xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision f71a261a)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <linux/module.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_ih.h"
29 #include "amdgpu_gfx.h"
30 #include "cikd.h"
31 #include "cik.h"
32 #include "cik_structs.h"
33 #include "atom.h"
34 #include "amdgpu_ucode.h"
35 #include "clearstate_ci.h"
36 
37 #include "dce/dce_8_0_d.h"
38 #include "dce/dce_8_0_sh_mask.h"
39 
40 #include "bif/bif_4_1_d.h"
41 #include "bif/bif_4_1_sh_mask.h"
42 
43 #include "gca/gfx_7_0_d.h"
44 #include "gca/gfx_7_2_enum.h"
45 #include "gca/gfx_7_2_sh_mask.h"
46 
47 #include "gmc/gmc_7_0_d.h"
48 #include "gmc/gmc_7_0_sh_mask.h"
49 
50 #include "oss/oss_2_0_d.h"
51 #include "oss/oss_2_0_sh_mask.h"
52 
53 #define NUM_SIMD_PER_CU 0x4 /* missing from the gfx_7 IP headers */
54 
55 #define GFX7_NUM_GFX_RINGS     1
56 #define GFX7_MEC_HPD_SIZE      2048
57 
58 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev);
59 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev);
60 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev);
61 
62 MODULE_FIRMWARE("amdgpu/bonaire_pfp.bin");
63 MODULE_FIRMWARE("amdgpu/bonaire_me.bin");
64 MODULE_FIRMWARE("amdgpu/bonaire_ce.bin");
65 MODULE_FIRMWARE("amdgpu/bonaire_rlc.bin");
66 MODULE_FIRMWARE("amdgpu/bonaire_mec.bin");
67 
68 MODULE_FIRMWARE("amdgpu/hawaii_pfp.bin");
69 MODULE_FIRMWARE("amdgpu/hawaii_me.bin");
70 MODULE_FIRMWARE("amdgpu/hawaii_ce.bin");
71 MODULE_FIRMWARE("amdgpu/hawaii_rlc.bin");
72 MODULE_FIRMWARE("amdgpu/hawaii_mec.bin");
73 
74 MODULE_FIRMWARE("amdgpu/kaveri_pfp.bin");
75 MODULE_FIRMWARE("amdgpu/kaveri_me.bin");
76 MODULE_FIRMWARE("amdgpu/kaveri_ce.bin");
77 MODULE_FIRMWARE("amdgpu/kaveri_rlc.bin");
78 MODULE_FIRMWARE("amdgpu/kaveri_mec.bin");
79 MODULE_FIRMWARE("amdgpu/kaveri_mec2.bin");
80 
81 MODULE_FIRMWARE("amdgpu/kabini_pfp.bin");
82 MODULE_FIRMWARE("amdgpu/kabini_me.bin");
83 MODULE_FIRMWARE("amdgpu/kabini_ce.bin");
84 MODULE_FIRMWARE("amdgpu/kabini_rlc.bin");
85 MODULE_FIRMWARE("amdgpu/kabini_mec.bin");
86 
87 MODULE_FIRMWARE("amdgpu/mullins_pfp.bin");
88 MODULE_FIRMWARE("amdgpu/mullins_me.bin");
89 MODULE_FIRMWARE("amdgpu/mullins_ce.bin");
90 MODULE_FIRMWARE("amdgpu/mullins_rlc.bin");
91 MODULE_FIRMWARE("amdgpu/mullins_mec.bin");
92 
93 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
94 {
95 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
96 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
97 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
98 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
99 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
100 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
101 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
102 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
103 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
104 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
105 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
106 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
107 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
108 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
109 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
110 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
111 };
112 
113 static const u32 spectre_rlc_save_restore_register_list[] =
114 {
115 	(0x0e00 << 16) | (0xc12c >> 2),
116 	0x00000000,
117 	(0x0e00 << 16) | (0xc140 >> 2),
118 	0x00000000,
119 	(0x0e00 << 16) | (0xc150 >> 2),
120 	0x00000000,
121 	(0x0e00 << 16) | (0xc15c >> 2),
122 	0x00000000,
123 	(0x0e00 << 16) | (0xc168 >> 2),
124 	0x00000000,
125 	(0x0e00 << 16) | (0xc170 >> 2),
126 	0x00000000,
127 	(0x0e00 << 16) | (0xc178 >> 2),
128 	0x00000000,
129 	(0x0e00 << 16) | (0xc204 >> 2),
130 	0x00000000,
131 	(0x0e00 << 16) | (0xc2b4 >> 2),
132 	0x00000000,
133 	(0x0e00 << 16) | (0xc2b8 >> 2),
134 	0x00000000,
135 	(0x0e00 << 16) | (0xc2bc >> 2),
136 	0x00000000,
137 	(0x0e00 << 16) | (0xc2c0 >> 2),
138 	0x00000000,
139 	(0x0e00 << 16) | (0x8228 >> 2),
140 	0x00000000,
141 	(0x0e00 << 16) | (0x829c >> 2),
142 	0x00000000,
143 	(0x0e00 << 16) | (0x869c >> 2),
144 	0x00000000,
145 	(0x0600 << 16) | (0x98f4 >> 2),
146 	0x00000000,
147 	(0x0e00 << 16) | (0x98f8 >> 2),
148 	0x00000000,
149 	(0x0e00 << 16) | (0x9900 >> 2),
150 	0x00000000,
151 	(0x0e00 << 16) | (0xc260 >> 2),
152 	0x00000000,
153 	(0x0e00 << 16) | (0x90e8 >> 2),
154 	0x00000000,
155 	(0x0e00 << 16) | (0x3c000 >> 2),
156 	0x00000000,
157 	(0x0e00 << 16) | (0x3c00c >> 2),
158 	0x00000000,
159 	(0x0e00 << 16) | (0x8c1c >> 2),
160 	0x00000000,
161 	(0x0e00 << 16) | (0x9700 >> 2),
162 	0x00000000,
163 	(0x0e00 << 16) | (0xcd20 >> 2),
164 	0x00000000,
165 	(0x4e00 << 16) | (0xcd20 >> 2),
166 	0x00000000,
167 	(0x5e00 << 16) | (0xcd20 >> 2),
168 	0x00000000,
169 	(0x6e00 << 16) | (0xcd20 >> 2),
170 	0x00000000,
171 	(0x7e00 << 16) | (0xcd20 >> 2),
172 	0x00000000,
173 	(0x8e00 << 16) | (0xcd20 >> 2),
174 	0x00000000,
175 	(0x9e00 << 16) | (0xcd20 >> 2),
176 	0x00000000,
177 	(0xae00 << 16) | (0xcd20 >> 2),
178 	0x00000000,
179 	(0xbe00 << 16) | (0xcd20 >> 2),
180 	0x00000000,
181 	(0x0e00 << 16) | (0x89bc >> 2),
182 	0x00000000,
183 	(0x0e00 << 16) | (0x8900 >> 2),
184 	0x00000000,
185 	0x3,
186 	(0x0e00 << 16) | (0xc130 >> 2),
187 	0x00000000,
188 	(0x0e00 << 16) | (0xc134 >> 2),
189 	0x00000000,
190 	(0x0e00 << 16) | (0xc1fc >> 2),
191 	0x00000000,
192 	(0x0e00 << 16) | (0xc208 >> 2),
193 	0x00000000,
194 	(0x0e00 << 16) | (0xc264 >> 2),
195 	0x00000000,
196 	(0x0e00 << 16) | (0xc268 >> 2),
197 	0x00000000,
198 	(0x0e00 << 16) | (0xc26c >> 2),
199 	0x00000000,
200 	(0x0e00 << 16) | (0xc270 >> 2),
201 	0x00000000,
202 	(0x0e00 << 16) | (0xc274 >> 2),
203 	0x00000000,
204 	(0x0e00 << 16) | (0xc278 >> 2),
205 	0x00000000,
206 	(0x0e00 << 16) | (0xc27c >> 2),
207 	0x00000000,
208 	(0x0e00 << 16) | (0xc280 >> 2),
209 	0x00000000,
210 	(0x0e00 << 16) | (0xc284 >> 2),
211 	0x00000000,
212 	(0x0e00 << 16) | (0xc288 >> 2),
213 	0x00000000,
214 	(0x0e00 << 16) | (0xc28c >> 2),
215 	0x00000000,
216 	(0x0e00 << 16) | (0xc290 >> 2),
217 	0x00000000,
218 	(0x0e00 << 16) | (0xc294 >> 2),
219 	0x00000000,
220 	(0x0e00 << 16) | (0xc298 >> 2),
221 	0x00000000,
222 	(0x0e00 << 16) | (0xc29c >> 2),
223 	0x00000000,
224 	(0x0e00 << 16) | (0xc2a0 >> 2),
225 	0x00000000,
226 	(0x0e00 << 16) | (0xc2a4 >> 2),
227 	0x00000000,
228 	(0x0e00 << 16) | (0xc2a8 >> 2),
229 	0x00000000,
230 	(0x0e00 << 16) | (0xc2ac  >> 2),
231 	0x00000000,
232 	(0x0e00 << 16) | (0xc2b0 >> 2),
233 	0x00000000,
234 	(0x0e00 << 16) | (0x301d0 >> 2),
235 	0x00000000,
236 	(0x0e00 << 16) | (0x30238 >> 2),
237 	0x00000000,
238 	(0x0e00 << 16) | (0x30250 >> 2),
239 	0x00000000,
240 	(0x0e00 << 16) | (0x30254 >> 2),
241 	0x00000000,
242 	(0x0e00 << 16) | (0x30258 >> 2),
243 	0x00000000,
244 	(0x0e00 << 16) | (0x3025c >> 2),
245 	0x00000000,
246 	(0x4e00 << 16) | (0xc900 >> 2),
247 	0x00000000,
248 	(0x5e00 << 16) | (0xc900 >> 2),
249 	0x00000000,
250 	(0x6e00 << 16) | (0xc900 >> 2),
251 	0x00000000,
252 	(0x7e00 << 16) | (0xc900 >> 2),
253 	0x00000000,
254 	(0x8e00 << 16) | (0xc900 >> 2),
255 	0x00000000,
256 	(0x9e00 << 16) | (0xc900 >> 2),
257 	0x00000000,
258 	(0xae00 << 16) | (0xc900 >> 2),
259 	0x00000000,
260 	(0xbe00 << 16) | (0xc900 >> 2),
261 	0x00000000,
262 	(0x4e00 << 16) | (0xc904 >> 2),
263 	0x00000000,
264 	(0x5e00 << 16) | (0xc904 >> 2),
265 	0x00000000,
266 	(0x6e00 << 16) | (0xc904 >> 2),
267 	0x00000000,
268 	(0x7e00 << 16) | (0xc904 >> 2),
269 	0x00000000,
270 	(0x8e00 << 16) | (0xc904 >> 2),
271 	0x00000000,
272 	(0x9e00 << 16) | (0xc904 >> 2),
273 	0x00000000,
274 	(0xae00 << 16) | (0xc904 >> 2),
275 	0x00000000,
276 	(0xbe00 << 16) | (0xc904 >> 2),
277 	0x00000000,
278 	(0x4e00 << 16) | (0xc908 >> 2),
279 	0x00000000,
280 	(0x5e00 << 16) | (0xc908 >> 2),
281 	0x00000000,
282 	(0x6e00 << 16) | (0xc908 >> 2),
283 	0x00000000,
284 	(0x7e00 << 16) | (0xc908 >> 2),
285 	0x00000000,
286 	(0x8e00 << 16) | (0xc908 >> 2),
287 	0x00000000,
288 	(0x9e00 << 16) | (0xc908 >> 2),
289 	0x00000000,
290 	(0xae00 << 16) | (0xc908 >> 2),
291 	0x00000000,
292 	(0xbe00 << 16) | (0xc908 >> 2),
293 	0x00000000,
294 	(0x4e00 << 16) | (0xc90c >> 2),
295 	0x00000000,
