 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1299 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1300 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1301 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1302 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1303 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1304 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1305 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1306 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1307 		tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1308 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1309 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1310 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1311 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1312 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1313 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1314 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1315 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1316 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1317 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1318 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1319 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1320 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1321 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1322 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1323 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1324 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1325 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1326 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1327 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1328 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1329 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1330 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1331 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1332 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1333 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1334 		tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1335 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1336 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1337 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1338 
1339 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1340 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1341 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1342 				NUM_BANKS(ADDR_SURF_16_BANK));
1343 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1344 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1345 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1346 				NUM_BANKS(ADDR_SURF_16_BANK));
1347 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1348 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1349 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1350 				NUM_BANKS(ADDR_SURF_16_BANK));
1351 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1352 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1353 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1354 				NUM_BANKS(ADDR_SURF_16_BANK));
1355 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1356 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1357 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1358 				NUM_BANKS(ADDR_SURF_8_BANK));
1359 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1360 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1361 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1362 				NUM_BANKS(ADDR_SURF_4_BANK));
1363 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1364 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1365 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1366 				NUM_BANKS(ADDR_SURF_4_BANK));
1367 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1368 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1369 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1370 				NUM_BANKS(ADDR_SURF_16_BANK));
1371 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1372 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1373 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1374 				NUM_BANKS(ADDR_SURF_16_BANK));
1375 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1376 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1377 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1378 				NUM_BANKS(ADDR_SURF_16_BANK));
1379 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1380 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1381 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1382 				NUM_BANKS(ADDR_SURF_8_BANK));
1383 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1384 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1385 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1386 				NUM_BANKS(ADDR_SURF_16_BANK));
1387 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1388 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1389 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1390 				NUM_BANKS(ADDR_SURF_8_BANK));
1391 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1392 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1393 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1394 				NUM_BANKS(ADDR_SURF_4_BANK));
1395 
1396 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1397 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1398 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1399 			if (reg_offset != 7)
1400 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1401 		break;
1402 	case CHIP_KABINI:
1403 	case CHIP_KAVERI:
1404 	case CHIP_MULLINS:
1405 	default:
1406 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1407 			   PIPE_CONFIG(ADDR_SURF_P2) |
1408 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1409 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1410 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1411 			   PIPE_CONFIG(ADDR_SURF_P2) |
1412 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1413 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1414 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1415 			   PIPE_CONFIG(ADDR_SURF_P2) |
1416 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1417 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1418 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1419 			   PIPE_CONFIG(ADDR_SURF_P2) |
1420 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1421 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1422 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1423 			   PIPE_CONFIG(ADDR_SURF_P2) |
1424 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1425 			   TILE_SPLIT(split_equal_to_row_size));
1426 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1427 			   PIPE_CONFIG(ADDR_SURF_P2) |
1428 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1429 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1430 			   PIPE_CONFIG(ADDR_SURF_P2) |
1431 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1432 			   TILE_SPLIT(split_equal_to_row_size));
1433 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1434 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1435 			   PIPE_CONFIG(ADDR_SURF_P2));
1436 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1437 			   PIPE_CONFIG(ADDR_SURF_P2) |
1438 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1439 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1440 			    PIPE_CONFIG(ADDR_SURF_P2) |
1441 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1442 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1443 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1444 			    PIPE_CONFIG(ADDR_SURF_P2) |
1445 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1446 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1447 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1448 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1449 			    PIPE_CONFIG(ADDR_SURF_P2) |
1450 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1451 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1452 			    PIPE_CONFIG(ADDR_SURF_P2) |
1453 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1454 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1455 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1456 			    PIPE_CONFIG(ADDR_SURF_P2) |
1457 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1458 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1459 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1460 			    PIPE_CONFIG(ADDR_SURF_P2) |
1461 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1462 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1463 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1464 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1465 			    PIPE_CONFIG(ADDR_SURF_P2) |
1466 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1467 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1468 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1469 			    PIPE_CONFIG(ADDR_SURF_P2) |
1470 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1471 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1472 			    PIPE_CONFIG(ADDR_SURF_P2) |
1473 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1474 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1475 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1476 			    PIPE_CONFIG(ADDR_SURF_P2) |
1477 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1478 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1479 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1480 			    PIPE_CONFIG(ADDR_SURF_P2) |
1481 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1482 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1483 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1484 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1485 			    PIPE_CONFIG(ADDR_SURF_P2) |
1486 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1487 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1488 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1489 			    PIPE_CONFIG(ADDR_SURF_P2) |
1490 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1491 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1492 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1493 			    PIPE_CONFIG(ADDR_SURF_P2) |
1494 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1495 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1496 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1497 			    PIPE_CONFIG(ADDR_SURF_P2) |
1498 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1499 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1500 			    PIPE_CONFIG(ADDR_SURF_P2) |
1501 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1502 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1503 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1504 			    PIPE_CONFIG(ADDR_SURF_P2) |
1505 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1506 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1507 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1508 
1509 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1510 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1511 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1512 				NUM_BANKS(ADDR_SURF_8_BANK));
1513 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1514 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1515 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1516 				NUM_BANKS(ADDR_SURF_8_BANK));
1517 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1518 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1519 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1520 				NUM_BANKS(ADDR_SURF_8_BANK));
1521 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1522 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1523 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1524 				NUM_BANKS(ADDR_SURF_8_BANK));
1525 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1526 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1527 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1528 				NUM_BANKS(ADDR_SURF_8_BANK));
1529 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1530 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1531 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1532 				NUM_BANKS(ADDR_SURF_8_BANK));
1533 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1534 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1535 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1536 				NUM_BANKS(ADDR_SURF_8_BANK));
1537 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1538 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1539 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1540 				NUM_BANKS(ADDR_SURF_16_BANK));
1541 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1542 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1543 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1544 				NUM_BANKS(ADDR_SURF_16_BANK));
1545 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1546 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1547 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1548 				NUM_BANKS(ADDR_SURF_16_BANK));
1549 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1550 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1551 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1552 				NUM_BANKS(ADDR_SURF_16_BANK));
1553 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1554 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1555 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1556 				NUM_BANKS(ADDR_SURF_16_BANK));
1557 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1558 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1559 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1560 				NUM_BANKS(ADDR_SURF_16_BANK));
1561 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1562 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1563 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1564 				NUM_BANKS(ADDR_SURF_8_BANK));
1565 
1566 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1567 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1568 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1569 			if (reg_offset != 7)
1570 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1571 		break;
1572 	}
1573 }
1574 
1575 /**
1576  * gfx_v7_0_select_se_sh - select which SE, SH to address
1577  *
1578  * @adev: amdgpu_device pointer
1579  * @se_num: shader engine to address
1580  * @sh_num: sh block to address
1581  *
1582  * Select which SE, SH combinations to address. Certain
1583  * registers are instanced per SE or SH.  0xffffffff means
1584  * broadcast to all SEs or SHs (CIK).
1585  */
1586 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev,
1587 				  u32 se_num, u32 sh_num, u32 instance)
1588 {
1589 	u32 data;
1590 
1591 	if (instance == 0xffffffff)
1592 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1593 	else
1594 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1595 
1596 	if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1597 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1598 			GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
1599 	else if (se_num == 0xffffffff)
1600 		data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK |
1601 			(sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT);
1602 	else if (sh_num == 0xffffffff)
1603 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1604 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1605 	else
1606 		data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) |
1607 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1608 	WREG32(mmGRBM_GFX_INDEX, data);
1609 }
1610 
1611 /**
1612  * gfx_v7_0_create_bitmask - create a bitmask
1613  *
1614  * @bit_width: length of the mask
1615  *
1616  * create a variable length bit mask (CIK).
1617  * Returns the bitmask.
1618  */
1619 static u32 gfx_v7_0_create_bitmask(u32 bit_width)
1620 {
1621 	return (u32)((1ULL << bit_width) - 1);
1622 }
1623 
1624 /**
1625  * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs
1626  *
1627  * @adev: amdgpu_device pointer
1628  *
1629  * Calculates the bitmask of enabled RBs (CIK).
1630  * Returns the enabled RB bitmask.
1631  */
1632 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1633 {
1634 	u32 data, mask;
1635 
1636 	data = RREG32(mmCC_RB_BACKEND_DISABLE);
1637 	data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1638 
1639 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1640 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1641 
1642 	mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_backends_per_se /
1643 				       adev->gfx.config.max_sh_per_se);
1644 
1645 	return (~data) & mask;
1646 }
1647 
1648 static void
1649 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
1650 {
1651 	switch (adev->asic_type) {
1652 	case CHIP_BONAIRE:
1653 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
1654 			  SE_XSEL(1) | SE_YSEL(1);
1655 		*rconf1 |= 0x0;
1656 		break;
1657 	case CHIP_HAWAII:
1658 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
1659 			  RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) |
1660 			  PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) |
1661 			  SE_YSEL(3);
1662 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
1663 			   SE_PAIR_YSEL(2);
1664 		break;
1665 	case CHIP_KAVERI:
1666 		*rconf |= RB_MAP_PKR0(2);
1667 		*rconf1 |= 0x0;
1668 		break;
1669 	case CHIP_KABINI:
1670 	case CHIP_MULLINS:
1671 		*rconf |= 0x0;
1672 		*rconf1 |= 0x0;
1673 		break;
1674 	default:
1675 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
1676 		break;
1677 	}
1678 }
1679 
1680 static void
1681 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev,
1682 					u32 raster_config, u32 raster_config_1,
1683 					unsigned rb_mask, unsigned num_rb)
1684 {
1685 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
1686 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
1687 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
1688 	unsigned rb_per_se = num_rb / num_se;
1689 	unsigned se_mask[4];
1690 	unsigned se;
1691 
1692 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
1693 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
1694 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
1695 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
1696 
1697 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
1698 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
1699 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
1700 
1701 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
1702 			     (!se_mask[2] && !se_mask[3]))) {
1703 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
1704 
1705 		if (!se_mask[0] && !se_mask[1]) {
1706 			raster_config_1 |=
1707 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
1708 		} else {
1709 			raster_config_1 |=
1710 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
1711 		}
1712 	}
1713 
1714 	for (se = 0; se < num_se; se++) {
1715 		unsigned raster_config_se = raster_config;
1716 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
1717 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
1718 		int idx = (se / 2) * 2;
1719 
1720 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
1721 			raster_config_se &= ~SE_MAP_MASK;
1722 
1723 			if (!se_mask[idx]) {
1724 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
1725 			} else {
1726 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
1727 			}
1728 		}
1729 
1730 		pkr0_mask &= rb_mask;
1731 		pkr1_mask &= rb_mask;
1732 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
1733 			raster_config_se &= ~PKR_MAP_MASK;
1734 
1735 			if (!pkr0_mask) {
1736 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
1737 			} else {
1738 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
1739 			}
1740 		}
1741 
1742 		if (rb_per_se >= 2) {
1743 			unsigned rb0_mask = 1 << (se * rb_per_se);
1744 			unsigned rb1_mask = rb0_mask << 1;
1745 
1746 			rb0_mask &= rb_mask;
1747 			rb1_mask &= rb_mask;
1748 			if (!rb0_mask || !rb1_mask) {
1749 				raster_config_se &= ~RB_MAP_PKR0_MASK;
1750 
1751 				if (!rb0_mask) {
1752 					raster_config_se |=
1753 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
1754 				} else {
1755 					raster_config_se |=
1756 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
1757 				}
1758 			}
1759 
1760 			if (rb_per_se > 2) {
1761 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
1762 				rb1_mask = rb0_mask << 1;
1763 				rb0_mask &= rb_mask;
1764 				rb1_mask &= rb_mask;
1765 				if (!rb0_mask || !rb1_mask) {
1766 					raster_config_se &= ~RB_MAP_PKR1_MASK;
1767 
1768 					if (!rb0_mask) {
1769 						raster_config_se |=
1770 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
1771 					} else {
1772 						raster_config_se |=
1773 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
1774 					}
1775 				}
1776 			}
1777 		}
1778 
1779 		/* GRBM_GFX_INDEX has a different offset on CI+ */
1780 		gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
1781 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
1782 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1783 	}
1784 
1785 	/* GRBM_GFX_INDEX has a different offset on CI+ */
1786 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1787 }
1788 
1789 /**
1790  * gfx_v7_0_setup_rb - setup the RBs on the asic
1791  *
1792  * @adev: amdgpu_device pointer
1793  * @se_num: number of SEs (shader engines) for the asic
1794  * @sh_per_se: number of SH blocks per SE for the asic
1795  *
1796  * Configures per-SE/SH RB registers (CIK).
1797  */
1798 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1799 {
1800 	int i, j;
1801 	u32 data;
1802 	u32 raster_config = 0, raster_config_1 = 0;
1803 	u32 active_rbs = 0;
1804 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1805 					adev->gfx.config.max_sh_per_se;
1806 	unsigned num_rb_pipes;
1807 
1808 	mutex_lock(&adev->grbm_idx_mutex);
1809 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1810 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1811 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1812 			data = gfx_v7_0_get_rb_active_bitmap(adev);
1813 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1814 					       rb_bitmap_width_per_sh);
1815 		}
1816 	}
1817 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1818 
1819 	adev->gfx.config.backend_enable_mask = active_rbs;
1820 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1821 
1822 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1823 			     adev->gfx.config.max_shader_engines, 16);
1824 
1825 	gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1826 
1827 	if (!adev->gfx.config.backend_enable_mask ||
1828 			adev->gfx.config.num_rbs >= num_rb_pipes) {
1829 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1830 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1831 	} else {
1832 		gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1833 							adev->gfx.config.backend_enable_mask,
1834 							num_rb_pipes);
1835 	}
1836 	mutex_unlock(&adev->grbm_idx_mutex);
1837 }
1838 
1839 /**
1840  * gmc_v7_0_init_compute_vmid - gart enable
1841  *
1842  * @rdev: amdgpu_device pointer
1843  *
1844  * Initialize compute vmid sh_mem registers
1845  *
1846  */
1847 #define DEFAULT_SH_MEM_BASES	(0x6000)
1848 #define FIRST_COMPUTE_VMID	(8)
1849 #define LAST_COMPUTE_VMID	(16)
1850 static void gmc_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1851 {
1852 	int i;
1853 	uint32_t sh_mem_config;
1854 	uint32_t sh_mem_bases;
1855 
1856 	/*
1857 	 * Configure apertures:
1858 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1859 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1860 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1861 	*/
1862 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1863 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1864 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1865 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1866 	mutex_lock(&adev->srbm_mutex);
1867 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1868 		cik_srbm_select(adev, 0, 0, 0, i);
1869 		/* CP and shaders */
1870 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1871 		WREG32(mmSH_MEM_APE1_BASE, 1);
1872 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1873 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1874 	}
1875 	cik_srbm_select(adev, 0, 0, 0, 0);
1876 	mutex_unlock(&adev->srbm_mutex);
1877 }
1878 
1879 /**
1880  * gfx_v7_0_gpu_init - setup the 3D engine
1881  *
1882  * @adev: amdgpu_device pointer
1883  *
1884  * Configures the 3D engine and tiling configuration
1885  * registers so that the 3D engine is usable.
1886  */
1887 static void gfx_v7_0_gpu_init(struct amdgpu_device *adev)
1888 {
1889 	u32 tmp, sh_mem_cfg;
1890 	int i;
1891 
1892 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1893 
1894 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1895 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1896 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1897 
1898 	gfx_v7_0_tiling_mode_table_init(adev);
1899 
1900 	gfx_v7_0_setup_rb(adev);
1901 	gfx_v7_0_get_cu_info(adev);
1902 
1903 	/* set HW defaults for 3D engine */
1904 	WREG32(mmCP_MEQ_THRESHOLDS,
1905 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1906 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1907 
1908 	mutex_lock(&adev->grbm_idx_mutex);
1909 	/*
1910 	 * making sure that the following register writes will be broadcasted
1911 	 * to all the shaders
1912 	 */
1913 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1914 
1915 	/* XXX SH_MEM regs */
1916 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1917 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1918 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1919 
1920 	mutex_lock(&adev->srbm_mutex);
1921 	for (i = 0; i < 16; i++) {
1922 		cik_srbm_select(adev, 0, 0, 0, i);
1923 		/* CP and shaders */
1924 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1925 		WREG32(mmSH_MEM_APE1_BASE, 1);
1926 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1927 		WREG32(mmSH_MEM_BASES, 0);
1928 	}
1929 	cik_srbm_select(adev, 0, 0, 0, 0);
1930 	mutex_unlock(&adev->srbm_mutex);
1931 
1932 	gmc_v7_0_init_compute_vmid(adev);
1933 
1934 	WREG32(mmSX_DEBUG_1, 0x20);
1935 
1936 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1937 
1938 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1939 	tmp |= 0x03000000;
1940 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1941 
1942 	WREG32(mmSQ_CONFIG, 1);
1943 
1944 	WREG32(mmDB_DEBUG, 0);
1945 
1946 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
1947 	tmp |= 0x00000400;
1948 	WREG32(mmDB_DEBUG2, tmp);
1949 
1950 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
1951 	tmp |= 0x00020200;
1952 	WREG32(mmDB_DEBUG3, tmp);
1953 
1954 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
1955 	tmp |= 0x00018208;
1956 	WREG32(mmCB_HW_CONTROL, tmp);
1957 
1958 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
1959 
1960 	WREG32(mmPA_SC_FIFO_SIZE,
1961 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
1962 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
1963 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
1964 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
1965 
1966 	WREG32(mmVGT_NUM_INSTANCES, 1);
1967 
1968 	WREG32(mmCP_PERFMON_CNTL, 0);
1969 
1970 	WREG32(mmSQ_CONFIG, 0);
1971 
1972 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
1973 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
1974 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
1975 
1976 	WREG32(mmVGT_CACHE_INVALIDATION,
1977 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
1978 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
1979 
1980 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
1981 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
1982 
1983 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
1984 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
1985 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
1986 
1987 	tmp = RREG32(mmSPI_ARB_PRIORITY);
1988 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
1989 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
1990 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
1991 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
1992 	WREG32(mmSPI_ARB_PRIORITY, tmp);
1993 
1994 	mutex_unlock(&adev->grbm_idx_mutex);
1995 
1996 	udelay(50);
1997 }
1998 
1999 /*
2000  * GPU scratch registers helpers function.
2001  */
2002 /**
2003  * gfx_v7_0_scratch_init - setup driver info for CP scratch regs
2004  *
2005  * @adev: amdgpu_device pointer
2006  *
2007  * Set up the number and offset of the CP scratch registers.
2008  * NOTE: use of CP scratch registers is a legacy inferface and
2009  * is not used by default on newer asics (r6xx+).  On newer asics,
2010  * memory buffers are used for fences rather than scratch regs.
2011  */
2012 static void gfx_v7_0_scratch_init(struct amdgpu_device *adev)
2013 {
2014 	adev->gfx.scratch.num_reg = 7;
2015 	adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
2016 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
2017 }
2018 
2019 /**
2020  * gfx_v7_0_ring_test_ring - basic gfx ring test
2021  *
2022  * @adev: amdgpu_device pointer
2023  * @ring: amdgpu_ring structure holding ring information
2024  *
2025  * Allocate a scratch register and write to it using the gfx ring (CIK).
2026  * Provides a basic gfx ring test to verify that the ring is working.
2027  * Used by gfx_v7_0_cp_gfx_resume();
2028  * Returns 0 on success, error on failure.
2029  */
2030 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2031 {
2032 	struct amdgpu_device *adev = ring->adev;
2033 	uint32_t scratch;
2034 	uint32_t tmp = 0;
2035 	unsigned i;
2036 	int r;
2037 
2038 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2039 	if (r) {
2040 		DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r);
2041 		return r;
2042 	}
2043 	WREG32(scratch, 0xCAFEDEAD);
2044 	r = amdgpu_ring_alloc(ring, 3);
2045 	if (r) {
2046 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r);
2047 		amdgpu_gfx_scratch_free(adev, scratch);
2048 		return r;
2049 	}
2050 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2051 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
2052 	amdgpu_ring_write(ring, 0xDEADBEEF);
2053 	amdgpu_ring_commit(ring);
2054 
2055 	for (i = 0; i < adev->usec_timeout; i++) {
2056 		tmp = RREG32(scratch);
2057 		if (tmp == 0xDEADBEEF)
2058 			break;
2059 		DRM_UDELAY(1);
2060 	}
2061 	if (i < adev->usec_timeout) {
2062 		DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
2063 	} else {
2064 		DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
2065 			  ring->idx, scratch, tmp);
2066 		r = -EINVAL;
2067 	}
2068 	amdgpu_gfx_scratch_free(adev, scratch);
2069 	return r;
2070 }
2071 
2072 /**
2073  * gfx_v7_0_ring_emit_hdp - emit an hdp flush on the cp
2074  *
2075  * @adev: amdgpu_device pointer
2076  * @ridx: amdgpu ring index
2077  *
2078  * Emits an hdp flush on the cp.
2079  */
2080 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2081 {
2082 	u32 ref_and_mask;
2083 	int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2084 
2085 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2086 		switch (ring->me) {
2087 		case 1:
2088 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2089 			break;
2090 		case 2:
2091 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2092 			break;
2093 		default:
2094 			return;
2095 		}
2096 	} else {
2097 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2098 	}
2099 
2100 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2101 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2102 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2103 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2104 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2105 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2106 	amdgpu_ring_write(ring, ref_and_mask);
2107 	amdgpu_ring_write(ring, ref_and_mask);
2108 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2109 }
2110 
2111 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2112 {
2113 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2114 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2115 		EVENT_INDEX(4));
2116 
2117 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2118 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2119 		EVENT_INDEX(0));
2120 }
2121 
2122 
2123 /**
2124  * gfx_v7_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp
2125  *
2126  * @adev: amdgpu_device pointer
2127  * @ridx: amdgpu ring index
2128  *
2129  * Emits an hdp invalidate on the cp.
2130  */
2131 static void gfx_v7_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
2132 {
2133 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
2134 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
2135 				 WRITE_DATA_DST_SEL(0) |
2136 				 WR_CONFIRM));
2137 	amdgpu_ring_write(ring, mmHDP_DEBUG0);
2138 	amdgpu_ring_write(ring, 0);
2139 	amdgpu_ring_write(ring, 1);
2140 }
2141 
2142 /**
2143  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2144  *
2145  * @adev: amdgpu_device pointer
2146  * @fence: amdgpu fence object
2147  *
2148  * Emits a fence sequnce number on the gfx ring and flushes
2149  * GPU caches.
2150  */
2151 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2152 					 u64 seq, unsigned flags)
2153 {
2154 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2155 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2156 	/* Workaround for cache flush problems. First send a dummy EOP
2157 	 * event down the pipe with seq one below.
2158 	 */
2159 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2160 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2161 				 EOP_TC_ACTION_EN |
2162 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2163 				 EVENT_INDEX(5)));
2164 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2165 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2166 				DATA_SEL(1) | INT_SEL(0));
2167 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2168 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2169 
2170 	/* Then send the real EOP event down the pipe. */
2171 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2172 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2173 				 EOP_TC_ACTION_EN |
2174 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2175 				 EVENT_INDEX(5)));
2176 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2177 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2178 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2179 	amdgpu_ring_write(ring, lower_32_bits(seq));
2180 	amdgpu_ring_write(ring, upper_32_bits(seq));
2181 }
2182 
2183 /**
2184  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2185  *
2186  * @adev: amdgpu_device pointer
2187  * @fence: amdgpu fence object
2188  *
2189  * Emits a fence sequnce number on the compute ring and flushes
2190  * GPU caches.
2191  */
2192 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2193 					     u64 addr, u64 seq,
2194 					     unsigned flags)
2195 {
2196 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2197 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2198 
2199 	/* RELEASE_MEM - flush caches, send int */
2200 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2201 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2202 				 EOP_TC_ACTION_EN |
2203 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2204 				 EVENT_INDEX(5)));
2205 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2206 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2207 	amdgpu_ring_write(ring, upper_32_bits(addr));
2208 	amdgpu_ring_write(ring, lower_32_bits(seq));
2209 	amdgpu_ring_write(ring, upper_32_bits(seq));
2210 }
2211 
2212 /*
2213  * IB stuff
2214  */
2215 /**
2216  * gfx_v7_0_ring_emit_ib - emit an IB (Indirect Buffer) on the ring
2217  *
2218  * @ring: amdgpu_ring structure holding ring information
2219  * @ib: amdgpu indirect buffer object
2220  *
2221  * Emits an DE (drawing engine) or CE (constant engine) IB
2222  * on the gfx ring.  IBs are usually generated by userspace
2223  * acceleration drivers and submitted to the kernel for
2224  * sheduling on the ring.  This function schedules the IB
2225  * on the gfx ring for execution by the GPU.
2226  */
2227 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2228 				      struct amdgpu_ib *ib,
2229 				      unsigned vm_id, bool ctx_switch)
2230 {
2231 	u32 header, control = 0;
2232 
2233 	/* insert SWITCH_BUFFER packet before first IB in the ring frame */
2234 	if (ctx_switch) {
2235 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2236 		amdgpu_ring_write(ring, 0);
2237 	}
2238 
2239 	if (ib->flags & AMDGPU_IB_FLAG_CE)
2240 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2241 	else
2242 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2243 
2244 	control |= ib->length_dw | (vm_id << 24);
2245 
2246 	amdgpu_ring_write(ring, header);
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_ib_compute(struct amdgpu_ring *ring,
2257 					  struct amdgpu_ib *ib,
2258 					  unsigned vm_id, bool ctx_switch)
2259 {
2260 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vm_id << 24);
2261 
2262 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2263 	amdgpu_ring_write(ring,
2264 #ifdef __BIG_ENDIAN
2265 					  (2 << 0) |
2266 #endif
2267 					  (ib->gpu_addr & 0xFFFFFFFC));
2268 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2269 	amdgpu_ring_write(ring, control);
2270 }
2271 
2272 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2273 {
2274 	uint32_t dw2 = 0;
2275 
2276 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2277 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2278 		gfx_v7_0_ring_emit_vgt_flush(ring);
2279 		/* set load_global_config & load_global_uconfig */
2280 		dw2 |= 0x8001;
2281 		/* set load_cs_sh_regs */
2282 		dw2 |= 0x01000000;
2283 		/* set load_per_context_state & load_gfx_sh_regs */
2284 		dw2 |= 0x10002;
2285 	}
2286 
2287 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2288 	amdgpu_ring_write(ring, dw2);
2289 	amdgpu_ring_write(ring, 0);
2290 }
2291 
2292 /**
2293  * gfx_v7_0_ring_test_ib - basic ring IB test
2294  *
2295  * @ring: amdgpu_ring structure holding ring information
2296  *
2297  * Allocate an IB and execute it on the gfx ring (CIK).
2298  * Provides a basic gfx ring test to verify that IBs are working.
2299  * Returns 0 on success, error on failure.
2300  */
2301 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2302 {
2303 	struct amdgpu_device *adev = ring->adev;
2304 	struct amdgpu_ib ib;
2305 	struct dma_fence *f = NULL;
2306 	uint32_t scratch;
2307 	uint32_t tmp = 0;
2308 	long r;
2309 
2310 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2311 	if (r) {
2312 		DRM_ERROR("amdgpu: failed to get scratch reg (%ld).\n", r);
2313 		return r;
2314 	}
2315 	WREG32(scratch, 0xCAFEDEAD);
2316 	memset(&ib, 0, sizeof(ib));
2317 	r = amdgpu_ib_get(adev, NULL, 256, &ib);
2318 	if (r) {
2319 		DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
2320 		goto err1;
2321 	}
2322 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2323 	ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
2324 	ib.ptr[2] = 0xDEADBEEF;
2325 	ib.length_dw = 3;
2326 
2327 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2328 	if (r)
2329 		goto err2;
2330 
2331 	r = dma_fence_wait_timeout(f, false, timeout);
2332 	if (r == 0) {
2333 		DRM_ERROR("amdgpu: IB test timed out\n");
2334 		r = -ETIMEDOUT;
2335 		goto err2;
2336 	} else if (r < 0) {
2337 		DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
2338 		goto err2;
2339 	}
2340 	tmp = RREG32(scratch);
2341 	if (tmp == 0xDEADBEEF) {
2342 		DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
2343 		r = 0;
2344 	} else {
2345 		DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n",
2346 			  scratch, tmp);
2347 		r = -EINVAL;
2348 	}
2349 
2350 err2:
2351 	amdgpu_ib_free(adev, &ib, NULL);
2352 	dma_fence_put(f);
2353 err1:
2354 	amdgpu_gfx_scratch_free(adev, scratch);
2355 	return r;
2356 }
2357 
2358 /*
2359  * CP.
2360  * On CIK, gfx and compute now have independant command processors.
2361  *
2362  * GFX
2363  * Gfx consists of a single ring and can process both gfx jobs and
2364  * compute jobs.  The gfx CP consists of three microengines (ME):
2365  * PFP - Pre-Fetch Parser
2366  * ME - Micro Engine
2367  * CE - Constant Engine
2368  * The PFP and ME make up what is considered the Drawing Engine (DE).
2369  * The CE is an asynchronous engine used for updating buffer desciptors
2370  * used by the DE so that they can be loaded into cache in parallel
2371  * while the DE is processing state update packets.
2372  *
2373  * Compute
2374  * The compute CP consists of two microengines (ME):
2375  * MEC1 - Compute MicroEngine 1
2376  * MEC2 - Compute MicroEngine 2
2377  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2378  * The queues are exposed to userspace and are programmed directly
2379  * by the compute runtime.
2380  */
2381 /**
2382  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2383  *
2384  * @adev: amdgpu_device pointer
2385  * @enable: enable or disable the MEs
2386  *
2387  * Halts or unhalts the gfx MEs.
2388  */
2389 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2390 {
2391 	int i;
2392 
2393 	if (enable) {
2394 		WREG32(mmCP_ME_CNTL, 0);
2395 	} else {
2396 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK));
2397 		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2398 			adev->gfx.gfx_ring[i].ready = false;
2399 	}
2400 	udelay(50);
2401 }
2402 
2403 /**
2404  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2405  *
2406  * @adev: amdgpu_device pointer
2407  *
2408  * Loads the gfx PFP, ME, and CE ucode.
2409  * Returns 0 for success, -EINVAL if the ucode is not available.