296 	(0x5e00 << 16) | (0xc90c >> 2),
297 	0x00000000,
298 	(0x6e00 << 16) | (0xc90c >> 2),
299 	0x00000000,
300 	(0x7e00 << 16) | (0xc90c >> 2),
301 	0x00000000,
302 	(0x8e00 << 16) | (0xc90c >> 2),
303 	0x00000000,
304 	(0x9e00 << 16) | (0xc90c >> 2),
305 	0x00000000,
306 	(0xae00 << 16) | (0xc90c >> 2),
307 	0x00000000,
308 	(0xbe00 << 16) | (0xc90c >> 2),
309 	0x00000000,
310 	(0x4e00 << 16) | (0xc910 >> 2),
311 	0x00000000,
312 	(0x5e00 << 16) | (0xc910 >> 2),
313 	0x00000000,
314 	(0x6e00 << 16) | (0xc910 >> 2),
315 	0x00000000,
316 	(0x7e00 << 16) | (0xc910 >> 2),
317 	0x00000000,
318 	(0x8e00 << 16) | (0xc910 >> 2),
319 	0x00000000,
320 	(0x9e00 << 16) | (0xc910 >> 2),
321 	0x00000000,
322 	(0xae00 << 16) | (0xc910 >> 2),
323 	0x00000000,
324 	(0xbe00 << 16) | (0xc910 >> 2),
325 	0x00000000,
326 	(0x0e00 << 16) | (0xc99c >> 2),
327 	0x00000000,
328 	(0x0e00 << 16) | (0x9834 >> 2),
329 	0x00000000,
330 	(0x0000 << 16) | (0x30f00 >> 2),
331 	0x00000000,
332 	(0x0001 << 16) | (0x30f00 >> 2),
333 	0x00000000,
334 	(0x0000 << 16) | (0x30f04 >> 2),
335 	0x00000000,
336 	(0x0001 << 16) | (0x30f04 >> 2),
337 	0x00000000,
338 	(0x0000 << 16) | (0x30f08 >> 2),
339 	0x00000000,
340 	(0x0001 << 16) | (0x30f08 >> 2),
341 	0x00000000,
342 	(0x0000 << 16) | (0x30f0c >> 2),
343 	0x00000000,
344 	(0x0001 << 16) | (0x30f0c >> 2),
345 	0x00000000,
346 	(0x0600 << 16) | (0x9b7c >> 2),
347 	0x00000000,
348 	(0x0e00 << 16) | (0x8a14 >> 2),
349 	0x00000000,
350 	(0x0e00 << 16) | (0x8a18 >> 2),
351 	0x00000000,
352 	(0x0600 << 16) | (0x30a00 >> 2),
353 	0x00000000,
354 	(0x0e00 << 16) | (0x8bf0 >> 2),
355 	0x00000000,
356 	(0x0e00 << 16) | (0x8bcc >> 2),
357 	0x00000000,
358 	(0x0e00 << 16) | (0x8b24 >> 2),
359 	0x00000000,
360 	(0x0e00 << 16) | (0x30a04 >> 2),
361 	0x00000000,
362 	(0x0600 << 16) | (0x30a10 >> 2),
363 	0x00000000,
364 	(0x0600 << 16) | (0x30a14 >> 2),
365 	0x00000000,
366 	(0x0600 << 16) | (0x30a18 >> 2),
367 	0x00000000,
368 	(0x0600 << 16) | (0x30a2c >> 2),
369 	0x00000000,
370 	(0x0e00 << 16) | (0xc700 >> 2),
371 	0x00000000,
372 	(0x0e00 << 16) | (0xc704 >> 2),
373 	0x00000000,
374 	(0x0e00 << 16) | (0xc708 >> 2),
375 	0x00000000,
376 	(0x0e00 << 16) | (0xc768 >> 2),
377 	0x00000000,
378 	(0x0400 << 16) | (0xc770 >> 2),
379 	0x00000000,
380 	(0x0400 << 16) | (0xc774 >> 2),
381 	0x00000000,
382 	(0x0400 << 16) | (0xc778 >> 2),
383 	0x00000000,
384 	(0x0400 << 16) | (0xc77c >> 2),
385 	0x00000000,
386 	(0x0400 << 16) | (0xc780 >> 2),
387 	0x00000000,
388 	(0x0400 << 16) | (0xc784 >> 2),
389 	0x00000000,
390 	(0x0400 << 16) | (0xc788 >> 2),
391 	0x00000000,
392 	(0x0400 << 16) | (0xc78c >> 2),
393 	0x00000000,
394 	(0x0400 << 16) | (0xc798 >> 2),
395 	0x00000000,
396 	(0x0400 << 16) | (0xc79c >> 2),
397 	0x00000000,
398 	(0x0400 << 16) | (0xc7a0 >> 2),
399 	0x00000000,
400 	(0x0400 << 16) | (0xc7a4 >> 2),
401 	0x00000000,
402 	(0x0400 << 16) | (0xc7a8 >> 2),
403 	0x00000000,
404 	(0x0400 << 16) | (0xc7ac >> 2),
405 	0x00000000,
406 	(0x0400 << 16) | (0xc7b0 >> 2),
407 	0x00000000,
408 	(0x0400 << 16) | (0xc7b4 >> 2),
409 	0x00000000,
410 	(0x0e00 << 16) | (0x9100 >> 2),
411 	0x00000000,
412 	(0x0e00 << 16) | (0x3c010 >> 2),
413 	0x00000000,
414 	(0x0e00 << 16) | (0x92a8 >> 2),
415 	0x00000000,
416 	(0x0e00 << 16) | (0x92ac >> 2),
417 	0x00000000,
418 	(0x0e00 << 16) | (0x92b4 >> 2),
419 	0x00000000,
420 	(0x0e00 << 16) | (0x92b8 >> 2),
421 	0x00000000,
422 	(0x0e00 << 16) | (0x92bc >> 2),
423 	0x00000000,
424 	(0x0e00 << 16) | (0x92c0 >> 2),
425 	0x00000000,
426 	(0x0e00 << 16) | (0x92c4 >> 2),
427 	0x00000000,
428 	(0x0e00 << 16) | (0x92c8 >> 2),
429 	0x00000000,
430 	(0x0e00 << 16) | (0x92cc >> 2),
431 	0x00000000,
432 	(0x0e00 << 16) | (0x92d0 >> 2),
433 	0x00000000,
434 	(0x0e00 << 16) | (0x8c00 >> 2),
435 	0x00000000,
436 	(0x0e00 << 16) | (0x8c04 >> 2),
437 	0x00000000,
438 	(0x0e00 << 16) | (0x8c20 >> 2),
439 	0x00000000,
440 	(0x0e00 << 16) | (0x8c38 >> 2),
441 	0x00000000,
442 	(0x0e00 << 16) | (0x8c3c >> 2),
443 	0x00000000,
444 	(0x0e00 << 16) | (0xae00 >> 2),
445 	0x00000000,
446 	(0x0e00 << 16) | (0x9604 >> 2),
447 	0x00000000,
448 	(0x0e00 << 16) | (0xac08 >> 2),
449 	0x00000000,
450 	(0x0e00 << 16) | (0xac0c >> 2),
451 	0x00000000,
452 	(0x0e00 << 16) | (0xac10 >> 2),
453 	0x00000000,
454 	(0x0e00 << 16) | (0xac14 >> 2),
455 	0x00000000,
456 	(0x0e00 << 16) | (0xac58 >> 2),
457 	0x00000000,
458 	(0x0e00 << 16) | (0xac68 >> 2),
459 	0x00000000,
460 	(0x0e00 << 16) | (0xac6c >> 2),
461 	0x00000000,
462 	(0x0e00 << 16) | (0xac70 >> 2),
463 	0x00000000,
464 	(0x0e00 << 16) | (0xac74 >> 2),
465 	0x00000000,
466 	(0x0e00 << 16) | (0xac78 >> 2),
467 	0x00000000,
468 	(0x0e00 << 16) | (0xac7c >> 2),
469 	0x00000000,
470 	(0x0e00 << 16) | (0xac80 >> 2),
471 	0x00000000,
472 	(0x0e00 << 16) | (0xac84 >> 2),
473 	0x00000000,
474 	(0x0e00 << 16) | (0xac88 >> 2),
475 	0x00000000,
476 	(0x0e00 << 16) | (0xac8c >> 2),
477 	0x00000000,
478 	(0x0e00 << 16) | (0x970c >> 2),
479 	0x00000000,
480 	(0x0e00 << 16) | (0x9714 >> 2),
481 	0x00000000,
482 	(0x0e00 << 16) | (0x9718 >> 2),
483 	0x00000000,
484 	(0x0e00 << 16) | (0x971c >> 2),
485 	0x00000000,
486 	(0x0e00 << 16) | (0x31068 >> 2),
487 	0x00000000,
488 	(0x4e00 << 16) | (0x31068 >> 2),
489 	0x00000000,
490 	(0x5e00 << 16) | (0x31068 >> 2),
491 	0x00000000,
492 	(0x6e00 << 16) | (0x31068 >> 2),
493 	0x00000000,
494 	(0x7e00 << 16) | (0x31068 >> 2),
495 	0x00000000,
496 	(0x8e00 << 16) | (0x31068 >> 2),
497 	0x00000000,
498 	(0x9e00 << 16) | (0x31068 >> 2),
499 	0x00000000,
500 	(0xae00 << 16) | (0x31068 >> 2),
501 	0x00000000,
502 	(0xbe00 << 16) | (0x31068 >> 2),
503 	0x00000000,
504 	(0x0e00 << 16) | (0xcd10 >> 2),
505 	0x00000000,
506 	(0x0e00 << 16) | (0xcd14 >> 2),
507 	0x00000000,
508 	(0x0e00 << 16) | (0x88b0 >> 2),
509 	0x00000000,
510 	(0x0e00 << 16) | (0x88b4 >> 2),
511 	0x00000000,
512 	(0x0e00 << 16) | (0x88b8 >> 2),
513 	0x00000000,
514 	(0x0e00 << 16) | (0x88bc >> 2),
515 	0x00000000,
516 	(0x0400 << 16) | (0x89c0 >> 2),
517 	0x00000000,
518 	(0x0e00 << 16) | (0x88c4 >> 2),
519 	0x00000000,
520 	(0x0e00 << 16) | (0x88c8 >> 2),
521 	0x00000000,
522 	(0x0e00 << 16) | (0x88d0 >> 2),
523 	0x00000000,
524 	(0x0e00 << 16) | (0x88d4 >> 2),
525 	0x00000000,
526 	(0x0e00 << 16) | (0x88d8 >> 2),
527 	0x00000000,
528 	(0x0e00 << 16) | (0x8980 >> 2),
529 	0x00000000,
530 	(0x0e00 << 16) | (0x30938 >> 2),
531 	0x00000000,
532 	(0x0e00 << 16) | (0x3093c >> 2),
533 	0x00000000,
534 	(0x0e00 << 16) | (0x30940 >> 2),
535 	0x00000000,
536 	(0x0e00 << 16) | (0x89a0 >> 2),
537 	0x00000000,
538 	(0x0e00 << 16) | (0x30900 >> 2),
539 	0x00000000,
540 	(0x0e00 << 16) | (0x30904 >> 2),
541 	0x00000000,
542 	(0x0e00 << 16) | (0x89b4 >> 2),
543 	0x00000000,
544 	(0x0e00 << 16) | (0x3c210 >> 2),
545 	0x00000000,
546 	(0x0e00 << 16) | (0x3c214 >> 2),
547 	0x00000000,
548 	(0x0e00 << 16) | (0x3c218 >> 2),
549 	0x00000000,
550 	(0x0e00 << 16) | (0x8904 >> 2),
551 	0x00000000,
552 	0x5,
553 	(0x0e00 << 16) | (0x8c28 >> 2),
554 	(0x0e00 << 16) | (0x8c2c >> 2),
555 	(0x0e00 << 16) | (0x8c30 >> 2),
556 	(0x0e00 << 16) | (0x8c34 >> 2),
557 	(0x0e00 << 16) | (0x9600 >> 2),
558 };
559 
560 static const u32 kalindi_rlc_save_restore_register_list[] =
561 {
562 	(0x0e00 << 16) | (0xc12c >> 2),
563 	0x00000000,
564 	(0x0e00 << 16) | (0xc140 >> 2),
565 	0x00000000,
566 	(0x0e00 << 16) | (0xc150 >> 2),
567 	0x00000000,
568 	(0x0e00 << 16) | (0xc15c >> 2),
569 	0x00000000,
570 	(0x0e00 << 16) | (0xc168 >> 2),
571 	0x00000000,
572 	(0x0e00 << 16) | (0xc170 >> 2),
573 	0x00000000,
574 	(0x0e00 << 16) | (0xc204 >> 2),
575 	0x00000000,
576 	(0x0e00 << 16) | (0xc2b4 >> 2),
577 	0x00000000,
578 	(0x0e00 << 16) | (0xc2b8 >> 2),
579 	0x00000000,
580 	(0x0e00 << 16) | (0xc2bc >> 2),
581 	0x00000000,
582 	(0x0e00 << 16) | (0xc2c0 >> 2),
583 	0x00000000,
584 	(0x0e00 << 16) | (0x8228 >> 2),
585 	0x00000000,
586 	(0x0e00 << 16) | (0x829c >> 2),
587 	0x00000000,
588 	(0x0e00 << 16) | (0x869c >> 2),
589 	0x00000000,
590 	(0x0600 << 16) | (0x98f4 >> 2),
591 	0x00000000,
592 	(0x0e00 << 16) | (0x98f8 >> 2),
593 	0x00000000,
594 	(0x0e00 << 16) | (0x9900 >> 2),
595 	0x00000000,
596 	(0x0e00 << 16) | (0xc260 >> 2),
597 	0x00000000,
598 	(0x0e00 << 16) | (0x90e8 >> 2),