2410  */
2411 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2412 {
2413 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2414 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2415 	const struct gfx_firmware_header_v1_0 *me_hdr;
2416 	const __le32 *fw_data;
2417 	unsigned i, fw_size;
2418 
2419 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2420 		return -EINVAL;
2421 
2422 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2423 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2424 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2425 
2426 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2427 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2428 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2429 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2430 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2431 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2432 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2433 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2434 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2435 
2436 	gfx_v7_0_cp_gfx_enable(adev, false);
2437 
2438 	/* PFP */
2439 	fw_data = (const __le32 *)
2440 		(adev->gfx.pfp_fw->data +
2441 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2442 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2443 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2444 	for (i = 0; i < fw_size; i++)
2445 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2446 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2447 
2448 	/* CE */
2449 	fw_data = (const __le32 *)
2450 		(adev->gfx.ce_fw->data +
2451 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2452 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2453 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2454 	for (i = 0; i < fw_size; i++)
2455 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2456 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2457 
2458 	/* ME */
2459 	fw_data = (const __le32 *)
2460 		(adev->gfx.me_fw->data +
2461 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2462 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2463 	WREG32(mmCP_ME_RAM_WADDR, 0);
2464 	for (i = 0; i < fw_size; i++)
2465 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2466 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2467 
2468 	return 0;
2469 }
2470 
2471 /**
2472  * gfx_v7_0_cp_gfx_start - start the gfx ring
2473  *
2474  * @adev: amdgpu_device pointer
2475  *
2476  * Enables the ring and loads the clear state context and other
2477  * packets required to init the ring.
2478  * Returns 0 for success, error for failure.
2479  */
2480 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2481 {
2482 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2483 	const struct cs_section_def *sect = NULL;
2484 	const struct cs_extent_def *ext = NULL;
2485 	int r, i;
2486 
2487 	/* init the CP */
2488 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2489 	WREG32(mmCP_ENDIAN_SWAP, 0);
2490 	WREG32(mmCP_DEVICE_ID, 1);
2491 
2492 	gfx_v7_0_cp_gfx_enable(adev, true);
2493 
2494 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2495 	if (r) {
2496 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2497 		return r;
2498 	}
2499 
2500 	/* init the CE partitions.  CE only used for gfx on CIK */
2501 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2502 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2503 	amdgpu_ring_write(ring, 0x8000);
2504 	amdgpu_ring_write(ring, 0x8000);
2505 
2506 	/* clear state buffer */
2507 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2508 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2509 
2510 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2511 	amdgpu_ring_write(ring, 0x80000000);
2512 	amdgpu_ring_write(ring, 0x80000000);
2513 
2514 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2515 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2516 			if (sect->id == SECT_CONTEXT) {
2517 				amdgpu_ring_write(ring,
2518 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2519 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2520 				for (i = 0; i < ext->reg_count; i++)
2521 					amdgpu_ring_write(ring, ext->extent[i]);
2522 			}
2523 		}
2524 	}
2525 
2526 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2527 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2528 	switch (adev->asic_type) {
2529 	case CHIP_BONAIRE:
2530 		amdgpu_ring_write(ring, 0x16000012);
2531 		amdgpu_ring_write(ring, 0x00000000);
2532 		break;
2533 	case CHIP_KAVERI:
2534 		amdgpu_ring_write(ring, 0x00000000); /* XXX */
2535 		amdgpu_ring_write(ring, 0x00000000);
2536 		break;
2537 	case CHIP_KABINI:
2538 	case CHIP_MULLINS:
2539 		amdgpu_ring_write(ring, 0x00000000); /* XXX */
2540 		amdgpu_ring_write(ring, 0x00000000);
2541 		break;
2542 	case CHIP_HAWAII:
2543 		amdgpu_ring_write(ring, 0x3a00161a);
2544 		amdgpu_ring_write(ring, 0x0000002e);
2545 		break;
2546 	default:
2547 		amdgpu_ring_write(ring, 0x00000000);
2548 		amdgpu_ring_write(ring, 0x00000000);
2549 		break;
2550 	}
2551 
2552 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2553 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2554 
2555 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2556 	amdgpu_ring_write(ring, 0);
2557 
2558 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2559 	amdgpu_ring_write(ring, 0x00000316);
2560 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2561 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2562 
2563 	amdgpu_ring_commit(ring);
2564 
2565 	return 0;
2566 }
2567 
2568 /**
2569  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2570  *
2571  * @adev: amdgpu_device pointer
2572  *
2573  * Program the location and size of the gfx ring buffer
2574  * and test it to make sure it's working.
2575  * Returns 0 for success, error for failure.
2576  */
2577 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2578 {
2579 	struct amdgpu_ring *ring;
2580 	u32 tmp;
2581 	u32 rb_bufsz;
2582 	u64 rb_addr, rptr_addr;
2583 	int r;
2584 
2585 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2586 	if (adev->asic_type != CHIP_HAWAII)
2587 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2588 
2589 	/* Set the write pointer delay */
2590 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2591 
2592 	/* set the RB to use vmid 0 */
2593 	WREG32(mmCP_RB_VMID, 0);
2594 
2595 	WREG32(mmSCRATCH_ADDR, 0);
2596 
2597 	/* ring 0 - compute and gfx */
2598 	/* Set ring buffer size */
2599 	ring = &adev->gfx.gfx_ring[0];
2600 	rb_bufsz = order_base_2(ring->ring_size / 8);
2601 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2602 #ifdef __BIG_ENDIAN
2603 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2604 #endif
2605 	WREG32(mmCP_RB0_CNTL, tmp);
2606 
2607 	/* Initialize the ring buffer's read and write pointers */
2608 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2609 	ring->wptr = 0;
2610 	WREG32(mmCP_RB0_WPTR, ring->wptr);
2611 
2612 	/* set the wb address wether it's enabled or not */
2613 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2614 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2615 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2616 
2617 	/* scratch register shadowing is no longer supported */
2618 	WREG32(mmSCRATCH_UMSK, 0);
2619 
2620 	mdelay(1);
2621 	WREG32(mmCP_RB0_CNTL, tmp);
2622 
2623 	rb_addr = ring->gpu_addr >> 8;
2624 	WREG32(mmCP_RB0_BASE, rb_addr);
2625 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2626 
2627 	/* start the ring */
2628 	gfx_v7_0_cp_gfx_start(adev);
2629 	ring->ready = true;
2630 	r = amdgpu_ring_test_ring(ring);
2631 	if (r) {
2632 		ring->ready = false;
2633 		return r;
2634 	}
2635 
2636 	return 0;
2637 }
2638 
2639 static u32 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2640 {
2641 	return ring->adev->wb.wb[ring->rptr_offs];
2642 }
2643 
2644 static u32 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2645 {
2646 	struct amdgpu_device *adev = ring->adev;
2647 
2648 	return RREG32(mmCP_RB0_WPTR);
2649 }
2650 
2651 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2652 {
2653 	struct amdgpu_device *adev = ring->adev;
2654 
2655 	WREG32(mmCP_RB0_WPTR, ring->wptr);
2656 	(void)RREG32(mmCP_RB0_WPTR);
2657 }
2658 
2659 static u32 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2660 {
2661 	/* XXX check if swapping is necessary on BE */
2662 	return ring->adev->wb.wb[ring->wptr_offs];
2663 }
2664 
2665 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2666 {
2667 	struct amdgpu_device *adev = ring->adev;
2668 
2669 	/* XXX check if swapping is necessary on BE */
2670 	adev->wb.wb[ring->wptr_offs] = ring->wptr;
2671 	WDOORBELL32(ring->doorbell_index, ring->wptr);
2672 }
2673 
2674 /**
2675  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2676  *
2677  * @adev: amdgpu_device pointer
2678  * @enable: enable or disable the MEs
2679  *
2680  * Halts or unhalts the compute MEs.
2681  */
2682 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2683 {
2684 	int i;
2685 
2686 	if (enable) {
2687 		WREG32(mmCP_MEC_CNTL, 0);
2688 	} else {
2689 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2690 		for (i = 0; i < adev->gfx.num_compute_rings; i++)
2691 			adev->gfx.compute_ring[i].ready = false;
2692 	}
2693 	udelay(50);
2694 }
2695 
2696 /**
2697  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2698  *
2699  * @adev: amdgpu_device pointer
2700  *
2701  * Loads the compute MEC1&2 ucode.
2702  * Returns 0 for success, -EINVAL if the ucode is not available.
2703  */
2704 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2705 {
2706 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2707 	const __le32 *fw_data;
2708 	unsigned i, fw_size;
2709 
2710 	if (!adev->gfx.mec_fw)
2711 		return -EINVAL;
2712 
2713 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2714 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2715 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2716 	adev->gfx.mec_feature_version = le32_to_cpu(
2717 					mec_hdr->ucode_feature_version);
2718 
2719 	gfx_v7_0_cp_compute_enable(adev, false);
2720 
2721 	/* MEC1 */
2722 	fw_data = (const __le32 *)
2723 		(adev->gfx.mec_fw->data +
2724 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2725 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2726 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2727 	for (i = 0; i < fw_size; i++)
2728 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2729 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2730 
2731 	if (adev->asic_type == CHIP_KAVERI) {
2732 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2733 
2734 		if (!adev->gfx.mec2_fw)
2735 			return -EINVAL;
2736 
2737 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2738 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2739 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2740 		adev->gfx.mec2_feature_version = le32_to_cpu(
2741 				mec2_hdr->ucode_feature_version);
2742 
2743 		/* MEC2 */
2744 		fw_data = (const __le32 *)
2745 			(adev->gfx.mec2_fw->data +
2746 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2747 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2748 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2749 		for (i = 0; i < fw_size; i++)
2750 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2751 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2752 	}
2753 
2754 	return 0;
2755 }
2756 
2757 /**
2758  * gfx_v7_0_cp_compute_fini - stop the compute queues
2759  *
2760  * @adev: amdgpu_device pointer
2761  *
2762  * Stop the compute queues and tear down the driver queue
2763  * info.
2764  */
2765 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2766 {
2767 	int i, r;
2768 
2769 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2770 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2771 
2772 		if (ring->mqd_obj) {
2773 			r = amdgpu_bo_reserve(ring->mqd_obj, false);
2774 			if (unlikely(r != 0))
2775 				dev_warn(adev->dev, "(%d) reserve MQD bo failed\n", r);
2776 
2777 			amdgpu_bo_unpin(ring->mqd_obj);
2778 			amdgpu_bo_unreserve(ring->mqd_obj);
2779 
2780 			amdgpu_bo_unref(&ring->mqd_obj);
2781 			ring->mqd_obj = NULL;
2782 		}
2783 	}
2784 }
2785 
2786 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2787 {
2788 	int r;
2789 
2790 	if (adev->gfx.mec.hpd_eop_obj) {
2791 		r = amdgpu_bo_reserve(adev->gfx.mec.hpd_eop_obj, false);
2792 		if (unlikely(r != 0))
2793 			dev_warn(adev->dev, "(%d) reserve HPD EOP bo failed\n", r);
2794 		amdgpu_bo_unpin(adev->gfx.mec.hpd_eop_obj);
2795 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2796 
2797 		amdgpu_bo_unref(&adev->gfx.mec.hpd_eop_obj);
2798 		adev->gfx.mec.hpd_eop_obj = NULL;
2799 	}
2800 }
2801 
2802 #define MEC_HPD_SIZE 2048
2803 
2804 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2805 {
2806 	int r;
2807 	u32 *hpd;
2808 
2809 	/*
2810 	 * KV:    2 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 64 Queues total
2811 	 * CI/KB: 1 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 32 Queues total
2812 	 * Nonetheless, we assign only 1 pipe because all other pipes will
2813 	 * be handled by KFD
2814 	 */
2815 	adev->gfx.mec.num_mec = 1;
2816 	adev->gfx.mec.num_pipe = 1;
2817 	adev->gfx.mec.num_queue = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe * 8;
2818 
2819 	if (adev->gfx.mec.hpd_eop_obj == NULL) {
2820 		r = amdgpu_bo_create(adev,
2821 				     adev->gfx.mec.num_mec *adev->gfx.mec.num_pipe * MEC_HPD_SIZE * 2,
2822 				     PAGE_SIZE, true,
2823 				     AMDGPU_GEM_DOMAIN_GTT, 0, NULL, NULL,
2824 				     &adev->gfx.mec.hpd_eop_obj);
2825 		if (r) {
2826 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
2827 			return r;
2828 		}
2829 	}
2830 
2831 	r = amdgpu_bo_reserve(adev->gfx.mec.hpd_eop_obj, false);
2832 	if (unlikely(r != 0)) {
2833 		gfx_v7_0_mec_fini(adev);
2834 		return r;
2835 	}
2836 	r = amdgpu_bo_pin(adev->gfx.mec.hpd_eop_obj, AMDGPU_GEM_DOMAIN_GTT,
2837 			  &adev->gfx.mec.hpd_eop_gpu_addr);
2838 	if (r) {
2839 		dev_warn(adev->dev, "(%d) pin HDP EOP bo failed\n", r);
2840 		gfx_v7_0_mec_fini(adev);
2841 		return r;
2842 	}
2843 	r = amdgpu_bo_kmap(adev->gfx.mec.hpd_eop_obj, (void **)&hpd);
2844 	if (r) {
2845 		dev_warn(adev->dev, "(%d) map HDP EOP bo failed\n", r);
2846 		gfx_v7_0_mec_fini(adev);
2847 		return r;
2848 	}
2849 
2850 	/* clear memory.  Not sure if this is required or not */
2851 	memset(hpd, 0, adev->gfx.mec.num_mec *adev->gfx.mec.num_pipe * MEC_HPD_SIZE * 2);
2852 
2853 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2854 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2855 
2856 	return 0;
2857 }
2858 
2859 struct hqd_registers
2860 {
2861 	u32 cp_mqd_base_addr;
2862 	u32 cp_mqd_base_addr_hi;
2863 	u32 cp_hqd_active;
2864 	u32 cp_hqd_vmid;
2865 	u32 cp_hqd_persistent_state;
2866 	u32 cp_hqd_pipe_priority;
2867 	u32 cp_hqd_queue_priority;
2868 	u32 cp_hqd_quantum;
2869 	u32 cp_hqd_pq_base;
2870 	u32 cp_hqd_pq_base_hi;
2871 	u32 cp_hqd_pq_rptr;
2872 	u32 cp_hqd_pq_rptr_report_addr;
2873 	u32 cp_hqd_pq_rptr_report_addr_hi;
2874 	u32 cp_hqd_pq_wptr_poll_addr;
2875 	u32 cp_hqd_pq_wptr_poll_addr_hi;
2876 	u32 cp_hqd_pq_doorbell_control;
2877 	u32 cp_hqd_pq_wptr;
2878 	u32 cp_hqd_pq_control;
2879 	u32 cp_hqd_ib_base_addr;
2880 	u32 cp_hqd_ib_base_addr_hi;
2881 	u32 cp_hqd_ib_rptr;
2882 	u32 cp_hqd_ib_control;
2883 	u32 cp_hqd_iq_timer;
2884 	u32 cp_hqd_iq_rptr;
2885 	u32 cp_hqd_dequeue_request;
2886 	u32 cp_hqd_dma_offload;
2887 	u32 cp_hqd_sema_cmd;
2888 	u32 cp_hqd_msg_type;
2889 	u32 cp_hqd_atomic0_preop_lo;
2890 	u32 cp_hqd_atomic0_preop_hi;
2891 	u32 cp_hqd_atomic1_preop_lo;
2892 	u32 cp_hqd_atomic1_preop_hi;
2893 	u32 cp_hqd_hq_scheduler0;
2894 	u32 cp_hqd_hq_scheduler1;
2895 	u32 cp_mqd_control;
2896 };
2897 
2898 struct bonaire_mqd
2899 {
2900 	u32 header;
2901 	u32 dispatch_initiator;
2902 	u32 dimensions[3];
2903 	u32 start_idx[3];
2904 	u32 num_threads[3];
2905 	u32 pipeline_stat_enable;
2906 	u32 perf_counter_enable;
2907 	u32 pgm[2];
2908 	u32 tba[2];
2909 	u32 tma[2];
2910 	u32 pgm_rsrc[2];
2911 	u32 vmid;
2912 	u32 resource_limits;
2913 	u32 static_thread_mgmt01[2];
2914 	u32 tmp_ring_size;
2915 	u32 static_thread_mgmt23[2];
2916 	u32 restart[3];
2917 	u32 thread_trace_enable;
2918 	u32 reserved1;
2919 	u32 user_data[16];
2920 	u32 vgtcs_invoke_count[2];
2921 	struct hqd_registers queue_state;
2922 	u32 dequeue_cntr;
2923 	u32 interrupt_queue[64];
2924 };
2925 
2926 /**
2927  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
2928  *
2929  * @adev: amdgpu_device pointer
2930  *
2931  * Program the compute queues and test them to make sure they
2932  * are working.
2933  * Returns 0 for success, error for failure.
2934  */
2935 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
2936 {
2937 	int r, i, j;
2938 	u32 tmp;
2939 	bool use_doorbell = true;
2940 	u64 hqd_gpu_addr;
2941 	u64 mqd_gpu_addr;
2942 	u64 eop_gpu_addr;
2943 	u64 wb_gpu_addr;
2944 	u32 *buf;
2945 	struct bonaire_mqd *mqd;
2946 	struct amdgpu_ring *ring;
2947 
2948 	/* fix up chicken bits */
2949 	tmp = RREG32(mmCP_CPF_DEBUG);
2950 	tmp |= (1 << 23);
2951 	WREG32(mmCP_CPF_DEBUG, tmp);
2952 
2953 	/* init the pipes */
2954 	mutex_lock(&adev->srbm_mutex);
2955 	for (i = 0; i < (adev->gfx.mec.num_pipe * adev->gfx.mec.num_mec); i++) {
2956 		int me = (i < 4) ? 1 : 2;
2957 		int pipe = (i < 4) ? i : (i - 4);
2958 
2959 		eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (i * MEC_HPD_SIZE * 2);
2960 
2961 		cik_srbm_select(adev, me, pipe, 0, 0);
2962 
2963 		/* write the EOP addr */
2964 		WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2965 		WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2966 
2967 		/* set the VMID assigned */
2968 		WREG32(mmCP_HPD_EOP_VMID, 0);
2969 
2970 		/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2971 		tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2972 		tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2973 		tmp |= order_base_2(MEC_HPD_SIZE / 8);
2974 		WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2975 	}
2976 	cik_srbm_select(adev, 0, 0, 0, 0);
2977 	mutex_unlock(&adev->srbm_mutex);
2978 
2979 	/* init the queues.  Just two for now. */
2980 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2981 		ring = &adev->gfx.compute_ring[i];
2982 
2983 		if (ring->mqd_obj == NULL) {
2984 			r = amdgpu_bo_create(adev,
2985 					     sizeof(struct bonaire_mqd),
2986 					     PAGE_SIZE, true,
2987 					     AMDGPU_GEM_DOMAIN_GTT, 0, NULL, NULL,
2988 					     &ring->mqd_obj);
2989 			if (r) {
2990 				dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
2991 				return r;
2992 			}
2993 		}
2994 
2995 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
2996 		if (unlikely(r != 0)) {
2997 			gfx_v7_0_cp_compute_fini(adev);
2998 			return r;
2999 		}
3000 		r = amdgpu_bo_pin(ring->mqd_obj, AMDGPU_GEM_DOMAIN_GTT,
3001 				  &mqd_gpu_addr);
3002 		if (r) {
3003 			dev_warn(adev->dev, "(%d) pin MQD bo failed\n", r);
3004 			gfx_v7_0_cp_compute_fini(adev);
3005 			return r;
3006 		}
3007 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&buf);
3008 		if (r) {
3009 			dev_warn(adev->dev, "(%d) map MQD bo failed\n", r);
3010 			gfx_v7_0_cp_compute_fini(adev);
3011 			return r;
3012 		}
3013 
3014 		/* init the mqd struct */
3015 		memset(buf, 0, sizeof(struct bonaire_mqd));
3016 
3017 		mqd = (struct bonaire_mqd *)buf;
3018 		mqd->header = 0xC0310800;
3019 		mqd->static_thread_mgmt01[0] = 0xffffffff;
3020 		mqd->static_thread_mgmt01[1] = 0xffffffff;
3021 		mqd->static_thread_mgmt23[0] = 0xffffffff;
3022 		mqd->static_thread_mgmt23[1] = 0xffffffff;
3023 
3024 		mutex_lock(&adev->srbm_mutex);
3025 		cik_srbm_select(adev, ring->me,
3026 				ring->pipe,
3027 				ring->queue, 0);
3028 
3029 		/* disable wptr polling */
3030 		tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
3031 		tmp &= ~CP_PQ_WPTR_POLL_CNTL__EN_MASK;
3032 		WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
3033 
3034 		/* enable doorbell? */
3035 		mqd->queue_state.cp_hqd_pq_doorbell_control =
3036 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
3037 		if (use_doorbell)
3038 			mqd->queue_state.cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3039 		else
3040 			mqd->queue_state.cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3041 		WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL,
3042 		       mqd->queue_state.cp_hqd_pq_doorbell_control);
3043 
3044 		/* disable the queue if it's active */
3045 		mqd->queue_state.cp_hqd_dequeue_request = 0;
3046 		mqd->queue_state.cp_hqd_pq_rptr = 0;
3047 		mqd->queue_state.cp_hqd_pq_wptr= 0;
3048 		if (RREG32(mmCP_HQD_ACTIVE) & 1) {
3049 			WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
3050 			for (j = 0; j < adev->usec_timeout; j++) {
3051 				if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
3052 					break;
3053 				udelay(1);
3054 			}
3055 			WREG32(mmCP_HQD_DEQUEUE_REQUEST, mqd->queue_state.cp_hqd_dequeue_request);
3056 			WREG32(mmCP_HQD_PQ_RPTR, mqd->queue_state.cp_hqd_pq_rptr);
3057 			WREG32(mmCP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
3058 		}
3059 
3060 		/* set the pointer to the MQD */
3061 		mqd->queue_state.cp_mqd_base_addr = mqd_gpu_addr & 0xfffffffc;
3062 		mqd->queue_state.cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
3063 		WREG32(mmCP_MQD_BASE_ADDR, mqd->queue_state.cp_mqd_base_addr);
3064 		WREG32(mmCP_MQD_BASE_ADDR_HI, mqd->queue_state.cp_mqd_base_addr_hi);
3065 		/* set MQD vmid to 0 */
3066 		mqd->queue_state.cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
3067 		mqd->queue_state.cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
3068 		WREG32(mmCP_MQD_CONTROL, mqd->queue_state.cp_mqd_control);
3069 
3070 		/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3071 		hqd_gpu_addr = ring->gpu_addr >> 8;
3072 		mqd->queue_state.cp_hqd_pq_base = hqd_gpu_addr;
3073 		mqd->queue_state.cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3074 		WREG32(mmCP_HQD_PQ_BASE, mqd->queue_state.cp_hqd_pq_base);
3075 		WREG32(mmCP_HQD_PQ_BASE_HI, mqd->queue_state.cp_hqd_pq_base_hi);
3076 
3077 		/* set up the HQD, this is similar to CP_RB0_CNTL */
3078 		mqd->queue_state.cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
3079 		mqd->queue_state.cp_hqd_pq_control &=
3080 			~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
3081 					CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
3082 
3083 		mqd->queue_state.cp_hqd_pq_control |=
3084 			order_base_2(ring->ring_size / 8);
3085 		mqd->queue_state.cp_hqd_pq_control |=
3086 			(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
3087 #ifdef __BIG_ENDIAN
3088 		mqd->queue_state.cp_hqd_pq_control |=
3089 			2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
3090 #endif
3091 		mqd->queue_state.cp_hqd_pq_control &=
3092 			~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
3093 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
3094 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
3095 		mqd->queue_state.cp_hqd_pq_control |=
3096 			CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
3097 			CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
3098 		WREG32(mmCP_HQD_PQ_CONTROL, mqd->queue_state.cp_hqd_pq_control);
3099 
3100 		/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3101 		wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3102 		mqd->queue_state.cp_hqd_pq_wptr_poll_addr = wb_gpu_addr & 0xfffffffc;
3103 		mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3104 		WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR, mqd->queue_state.cp_hqd_pq_wptr_poll_addr);
3105 		WREG32(mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3106 		       mqd->queue_state.cp_hqd_pq_wptr_poll_addr_hi);
3107 
3108 		/* set the wb address wether it's enabled or not */
3109 		wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3110 		mqd->queue_state.cp_hqd_pq_rptr_report_addr = wb_gpu_addr & 0xfffffffc;
3111 		mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi =
3112 			upper_32_bits(wb_gpu_addr) & 0xffff;
3113 		WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3114 		       mqd->queue_state.cp_hqd_pq_rptr_report_addr);
3115 		WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3116 		       mqd->queue_state.cp_hqd_pq_rptr_report_addr_hi);
3117 
3118 		/* enable the doorbell if requested */
3119 		if (use_doorbell) {
3120 			mqd->queue_state.cp_hqd_pq_doorbell_control =
3121 				RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
3122 			mqd->queue_state.cp_hqd_pq_doorbell_control &=
3123 				~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
3124 			mqd->queue_state.cp_hqd_pq_doorbell_control |=
3125 				(ring->doorbell_index <<
3126 				 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
3127 			mqd->queue_state.cp_hqd_pq_doorbell_control |=
3128 				CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3129 			mqd->queue_state.cp_hqd_pq_doorbell_control &=
3130 				~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
3131 				CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
3132 
3133 		} else {
3134 			mqd->queue_state.cp_hqd_pq_doorbell_control = 0;
3135 		}
3136 		WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL,
3137 		       mqd->queue_state.cp_hqd_pq_doorbell_control);
3138 
3139 		/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3140 		ring->wptr = 0;
3141 		mqd->queue_state.cp_hqd_pq_wptr = ring->wptr;
3142 		WREG32(mmCP_HQD_PQ_WPTR, mqd->queue_state.cp_hqd_pq_wptr);
3143 		mqd->queue_state.cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
3144 
3145 		/* set the vmid for the queue */
3146 		mqd->queue_state.cp_hqd_vmid = 0;
3147 		WREG32(mmCP_HQD_VMID, mqd->queue_state.cp_hqd_vmid);
3148 
3149 		/* activate the queue */
3150 		mqd->queue_state.cp_hqd_active = 1;
3151 		WREG32(mmCP_HQD_ACTIVE, mqd->queue_state.cp_hqd_active);
3152 
3153 		cik_srbm_select(adev, 0, 0, 0, 0);
3154 		mutex_unlock(&adev->srbm_mutex);
3155 
3156 		amdgpu_bo_kunmap(ring->mqd_obj);
3157 		amdgpu_bo_unreserve(ring->mqd_obj);
3158 
3159 		ring->ready = true;
3160 	}
3161 
3162 	gfx_v7_0_cp_compute_enable(adev, true);
3163 
3164 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3165 		ring = &adev->gfx.compute_ring[i];
3166 
3167 		r = amdgpu_ring_test_ring(ring);
3168 		if (r)
3169 			ring->ready = false;
3170 	}
3171 
3172 	return 0;
3173 }
3174 
3175 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3176 {
3177 	gfx_v7_0_cp_gfx_enable(adev, enable);
3178 	gfx_v7_0_cp_compute_enable(adev, enable);
3179 }
3180 
3181 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3182 {
3183 	int r;
3184 
3185 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3186 	if (r)
3187 		return r;
3188 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3189 	if (r)
3190 		return r;
3191 
3192 	return 0;
3193 }
3194 
3195 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3196 					       bool enable)
3197 {
3198 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3199 
3200 	if (enable)
3201 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3202 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3203 	else
3204 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3205 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3206 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3207 }
3208 
3209 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3210 {
3211 	int r;
3212 
3213 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3214 
3215 	r = gfx_v7_0_cp_load_microcode(adev);
3216 	if (r)
3217 		return r;
3218 
3219 	r = gfx_v7_0_cp_gfx_resume(adev);
3220 	if (r)
3221 		return r;
3222 	r = gfx_v7_0_cp_compute_resume(adev);
3223 	if (r)
3224 		return r;
3225 
3226 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3227 
3228 	return 0;
3229 }
3230 
3231 /**
3232  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3233  *
3234  * @ring: the ring to emmit the commands to
3235  *
3236  * Sync the command pipeline with the PFP. E.g. wait for everything
3237  * to be completed.
3238  */
3239 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3240 {
3241 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3242 	uint32_t seq = ring->fence_drv.sync_seq;
3243 	uint64_t addr = ring->fence_drv.gpu_addr;
3244 
3245 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3246 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3247 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3248 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3249 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3250 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3251 	amdgpu_ring_write(ring, seq);
3252 	amdgpu_ring_write(ring, 0xffffffff);
3253 	amdgpu_ring_write(ring, 4); /* poll interval */
3254 
3255 	if (usepfp) {
3256 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3257 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3258 		amdgpu_ring_write(ring, 0);
3259 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3260 		amdgpu_ring_write(ring, 0);
3261 	}
3262 }
3263 
3264 /*
3265  * vm
3266  * VMID 0 is the physical GPU addresses as used by the kernel.
3267  * VMIDs 1-15 are used for userspace clients and are handled
3268  * by the amdgpu vm/hsa code.
3269  */
3270 /**
3271  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3272  *
3273  * @adev: amdgpu_device pointer
3274  *
3275  * Update the page table base and flush the VM TLB
3276  * using the CP (CIK).
3277  */
3278 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3279 					unsigned vm_id, uint64_t pd_addr)
3280 {
3281 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3282 
3283 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3284 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3285 				 WRITE_DATA_DST_SEL(0)));
3286 	if (vm_id < 8) {
3287 		amdgpu_ring_write(ring,
3288 				  (mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vm_id));
3289 	} else {
3290 		amdgpu_ring_write(ring,
3291 				  (mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vm_id - 8));
3292 	}
3293 	amdgpu_ring_write(ring, 0);
3294 	amdgpu_ring_write(ring, pd_addr >> 12);
3295 
3296 	/* bits 0-15 are the VM contexts0-15 */
3297 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3298 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3299 				 WRITE_DATA_DST_SEL(0)));
3300 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3301 	amdgpu_ring_write(ring, 0);
3302 	amdgpu_ring_write(ring, 1 << vm_id);
3303 
3304 	/* wait for the invalidate to complete */
3305 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3306 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3307 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3308 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3309 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3310 	amdgpu_ring_write(ring, 0);
3311 	amdgpu_ring_write(ring, 0); /* ref */
3312 	amdgpu_ring_write(ring, 0); /* mask */
3313 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3314 
3315 	/* compute doesn't have PFP */
3316 	if (usepfp) {
3317 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3318 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3319 		amdgpu_ring_write(ring, 0x0);
3320 
3321 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3322 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3323 		amdgpu_ring_write(ring, 0);
3324 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3325 		amdgpu_ring_write(ring, 0);
3326 	}
3327 }
3328 
3329 /*
3330  * RLC
3331  * The RLC is a multi-purpose microengine that handles a
3332  * variety of functions.