599 	0x00000000,
600 	(0x0e00 << 16) | (0x3c000 >> 2),
601 	0x00000000,
602 	(0x0e00 << 16) | (0x3c00c >> 2),
603 	0x00000000,
604 	(0x0e00 << 16) | (0x8c1c >> 2),
605 	0x00000000,
606 	(0x0e00 << 16) | (0x9700 >> 2),
607 	0x00000000,
608 	(0x0e00 << 16) | (0xcd20 >> 2),
609 	0x00000000,
610 	(0x4e00 << 16) | (0xcd20 >> 2),
611 	0x00000000,
612 	(0x5e00 << 16) | (0xcd20 >> 2),
613 	0x00000000,
614 	(0x6e00 << 16) | (0xcd20 >> 2),
615 	0x00000000,
616 	(0x7e00 << 16) | (0xcd20 >> 2),
617 	0x00000000,
618 	(0x0e00 << 16) | (0x89bc >> 2),
619 	0x00000000,
620 	(0x0e00 << 16) | (0x8900 >> 2),
621 	0x00000000,
622 	0x3,
623 	(0x0e00 << 16) | (0xc130 >> 2),
624 	0x00000000,
625 	(0x0e00 << 16) | (0xc134 >> 2),
626 	0x00000000,
627 	(0x0e00 << 16) | (0xc1fc >> 2),
628 	0x00000000,
629 	(0x0e00 << 16) | (0xc208 >> 2),
630 	0x00000000,
631 	(0x0e00 << 16) | (0xc264 >> 2),
632 	0x00000000,
633 	(0x0e00 << 16) | (0xc268 >> 2),
634 	0x00000000,
635 	(0x0e00 << 16) | (0xc26c >> 2),
636 	0x00000000,
637 	(0x0e00 << 16) | (0xc270 >> 2),
638 	0x00000000,
639 	(0x0e00 << 16) | (0xc274 >> 2),
640 	0x00000000,
641 	(0x0e00 << 16) | (0xc28c >> 2),
642 	0x00000000,
643 	(0x0e00 << 16) | (0xc290 >> 2),
644 	0x00000000,
645 	(0x0e00 << 16) | (0xc294 >> 2),
646 	0x00000000,
647 	(0x0e00 << 16) | (0xc298 >> 2),
648 	0x00000000,
649 	(0x0e00 << 16) | (0xc2a0 >> 2),
650 	0x00000000,
651 	(0x0e00 << 16) | (0xc2a4 >> 2),
652 	0x00000000,
653 	(0x0e00 << 16) | (0xc2a8 >> 2),
654 	0x00000000,
655 	(0x0e00 << 16) | (0xc2ac >> 2),
656 	0x00000000,
657 	(0x0e00 << 16) | (0x301d0 >> 2),
658 	0x00000000,
659 	(0x0e00 << 16) | (0x30238 >> 2),
660 	0x00000000,
661 	(0x0e00 << 16) | (0x30250 >> 2),
662 	0x00000000,
663 	(0x0e00 << 16) | (0x30254 >> 2),
664 	0x00000000,
665 	(0x0e00 << 16) | (0x30258 >> 2),
666 	0x00000000,
667 	(0x0e00 << 16) | (0x3025c >> 2),
668 	0x00000000,
669 	(0x4e00 << 16) | (0xc900 >> 2),
670 	0x00000000,
671 	(0x5e00 << 16) | (0xc900 >> 2),
672 	0x00000000,
673 	(0x6e00 << 16) | (0xc900 >> 2),
674 	0x00000000,
675 	(0x7e00 << 16) | (0xc900 >> 2),
676 	0x00000000,
677 	(0x4e00 << 16) | (0xc904 >> 2),
678 	0x00000000,
679 	(0x5e00 << 16) | (0xc904 >> 2),
680 	0x00000000,
681 	(0x6e00 << 16) | (0xc904 >> 2),
682 	0x00000000,
683 	(0x7e00 << 16) | (0xc904 >> 2),
684 	0x00000000,
685 	(0x4e00 << 16) | (0xc908 >> 2),
686 	0x00000000,
687 	(0x5e00 << 16) | (0xc908 >> 2),
688 	0x00000000,
689 	(0x6e00 << 16) | (0xc908 >> 2),
690 	0x00000000,
691 	(0x7e00 << 16) | (0xc908 >> 2),
692 	0x00000000,
693 	(0x4e00 << 16) | (0xc90c >> 2),
694 	0x00000000,
695 	(0x5e00 << 16) | (0xc90c >> 2),
696 	0x00000000,
697 	(0x6e00 << 16) | (0xc90c >> 2),
698 	0x00000000,
699 	(0x7e00 << 16) | (0xc90c >> 2),
700 	0x00000000,
701 	(0x4e00 << 16) | (0xc910 >> 2),
702 	0x00000000,
703 	(0x5e00 << 16) | (0xc910 >> 2),
704 	0x00000000,
705 	(0x6e00 << 16) | (0xc910 >> 2),
706 	0x00000000,
707 	(0x7e00 << 16) | (0xc910 >> 2),
708 	0x00000000,
709 	(0x0e00 << 16) | (0xc99c >> 2),
710 	0x00000000,
711 	(0x0e00 << 16) | (0x9834 >> 2),
712 	0x00000000,
713 	(0x0000 << 16) | (0x30f00 >> 2),
714 	0x00000000,
715 	(0x0000 << 16) | (0x30f04 >> 2),
716 	0x00000000,
717 	(0x0000 << 16) | (0x30f08 >> 2),
718 	0x00000000,
719 	(0x0000 << 16) | (0x30f0c >> 2),
720 	0x00000000,
721 	(0x0600 << 16) | (0x9b7c >> 2),
722 	0x00000000,
723 	(0x0e00 << 16) | (0x8a14 >> 2),
724 	0x00000000,
725 	(0x0e00 << 16) | (0x8a18 >> 2),
726 	0x00000000,
727 	(0x0600 << 16) | (0x30a00 >> 2),
728 	0x00000000,
729 	(0x0e00 << 16) | (0x8bf0 >> 2),
730 	0x00000000,
731 	(0x0e00 << 16) | (0x8bcc >> 2),
732 	0x00000000,
733 	(0x0e00 << 16) | (0x8b24 >> 2),
734 	0x00000000,
735 	(0x0e00 << 16) | (0x30a04 >> 2),
736 	0x00000000,
737 	(0x0600 << 16) | (0x30a10 >> 2),
738 	0x00000000,
739 	(0x0600 << 16) | (0x30a14 >> 2),
740 	0x00000000,
741 	(0x0600 << 16) | (0x30a18 >> 2),
742 	0x00000000,
743 	(0x0600 << 16) | (0x30a2c >> 2),
744 	0x00000000,
745 	(0x0e00 << 16) | (0xc700 >> 2),
746 	0x00000000,
747 	(0x0e00 << 16) | (0xc704 >> 2),
748 	0x00000000,
749 	(0x0e00 << 16) | (0xc708 >> 2),
750 	0x00000000,
751 	(0x0e00 << 16) | (0xc768 >> 2),
752 	0x00000000,
753 	(0x0400 << 16) | (0xc770 >> 2),
754 	0x00000000,
755 	(0x0400 << 16) | (0xc774 >> 2),
756 	0x00000000,
757 	(0x0400 << 16) | (0xc798 >> 2),
758 	0x00000000,
759 	(0x0400 << 16) | (0xc79c >> 2),
760 	0x00000000,
761 	(0x0e00 << 16) | (0x9100 >> 2),
762 	0x00000000,
763 	(0x0e00 << 16) | (0x3c010 >> 2),
764 	0x00000000,
765 	(0x0e00 << 16) | (0x8c00 >> 2),
766 	0x00000000,
767 	(0x0e00 << 16) | (0x8c04 >> 2),
768 	0x00000000,
769 	(0x0e00 << 16) | (0x8c20 >> 2),
770 	0x00000000,
771 	(0x0e00 << 16) | (0x8c38 >> 2),
772 	0x00000000,
773 	(0x0e00 << 16) | (0x8c3c >> 2),
774 	0x00000000,
775 	(0x0e00 << 16) | (0xae00 >> 2),
776 	0x00000000,
777 	(0x0e00 << 16) | (0x9604 >> 2),
778 	0x00000000,
779 	(0x0e00 << 16) | (0xac08 >> 2),
780 	0x00000000,
781 	(0x0e00 << 16) | (0xac0c >> 2),
782 	0x00000000,
783 	(0x0e00 << 16) | (0xac10 >> 2),
784 	0x00000000,
785 	(0x0e00 << 16) | (0xac14 >> 2),
786 	0x00000000,
787 	(0x0e00 << 16) | (0xac58 >> 2),
788 	0x00000000,
789 	(0x0e00 << 16) | (0xac68 >> 2),
790 	0x00000000,
791 	(0x0e00 << 16) | (0xac6c >> 2),
792 	0x00000000,
793 	(0x0e00 << 16) | (0xac70 >> 2),
794 	0x00000000,
795 	(0x0e00 << 16) | (0xac74 >> 2),
796 	0x00000000,
797 	(0x0e00 << 16) | (0xac78 >> 2),
798 	0x00000000,
799 	(0x0e00 << 16) | (0xac7c >> 2),
800 	0x00000000,
801 	(0x0e00 << 16) | (0xac80 >> 2),
802 	0x00000000,
803 	(0x0e00 << 16) | (0xac84 >> 2),
804 	0x00000000,
805 	(0x0e00 << 16) | (0xac88 >> 2),
806 	0x00000000,
807 	(0x0e00 << 16) | (0xac8c >> 2),
808 	0x00000000,
809 	(0x0e00 << 16) | (0x970c >> 2),
810 	0x00000000,
811 	(0x0e00 << 16) | (0x9714 >> 2),
812 	0x00000000,
813 	(0x0e00 << 16) | (0x9718 >> 2),
814 	0x00000000,
815 	(0x0e00 << 16) | (0x971c >> 2),
816 	0x00000000,
817 	(0x0e00 << 16) | (0x31068 >> 2),
818 	0x00000000,
819 	(0x4e00 << 16) | (0x31068 >> 2),
820 	0x00000000,
821 	(0x5e00 << 16) | (0x31068 >> 2),
822 	0x00000000,
823 	(0x6e00 << 16) | (0x31068 >> 2),
824 	0x00000000,
825 	(0x7e00 << 16) | (0x31068 >> 2),
826 	0x00000000,
827 	(0x0e00 << 16) | (0xcd10 >> 2),
828 	0x00000000,
829 	(0x0e00 << 16) | (0xcd14 >> 2),
830 	0x00000000,
831 	(0x0e00 << 16) | (0x88b0 >> 2),
832 	0x00000000,
833 	(0x0e00 << 16) | (0x88b4 >> 2),
834 	0x00000000,
835 	(0x0e00 << 16) | (0x88b8 >> 2),
836 	0x00000000,
837 	(0x0e00 << 16) | (0x88bc >> 2),
838 	0x00000000,
839 	(0x0400 << 16) | (0x89c0 >> 2),
840 	0x00000000,
841 	(0x0e00 << 16) | (0x88c4 >> 2),
842 	0x00000000,
843 	(0x0e00 << 16) | (0x88c8 >> 2),
844 	0x00000000,
845 	(0x0e00 << 16) | (0x88d0 >> 2),
846 	0x00000000,
847 	(0x0e00 << 16) | (0x88d4 >> 2),
848 	0x00000000,
849 	(0x0e00 << 16) | (0x88d8 >> 2),
850 	0x00000000,
851 	(0x0e00 << 16) | (0x8980 >> 2),
852 	0x00000000,
853 	(0x0e00 << 16) | (0x30938 >> 2),
854 	0x00000000,
855 	(0x0e00 << 16) | (0x3093c >> 2),
856 	0x00000000,
857 	(0x0e00 << 16) | (0x30940 >> 2),
858 	0x00000000,
859 	(0x0e00 << 16) | (0x89a0 >> 2),
860 	0x00000000,
861 	(0x0e00 << 16) | (0x30900 >> 2),
862 	0x00000000,
863 	(0x0e00 << 16) | (0x30904 >> 2),
864 	0x00000000,
865 	(0x0e00 << 16) | (0x89b4 >> 2),
866 	0x00000000,
867 	(0x0e00 << 16) | (0x3e1fc >> 2),
868 	0x00000000,
869 	(0x0e00 << 16) | (0x3c210 >> 2),
870 	0x00000000,
871 	(0x0e00 << 16) | (0x3c214 >> 2),
872 	0x00000000,
873 	(0x0e00 << 16) | (0x3c218 >> 2),
874 	0x00000000,
875 	(0x0e00 << 16) | (0x8904 >> 2),
876 	0x00000000,
877 	0x5,
878 	(0x0e00 << 16) | (0x8c28 >> 2),
879 	(0x0e00 << 16) | (0x8c2c >> 2),
880 	(0x0e00 << 16) | (0x8c30 >> 2),
881 	(0x0e00 << 16) | (0x8c34 >> 2),
882 	(0x0e00 << 16) | (0x9600 >> 2),
883 };
884 
885 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev);
886 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer);
887 static void gfx_v7_0_init_pg(struct amdgpu_device *adev);
888 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev);
889 
890 /*
891  * Core functions
892  */
893 /**
894  * gfx_v7_0_init_microcode - load ucode images from disk
895  *
896  * @adev: amdgpu_device pointer
897  *
898  * Use the firmware interface to load the ucode images into
899  * the driver (not loaded into hw).