3333  */
3334 static void gfx_v7_0_rlc_fini(struct amdgpu_device *adev)
3335 {
3336 	int r;
3337 
3338 	/* save restore block */
3339 	if (adev->gfx.rlc.save_restore_obj) {
3340 		r = amdgpu_bo_reserve(adev->gfx.rlc.save_restore_obj, false);
3341 		if (unlikely(r != 0))
3342 			dev_warn(adev->dev, "(%d) reserve RLC sr bo failed\n", r);
3343 		amdgpu_bo_unpin(adev->gfx.rlc.save_restore_obj);
3344 		amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj);
3345 
3346 		amdgpu_bo_unref(&adev->gfx.rlc.save_restore_obj);
3347 		adev->gfx.rlc.save_restore_obj = NULL;
3348 	}
3349 
3350 	/* clear state block */
3351 	if (adev->gfx.rlc.clear_state_obj) {
3352 		r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, false);
3353 		if (unlikely(r != 0))
3354 			dev_warn(adev->dev, "(%d) reserve RLC c bo failed\n", r);
3355 		amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj);
3356 		amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
3357 
3358 		amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj);
3359 		adev->gfx.rlc.clear_state_obj = NULL;
3360 	}
3361 
3362 	/* clear state block */
3363 	if (adev->gfx.rlc.cp_table_obj) {
3364 		r = amdgpu_bo_reserve(adev->gfx.rlc.cp_table_obj, false);
3365 		if (unlikely(r != 0))
3366 			dev_warn(adev->dev, "(%d) reserve RLC cp table bo failed\n", r);
3367 		amdgpu_bo_unpin(adev->gfx.rlc.cp_table_obj);
3368 		amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj);
3369 
3370 		amdgpu_bo_unref(&adev->gfx.rlc.cp_table_obj);
3371 		adev->gfx.rlc.cp_table_obj = NULL;
3372 	}
3373 }
3374 
3375 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3376 {
3377 	const u32 *src_ptr;
3378 	volatile u32 *dst_ptr;
3379 	u32 dws, i;
3380 	const struct cs_section_def *cs_data;
3381 	int r;
3382 
3383 	/* allocate rlc buffers */
3384 	if (adev->flags & AMD_IS_APU) {
3385 		if (adev->asic_type == CHIP_KAVERI) {
3386 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3387 			adev->gfx.rlc.reg_list_size =
3388 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3389 		} else {
3390 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3391 			adev->gfx.rlc.reg_list_size =
3392 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3393 		}
3394 	}
3395 	adev->gfx.rlc.cs_data = ci_cs_data;
3396 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3397 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3398 
3399 	src_ptr = adev->gfx.rlc.reg_list;
3400 	dws = adev->gfx.rlc.reg_list_size;
3401 	dws += (5 * 16) + 48 + 48 + 64;
3402 
3403 	cs_data = adev->gfx.rlc.cs_data;
3404 
3405 	if (src_ptr) {
3406 		/* save restore block */
3407 		if (adev->gfx.rlc.save_restore_obj == NULL) {
3408 			r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
3409 					     AMDGPU_GEM_DOMAIN_VRAM,
3410 					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
3411 					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
3412 					     NULL, NULL,
3413 					     &adev->gfx.rlc.save_restore_obj);
3414 			if (r) {
3415 				dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", r);
3416 				return r;
3417 			}
3418 		}
3419 
3420 		r = amdgpu_bo_reserve(adev->gfx.rlc.save_restore_obj, false);
3421 		if (unlikely(r != 0)) {
3422 			gfx_v7_0_rlc_fini(adev);
3423 			return r;
3424 		}
3425 		r = amdgpu_bo_pin(adev->gfx.rlc.save_restore_obj, AMDGPU_GEM_DOMAIN_VRAM,
3426 				  &adev->gfx.rlc.save_restore_gpu_addr);
3427 		if (r) {
3428 			amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj);
3429 			dev_warn(adev->dev, "(%d) pin RLC sr bo failed\n", r);
3430 			gfx_v7_0_rlc_fini(adev);
3431 			return r;
3432 		}
3433 
3434 		r = amdgpu_bo_kmap(adev->gfx.rlc.save_restore_obj, (void **)&adev->gfx.rlc.sr_ptr);
3435 		if (r) {
3436 			dev_warn(adev->dev, "(%d) map RLC sr bo failed\n", r);
3437 			gfx_v7_0_rlc_fini(adev);
3438 			return r;
3439 		}
3440 		/* write the sr buffer */
3441 		dst_ptr = adev->gfx.rlc.sr_ptr;
3442 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3443 			dst_ptr[i] = cpu_to_le32(src_ptr[i]);
3444 		amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj);
3445 		amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj);
3446 	}
3447 
3448 	if (cs_data) {
3449 		/* clear state block */
3450 		adev->gfx.rlc.clear_state_size = dws = gfx_v7_0_get_csb_size(adev);
3451 
3452 		if (adev->gfx.rlc.clear_state_obj == NULL) {
3453 			r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
3454 					     AMDGPU_GEM_DOMAIN_VRAM,
3455 					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
3456 					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
3457 					     NULL, NULL,
3458 					     &adev->gfx.rlc.clear_state_obj);
3459 			if (r) {
3460 				dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r);
3461 				gfx_v7_0_rlc_fini(adev);
3462 				return r;
3463 			}
3464 		}
3465 		r = amdgpu_bo_reserve(adev->gfx.rlc.clear_state_obj, false);
3466 		if (unlikely(r != 0)) {
3467 			gfx_v7_0_rlc_fini(adev);
3468 			return r;
3469 		}
3470 		r = amdgpu_bo_pin(adev->gfx.rlc.clear_state_obj, AMDGPU_GEM_DOMAIN_VRAM,
3471 				  &adev->gfx.rlc.clear_state_gpu_addr);
3472 		if (r) {
3473 			amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
3474 			dev_warn(adev->dev, "(%d) pin RLC c bo failed\n", r);
3475 			gfx_v7_0_rlc_fini(adev);
3476 			return r;
3477 		}
3478 
3479 		r = amdgpu_bo_kmap(adev->gfx.rlc.clear_state_obj, (void **)&adev->gfx.rlc.cs_ptr);
3480 		if (r) {
3481 			dev_warn(adev->dev, "(%d) map RLC c bo failed\n", r);
3482 			gfx_v7_0_rlc_fini(adev);
3483 			return r;
3484 		}
3485 		/* set up the cs buffer */
3486 		dst_ptr = adev->gfx.rlc.cs_ptr;
3487 		gfx_v7_0_get_csb_buffer(adev, dst_ptr);
3488 		amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj);
3489 		amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj);
3490 	}
3491 
3492 	if (adev->gfx.rlc.cp_table_size) {
3493 		if (adev->gfx.rlc.cp_table_obj == NULL) {
3494 			r = amdgpu_bo_create(adev, adev->gfx.rlc.cp_table_size, PAGE_SIZE, true,
3495 					     AMDGPU_GEM_DOMAIN_VRAM,
3496 					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
3497 					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
3498 					     NULL, NULL,
3499 					     &adev->gfx.rlc.cp_table_obj);
3500 			if (r) {
3501 				dev_warn(adev->dev, "(%d) create RLC cp table bo failed\n", r);
3502 				gfx_v7_0_rlc_fini(adev);
3503 				return r;
3504 			}
3505 		}
3506 
3507 		r = amdgpu_bo_reserve(adev->gfx.rlc.cp_table_obj, false);
3508 		if (unlikely(r != 0)) {
3509 			dev_warn(adev->dev, "(%d) reserve RLC cp table bo failed\n", r);
3510 			gfx_v7_0_rlc_fini(adev);
3511 			return r;
3512 		}
3513 		r = amdgpu_bo_pin(adev->gfx.rlc.cp_table_obj, AMDGPU_GEM_DOMAIN_VRAM,
3514 				  &adev->gfx.rlc.cp_table_gpu_addr);
3515 		if (r) {
3516 			amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj);
3517 			dev_warn(adev->dev, "(%d) pin RLC cp_table bo failed\n", r);
3518 			gfx_v7_0_rlc_fini(adev);
3519 			return r;
3520 		}
3521 		r = amdgpu_bo_kmap(adev->gfx.rlc.cp_table_obj, (void **)&adev->gfx.rlc.cp_table_ptr);
3522 		if (r) {
3523 			dev_warn(adev->dev, "(%d) map RLC cp table bo failed\n", r);
3524 			gfx_v7_0_rlc_fini(adev);
3525 			return r;
3526 		}
3527 
3528 		gfx_v7_0_init_cp_pg_table(adev);
3529 
3530 		amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj);
3531 		amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj);
3532 
3533 	}
3534 
3535 	return 0;
3536 }
3537 
3538 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3539 {
3540 	u32 tmp;
3541 
3542 	tmp = RREG32(mmRLC_LB_CNTL);
3543 	if (enable)
3544 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3545 	else
3546 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3547 	WREG32(mmRLC_LB_CNTL, tmp);
3548 }
3549 
3550 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3551 {
3552 	u32 i, j, k;
3553 	u32 mask;
3554 
3555 	mutex_lock(&adev->grbm_idx_mutex);
3556 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3557 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3558 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
3559 			for (k = 0; k < adev->usec_timeout; k++) {
3560 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3561 					break;
3562 				udelay(1);
3563 			}
3564 		}
3565 	}
3566 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3567 	mutex_unlock(&adev->grbm_idx_mutex);
3568 
3569 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3570 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3571 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3572 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3573 	for (k = 0; k < adev->usec_timeout; k++) {
3574 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3575 			break;
3576 		udelay(1);
3577 	}
3578 }
3579 
3580 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3581 {
3582 	u32 tmp;
3583 
3584 	tmp = RREG32(mmRLC_CNTL);
3585 	if (tmp != rlc)
3586 		WREG32(mmRLC_CNTL, rlc);
3587 }
3588 
3589 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3590 {
3591 	u32 data, orig;
3592 
3593 	orig = data = RREG32(mmRLC_CNTL);
3594 
3595 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3596 		u32 i;
3597 
3598 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3599 		WREG32(mmRLC_CNTL, data);
3600 
3601 		for (i = 0; i < adev->usec_timeout; i++) {
3602 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3603 				break;
3604 			udelay(1);
3605 		}
3606 
3607 		gfx_v7_0_wait_for_rlc_serdes(adev);
3608 	}
3609 
3610 	return orig;
3611 }
3612 
3613 static void gfx_v7_0_enter_rlc_safe_mode(struct amdgpu_device *adev)
3614 {
3615 	u32 tmp, i, mask;
3616 
3617 	tmp = 0x1 | (1 << 1);
3618 	WREG32(mmRLC_GPR_REG2, tmp);
3619 
3620 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3621 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3622 	for (i = 0; i < adev->usec_timeout; i++) {
3623 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3624 			break;
3625 		udelay(1);
3626 	}
3627 
3628 	for (i = 0; i < adev->usec_timeout; i++) {
3629 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3630 			break;
3631 		udelay(1);
3632 	}
3633 }
3634 
3635 static void gfx_v7_0_exit_rlc_safe_mode(struct amdgpu_device *adev)
3636 {
3637 	u32 tmp;
3638 
3639 	tmp = 0x1 | (0 << 1);
3640 	WREG32(mmRLC_GPR_REG2, tmp);
3641 }
3642 
3643 /**
3644  * gfx_v7_0_rlc_stop - stop the RLC ME
3645  *
3646  * @adev: amdgpu_device pointer
3647  *
3648  * Halt the RLC ME (MicroEngine) (CIK).
3649  */
3650 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3651 {
3652 	WREG32(mmRLC_CNTL, 0);
3653 
3654 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3655 
3656 	gfx_v7_0_wait_for_rlc_serdes(adev);
3657 }
3658 
3659 /**
3660  * gfx_v7_0_rlc_start - start the RLC ME
3661  *
3662  * @adev: amdgpu_device pointer
3663  *
3664  * Unhalt the RLC ME (MicroEngine) (CIK).
3665  */
3666 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3667 {
3668 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3669 
3670 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3671 
3672 	udelay(50);
3673 }
3674 
3675 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3676 {
3677 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3678 
3679 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3680 	WREG32(mmGRBM_SOFT_RESET, tmp);
3681 	udelay(50);
3682 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3683 	WREG32(mmGRBM_SOFT_RESET, tmp);
3684 	udelay(50);
3685 }
3686 
3687 /**
3688  * gfx_v7_0_rlc_resume - setup the RLC hw
3689  *
3690  * @adev: amdgpu_device pointer
3691  *
3692  * Initialize the RLC registers, load the ucode,
3693  * and start the RLC (CIK).
3694  * Returns 0 for success, -EINVAL if the ucode is not available.
3695  */
3696 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3697 {
3698 	const struct rlc_firmware_header_v1_0 *hdr;
3699 	const __le32 *fw_data;
3700 	unsigned i, fw_size;
3701 	u32 tmp;
3702 
3703 	if (!adev->gfx.rlc_fw)
3704 		return -EINVAL;
3705 
3706 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3707 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3708 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3709 	adev->gfx.rlc_feature_version = le32_to_cpu(
3710 					hdr->ucode_feature_version);
3711 
3712 	gfx_v7_0_rlc_stop(adev);
3713 
3714 	/* disable CG */
3715 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3716 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3717 
3718 	gfx_v7_0_rlc_reset(adev);
3719 
3720 	gfx_v7_0_init_pg(adev);
3721 
3722 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3723 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3724 
3725 	mutex_lock(&adev->grbm_idx_mutex);
3726 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3727 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3728 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3729 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3730 	mutex_unlock(&adev->grbm_idx_mutex);
3731 
3732 	WREG32(mmRLC_MC_CNTL, 0);
3733 	WREG32(mmRLC_UCODE_CNTL, 0);
3734 
3735 	fw_data = (const __le32 *)
3736 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3737 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3738 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3739 	for (i = 0; i < fw_size; i++)
3740 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3741 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3742 
3743 	/* XXX - find out what chips support lbpw */
3744 	gfx_v7_0_enable_lbpw(adev, false);
3745 
3746 	if (adev->asic_type == CHIP_BONAIRE)
3747 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3748 
3749 	gfx_v7_0_rlc_start(adev);
3750 
3751 	return 0;
3752 }
3753 
3754 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3755 {
3756 	u32 data, orig, tmp, tmp2;
3757 
3758 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3759 
3760 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3761 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3762 
3763 		tmp = gfx_v7_0_halt_rlc(adev);
3764 
3765 		mutex_lock(&adev->grbm_idx_mutex);
3766 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3767 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3768 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3769 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3770 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3771 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3772 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3773 		mutex_unlock(&adev->grbm_idx_mutex);
3774 
3775 		gfx_v7_0_update_rlc(adev, tmp);
3776 
3777 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3778 	} else {
3779 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3780 
3781 		RREG32(mmCB_CGTT_SCLK_CTRL);
3782 		RREG32(mmCB_CGTT_SCLK_CTRL);
3783 		RREG32(mmCB_CGTT_SCLK_CTRL);
3784 		RREG32(mmCB_CGTT_SCLK_CTRL);
3785 
3786 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3787 	}
3788 
3789 	if (orig != data)
3790 		WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3791 
3792 }
3793 
3794 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3795 {
3796 	u32 data, orig, tmp = 0;
3797 
3798 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3799 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3800 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3801 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3802 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3803 				if (orig != data)
3804 					WREG32(mmCP_MEM_SLP_CNTL, data);
3805 			}
3806 		}
3807 
3808 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3809 		data |= 0x00000001;
3810 		data &= 0xfffffffd;
3811 		if (orig != data)
3812 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3813 
3814 		tmp = gfx_v7_0_halt_rlc(adev);
3815 
3816 		mutex_lock(&adev->grbm_idx_mutex);
3817 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3818 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3819 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3820 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3821 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3822 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3823 		mutex_unlock(&adev->grbm_idx_mutex);
3824 
3825 		gfx_v7_0_update_rlc(adev, tmp);
3826 
3827 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3828 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3829 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3830 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3831 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3832 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3833 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3834 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3835 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3836 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3837 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3838 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3839 			if (orig != data)
3840 				WREG32(mmCGTS_SM_CTRL_REG, data);
3841 		}
3842 	} else {
3843 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3844 		data |= 0x00000003;
3845 		if (orig != data)
3846 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3847 
3848 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3849 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3850 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3851 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3852 		}
3853 
3854 		data = RREG32(mmCP_MEM_SLP_CNTL);
3855 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3856 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3857 			WREG32(mmCP_MEM_SLP_CNTL, data);
3858 		}
3859 
3860 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3861 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3862 		if (orig != data)
3863 			WREG32(mmCGTS_SM_CTRL_REG, data);
3864 
3865 		tmp = gfx_v7_0_halt_rlc(adev);
3866 
3867 		mutex_lock(&adev->grbm_idx_mutex);
3868 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3869 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3870 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3871 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3872 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3873 		mutex_unlock(&adev->grbm_idx_mutex);
3874 
3875 		gfx_v7_0_update_rlc(adev, tmp);
3876 	}
3877 }
3878 
3879 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3880 			       bool enable)
3881 {
3882 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3883 	/* order matters! */
3884 	if (enable) {
3885 		gfx_v7_0_enable_mgcg(adev, true);
3886 		gfx_v7_0_enable_cgcg(adev, true);
3887 	} else {
3888 		gfx_v7_0_enable_cgcg(adev, false);
3889 		gfx_v7_0_enable_mgcg(adev, false);
3890 	}
3891 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3892 }
3893 
3894 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3895 						bool enable)
3896 {
3897 	u32 data, orig;
3898 
3899 	orig = data = RREG32(mmRLC_PG_CNTL);
3900 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3901 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3902 	else
3903 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3904 	if (orig != data)
3905 		WREG32(mmRLC_PG_CNTL, data);
3906 }
3907 
3908 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3909 						bool enable)
3910 {
3911 	u32 data, orig;
3912 
3913 	orig = data = RREG32(mmRLC_PG_CNTL);
3914 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3915 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3916 	else
3917 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3918 	if (orig != data)
3919 		WREG32(mmRLC_PG_CNTL, data);
3920 }
3921 
3922 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3923 {
3924 	u32 data, orig;
3925 
3926 	orig = data = RREG32(mmRLC_PG_CNTL);
3927 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3928 		data &= ~0x8000;
3929 	else
3930 		data |= 0x8000;
3931 	if (orig != data)
3932 		WREG32(mmRLC_PG_CNTL, data);
3933 }
3934 
3935 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3936 {
3937 	u32 data, orig;
3938 
3939 	orig = data = RREG32(mmRLC_PG_CNTL);
3940 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3941 		data &= ~0x2000;
3942 	else
3943 		data |= 0x2000;
3944 	if (orig != data)
3945 		WREG32(mmRLC_PG_CNTL, data);
3946 }
3947 
3948 static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev)
3949 {
3950 	const __le32 *fw_data;
3951 	volatile u32 *dst_ptr;
3952 	int me, i, max_me = 4;
3953 	u32 bo_offset = 0;
3954 	u32 table_offset, table_size;
3955 
3956 	if (adev->asic_type == CHIP_KAVERI)
3957 		max_me = 5;
3958 
3959 	if (adev->gfx.rlc.cp_table_ptr == NULL)
3960 		return;
3961 
3962 	/* write the cp table buffer */
3963 	dst_ptr = adev->gfx.rlc.cp_table_ptr;
3964 	for (me = 0; me < max_me; me++) {
3965 		if (me == 0) {
3966 			const struct gfx_firmware_header_v1_0 *hdr =
3967 				(const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
3968 			fw_data = (const __le32 *)
3969 				(adev->gfx.ce_fw->data +
3970 				 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3971 			table_offset = le32_to_cpu(hdr->jt_offset);
3972 			table_size = le32_to_cpu(hdr->jt_size);
3973 		} else if (me == 1) {
3974 			const struct gfx_firmware_header_v1_0 *hdr =
3975 				(const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
3976 			fw_data = (const __le32 *)
3977 				(adev->gfx.pfp_fw->data +
3978 				 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3979 			table_offset = le32_to_cpu(hdr->jt_offset);
3980 			table_size = le32_to_cpu(hdr->jt_size);
3981 		} else if (me == 2) {
3982 			const struct gfx_firmware_header_v1_0 *hdr =
3983 				(const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
3984 			fw_data = (const __le32 *)
3985 				(adev->gfx.me_fw->data +
3986 				 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3987 			table_offset = le32_to_cpu(hdr->jt_offset);
3988 			table_size = le32_to_cpu(hdr->jt_size);
3989 		} else if (me == 3) {
3990 			const struct gfx_firmware_header_v1_0 *hdr =
3991 				(const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3992 			fw_data = (const __le32 *)
3993 				(adev->gfx.mec_fw->data +
3994 				 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3995 			table_offset = le32_to_cpu(hdr->jt_offset);
3996 			table_size = le32_to_cpu(hdr->jt_size);
3997 		} else {
3998 			const struct gfx_firmware_header_v1_0 *hdr =
3999 				(const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
4000 			fw_data = (const __le32 *)
4001 				(adev->gfx.mec2_fw->data +
4002 				 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
4003 			table_offset = le32_to_cpu(hdr->jt_offset);
4004 			table_size = le32_to_cpu(hdr->jt_size);
4005 		}
4006 
4007 		for (i = 0; i < table_size; i ++) {
4008 			dst_ptr[bo_offset + i] =
4009 				cpu_to_le32(le32_to_cpu(fw_data[table_offset + i]));
4010 		}
4011 
4012 		bo_offset += table_size;
4013 	}
4014 }
4015 
4016 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
4017 				     bool enable)
4018 {
4019 	u32 data, orig;
4020 
4021 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
4022 		orig = data = RREG32(mmRLC_PG_CNTL);
4023 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
4024 		if (orig != data)
4025 			WREG32(mmRLC_PG_CNTL, data);
4026 
4027 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
4028 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
4029 		if (orig != data)
4030 			WREG32(mmRLC_AUTO_PG_CTRL, data);
4031 	} else {
4032 		orig = data = RREG32(mmRLC_PG_CNTL);
4033 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
4034 		if (orig != data)
4035 			WREG32(mmRLC_PG_CNTL, data);
4036 
4037 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
4038 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
4039 		if (orig != data)
4040 			WREG32(mmRLC_AUTO_PG_CTRL, data);
4041 
4042 		data = RREG32(mmDB_RENDER_CONTROL);
4043 	}
4044 }
4045 
4046 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
4047 						 u32 bitmap)
4048 {
4049 	u32 data;
4050 
4051 	if (!bitmap)
4052 		return;
4053 
4054 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4055 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4056 
4057 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
4058 }
4059 
4060 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
4061 {
4062 	u32 data, mask;
4063 
4064 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
4065 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
4066 
4067 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4068 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4069 
4070 	mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_cu_per_sh);
4071 
4072 	return (~data) & mask;
4073 }
4074 
4075 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
4076 {
4077 	u32 tmp;
4078 
4079 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4080 
4081 	tmp = RREG32(mmRLC_MAX_PG_CU);
4082 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
4083 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
4084 	WREG32(mmRLC_MAX_PG_CU, tmp);
4085 }
4086 
4087 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
4088 					    bool enable)
4089 {
4090 	u32 data, orig;
4091 
4092 	orig = data = RREG32(mmRLC_PG_CNTL);
4093 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
4094 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
4095 	else
4096 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
4097 	if (orig != data)
4098 		WREG32(mmRLC_PG_CNTL, data);
4099 }
4100 
4101 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
4102 					     bool enable)
4103 {
4104 	u32 data, orig;
4105 
4106 	orig = data = RREG32(mmRLC_PG_CNTL);
4107 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
4108 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
4109 	else
4110 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
4111 	if (orig != data)
4112 		WREG32(mmRLC_PG_CNTL, data);
4113 }
4114 
4115 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
4116 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
4117 
4118 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
4119 {
4120 	u32 data, orig;
4121 	u32 i;
4122 
4123 	if (adev->gfx.rlc.cs_data) {
4124 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
4125 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
4126 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
4127 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
4128 	} else {
4129 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
4130 		for (i = 0; i < 3; i++)
4131 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
4132 	}
4133 	if (adev->gfx.rlc.reg_list) {
4134 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
4135 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
4136 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
4137 	}
4138 
4139 	orig = data = RREG32(mmRLC_PG_CNTL);
4140 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
4141 	if (orig != data)
4142 		WREG32(mmRLC_PG_CNTL, data);
4143 
4144 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
4145 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4146 
4147 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
4148 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
4149 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4150 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
4151 
4152 	data = 0x10101010;
4153 	WREG32(mmRLC_PG_DELAY, data);
4154 
4155 	data = RREG32(mmRLC_PG_DELAY_2);
4156 	data &= ~0xff;
4157 	data |= 0x3;
4158 	WREG32(mmRLC_PG_DELAY_2, data);
4159 
4160 	data = RREG32(mmRLC_AUTO_PG_CTRL);
4161 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
4162 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
4163 	WREG32(mmRLC_AUTO_PG_CTRL, data);
4164 
4165 }
4166 
4167 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
4168 {
4169 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
4170 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
4171 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
4172 }
4173 
4174 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
4175 {
4176 	u32 count = 0;
4177 	const struct cs_section_def *sect = NULL;
4178 	const struct cs_extent_def *ext = NULL;
4179 
4180 	if (adev->gfx.rlc.cs_data == NULL)
4181 		return 0;
4182 
4183 	/* begin clear state */
4184 	count += 2;
4185 	/* context control state */
4186 	count += 3;
4187 
4188 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
4189 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4190 			if (sect->id == SECT_CONTEXT)
4191 				count += 2 + ext->reg_count;
4192 			else
4193 				return 0;
4194 		}
4195 	}
4196 	/* pa_sc_raster_config/pa_sc_raster_config1 */
4197 	count += 4;
4198 	/* end clear state */
4199 	count += 2;
4200 	/* clear state */
4201 	count += 2;
4202 
4203 	return count;
4204 }
4205 
4206 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
4207 				    volatile u32 *buffer)
4208 {
4209 	u32 count = 0, i;
4210 	const struct cs_section_def *sect = NULL;
4211 	const struct cs_extent_def *ext = NULL;
4212 
4213 	if (adev->gfx.rlc.cs_data == NULL)
4214 		return;
4215 	if (buffer == NULL)
4216 		return;
4217 
4218 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4219 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4220 
4221 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4222 	buffer[count++] = cpu_to_le32(0x80000000);
4223 	buffer[count++] = cpu_to_le32(0x80000000);
4224 
4225 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
4226 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4227 			if (sect->id == SECT_CONTEXT) {
4228 				buffer[count++] =
4229 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
4230 				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4231 				for (i = 0; i < ext->reg_count; i++)
4232 					buffer[count++] = cpu_to_le32(ext->extent[i]);
4233 			} else {
4234 				return;
4235 			}
4236 		}
4237 	}
4238 
4239 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4240 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4241 	switch (adev->asic_type) {
4242 	case CHIP_BONAIRE:
4243 		buffer[count++] = cpu_to_le32(0x16000012);
4244 		buffer[count++] = cpu_to_le32(0x00000000);
4245 		break;
4246 	case CHIP_KAVERI:
4247 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4248 		buffer[count++] = cpu_to_le32(0x00000000);
4249 		break;
4250 	case CHIP_KABINI:
4251 	case CHIP_MULLINS:
4252 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4253 		buffer[count++] = cpu_to_le32(0x00000000);
4254 		break;
4255 	case CHIP_HAWAII:
4256 		buffer[count++] = cpu_to_le32(0x3a00161a);
4257 		buffer[count++] = cpu_to_le32(0x0000002e);
4258 		break;
4259 	default:
4260 		buffer[count++] = cpu_to_le32(0x00000000);
4261 		buffer[count++] = cpu_to_le32(0x00000000);
4262 		break;
4263 	}
4264 
4265 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4266 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
4267 
4268 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
4269 	buffer[count++] = cpu_to_le32(0);
4270 }
4271 
4272 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
4273 {
4274 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4275 			      AMD_PG_SUPPORT_GFX_SMG |
4276 			      AMD_PG_SUPPORT_GFX_DMG |
4277 			      AMD_PG_SUPPORT_CP |
4278 			      AMD_PG_SUPPORT_GDS |
4279 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4280 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
4281 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
4282 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4283 			gfx_v7_0_init_gfx_cgpg(adev);
4284 			gfx_v7_0_enable_cp_pg(adev, true);
4285 			gfx_v7_0_enable_gds_pg(adev, true);
4286 		}
4287 		gfx_v7_0_init_ao_cu_mask(adev);
4288 		gfx_v7_0_update_gfx_pg(adev, true);
4289 	}
4290 }
4291 
4292 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4293 {
4294 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4295 			      AMD_PG_SUPPORT_GFX_SMG |
4296 			      AMD_PG_SUPPORT_GFX_DMG |
4297 			      AMD_PG_SUPPORT_CP |
4298 			      AMD_PG_SUPPORT_GDS |
4299 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4300 		gfx_v7_0_update_gfx_pg(adev, false);
4301 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4302 			gfx_v7_0_enable_cp_pg(adev, false);
4303 			gfx_v7_0_enable_gds_pg(adev, false);
4304 		}
4305 	}
4306 }
4307 
4308 /**
4309  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4310  *
4311  * @adev: amdgpu_device pointer
4312  *
4313  * Fetches a GPU clock counter snapshot (SI).
4314  * Returns the 64 bit clock counter snapshot.