900  * Returns 0 on success, error on failure.
901  */
902 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev)
903 {
904 	const char *chip_name;
905 	char fw_name[30];
906 	int err;
907 
908 	DRM_DEBUG("\n");
909 
910 	switch (adev->asic_type) {
911 	case CHIP_BONAIRE:
912 		chip_name = "bonaire";
913 		break;
914 	case CHIP_HAWAII:
915 		chip_name = "hawaii";
916 		break;
917 	case CHIP_KAVERI:
918 		chip_name = "kaveri";
919 		break;
920 	case CHIP_KABINI:
921 		chip_name = "kabini";
922 		break;
923 	case CHIP_MULLINS:
924 		chip_name = "mullins";
925 		break;
926 	default: BUG();
927 	}
928 
929 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
930 	err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
931 	if (err)
932 		goto out;
933 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
934 	if (err)
935 		goto out;
936 
937 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
938 	err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
939 	if (err)
940 		goto out;
941 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
942 	if (err)
943 		goto out;
944 
945 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
946 	err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
947 	if (err)
948 		goto out;
949 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
950 	if (err)
951 		goto out;
952 
953 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
954 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
955 	if (err)
956 		goto out;
957 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
958 	if (err)
959 		goto out;
960 
961 	if (adev->asic_type == CHIP_KAVERI) {
962 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
963 		err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
964 		if (err)
965 			goto out;
966 		err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
967 		if (err)
968 			goto out;
969 	}
970 
971 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
972 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
973 	if (err)
974 		goto out;
975 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
976 
977 out:
978 	if (err) {
979 		pr_err("gfx7: Failed to load firmware \"%s\"\n", fw_name);
980 		release_firmware(adev->gfx.pfp_fw);
981 		adev->gfx.pfp_fw = NULL;
982 		release_firmware(adev->gfx.me_fw);
983 		adev->gfx.me_fw = NULL;
984 		release_firmware(adev->gfx.ce_fw);
985 		adev->gfx.ce_fw = NULL;
986 		release_firmware(adev->gfx.mec_fw);
987 		adev->gfx.mec_fw = NULL;
988 		release_firmware(adev->gfx.mec2_fw);
989 		adev->gfx.mec2_fw = NULL;
990 		release_firmware(adev->gfx.rlc_fw);
991 		adev->gfx.rlc_fw = NULL;
992 	}
993 	return err;
994 }
995 
996 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev)
997 {
998 	release_firmware(adev->gfx.pfp_fw);
999 	adev->gfx.pfp_fw = NULL;
1000 	release_firmware(adev->gfx.me_fw);
1001 	adev->gfx.me_fw = NULL;
1002 	release_firmware(adev->gfx.ce_fw);
1003 	adev->gfx.ce_fw = NULL;
1004 	release_firmware(adev->gfx.mec_fw);
1005 	adev->gfx.mec_fw = NULL;
1006 	release_firmware(adev->gfx.mec2_fw);
1007 	adev->gfx.mec2_fw = NULL;
1008 	release_firmware(adev->gfx.rlc_fw);
1009 	adev->gfx.rlc_fw = NULL;
1010 }
1011 
1012 /**
1013  * gfx_v7_0_tiling_mode_table_init - init the hw tiling table
1014  *
1015  * @adev: amdgpu_device pointer
1016  *
1017  * Starting with SI, the tiling setup is done globally in a
1018  * set of 32 tiling modes.  Rather than selecting each set of
1019  * parameters per surface as on older asics, we just select
1020  * which index in the tiling table we want to use, and the
1021  * surface uses those parameters (CIK).
1022  */
1023 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev)
1024 {
1025 	const u32 num_tile_mode_states =
1026 			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
1027 	const u32 num_secondary_tile_mode_states =
1028 			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
1029 	u32 reg_offset, split_equal_to_row_size;
1030 	uint32_t *tile, *macrotile;
1031 
1032 	tile = adev->gfx.config.tile_mode_array;
1033 	macrotile = adev->gfx.config.macrotile_mode_array;
1034 
1035 	switch (adev->gfx.config.mem_row_size_in_kb) {
1036 	case 1:
1037 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1038 		break;
1039 	case 2:
1040 	default:
1041 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1042 		break;
1043 	case 4:
1044 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1045 		break;
1046 	}
1047 
1048 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1049 		tile[reg_offset] = 0;
1050 	for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1051 		macrotile[reg_offset] = 0;
1052 
1053 	switch (adev->asic_type) {
1054 	case CHIP_BONAIRE:
1055 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1056 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1057 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1058 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1059 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1060 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1061 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1062 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1063 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1064 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1065 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1066 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1067 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1068 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1069 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1070 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1071 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1072 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1073 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1074 			   TILE_SPLIT(split_equal_to_row_size));
1075 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1076 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1077 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1078 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1079 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1080 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1081 			   TILE_SPLIT(split_equal_to_row_size));
1082 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1083 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1084 			   PIPE_CONFIG(ADDR_SURF_P4_16x16));
1085 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1086 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1087 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1088 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1089 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1090 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1091 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1092 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1093 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1094 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1095 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1096 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1097 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1098 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1099 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1100 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1101 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1102 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1103 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1104 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1105 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1106 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1107 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1108 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1109 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1110 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1111 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1112 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1113 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1114 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1115 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1116 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1117 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1118 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1119 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1120 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1121 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1122 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1123 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1124 		tile[21] =  (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1125 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1126 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1127 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1128 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1129 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1130 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1131 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1132 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1133 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1134 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1135 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1136 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1137 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1138 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1139 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1140 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1141 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1142 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1143 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1144 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1145 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1146 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1147 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1148 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1149 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1150 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1151 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1152 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1153 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1154 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1155 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1156 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1157 
1158 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1159 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1160 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1161 				NUM_BANKS(ADDR_SURF_16_BANK));
1162 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1163 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1164 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1165 				NUM_BANKS(ADDR_SURF_16_BANK));
1166 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1167 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1168 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1169 				NUM_BANKS(ADDR_SURF_16_BANK));
1170 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1171 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1172 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1173 				NUM_BANKS(ADDR_SURF_16_BANK));
1174 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1175 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1176 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1177 				NUM_BANKS(ADDR_SURF_16_BANK));
1178 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1179 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1180 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1181 				NUM_BANKS(ADDR_SURF_8_BANK));
1182 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1183 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1184 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1185 				NUM_BANKS(ADDR_SURF_4_BANK));
1186 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1187 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1188 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1189 				NUM_BANKS(ADDR_SURF_16_BANK));
1190 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1191 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1192 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1193 				NUM_BANKS(ADDR_SURF_16_BANK));
1194 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1195 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1196 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1197 				NUM_BANKS(ADDR_SURF_16_BANK));
1198 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1199 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1200 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1201 				NUM_BANKS(ADDR_SURF_16_BANK));
1202 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1203 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1204 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1205 				NUM_BANKS(ADDR_SURF_16_BANK));
1206 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1207 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1208 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1209 				NUM_BANKS(ADDR_SURF_8_BANK));
1210 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1211 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1212 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1213 				NUM_BANKS(ADDR_SURF_4_BANK));
1214 
1215 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1216 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1217 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1218 			if (reg_offset != 7)
1219 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1220 		break;
1221 	case CHIP_HAWAII:
1222 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1223 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1224 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1225 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1226 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1227 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1228 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1229 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1230 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1231 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1232 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1233 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1234 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1235 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1236 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1237 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1238 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1239 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1240 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1241 			   TILE_SPLIT(split_equal_to_row_size));
1242 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1243 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1244 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1245 			   TILE_SPLIT(split_equal_to_row_size));
1246 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1247 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1248 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1249 			   TILE_SPLIT(split_equal_to_row_size));
1250 		tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1251 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1252 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1253 			   TILE_SPLIT(split_equal_to_row_size));
1254 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1255 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
1256 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1257 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1258 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1259 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1260 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1261 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1262 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1263 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1264 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1265 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1266 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1267 		tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1268 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1269 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1270 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1271 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1272 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1273 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1274 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1275 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1276 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1277 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1278 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1279 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1280 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1281 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1282 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1283 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1284 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1285 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1286 		tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1287 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1288 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1289 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1290 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1291 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1292 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1293 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1294 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1295 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1296 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1297 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1298 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1299 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1300 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1301 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1302 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1303 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1304 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1305 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1306 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1307 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1308 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1309 		tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1310 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1311 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1312 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1313 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1314 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1315 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1316 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1317 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1318 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1319 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1320 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1321 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1322 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1323 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1324 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1325 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1326 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1327 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1328 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1329 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1330 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1331 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1332 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1333 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1334 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1335 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1336 		tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1337 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1338 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1339 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1340 
1341 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1342 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1343 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1344 				NUM_BANKS(ADDR_SURF_16_BANK));
1345 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1346 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1347 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1348 				NUM_BANKS(ADDR_SURF_16_BANK));
1349 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1350 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1351 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1352 				NUM_BANKS(ADDR_SURF_16_BANK));
1353 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1354 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1355 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1356 				NUM_BANKS(ADDR_SURF_16_BANK));
1357 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1358 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1359 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1360 				NUM_BANKS(ADDR_SURF_8_BANK));
1361 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1362 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1363 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1364 				NUM_BANKS(ADDR_SURF_4_BANK));
1365 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1366 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1367 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1368 				NUM_BANKS(ADDR_SURF_4_BANK));
1369 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1370 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1371 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1372 				NUM_BANKS(ADDR_SURF_16_BANK));
1373 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1374 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1375 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1376 				NUM_BANKS(ADDR_SURF_16_BANK));
1377 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1378 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1379 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1380 				NUM_BANKS(ADDR_SURF_16_BANK));
1381 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1382 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1383 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1384 				NUM_BANKS(ADDR_SURF_8_BANK));
1385 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1386 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1387 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1388 				NUM_BANKS(ADDR_SURF_16_BANK));
1389 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1390 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1391 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1392 				NUM_BANKS(ADDR_SURF_8_BANK));
1393 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1394 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1395 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1396 				NUM_BANKS(ADDR_SURF_4_BANK));
1397 
1398 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1399 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1400 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1401 			if (reg_offset != 7)
1402 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1403 		break;
1404 	case CHIP_KABINI:
1405 	case CHIP_KAVERI:
1406 	case CHIP_MULLINS:
1407 	default:
1408 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1409 			   PIPE_CONFIG(ADDR_SURF_P2) |
1410 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1411 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1412 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1413 			   PIPE_CONFIG(ADDR_SURF_P2) |
1414 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1415 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1416 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1417 			   PIPE_CONFIG(ADDR_SURF_P2) |
1418 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1419 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1420 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1421 			   PIPE_CONFIG(ADDR_SURF_P2) |
1422 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1423 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1424 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1425 			   PIPE_CONFIG(ADDR_SURF_P2) |
1426 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1427 			   TILE_SPLIT(split_equal_to_row_size));
1428 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1429 			   PIPE_CONFIG(ADDR_SURF_P2) |
1430 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1431 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1432 			   PIPE_CONFIG(ADDR_SURF_P2) |
1433 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1434 			   TILE_SPLIT(split_equal_to_row_size));
1435 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1436 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1437 			   PIPE_CONFIG(ADDR_SURF_P2));
1438 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1439 			   PIPE_CONFIG(ADDR_SURF_P2) |
1440 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1441 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1442 			    PIPE_CONFIG(ADDR_SURF_P2) |
1443 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1444 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1445 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1446 			    PIPE_CONFIG(ADDR_SURF_P2) |
1447 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1448 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1449 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1450 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1451 			    PIPE_CONFIG(ADDR_SURF_P2) |
1452 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1453 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1454 			    PIPE_CONFIG(ADDR_SURF_P2) |
1455 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1456 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1457 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1458 			    PIPE_CONFIG(ADDR_SURF_P2) |
1459 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1460 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1461 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1462 			    PIPE_CONFIG(ADDR_SURF_P2) |
1463 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1464 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1465 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1466 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1467 			    PIPE_CONFIG(ADDR_SURF_P2) |
1468 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1469 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1470 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1471 			    PIPE_CONFIG(ADDR_SURF_P2) |
1472 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1473 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1474 			    PIPE_CONFIG(ADDR_SURF_P2) |
1475 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1476 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1477 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1478 			    PIPE_CONFIG(ADDR_SURF_P2) |
1479 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1480 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1481 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1482 			    PIPE_CONFIG(ADDR_SURF_P2) |
1483 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1484 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1485 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1486 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1487 			    PIPE_CONFIG(ADDR_SURF_P2) |
1488 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1489 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1490 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1491 			    PIPE_CONFIG(ADDR_SURF_P2) |
1492 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1493 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1494 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1495 			    PIPE_CONFIG(ADDR_SURF_P2) |
1496 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1497 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1498 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1499 			    PIPE_CONFIG(ADDR_SURF_P2) |
1500 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1501 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1502 			    PIPE_CONFIG(ADDR_SURF_P2) |
1503 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1504 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1505 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1506 			    PIPE_CONFIG(ADDR_SURF_P2) |
1507 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1508 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1509 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1510 
1511 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1512 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1513 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1514 				NUM_BANKS(ADDR_SURF_8_BANK));
1515 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1516 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1517 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1518 				NUM_BANKS(ADDR_SURF_8_BANK));
1519 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1520 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1521 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1522 				NUM_BANKS(ADDR_SURF_8_BANK));
1523 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1524 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1525 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1526 				NUM_BANKS(ADDR_SURF_8_BANK));
1527 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1528 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1529 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1530 				NUM_BANKS(ADDR_SURF_8_BANK));
1531 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1532 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1533 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1534 				NUM_BANKS(ADDR_SURF_8_BANK));
1535 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1536 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1537 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1538 				NUM_BANKS(ADDR_SURF_8_BANK));
1539 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1540 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1541 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1542 				NUM_BANKS(ADDR_SURF_16_BANK));
1543 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1544 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1545 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1546 				NUM_BANKS(ADDR_SURF_16_BANK));
1547 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1548 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1549 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1550 				NUM_BANKS(ADDR_SURF_16_BANK));
1551 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1552 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1553 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1554 				NUM_BANKS(ADDR_SURF_16_BANK));
1555 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1556 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1557 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1558 				NUM_BANKS(ADDR_SURF_16_BANK));
1559 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1560 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1561 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1562 				NUM_BANKS(ADDR_SURF_16_BANK));
1563 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1564 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1565 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1566 				NUM_BANKS(ADDR_SURF_8_BANK));
1567 
1568 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1569 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1570 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1571 			if (reg_offset != 7)
1572 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1573 		break;
1574 	}
1575 }
1576 
1577 /**
1578  * gfx_v7_0_select_se_sh - select which SE, SH to address
1579  *
1580  * @adev: amdgpu_device pointer
1581  * @se_num: shader engine to address
1582  * @sh_num: sh block to address
1583  * @instance: Certain registers are instanced per SE or SH.
1584  *            0xffffffff means broadcast to all SEs or SHs (CIK).
1585  *
1586  * Select which SE, SH combinations to address.
1587  */
1588 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev,
1589 				  u32 se_num, u32 sh_num, u32 instance)
1590 {
1591 	u32 data;
1592 
1593 	if (instance == 0xffffffff)
1594 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1595 	else
1596 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1597 
1598 	if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1599 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1600 			GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
1601 	else if (se_num == 0xffffffff)
1602 		data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK |
1603 			(sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT);
1604 	else if (sh_num == 0xffffffff)
1605 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1606 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1607 	else
1608 		data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) |
1609 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1610 	WREG32(mmGRBM_GFX_INDEX, data);
1611 }
1612 
1613 /**
1614  * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs
1615  *
1616  * @adev: amdgpu_device pointer
1617  *
1618  * Calculates the bitmask of enabled RBs (CIK).
1619  * Returns the enabled RB bitmask.
1620  */
1621 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1622 {
1623 	u32 data, mask;
1624 
1625 	data = RREG32(mmCC_RB_BACKEND_DISABLE);
1626 	data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1627 
1628 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1629 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1630 
1631 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1632 					 adev->gfx.config.max_sh_per_se);
1633 
1634 	return (~data) & mask;
1635 }
1636 
1637 static void
1638 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
1639 {
1640 	switch (adev->asic_type) {
1641 	case CHIP_BONAIRE:
1642 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
1643 			  SE_XSEL(1) | SE_YSEL(1);
1644 		*rconf1 |= 0x0;
1645 		break;
1646 	case CHIP_HAWAII:
1647 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
1648 			  RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) |
1649 			  PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) |
1650 			  SE_YSEL(3);
1651 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
1652 			   SE_PAIR_YSEL(2);
1653 		break;
1654 	case CHIP_KAVERI:
1655 		*rconf |= RB_MAP_PKR0(2);
1656 		*rconf1 |= 0x0;
1657 		break;
1658 	case CHIP_KABINI:
1659 	case CHIP_MULLINS:
1660 		*rconf |= 0x0;
1661 		*rconf1 |= 0x0;
1662 		break;
1663 	default:
1664 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
1665 		break;
1666 	}
1667 }
1668 
1669 static void
1670 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev,
1671 					u32 raster_config, u32 raster_config_1,
1672 					unsigned rb_mask, unsigned num_rb)
1673 {
1674 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
1675 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
1676 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
1677 	unsigned rb_per_se = num_rb / num_se;
1678 	unsigned se_mask[4];
1679 	unsigned se;
1680 
1681 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
1682 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
1683 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
1684 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
1685 
1686 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
1687 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
1688 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
1689 
1690 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
1691 			     (!se_mask[2] && !se_mask[3]))) {
1692 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
1693 
1694 		if (!se_mask[0] && !se_mask[1]) {
1695 			raster_config_1 |=
1696 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
1697 		} else {
1698 			raster_config_1 |=
1699 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
1700 		}
1701 	}
1702 
1703 	for (se = 0; se < num_se; se++) {
1704 		unsigned raster_config_se = raster_config;
1705 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
1706 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
1707 		int idx = (se / 2) * 2;
1708 
1709 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
1710 			raster_config_se &= ~SE_MAP_MASK;
1711 
1712 			if (!se_mask[idx]) {
1713 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
1714 			} else {
1715 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
1716 			}
1717 		}
1718 
1719 		pkr0_mask &= rb_mask;
1720 		pkr1_mask &= rb_mask;
1721 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
1722 			raster_config_se &= ~PKR_MAP_MASK;
1723 
1724 			if (!pkr0_mask) {
1725 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
1726 			} else {
1727 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
1728 			}
1729 		}
1730 
1731 		if (rb_per_se >= 2) {
1732 			unsigned rb0_mask = 1 << (se * rb_per_se);
1733 			unsigned rb1_mask = rb0_mask << 1;
1734 
1735 			rb0_mask &= rb_mask;
1736 			rb1_mask &= rb_mask;
1737 			if (!rb0_mask || !rb1_mask) {
1738 				raster_config_se &= ~RB_MAP_PKR0_MASK;
1739 
1740 				if (!rb0_mask) {
1741 					raster_config_se |=
1742 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
1743 				} else {
1744 					raster_config_se |=
1745 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
1746 				}
1747 			}
1748 
1749 			if (rb_per_se > 2) {
1750 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
1751 				rb1_mask = rb0_mask << 1;
1752 				rb0_mask &= rb_mask;
1753 				rb1_mask &= rb_mask;
1754 				if (!rb0_mask || !rb1_mask) {
1755 					raster_config_se &= ~RB_MAP_PKR1_MASK;
1756 
1757 					if (!rb0_mask) {
1758 						raster_config_se |=
1759 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
1760 					} else {
1761 						raster_config_se |=
1762 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
1763 					}
1764 				}
1765 			}
1766 		}
1767 
1768 		/* GRBM_GFX_INDEX has a different offset on CI+ */
1769 		gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
1770 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
1771 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1772 	}
1773 
1774 	/* GRBM_GFX_INDEX has a different offset on CI+ */
1775 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1776 }
1777 
1778 /**
1779  * gfx_v7_0_setup_rb - setup the RBs on the asic
1780  *
1781  * @adev: amdgpu_device pointer
1782  *
1783  * Configures per-SE/SH RB registers (CIK).