4315  */
4316 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4317 {
4318 	uint64_t clock;
4319 
4320 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4321 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4322 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4323 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4324 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4325 	return clock;
4326 }
4327 
4328 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4329 					  uint32_t vmid,
4330 					  uint32_t gds_base, uint32_t gds_size,
4331 					  uint32_t gws_base, uint32_t gws_size,
4332 					  uint32_t oa_base, uint32_t oa_size)
4333 {
4334 	gds_base = gds_base >> AMDGPU_GDS_SHIFT;
4335 	gds_size = gds_size >> AMDGPU_GDS_SHIFT;
4336 
4337 	gws_base = gws_base >> AMDGPU_GWS_SHIFT;
4338 	gws_size = gws_size >> AMDGPU_GWS_SHIFT;
4339 
4340 	oa_base = oa_base >> AMDGPU_OA_SHIFT;
4341 	oa_size = oa_size >> AMDGPU_OA_SHIFT;
4342 
4343 	/* GDS Base */
4344 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4345 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4346 				WRITE_DATA_DST_SEL(0)));
4347 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4348 	amdgpu_ring_write(ring, 0);
4349 	amdgpu_ring_write(ring, gds_base);
4350 
4351 	/* GDS Size */
4352 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4353 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4354 				WRITE_DATA_DST_SEL(0)));
4355 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
4356 	amdgpu_ring_write(ring, 0);
4357 	amdgpu_ring_write(ring, gds_size);
4358 
4359 	/* GWS */
4360 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4361 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4362 				WRITE_DATA_DST_SEL(0)));
4363 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
4364 	amdgpu_ring_write(ring, 0);
4365 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4366 
4367 	/* OA */
4368 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4369 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4370 				WRITE_DATA_DST_SEL(0)));
4371 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
4372 	amdgpu_ring_write(ring, 0);
4373 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4374 }
4375 
4376 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4377 {
4378 	WREG32(mmSQ_IND_INDEX,
4379 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4380 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4381 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4382 		(SQ_IND_INDEX__FORCE_READ_MASK));
4383 	return RREG32(mmSQ_IND_DATA);
4384 }
4385 
4386 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4387 			   uint32_t wave, uint32_t thread,
4388 			   uint32_t regno, uint32_t num, uint32_t *out)
4389 {
4390 	WREG32(mmSQ_IND_INDEX,
4391 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4392 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4393 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4394 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4395 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4396 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4397 	while (num--)
4398 		*(out++) = RREG32(mmSQ_IND_DATA);
4399 }
4400 
4401 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4402 {
4403 	/* type 0 wave data */
4404 	dst[(*no_fields)++] = 0;
4405 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4406 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4407 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4408 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4409 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4410 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4411 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4412 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4413 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4414 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4415 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4416 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4417 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4418 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4419 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4420 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4421 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4422 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4423 }
4424 
4425 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
4426 				     uint32_t wave, uint32_t start,
4427 				     uint32_t size, uint32_t *dst)
4428 {
4429 	wave_read_regs(
4430 		adev, simd, wave, 0,
4431 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4432 }
4433 
4434 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4435 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4436 	.select_se_sh = &gfx_v7_0_select_se_sh,
4437 	.read_wave_data = &gfx_v7_0_read_wave_data,
4438 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4439 };
4440 
4441 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4442 	.enter_safe_mode = gfx_v7_0_enter_rlc_safe_mode,
4443 	.exit_safe_mode = gfx_v7_0_exit_rlc_safe_mode
4444 };
4445 
4446 static int gfx_v7_0_early_init(void *handle)
4447 {
4448 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4449 
4450 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4451 	adev->gfx.num_compute_rings = GFX7_NUM_COMPUTE_RINGS;
4452 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4453 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4454 	gfx_v7_0_set_ring_funcs(adev);
4455 	gfx_v7_0_set_irq_funcs(adev);
4456 	gfx_v7_0_set_gds_init(adev);
4457 
4458 	return 0;
4459 }
4460 
4461 static int gfx_v7_0_late_init(void *handle)
4462 {
4463 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4464 	int r;
4465 
4466 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4467 	if (r)
4468 		return r;
4469 
4470 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4471 	if (r)
4472 		return r;
4473 
4474 	return 0;
4475 }
4476 
4477 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4478 {
4479 	u32 gb_addr_config;
4480 	u32 mc_shared_chmap, mc_arb_ramcfg;
4481 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4482 	u32 tmp;
4483 
4484 	switch (adev->asic_type) {
4485 	case CHIP_BONAIRE:
4486 		adev->gfx.config.max_shader_engines = 2;
4487 		adev->gfx.config.max_tile_pipes = 4;
4488 		adev->gfx.config.max_cu_per_sh = 7;
4489 		adev->gfx.config.max_sh_per_se = 1;
4490 		adev->gfx.config.max_backends_per_se = 2;
4491 		adev->gfx.config.max_texture_channel_caches = 4;
4492 		adev->gfx.config.max_gprs = 256;
4493 		adev->gfx.config.max_gs_threads = 32;
4494 		adev->gfx.config.max_hw_contexts = 8;
4495 
4496 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4497 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4498 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4499 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4500 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4501 		break;
4502 	case CHIP_HAWAII:
4503 		adev->gfx.config.max_shader_engines = 4;
4504 		adev->gfx.config.max_tile_pipes = 16;
4505 		adev->gfx.config.max_cu_per_sh = 11;
4506 		adev->gfx.config.max_sh_per_se = 1;
4507 		adev->gfx.config.max_backends_per_se = 4;
4508 		adev->gfx.config.max_texture_channel_caches = 16;
4509 		adev->gfx.config.max_gprs = 256;
4510 		adev->gfx.config.max_gs_threads = 32;
4511 		adev->gfx.config.max_hw_contexts = 8;
4512 
4513 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4514 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4515 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4516 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4517 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4518 		break;
4519 	case CHIP_KAVERI:
4520 		adev->gfx.config.max_shader_engines = 1;
4521 		adev->gfx.config.max_tile_pipes = 4;
4522 		if ((adev->pdev->device == 0x1304) ||
4523 		    (adev->pdev->device == 0x1305) ||
4524 		    (adev->pdev->device == 0x130C) ||
4525 		    (adev->pdev->device == 0x130F) ||
4526 		    (adev->pdev->device == 0x1310) ||
4527 		    (adev->pdev->device == 0x1311) ||
4528 		    (adev->pdev->device == 0x131C)) {
4529 			adev->gfx.config.max_cu_per_sh = 8;
4530 			adev->gfx.config.max_backends_per_se = 2;
4531 		} else if ((adev->pdev->device == 0x1309) ||
4532 			   (adev->pdev->device == 0x130A) ||
4533 			   (adev->pdev->device == 0x130D) ||
4534 			   (adev->pdev->device == 0x1313) ||
4535 			   (adev->pdev->device == 0x131D)) {
4536 			adev->gfx.config.max_cu_per_sh = 6;
4537 			adev->gfx.config.max_backends_per_se = 2;
4538 		} else if ((adev->pdev->device == 0x1306) ||
4539 			   (adev->pdev->device == 0x1307) ||
4540 			   (adev->pdev->device == 0x130B) ||
4541 			   (adev->pdev->device == 0x130E) ||
4542 			   (adev->pdev->device == 0x1315) ||
4543 			   (adev->pdev->device == 0x131B)) {
4544 			adev->gfx.config.max_cu_per_sh = 4;
4545 			adev->gfx.config.max_backends_per_se = 1;
4546 		} else {
4547 			adev->gfx.config.max_cu_per_sh = 3;
4548 			adev->gfx.config.max_backends_per_se = 1;
4549 		}
4550 		adev->gfx.config.max_sh_per_se = 1;
4551 		adev->gfx.config.max_texture_channel_caches = 4;
4552 		adev->gfx.config.max_gprs = 256;
4553 		adev->gfx.config.max_gs_threads = 16;
4554 		adev->gfx.config.max_hw_contexts = 8;
4555 
4556 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4557 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4558 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4559 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4560 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4561 		break;
4562 	case CHIP_KABINI:
4563 	case CHIP_MULLINS:
4564 	default:
4565 		adev->gfx.config.max_shader_engines = 1;
4566 		adev->gfx.config.max_tile_pipes = 2;
4567 		adev->gfx.config.max_cu_per_sh = 2;
4568 		adev->gfx.config.max_sh_per_se = 1;
4569 		adev->gfx.config.max_backends_per_se = 1;
4570 		adev->gfx.config.max_texture_channel_caches = 2;
4571 		adev->gfx.config.max_gprs = 256;
4572 		adev->gfx.config.max_gs_threads = 16;
4573 		adev->gfx.config.max_hw_contexts = 8;
4574 
4575 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4576 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4577 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4578 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4579 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4580 		break;
4581 	}
4582 
4583 	mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP);
4584 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4585 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4586 
4587 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4588 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4589 	if (adev->flags & AMD_IS_APU) {
4590 		/* Get memory bank mapping mode. */
4591 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4592 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4593 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4594 
4595 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4596 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4597 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4598 
4599 		/* Validate settings in case only one DIMM installed. */
4600 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4601 			dimm00_addr_map = 0;
4602 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4603 			dimm01_addr_map = 0;
4604 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4605 			dimm10_addr_map = 0;
4606 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4607 			dimm11_addr_map = 0;
4608 
4609 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4610 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4611 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4612 			adev->gfx.config.mem_row_size_in_kb = 2;
4613 		else
4614 			adev->gfx.config.mem_row_size_in_kb = 1;
4615 	} else {
4616 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4617 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4618 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4619 			adev->gfx.config.mem_row_size_in_kb = 4;
4620 	}
4621 	/* XXX use MC settings? */
4622 	adev->gfx.config.shader_engine_tile_size = 32;
4623 	adev->gfx.config.num_gpus = 1;
4624 	adev->gfx.config.multi_gpu_tile_size = 64;
4625 
4626 	/* fix up row size */
4627 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4628 	switch (adev->gfx.config.mem_row_size_in_kb) {
4629 	case 1:
4630 	default:
4631 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4632 		break;
4633 	case 2:
4634 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4635 		break;
4636 	case 4:
4637 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4638 		break;
4639 	}
4640 	adev->gfx.config.gb_addr_config = gb_addr_config;
4641 }
4642 
4643 static int gfx_v7_0_sw_init(void *handle)
4644 {
4645 	struct amdgpu_ring *ring;
4646 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4647 	int i, r;
4648 
4649 	/* EOP Event */
4650 	r = amdgpu_irq_add_id(adev, 181, &adev->gfx.eop_irq);
4651 	if (r)
4652 		return r;
4653 
4654 	/* Privileged reg */
4655 	r = amdgpu_irq_add_id(adev, 184, &adev->gfx.priv_reg_irq);
4656 	if (r)
4657 		return r;
4658 
4659 	/* Privileged inst */
4660 	r = amdgpu_irq_add_id(adev, 185, &adev->gfx.priv_inst_irq);
4661 	if (r)
4662 		return r;
4663 
4664 	gfx_v7_0_scratch_init(adev);
4665 
4666 	r = gfx_v7_0_init_microcode(adev);
4667 	if (r) {
4668 		DRM_ERROR("Failed to load gfx firmware!\n");
4669 		return r;
4670 	}
4671 
4672 	r = gfx_v7_0_rlc_init(adev);
4673 	if (r) {
4674 		DRM_ERROR("Failed to init rlc BOs!\n");
4675 		return r;
4676 	}
4677 
4678 	/* allocate mec buffers */
4679 	r = gfx_v7_0_mec_init(adev);
4680 	if (r) {
4681 		DRM_ERROR("Failed to init MEC BOs!\n");
4682 		return r;
4683 	}
4684 
4685 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4686 		ring = &adev->gfx.gfx_ring[i];
4687 		ring->ring_obj = NULL;
4688 		sprintf(ring->name, "gfx");
4689 		r = amdgpu_ring_init(adev, ring, 1024,
4690 				     &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_EOP);
4691 		if (r)
4692 			return r;
4693 	}
4694 
4695 	/* set up the compute queues */
4696 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4697 		unsigned irq_type;
4698 
4699 		/* max 32 queues per MEC */
4700 		if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) {
4701 			DRM_ERROR("Too many (%d) compute rings!\n", i);
4702 			break;
4703 		}
4704 		ring = &adev->gfx.compute_ring[i];
4705 		ring->ring_obj = NULL;
4706 		ring->use_doorbell = true;
4707 		ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + i;
4708 		ring->me = 1; /* first MEC */
4709 		ring->pipe = i / 8;
4710 		ring->queue = i % 8;
4711 		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4712 		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe;
4713 		/* type-2 packets are deprecated on MEC, use type-3 instead */
4714 		r = amdgpu_ring_init(adev, ring, 1024,
4715 				     &adev->gfx.eop_irq, irq_type);
4716 		if (r)
4717 			return r;
4718 	}
4719 
4720 	/* reserve GDS, GWS and OA resource for gfx */
4721 	r = amdgpu_bo_create_kernel(adev, adev->gds.mem.gfx_partition_size,
4722 				    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GDS,
4723 				    &adev->gds.gds_gfx_bo, NULL, NULL);
4724 	if (r)
4725 		return r;
4726 
4727 	r = amdgpu_bo_create_kernel(adev, adev->gds.gws.gfx_partition_size,
4728 				    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GWS,
4729 				    &adev->gds.gws_gfx_bo, NULL, NULL);
4730 	if (r)
4731 		return r;
4732 
4733 	r = amdgpu_bo_create_kernel(adev, adev->gds.oa.gfx_partition_size,
4734 				    PAGE_SIZE, AMDGPU_GEM_DOMAIN_OA,
4735 				    &adev->gds.oa_gfx_bo, NULL, NULL);
4736 	if (r)
4737 		return r;
4738 
4739 	adev->gfx.ce_ram_size = 0x8000;
4740 
4741 	gfx_v7_0_gpu_early_init(adev);
4742 
4743 	return r;
4744 }
4745 
4746 static int gfx_v7_0_sw_fini(void *handle)
4747 {
4748 	int i;
4749 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4750 
4751 	amdgpu_bo_free_kernel(&adev->gds.oa_gfx_bo, NULL, NULL);
4752 	amdgpu_bo_free_kernel(&adev->gds.gws_gfx_bo, NULL, NULL);
4753 	amdgpu_bo_free_kernel(&adev->gds.gds_gfx_bo, NULL, NULL);
4754 
4755 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4756 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4757 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
4758 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4759 
4760 	gfx_v7_0_cp_compute_fini(adev);
4761 	gfx_v7_0_rlc_fini(adev);
4762 	gfx_v7_0_mec_fini(adev);
4763 	gfx_v7_0_free_microcode(adev);
4764 
4765 	return 0;
4766 }
4767 
4768 static int gfx_v7_0_hw_init(void *handle)
4769 {
4770 	int r;
4771 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4772 
4773 	gfx_v7_0_gpu_init(adev);
4774 
4775 	/* init rlc */
4776 	r = gfx_v7_0_rlc_resume(adev);
4777 	if (r)
4778 		return r;
4779 
4780 	r = gfx_v7_0_cp_resume(adev);
4781 	if (r)
4782 		return r;
4783 
4784 	return r;
4785 }
4786 
4787 static int gfx_v7_0_hw_fini(void *handle)
4788 {
4789 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4790 
4791 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4792 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4793 	gfx_v7_0_cp_enable(adev, false);
4794 	gfx_v7_0_rlc_stop(adev);
4795 	gfx_v7_0_fini_pg(adev);
4796 
4797 	return 0;
4798 }
4799 
4800 static int gfx_v7_0_suspend(void *handle)
4801 {
4802 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4803 
4804 	return gfx_v7_0_hw_fini(adev);
4805 }
4806 
4807 static int gfx_v7_0_resume(void *handle)
4808 {
4809 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4810 
4811 	return gfx_v7_0_hw_init(adev);
4812 }
4813 
4814 static bool gfx_v7_0_is_idle(void *handle)
4815 {
4816 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4817 
4818 	if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK)
4819 		return false;
4820 	else
4821 		return true;
4822 }
4823 
4824 static int gfx_v7_0_wait_for_idle(void *handle)
4825 {
4826 	unsigned i;
4827 	u32 tmp;
4828 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4829 
4830 	for (i = 0; i < adev->usec_timeout; i++) {
4831 		/* read MC_STATUS */
4832 		tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4833 
4834 		if (!tmp)
4835 			return 0;
4836 		udelay(1);
4837 	}
4838 	return -ETIMEDOUT;
4839 }
4840 
4841 static int gfx_v7_0_soft_reset(void *handle)
4842 {
4843 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4844 	u32 tmp;
4845 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4846 
4847 	/* GRBM_STATUS */
4848 	tmp = RREG32(mmGRBM_STATUS);
4849 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4850 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4851 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4852 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4853 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4854 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK))
4855 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK |
4856 			GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK;
4857 
4858 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4859 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK;
4860 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4861 	}
4862 
4863 	/* GRBM_STATUS2 */
4864 	tmp = RREG32(mmGRBM_STATUS2);
4865 	if (tmp & GRBM_STATUS2__RLC_BUSY_MASK)
4866 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
4867 
4868 	/* SRBM_STATUS */
4869 	tmp = RREG32(mmSRBM_STATUS);
4870 	if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK)
4871 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4872 
4873 	if (grbm_soft_reset || srbm_soft_reset) {
4874 		/* disable CG/PG */
4875 		gfx_v7_0_fini_pg(adev);
4876 		gfx_v7_0_update_cg(adev, false);
4877 
4878 		/* stop the rlc */
4879 		gfx_v7_0_rlc_stop(adev);
4880 
4881 		/* Disable GFX parsing/prefetching */
4882 		WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK);
4883 
4884 		/* Disable MEC parsing/prefetching */
4885 		WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK);
4886 
4887 		if (grbm_soft_reset) {
4888 			tmp = RREG32(mmGRBM_SOFT_RESET);
4889 			tmp |= grbm_soft_reset;
4890 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4891 			WREG32(mmGRBM_SOFT_RESET, tmp);
4892 			tmp = RREG32(mmGRBM_SOFT_RESET);
4893 
4894 			udelay(50);
4895 
4896 			tmp &= ~grbm_soft_reset;
4897 			WREG32(mmGRBM_SOFT_RESET, tmp);
4898 			tmp = RREG32(mmGRBM_SOFT_RESET);
4899 		}
4900 
4901 		if (srbm_soft_reset) {
4902 			tmp = RREG32(mmSRBM_SOFT_RESET);
4903 			tmp |= srbm_soft_reset;
4904 			dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4905 			WREG32(mmSRBM_SOFT_RESET, tmp);
4906 			tmp = RREG32(mmSRBM_SOFT_RESET);
4907 
4908 			udelay(50);
4909 
4910 			tmp &= ~srbm_soft_reset;
4911 			WREG32(mmSRBM_SOFT_RESET, tmp);
4912 			tmp = RREG32(mmSRBM_SOFT_RESET);
4913 		}
4914 		/* Wait a little for things to settle down */
4915 		udelay(50);
4916 	}
4917 	return 0;
4918 }
4919 
4920 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4921 						 enum amdgpu_interrupt_state state)
4922 {
4923 	u32 cp_int_cntl;
4924 
4925 	switch (state) {
4926 	case AMDGPU_IRQ_STATE_DISABLE:
4927 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4928 		cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4929 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4930 		break;
4931 	case AMDGPU_IRQ_STATE_ENABLE:
4932 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4933 		cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4934 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4935 		break;
4936 	default:
4937 		break;
4938 	}
4939 }
4940 
4941 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4942 						     int me, int pipe,
4943 						     enum amdgpu_interrupt_state state)
4944 {
4945 	u32 mec_int_cntl, mec_int_cntl_reg;
4946 
4947 	/*
4948 	 * amdgpu controls only pipe 0 of MEC1. That's why this function only
4949 	 * handles the setting of interrupts for this specific pipe. All other
4950 	 * pipes' interrupts are set by amdkfd.