1784  */
1785 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1786 {
1787 	int i, j;
1788 	u32 data;
1789 	u32 raster_config = 0, raster_config_1 = 0;
1790 	u32 active_rbs = 0;
1791 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1792 					adev->gfx.config.max_sh_per_se;
1793 	unsigned num_rb_pipes;
1794 
1795 	mutex_lock(&adev->grbm_idx_mutex);
1796 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1797 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1798 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1799 			data = gfx_v7_0_get_rb_active_bitmap(adev);
1800 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1801 					       rb_bitmap_width_per_sh);
1802 		}
1803 	}
1804 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1805 
1806 	adev->gfx.config.backend_enable_mask = active_rbs;
1807 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1808 
1809 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1810 			     adev->gfx.config.max_shader_engines, 16);
1811 
1812 	gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1813 
1814 	if (!adev->gfx.config.backend_enable_mask ||
1815 			adev->gfx.config.num_rbs >= num_rb_pipes) {
1816 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1817 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1818 	} else {
1819 		gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1820 							adev->gfx.config.backend_enable_mask,
1821 							num_rb_pipes);
1822 	}
1823 
1824 	/* cache the values for userspace */
1825 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1826 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1827 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1828 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
1829 				RREG32(mmCC_RB_BACKEND_DISABLE);
1830 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
1831 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1832 			adev->gfx.config.rb_config[i][j].raster_config =
1833 				RREG32(mmPA_SC_RASTER_CONFIG);
1834 			adev->gfx.config.rb_config[i][j].raster_config_1 =
1835 				RREG32(mmPA_SC_RASTER_CONFIG_1);
1836 		}
1837 	}
1838 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1839 	mutex_unlock(&adev->grbm_idx_mutex);
1840 }
1841 
1842 #define DEFAULT_SH_MEM_BASES	(0x6000)
1843 /**
1844  * gfx_v7_0_init_compute_vmid - gart enable
1845  *
1846  * @adev: amdgpu_device pointer
1847  *
1848  * Initialize compute vmid sh_mem registers
1849  *
1850  */
1851 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1852 {
1853 	int i;
1854 	uint32_t sh_mem_config;
1855 	uint32_t sh_mem_bases;
1856 
1857 	/*
1858 	 * Configure apertures:
1859 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1860 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1861 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1862 	*/
1863 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1864 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1865 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1866 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1867 	mutex_lock(&adev->srbm_mutex);
1868 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1869 		cik_srbm_select(adev, 0, 0, 0, i);
1870 		/* CP and shaders */
1871 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1872 		WREG32(mmSH_MEM_APE1_BASE, 1);
1873 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1874 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1875 	}
1876 	cik_srbm_select(adev, 0, 0, 0, 0);
1877 	mutex_unlock(&adev->srbm_mutex);
1878 
1879 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1880 	   access. These should be enabled by FW for target VMIDs. */
1881 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1882 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
1883 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
1884 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
1885 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
1886 	}
1887 }
1888 
1889 static void gfx_v7_0_init_gds_vmid(struct amdgpu_device *adev)
1890 {
1891 	int vmid;
1892 
1893 	/*
1894 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1895 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1896 	 * the driver can enable them for graphics. VMID0 should maintain
1897 	 * access so that HWS firmware can save/restore entries.
1898 	 */
1899 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
1900 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
1901 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
1902 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
1903 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
1904 	}
1905 }
1906 
1907 static void gfx_v7_0_config_init(struct amdgpu_device *adev)
1908 {
1909 	adev->gfx.config.double_offchip_lds_buf = 1;
1910 }
1911 
1912 /**
1913  * gfx_v7_0_constants_init - setup the 3D engine
1914  *
1915  * @adev: amdgpu_device pointer
1916  *
1917  * init the gfx constants such as the 3D engine, tiling configuration
1918  * registers, maximum number of quad pipes, render backends...
1919  */
1920 static void gfx_v7_0_constants_init(struct amdgpu_device *adev)
1921 {
1922 	u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base;
1923 	u32 tmp;
1924 	int i;
1925 
1926 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1927 
1928 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1929 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1930 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1931 
1932 	gfx_v7_0_tiling_mode_table_init(adev);
1933 
1934 	gfx_v7_0_setup_rb(adev);
1935 	gfx_v7_0_get_cu_info(adev);
1936 	gfx_v7_0_config_init(adev);
1937 
1938 	/* set HW defaults for 3D engine */
1939 	WREG32(mmCP_MEQ_THRESHOLDS,
1940 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1941 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1942 
1943 	mutex_lock(&adev->grbm_idx_mutex);
1944 	/*
1945 	 * making sure that the following register writes will be broadcasted
1946 	 * to all the shaders
1947 	 */
1948 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1949 
1950 	/* XXX SH_MEM regs */
1951 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1952 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1953 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1954 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE,
1955 				   MTYPE_NC);
1956 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE,
1957 				   MTYPE_UC);
1958 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0);
1959 
1960 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
1961 				   SWIZZLE_ENABLE, 1);
1962 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1963 				   ELEMENT_SIZE, 1);
1964 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1965 				   INDEX_STRIDE, 3);
1966 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
1967 
1968 	mutex_lock(&adev->srbm_mutex);
1969 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
1970 		if (i == 0)
1971 			sh_mem_base = 0;
1972 		else
1973 			sh_mem_base = adev->gmc.shared_aperture_start >> 48;
1974 		cik_srbm_select(adev, 0, 0, 0, i);
1975 		/* CP and shaders */
1976 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1977 		WREG32(mmSH_MEM_APE1_BASE, 1);
1978 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1979 		WREG32(mmSH_MEM_BASES, sh_mem_base);
1980 	}
1981 	cik_srbm_select(adev, 0, 0, 0, 0);
1982 	mutex_unlock(&adev->srbm_mutex);
1983 
1984 	gfx_v7_0_init_compute_vmid(adev);
1985 	gfx_v7_0_init_gds_vmid(adev);
1986 
1987 	WREG32(mmSX_DEBUG_1, 0x20);
1988 
1989 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1990 
1991 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1992 	tmp |= 0x03000000;
1993 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1994 
1995 	WREG32(mmSQ_CONFIG, 1);
1996 
1997 	WREG32(mmDB_DEBUG, 0);
1998 
1999 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
2000 	tmp |= 0x00000400;
2001 	WREG32(mmDB_DEBUG2, tmp);
2002 
2003 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
2004 	tmp |= 0x00020200;
2005 	WREG32(mmDB_DEBUG3, tmp);
2006 
2007 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
2008 	tmp |= 0x00018208;
2009 	WREG32(mmCB_HW_CONTROL, tmp);
2010 
2011 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
2012 
2013 	WREG32(mmPA_SC_FIFO_SIZE,
2014 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
2015 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
2016 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
2017 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
2018 
2019 	WREG32(mmVGT_NUM_INSTANCES, 1);
2020 
2021 	WREG32(mmCP_PERFMON_CNTL, 0);
2022 
2023 	WREG32(mmSQ_CONFIG, 0);
2024 
2025 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
2026 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
2027 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
2028 
2029 	WREG32(mmVGT_CACHE_INVALIDATION,
2030 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
2031 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
2032 
2033 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
2034 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
2035 
2036 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
2037 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
2038 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
2039 
2040 	tmp = RREG32(mmSPI_ARB_PRIORITY);
2041 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
2042 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
2043 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
2044 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
2045 	WREG32(mmSPI_ARB_PRIORITY, tmp);
2046 
2047 	mutex_unlock(&adev->grbm_idx_mutex);
2048 
2049 	udelay(50);
2050 }
2051 
2052 /**
2053  * gfx_v7_0_ring_test_ring - basic gfx ring test
2054  *
2055  * @ring: amdgpu_ring structure holding ring information
2056  *
2057  * Allocate a scratch register and write to it using the gfx ring (CIK).
2058  * Provides a basic gfx ring test to verify that the ring is working.
2059  * Used by gfx_v7_0_cp_gfx_resume();
2060  * Returns 0 on success, error on failure.
2061  */
2062 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2063 {
2064 	struct amdgpu_device *adev = ring->adev;
2065 	uint32_t tmp = 0;
2066 	unsigned i;
2067 	int r;
2068 
2069 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2070 	r = amdgpu_ring_alloc(ring, 3);
2071 	if (r)
2072 		return r;
2073 
2074 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2075 	amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START);
2076 	amdgpu_ring_write(ring, 0xDEADBEEF);
2077 	amdgpu_ring_commit(ring);
2078 
2079 	for (i = 0; i < adev->usec_timeout; i++) {
2080 		tmp = RREG32(mmSCRATCH_REG0);
2081 		if (tmp == 0xDEADBEEF)
2082 			break;
2083 		udelay(1);
2084 	}
2085 	if (i >= adev->usec_timeout)
2086 		r = -ETIMEDOUT;
2087 	return r;
2088 }
2089 
2090 /**
2091  * gfx_v7_0_ring_emit_hdp_flush - emit an hdp flush on the cp
2092  *
2093  * @ring: amdgpu_ring structure holding ring information
2094  *
2095  * Emits an hdp flush on the cp.
2096  */
2097 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2098 {
2099 	u32 ref_and_mask;
2100 	int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2101 
2102 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2103 		switch (ring->me) {
2104 		case 1:
2105 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2106 			break;
2107 		case 2:
2108 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2109 			break;
2110 		default:
2111 			return;
2112 		}
2113 	} else {
2114 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2115 	}
2116 
2117 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2118 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2119 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2120 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2121 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2122 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2123 	amdgpu_ring_write(ring, ref_and_mask);
2124 	amdgpu_ring_write(ring, ref_and_mask);
2125 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2126 }
2127 
2128 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2129 {
2130 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2131 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2132 		EVENT_INDEX(4));
2133 
2134 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2135 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2136 		EVENT_INDEX(0));
2137 }
2138 
2139 /**
2140  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2141  *
2142  * @ring: amdgpu_ring structure holding ring information
2143  * @addr: address
2144  * @seq: sequence number
2145  * @flags: fence related flags
2146  *
2147  * Emits a fence sequence number on the gfx ring and flushes
2148  * GPU caches.
2149  */
2150 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2151 					 u64 seq, unsigned flags)
2152 {
2153 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2154 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2155 	/* Workaround for cache flush problems. First send a dummy EOP
2156 	 * event down the pipe with seq one below.
2157 	 */
2158 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2159 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2160 				 EOP_TC_ACTION_EN |
2161 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2162 				 EVENT_INDEX(5)));
2163 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2164 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2165 				DATA_SEL(1) | INT_SEL(0));
2166 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2167 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2168 
2169 	/* Then send the real EOP event down the pipe. */
2170 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2171 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2172 				 EOP_TC_ACTION_EN |
2173 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2174 				 EVENT_INDEX(5)));
2175 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2176 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2177 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2178 	amdgpu_ring_write(ring, lower_32_bits(seq));
2179 	amdgpu_ring_write(ring, upper_32_bits(seq));
2180 }
2181 
2182 /**
2183  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2184  *
2185  * @ring: amdgpu_ring structure holding ring information
2186  * @addr: address
2187  * @seq: sequence number
2188  * @flags: fence related flags
2189  *
2190  * Emits a fence sequence number on the compute ring and flushes
2191  * GPU caches.
2192  */
2193 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2194 					     u64 addr, u64 seq,
2195 					     unsigned flags)
2196 {
2197 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2198 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2199 
2200 	/* RELEASE_MEM - flush caches, send int */
2201 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2202 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2203 				 EOP_TC_ACTION_EN |
2204 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2205 				 EVENT_INDEX(5)));
2206 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2207 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2208 	amdgpu_ring_write(ring, upper_32_bits(addr));
2209 	amdgpu_ring_write(ring, lower_32_bits(seq));
2210 	amdgpu_ring_write(ring, upper_32_bits(seq));
2211 }
2212 
2213 /*
2214  * IB stuff
2215  */
2216 /**
2217  * gfx_v7_0_ring_emit_ib_gfx - emit an IB (Indirect Buffer) on the ring
2218  *
2219  * @ring: amdgpu_ring structure holding ring information
2220  * @job: job to retrieve vmid from
2221  * @ib: amdgpu indirect buffer object
2222  * @flags: options (AMDGPU_HAVE_CTX_SWITCH)
2223  *
2224  * Emits an DE (drawing engine) or CE (constant engine) IB
2225  * on the gfx ring.  IBs are usually generated by userspace
2226  * acceleration drivers and submitted to the kernel for
2227  * scheduling on the ring.  This function schedules the IB
2228  * on the gfx ring for execution by the GPU.
2229  */
2230 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2231 					struct amdgpu_job *job,
2232 					struct amdgpu_ib *ib,
2233 					uint32_t flags)
2234 {
2235 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2236 	u32 header, control = 0;
2237 
2238 	/* insert SWITCH_BUFFER packet before first IB in the ring frame */
2239 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2240 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2241 		amdgpu_ring_write(ring, 0);
2242 	}
2243 
2244 	if (ib->flags & AMDGPU_IB_FLAG_CE)
2245 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2246 	else
2247 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2248 
2249 	control |= ib->length_dw | (vmid << 24);
2250 
2251 	amdgpu_ring_write(ring, header);
2252 	amdgpu_ring_write(ring,
2253 #ifdef __BIG_ENDIAN
2254 			  (2 << 0) |
2255 #endif
2256 			  (ib->gpu_addr & 0xFFFFFFFC));
2257 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2258 	amdgpu_ring_write(ring, control);
2259 }
2260 
2261 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
2262 					  struct amdgpu_job *job,
2263 					  struct amdgpu_ib *ib,
2264 					  uint32_t flags)
2265 {
2266 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2267 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
2268 
2269 	/* Currently, there is a high possibility to get wave ID mismatch
2270 	 * between ME and GDS, leading to a hw deadlock, because ME generates
2271 	 * different wave IDs than the GDS expects. This situation happens
2272 	 * randomly when at least 5 compute pipes use GDS ordered append.
2273 	 * The wave IDs generated by ME are also wrong after suspend/resume.
2274 	 * Those are probably bugs somewhere else in the kernel driver.
2275 	 *
2276 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
2277 	 * GDS to 0 for this ring (me/pipe).
2278 	 */
2279 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
2280 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2281 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
2282 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
2283 	}
2284 
2285 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2286 	amdgpu_ring_write(ring,
2287 #ifdef __BIG_ENDIAN
2288 					  (2 << 0) |
2289 #endif
2290 					  (ib->gpu_addr & 0xFFFFFFFC));
2291 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2292 	amdgpu_ring_write(ring, control);
2293 }
2294 
2295 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2296 {
2297 	uint32_t dw2 = 0;
2298 
2299 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2300 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2301 		gfx_v7_0_ring_emit_vgt_flush(ring);
2302 		/* set load_global_config & load_global_uconfig */
2303 		dw2 |= 0x8001;
2304 		/* set load_cs_sh_regs */
2305 		dw2 |= 0x01000000;
2306 		/* set load_per_context_state & load_gfx_sh_regs */
2307 		dw2 |= 0x10002;
2308 	}
2309 
2310 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2311 	amdgpu_ring_write(ring, dw2);
2312 	amdgpu_ring_write(ring, 0);
2313 }
2314 
2315 /**
2316  * gfx_v7_0_ring_test_ib - basic ring IB test
2317  *
2318  * @ring: amdgpu_ring structure holding ring information
2319  * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT
2320  *
2321  * Allocate an IB and execute it on the gfx ring (CIK).
2322  * Provides a basic gfx ring test to verify that IBs are working.
2323  * Returns 0 on success, error on failure.
2324  */
2325 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2326 {
2327 	struct amdgpu_device *adev = ring->adev;
2328 	struct amdgpu_ib ib;
2329 	struct dma_fence *f = NULL;
2330 	uint32_t tmp = 0;
2331 	long r;
2332 
2333 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2334 	memset(&ib, 0, sizeof(ib));
2335 	r = amdgpu_ib_get(adev, NULL, 256, AMDGPU_IB_POOL_DIRECT, &ib);
2336 	if (r)
2337 		return r;
2338 
2339 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2340 	ib.ptr[1] = mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START;
2341 	ib.ptr[2] = 0xDEADBEEF;
2342 	ib.length_dw = 3;
2343 
2344 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2345 	if (r)
2346 		goto error;
2347 
2348 	r = dma_fence_wait_timeout(f, false, timeout);
2349 	if (r == 0) {
2350 		r = -ETIMEDOUT;
2351 		goto error;
2352 	} else if (r < 0) {
2353 		goto error;
2354 	}
2355 	tmp = RREG32(mmSCRATCH_REG0);
2356 	if (tmp == 0xDEADBEEF)
2357 		r = 0;
2358 	else
2359 		r = -EINVAL;
2360 
2361 error:
2362 	amdgpu_ib_free(adev, &ib, NULL);
2363 	dma_fence_put(f);
2364 	return r;
2365 }
2366 
2367 /*
2368  * CP.
2369  * On CIK, gfx and compute now have independent command processors.
2370  *
2371  * GFX
2372  * Gfx consists of a single ring and can process both gfx jobs and
2373  * compute jobs.  The gfx CP consists of three microengines (ME):
2374  * PFP - Pre-Fetch Parser
2375  * ME - Micro Engine
2376  * CE - Constant Engine
2377  * The PFP and ME make up what is considered the Drawing Engine (DE).
2378  * The CE is an asynchronous engine used for updating buffer desciptors
2379  * used by the DE so that they can be loaded into cache in parallel
2380  * while the DE is processing state update packets.
2381  *
2382  * Compute
2383  * The compute CP consists of two microengines (ME):
2384  * MEC1 - Compute MicroEngine 1
2385  * MEC2 - Compute MicroEngine 2
2386  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2387  * The queues are exposed to userspace and are programmed directly
2388  * by the compute runtime.
2389  */
2390 /**
2391  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2392  *
2393  * @adev: amdgpu_device pointer
2394  * @enable: enable or disable the MEs
2395  *
2396  * Halts or unhalts the gfx MEs.
2397  */
2398 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2399 {
2400 	if (enable)
2401 		WREG32(mmCP_ME_CNTL, 0);
2402 	else
2403 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK |
2404 				      CP_ME_CNTL__PFP_HALT_MASK |
2405 				      CP_ME_CNTL__CE_HALT_MASK));
2406 	udelay(50);
2407 }
2408 
2409 /**
2410  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2411  *
2412  * @adev: amdgpu_device pointer
2413  *
2414  * Loads the gfx PFP, ME, and CE ucode.
2415  * Returns 0 for success, -EINVAL if the ucode is not available.
2416  */
2417 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2418 {
2419 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2420 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2421 	const struct gfx_firmware_header_v1_0 *me_hdr;
2422 	const __le32 *fw_data;
2423 	unsigned i, fw_size;
2424 
2425 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2426 		return -EINVAL;
2427 
2428 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2429 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2430 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2431 
2432 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2433 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2434 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2435 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2436 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2437 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2438 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2439 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2440 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2441 
2442 	gfx_v7_0_cp_gfx_enable(adev, false);
2443 
2444 	/* PFP */
2445 	fw_data = (const __le32 *)
2446 		(adev->gfx.pfp_fw->data +
2447 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2448 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2449 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2450 	for (i = 0; i < fw_size; i++)
2451 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2452 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2453 
2454 	/* CE */
2455 	fw_data = (const __le32 *)
2456 		(adev->gfx.ce_fw->data +
2457 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2458 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2459 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2460 	for (i = 0; i < fw_size; i++)
2461 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2462 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2463 
2464 	/* ME */
2465 	fw_data = (const __le32 *)
2466 		(adev->gfx.me_fw->data +
2467 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2468 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2469 	WREG32(mmCP_ME_RAM_WADDR, 0);
2470 	for (i = 0; i < fw_size; i++)
2471 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2472 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2473 
2474 	return 0;
2475 }
2476 
2477 /**
2478  * gfx_v7_0_cp_gfx_start - start the gfx ring
2479  *
2480  * @adev: amdgpu_device pointer
2481  *
2482  * Enables the ring and loads the clear state context and other
2483  * packets required to init the ring.
2484  * Returns 0 for success, error for failure.
2485  */
2486 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2487 {
2488 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2489 	const struct cs_section_def *sect = NULL;
2490 	const struct cs_extent_def *ext = NULL;
2491 	int r, i;
2492 
2493 	/* init the CP */
2494 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2495 	WREG32(mmCP_ENDIAN_SWAP, 0);
2496 	WREG32(mmCP_DEVICE_ID, 1);
2497 
2498 	gfx_v7_0_cp_gfx_enable(adev, true);
2499 
2500 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2501 	if (r) {
2502 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2503 		return r;
2504 	}
2505 
2506 	/* init the CE partitions.  CE only used for gfx on CIK */
2507 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2508 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2509 	amdgpu_ring_write(ring, 0x8000);
2510 	amdgpu_ring_write(ring, 0x8000);
2511 
2512 	/* clear state buffer */
2513 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2514 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2515 
2516 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2517 	amdgpu_ring_write(ring, 0x80000000);
2518 	amdgpu_ring_write(ring, 0x80000000);
2519 
2520 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2521 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2522 			if (sect->id == SECT_CONTEXT) {
2523 				amdgpu_ring_write(ring,
2524 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2525 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2526 				for (i = 0; i < ext->reg_count; i++)
2527 					amdgpu_ring_write(ring, ext->extent[i]);
2528 			}
2529 		}
2530 	}
2531 
2532 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2533 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2534 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
2535 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
2536 
2537 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2538 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2539 
2540 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2541 	amdgpu_ring_write(ring, 0);
2542 
2543 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2544 	amdgpu_ring_write(ring, 0x00000316);
2545 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2546 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2547 
2548 	amdgpu_ring_commit(ring);
2549 
2550 	return 0;
2551 }
2552 
2553 /**
2554  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2555  *
2556  * @adev: amdgpu_device pointer
2557  *
2558  * Program the location and size of the gfx ring buffer
2559  * and test it to make sure it's working.
2560  * Returns 0 for success, error for failure.
2561  */
2562 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2563 {
2564 	struct amdgpu_ring *ring;
2565 	u32 tmp;
2566 	u32 rb_bufsz;
2567 	u64 rb_addr, rptr_addr;
2568 	int r;
2569 
2570 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2571 	if (adev->asic_type != CHIP_HAWAII)
2572 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2573 
2574 	/* Set the write pointer delay */
2575 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2576 
2577 	/* set the RB to use vmid 0 */
2578 	WREG32(mmCP_RB_VMID, 0);
2579 
2580 	WREG32(mmSCRATCH_ADDR, 0);
2581 
2582 	/* ring 0 - compute and gfx */
2583 	/* Set ring buffer size */
2584 	ring = &adev->gfx.gfx_ring[0];
2585 	rb_bufsz = order_base_2(ring->ring_size / 8);
2586 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2587 #ifdef __BIG_ENDIAN
2588 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2589 #endif
2590 	WREG32(mmCP_RB0_CNTL, tmp);
2591 
2592 	/* Initialize the ring buffer's read and write pointers */
2593 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2594 	ring->wptr = 0;
2595 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2596 
2597 	/* set the wb address wether it's enabled or not */
2598 	rptr_addr = ring->rptr_gpu_addr;
2599 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2600 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2601 
2602 	/* scratch register shadowing is no longer supported */
2603 	WREG32(mmSCRATCH_UMSK, 0);
2604 
2605 	mdelay(1);
2606 	WREG32(mmCP_RB0_CNTL, tmp);
2607 
2608 	rb_addr = ring->gpu_addr >> 8;
2609 	WREG32(mmCP_RB0_BASE, rb_addr);
2610 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2611 
2612 	/* start the ring */
2613 	gfx_v7_0_cp_gfx_start(adev);
2614 	r = amdgpu_ring_test_helper(ring);
2615 	if (r)
2616 		return r;
2617 
2618 	return 0;
2619 }
2620 
2621 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2622 {
2623 	return *ring->rptr_cpu_addr;
2624 }
2625 
2626 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2627 {
2628 	struct amdgpu_device *adev = ring->adev;
2629 
2630 	return RREG32(mmCP_RB0_WPTR);
2631 }
2632 
2633 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2634 {
2635 	struct amdgpu_device *adev = ring->adev;
2636 
2637 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2638 	(void)RREG32(mmCP_RB0_WPTR);
2639 }
2640 
2641 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2642 {
2643 	/* XXX check if swapping is necessary on BE */
2644 	return *ring->wptr_cpu_addr;
2645 }
2646 
2647 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2648 {
2649 	struct amdgpu_device *adev = ring->adev;
2650 
2651 	/* XXX check if swapping is necessary on BE */
2652 	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
2653 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2654 }
2655 
2656 /**
2657  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2658  *
2659  * @adev: amdgpu_device pointer
2660  * @enable: enable or disable the MEs
2661  *
2662  * Halts or unhalts the compute MEs.
2663  */
2664 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2665 {
2666 	if (enable)
2667 		WREG32(mmCP_MEC_CNTL, 0);
2668 	else
2669 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK |
2670 				       CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2671 	udelay(50);
2672 }
2673 
2674 /**
2675  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2676  *
2677  * @adev: amdgpu_device pointer
2678  *
2679  * Loads the compute MEC1&2 ucode.
2680  * Returns 0 for success, -EINVAL if the ucode is not available.