4951 	 */
4952 
4953 	if (me == 1) {
4954 		switch (pipe) {
4955 		case 0:
4956 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4957 			break;
4958 		default:
4959 			DRM_DEBUG("invalid pipe %d\n", pipe);
4960 			return;
4961 		}
4962 	} else {
4963 		DRM_DEBUG("invalid me %d\n", me);
4964 		return;
4965 	}
4966 
4967 	switch (state) {
4968 	case AMDGPU_IRQ_STATE_DISABLE:
4969 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4970 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4971 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4972 		break;
4973 	case AMDGPU_IRQ_STATE_ENABLE:
4974 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4975 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4976 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4977 		break;
4978 	default:
4979 		break;
4980 	}
4981 }
4982 
4983 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4984 					     struct amdgpu_irq_src *src,
4985 					     unsigned type,
4986 					     enum amdgpu_interrupt_state state)
4987 {
4988 	u32 cp_int_cntl;
4989 
4990 	switch (state) {
4991 	case AMDGPU_IRQ_STATE_DISABLE:
4992 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4993 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4994 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4995 		break;
4996 	case AMDGPU_IRQ_STATE_ENABLE:
4997 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4998 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4999 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
5000 		break;
5001 	default:
5002 		break;
5003 	}
5004 
5005 	return 0;
5006 }
5007 
5008 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5009 					      struct amdgpu_irq_src *src,
5010 					      unsigned type,
5011 					      enum amdgpu_interrupt_state state)
5012 {
5013 	u32 cp_int_cntl;
5014 
5015 	switch (state) {
5016 	case AMDGPU_IRQ_STATE_DISABLE:
5017 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
5018 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
5019 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
5020 		break;
5021 	case AMDGPU_IRQ_STATE_ENABLE:
5022 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
5023 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
5024 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
5025 		break;
5026 	default:
5027 		break;
5028 	}
5029 
5030 	return 0;
5031 }
5032 
5033 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5034 					    struct amdgpu_irq_src *src,
5035 					    unsigned type,
5036 					    enum amdgpu_interrupt_state state)
5037 {
5038 	switch (type) {
5039 	case AMDGPU_CP_IRQ_GFX_EOP:
5040 		gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
5041 		break;
5042 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5043 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5044 		break;
5045 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5046 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5047 		break;
5048 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5049 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5050 		break;
5051 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5052 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5053 		break;
5054 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
5055 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
5056 		break;
5057 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
5058 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
5059 		break;
5060 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
5061 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
5062 		break;
5063 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
5064 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
5065 		break;
5066 	default:
5067 		break;
5068 	}
5069 	return 0;
5070 }
5071 
5072 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
5073 			    struct amdgpu_irq_src *source,
5074 			    struct amdgpu_iv_entry *entry)
5075 {
5076 	u8 me_id, pipe_id;
5077 	struct amdgpu_ring *ring;
5078 	int i;
5079 
5080 	DRM_DEBUG("IH: CP EOP\n");
5081 	me_id = (entry->ring_id & 0x0c) >> 2;
5082 	pipe_id = (entry->ring_id & 0x03) >> 0;
5083 	switch (me_id) {
5084 	case 0:
5085 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5086 		break;
5087 	case 1:
5088 	case 2:
5089 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5090 			ring = &adev->gfx.compute_ring[i];
5091 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
5092 				amdgpu_fence_process(ring);
5093 		}
5094 		break;
5095 	}
5096 	return 0;
5097 }
5098 
5099 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
5100 				 struct amdgpu_irq_src *source,
5101 				 struct amdgpu_iv_entry *entry)
5102 {
5103 	DRM_ERROR("Illegal register access in command stream\n");
5104 	schedule_work(&adev->reset_work);
5105 	return 0;
5106 }
5107 
5108 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
5109 				  struct amdgpu_irq_src *source,
5110 				  struct amdgpu_iv_entry *entry)
5111 {
5112 	DRM_ERROR("Illegal instruction in command stream\n");
5113 	// XXX soft reset the gfx block only
5114 	schedule_work(&adev->reset_work);
5115 	return 0;
5116 }
5117 
5118 static int gfx_v7_0_set_clockgating_state(void *handle,
5119 					  enum amd_clockgating_state state)
5120 {
5121 	bool gate = false;
5122 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5123 
5124 	if (state == AMD_CG_STATE_GATE)
5125 		gate = true;
5126 
5127 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
5128 	/* order matters! */
5129 	if (gate) {
5130 		gfx_v7_0_enable_mgcg(adev, true);
5131 		gfx_v7_0_enable_cgcg(adev, true);
5132 	} else {
5133 		gfx_v7_0_enable_cgcg(adev, false);
5134 		gfx_v7_0_enable_mgcg(adev, false);
5135 	}
5136 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
5137 
5138 	return 0;
5139 }
5140 
5141 static int gfx_v7_0_set_powergating_state(void *handle,
5142 					  enum amd_powergating_state state)
5143 {
5144 	bool gate = false;
5145 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5146 
5147 	if (state == AMD_PG_STATE_GATE)
5148 		gate = true;
5149 
5150 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
5151 			      AMD_PG_SUPPORT_GFX_SMG |
5152 			      AMD_PG_SUPPORT_GFX_DMG |
5153 			      AMD_PG_SUPPORT_CP |
5154 			      AMD_PG_SUPPORT_GDS |
5155 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
5156 		gfx_v7_0_update_gfx_pg(adev, gate);
5157 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
5158 			gfx_v7_0_enable_cp_pg(adev, gate);
5159 			gfx_v7_0_enable_gds_pg(adev, gate);
5160 		}
5161 	}
5162 
5163 	return 0;
5164 }
5165 
5166 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
5167 	.name = "gfx_v7_0",
5168 	.early_init = gfx_v7_0_early_init,
5169 	.late_init = gfx_v7_0_late_init,
5170 	.sw_init = gfx_v7_0_sw_init,
5171 	.sw_fini = gfx_v7_0_sw_fini,
5172 	.hw_init = gfx_v7_0_hw_init,
5173 	.hw_fini = gfx_v7_0_hw_fini,
5174 	.suspend = gfx_v7_0_suspend,
5175 	.resume = gfx_v7_0_resume,
5176 	.is_idle = gfx_v7_0_is_idle,
5177 	.wait_for_idle = gfx_v7_0_wait_for_idle,
5178 	.soft_reset = gfx_v7_0_soft_reset,
5179 	.set_clockgating_state = gfx_v7_0_set_clockgating_state,
5180 	.set_powergating_state = gfx_v7_0_set_powergating_state,
5181 };
5182 
5183 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
5184 	.type = AMDGPU_RING_TYPE_GFX,
5185 	.align_mask = 0xff,
5186 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5187 	.get_rptr = gfx_v7_0_ring_get_rptr,
5188 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
5189 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
5190 	.emit_frame_size =
5191 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5192 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5193 		5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
5194 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
5195 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
5196 		17 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
5197 		3 + 4, /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
5198 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
5199 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
5200 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
5201 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5202 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5203 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5204 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5205 	.emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate,
5206 	.test_ring = gfx_v7_0_ring_test_ring,
5207 	.test_ib = gfx_v7_0_ring_test_ib,
5208 	.insert_nop = amdgpu_ring_insert_nop,
5209 	.pad_ib = amdgpu_ring_generic_pad_ib,
5210 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
5211 };
5212 
5213 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5214 	.type = AMDGPU_RING_TYPE_COMPUTE,
5215 	.align_mask = 0xff,
5216 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5217 	.get_rptr = gfx_v7_0_ring_get_rptr,
5218 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5219 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5220 	.emit_frame_size =
5221 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5222 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5223 		5 + /* gfx_v7_0_ring_emit_hdp_invalidate */
5224 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5225 		17 + /* gfx_v7_0_ring_emit_vm_flush */
5226 		7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5227 	.emit_ib_size =	4, /* gfx_v7_0_ring_emit_ib_compute */
5228 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5229 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5230 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5231 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5232 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5233 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5234 	.emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate,
5235 	.test_ring = gfx_v7_0_ring_test_ring,
5236 	.test_ib = gfx_v7_0_ring_test_ib,
5237 	.insert_nop = amdgpu_ring_insert_nop,
5238 	.pad_ib = amdgpu_ring_generic_pad_ib,
5239 };
5240 
5241 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5242 {
5243 	int i;
5244 
5245 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5246 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5247 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5248 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5249 }
5250 
5251 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5252 	.set = gfx_v7_0_set_eop_interrupt_state,
5253 	.process = gfx_v7_0_eop_irq,
5254 };
5255 
5256 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5257 	.set = gfx_v7_0_set_priv_reg_fault_state,
5258 	.process = gfx_v7_0_priv_reg_irq,
5259 };
5260 
5261 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5262 	.set = gfx_v7_0_set_priv_inst_fault_state,
5263 	.process = gfx_v7_0_priv_inst_irq,
5264 };
5265 
5266 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5267 {
5268 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5269 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5270 
5271 	adev->gfx.priv_reg_irq.num_types = 1;
5272 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5273 
5274 	adev->gfx.priv_inst_irq.num_types = 1;
5275 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5276 }
5277 
5278 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5279 {
5280 	/* init asci gds info */
5281 	adev->gds.mem.total_size = RREG32(mmGDS_VMID0_SIZE);
5282 	adev->gds.gws.total_size = 64;
5283 	adev->gds.oa.total_size = 16;
5284 
5285 	if (adev->gds.mem.total_size == 64 * 1024) {
5286 		adev->gds.mem.gfx_partition_size = 4096;
5287 		adev->gds.mem.cs_partition_size = 4096;
5288 
5289 		adev->gds.gws.gfx_partition_size = 4;
5290 		adev->gds.gws.cs_partition_size = 4;
5291 
5292 		adev->gds.oa.gfx_partition_size = 4;
5293 		adev->gds.oa.cs_partition_size = 1;
5294 	} else {
5295 		adev->gds.mem.gfx_partition_size = 1024;
5296 		adev->gds.mem.cs_partition_size = 1024;
5297 
5298 		adev->gds.gws.gfx_partition_size = 16;
5299 		adev->gds.gws.cs_partition_size = 16;
5300 
5301 		adev->gds.oa.gfx_partition_size = 4;
5302 		adev->gds.oa.cs_partition_size = 4;
5303 	}
5304 }
5305 
5306 
5307 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5308 {
5309 	int i, j, k, counter, active_cu_number = 0;
5310 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5311 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5312 	unsigned disable_masks[4 * 2];
5313 
5314 	memset(cu_info, 0, sizeof(*cu_info));
5315 
5316 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5317 
5318 	mutex_lock(&adev->grbm_idx_mutex);
5319 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5320 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5321 			mask = 1;
5322 			ao_bitmap = 0;
5323 			counter = 0;
5324 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
5325 			if (i < 4 && j < 2)
5326 				gfx_v7_0_set_user_cu_inactive_bitmap(
5327 					adev, disable_masks[i * 2 + j]);
5328 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5329 			cu_info->bitmap[i][j] = bitmap;
5330 
5331 			for (k = 0; k < 16; k ++) {
5332 				if (bitmap & mask) {
5333 					if (counter < 2)
5334 						ao_bitmap |= mask;
5335 					counter ++;
5336 				}
5337 				mask <<= 1;
5338 			}
5339 			active_cu_number += counter;
5340 			ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5341 		}
5342 	}
5343 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5344 	mutex_unlock(&adev->grbm_idx_mutex);
5345 
5346 	cu_info->number = active_cu_number;
5347 	cu_info->ao_cu_mask = ao_cu_mask;
5348 }
5349 
5350 const struct amdgpu_ip_block_version gfx_v7_0_ip_block =
5351 {
5352 	.type = AMD_IP_BLOCK_TYPE_GFX,
5353 	.major = 7,
5354 	.minor = 0,
5355 	.rev = 0,
5356 	.funcs = &gfx_v7_0_ip_funcs,
5357 };
5358 
5359 const struct amdgpu_ip_block_version gfx_v7_1_ip_block =
5360 {
5361 	.type = AMD_IP_BLOCK_TYPE_GFX,
5362 	.major = 7,
5363 	.minor = 1,
5364 	.rev = 0,
5365 	.funcs = &gfx_v7_0_ip_funcs,
5366 };
5367 
5368 const struct amdgpu_ip_block_version gfx_v7_2_ip_block =
5369 {
5370 	.type = AMD_IP_BLOCK_TYPE_GFX,
5371 	.major = 7,
5372 	.minor = 2,
5373 	.rev = 0,
5374 	.funcs = &gfx_v7_0_ip_funcs,
5375 };
5376 
5377 const struct amdgpu_ip_block_version gfx_v7_3_ip_block =
5378 {
5379 	.type = AMD_IP_BLOCK_TYPE_GFX,
5380 	.major = 7,
5381 	.minor = 3,
5382 	.rev = 0,
5383 	.funcs = &gfx_v7_0_ip_funcs,
5384 };
5385