2681  */
2682 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2683 {
2684 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2685 	const __le32 *fw_data;
2686 	unsigned i, fw_size;
2687 
2688 	if (!adev->gfx.mec_fw)
2689 		return -EINVAL;
2690 
2691 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2692 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2693 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2694 	adev->gfx.mec_feature_version = le32_to_cpu(
2695 					mec_hdr->ucode_feature_version);
2696 
2697 	gfx_v7_0_cp_compute_enable(adev, false);
2698 
2699 	/* MEC1 */
2700 	fw_data = (const __le32 *)
2701 		(adev->gfx.mec_fw->data +
2702 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2703 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2704 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2705 	for (i = 0; i < fw_size; i++)
2706 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2707 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2708 
2709 	if (adev->asic_type == CHIP_KAVERI) {
2710 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2711 
2712 		if (!adev->gfx.mec2_fw)
2713 			return -EINVAL;
2714 
2715 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2716 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2717 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2718 		adev->gfx.mec2_feature_version = le32_to_cpu(
2719 				mec2_hdr->ucode_feature_version);
2720 
2721 		/* MEC2 */
2722 		fw_data = (const __le32 *)
2723 			(adev->gfx.mec2_fw->data +
2724 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2725 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2726 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2727 		for (i = 0; i < fw_size; i++)
2728 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2729 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2730 	}
2731 
2732 	return 0;
2733 }
2734 
2735 /**
2736  * gfx_v7_0_cp_compute_fini - stop the compute queues
2737  *
2738  * @adev: amdgpu_device pointer
2739  *
2740  * Stop the compute queues and tear down the driver queue
2741  * info.
2742  */
2743 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2744 {
2745 	int i;
2746 
2747 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2748 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2749 
2750 		amdgpu_bo_free_kernel(&ring->mqd_obj, NULL, NULL);
2751 	}
2752 }
2753 
2754 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2755 {
2756 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
2757 }
2758 
2759 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2760 {
2761 	int r;
2762 	u32 *hpd;
2763 	size_t mec_hpd_size;
2764 
2765 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
2766 
2767 	/* take ownership of the relevant compute queues */
2768 	amdgpu_gfx_compute_queue_acquire(adev);
2769 
2770 	/* allocate space for ALL pipes (even the ones we don't own) */
2771 	mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec
2772 		* GFX7_MEC_HPD_SIZE * 2;
2773 
2774 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
2775 				      AMDGPU_GEM_DOMAIN_VRAM,
2776 				      &adev->gfx.mec.hpd_eop_obj,
2777 				      &adev->gfx.mec.hpd_eop_gpu_addr,
2778 				      (void **)&hpd);
2779 	if (r) {
2780 		dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r);
2781 		gfx_v7_0_mec_fini(adev);
2782 		return r;
2783 	}
2784 
2785 	/* clear memory.  Not sure if this is required or not */
2786 	memset(hpd, 0, mec_hpd_size);
2787 
2788 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2789 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2790 
2791 	return 0;
2792 }
2793 
2794 struct hqd_registers
2795 {
2796 	u32 cp_mqd_base_addr;
2797 	u32 cp_mqd_base_addr_hi;
2798 	u32 cp_hqd_active;
2799 	u32 cp_hqd_vmid;
2800 	u32 cp_hqd_persistent_state;
2801 	u32 cp_hqd_pipe_priority;
2802 	u32 cp_hqd_queue_priority;
2803 	u32 cp_hqd_quantum;
2804 	u32 cp_hqd_pq_base;
2805 	u32 cp_hqd_pq_base_hi;
2806 	u32 cp_hqd_pq_rptr;
2807 	u32 cp_hqd_pq_rptr_report_addr;
2808 	u32 cp_hqd_pq_rptr_report_addr_hi;
2809 	u32 cp_hqd_pq_wptr_poll_addr;
2810 	u32 cp_hqd_pq_wptr_poll_addr_hi;
2811 	u32 cp_hqd_pq_doorbell_control;
2812 	u32 cp_hqd_pq_wptr;
2813 	u32 cp_hqd_pq_control;
2814 	u32 cp_hqd_ib_base_addr;
2815 	u32 cp_hqd_ib_base_addr_hi;
2816 	u32 cp_hqd_ib_rptr;
2817 	u32 cp_hqd_ib_control;
2818 	u32 cp_hqd_iq_timer;
2819 	u32 cp_hqd_iq_rptr;
2820 	u32 cp_hqd_dequeue_request;
2821 	u32 cp_hqd_dma_offload;
2822 	u32 cp_hqd_sema_cmd;
2823 	u32 cp_hqd_msg_type;
2824 	u32 cp_hqd_atomic0_preop_lo;
2825 	u32 cp_hqd_atomic0_preop_hi;
2826 	u32 cp_hqd_atomic1_preop_lo;
2827 	u32 cp_hqd_atomic1_preop_hi;
2828 	u32 cp_hqd_hq_scheduler0;
2829 	u32 cp_hqd_hq_scheduler1;
2830 	u32 cp_mqd_control;
2831 };
2832 
2833 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev,
2834 				       int mec, int pipe)
2835 {
2836 	u64 eop_gpu_addr;
2837 	u32 tmp;
2838 	size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe)
2839 			    * GFX7_MEC_HPD_SIZE * 2;
2840 
2841 	mutex_lock(&adev->srbm_mutex);
2842 	eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset;
2843 
2844 	cik_srbm_select(adev, mec + 1, pipe, 0, 0);
2845 
2846 	/* write the EOP addr */
2847 	WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2848 	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2849 
2850 	/* set the VMID assigned */
2851 	WREG32(mmCP_HPD_EOP_VMID, 0);
2852 
2853 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2854 	tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2855 	tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2856 	tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8);
2857 	WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2858 
2859 	cik_srbm_select(adev, 0, 0, 0, 0);
2860 	mutex_unlock(&adev->srbm_mutex);
2861 }
2862 
2863 static int gfx_v7_0_mqd_deactivate(struct amdgpu_device *adev)
2864 {
2865 	int i;
2866 
2867 	/* disable the queue if it's active */
2868 	if (RREG32(mmCP_HQD_ACTIVE) & 1) {
2869 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
2870 		for (i = 0; i < adev->usec_timeout; i++) {
2871 			if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
2872 				break;
2873 			udelay(1);
2874 		}
2875 
2876 		if (i == adev->usec_timeout)
2877 			return -ETIMEDOUT;
2878 
2879 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
2880 		WREG32(mmCP_HQD_PQ_RPTR, 0);
2881 		WREG32(mmCP_HQD_PQ_WPTR, 0);
2882 	}
2883 
2884 	return 0;
2885 }
2886 
2887 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev,
2888 			     struct cik_mqd *mqd,
2889 			     uint64_t mqd_gpu_addr,
2890 			     struct amdgpu_ring *ring)
2891 {
2892 	u64 hqd_gpu_addr;
2893 	u64 wb_gpu_addr;
2894 
2895 	/* init the mqd struct */
2896 	memset(mqd, 0, sizeof(struct cik_mqd));
2897 
2898 	mqd->header = 0xC0310800;
2899 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2900 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2901 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2902 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2903 
2904 	/* enable doorbell? */
2905 	mqd->cp_hqd_pq_doorbell_control =
2906 		RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2907 	if (ring->use_doorbell)
2908 		mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2909 	else
2910 		mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2911 
2912 	/* set the pointer to the MQD */
2913 	mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc;
2914 	mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
2915 
2916 	/* set MQD vmid to 0 */
2917 	mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
2918 	mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
2919 
2920 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2921 	hqd_gpu_addr = ring->gpu_addr >> 8;
2922 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2923 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2924 
2925 	/* set up the HQD, this is similar to CP_RB0_CNTL */
2926 	mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
2927 	mqd->cp_hqd_pq_control &=
2928 		~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
2929 				CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
2930 
2931 	mqd->cp_hqd_pq_control |=
2932 		order_base_2(ring->ring_size / 8);
2933 	mqd->cp_hqd_pq_control |=
2934 		(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
2935 #ifdef __BIG_ENDIAN
2936 	mqd->cp_hqd_pq_control |=
2937 		2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
2938 #endif
2939 	mqd->cp_hqd_pq_control &=
2940 		~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
2941 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
2942 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
2943 	mqd->cp_hqd_pq_control |=
2944 		CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
2945 		CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
2946 
2947 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2948 	wb_gpu_addr = ring->wptr_gpu_addr;
2949 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2950 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2951 
2952 	/* set the wb address wether it's enabled or not */
2953 	wb_gpu_addr = ring->rptr_gpu_addr;
2954 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2955 	mqd->cp_hqd_pq_rptr_report_addr_hi =
2956 		upper_32_bits(wb_gpu_addr) & 0xffff;
2957 
2958 	/* enable the doorbell if requested */
2959 	if (ring->use_doorbell) {
2960 		mqd->cp_hqd_pq_doorbell_control =
2961 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2962 		mqd->cp_hqd_pq_doorbell_control &=
2963 			~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
2964 		mqd->cp_hqd_pq_doorbell_control |=
2965 			(ring->doorbell_index <<
2966 			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
2967 		mqd->cp_hqd_pq_doorbell_control |=
2968 			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2969 		mqd->cp_hqd_pq_doorbell_control &=
2970 			~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
2971 					CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
2972 
2973 	} else {
2974 		mqd->cp_hqd_pq_doorbell_control = 0;
2975 	}
2976 
2977 	/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
2978 	ring->wptr = 0;
2979 	mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr);
2980 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2981 
2982 	/* set the vmid for the queue */
2983 	mqd->cp_hqd_vmid = 0;
2984 
2985 	/* defaults */
2986 	mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL);
2987 	mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR);
2988 	mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI);
2989 	mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR);
2990 	mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE);
2991 	mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD);
2992 	mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE);
2993 	mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO);
2994 	mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI);
2995 	mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO);
2996 	mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI);
2997 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2998 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
2999 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
3000 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
3001 	mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR);
3002 
3003 	/* activate the queue */
3004 	mqd->cp_hqd_active = 1;
3005 }
3006 
3007 static int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd)
3008 {
3009 	uint32_t tmp;
3010 	uint32_t mqd_reg;
3011 	uint32_t *mqd_data;
3012 
3013 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */
3014 	mqd_data = &mqd->cp_mqd_base_addr_lo;
3015 
3016 	/* disable wptr polling */
3017 	tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
3018 	tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3019 	WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
3020 
3021 	/* program all HQD registers */
3022 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++)
3023 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3024 
3025 	/* activate the HQD */
3026 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
3027 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3028 
3029 	return 0;
3030 }
3031 
3032 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id)
3033 {
3034 	int r;
3035 	u64 mqd_gpu_addr;
3036 	struct cik_mqd *mqd;
3037 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
3038 
3039 	r = amdgpu_bo_create_reserved(adev, sizeof(struct cik_mqd), PAGE_SIZE,
3040 				      AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj,
3041 				      &mqd_gpu_addr, (void **)&mqd);
3042 	if (r) {
3043 		dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
3044 		return r;
3045 	}
3046 
3047 	mutex_lock(&adev->srbm_mutex);
3048 	cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3049 
3050 	gfx_v7_0_mqd_init(adev, mqd, mqd_gpu_addr, ring);
3051 	gfx_v7_0_mqd_deactivate(adev);
3052 	gfx_v7_0_mqd_commit(adev, mqd);
3053 
3054 	cik_srbm_select(adev, 0, 0, 0, 0);
3055 	mutex_unlock(&adev->srbm_mutex);
3056 
3057 	amdgpu_bo_kunmap(ring->mqd_obj);
3058 	amdgpu_bo_unreserve(ring->mqd_obj);
3059 	return 0;
3060 }
3061 
3062 /**
3063  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
3064  *
3065  * @adev: amdgpu_device pointer
3066  *
3067  * Program the compute queues and test them to make sure they
3068  * are working.
3069  * Returns 0 for success, error for failure.
3070  */
3071 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
3072 {
3073 	int r, i, j;
3074 	u32 tmp;
3075 	struct amdgpu_ring *ring;
3076 
3077 	/* fix up chicken bits */
3078 	tmp = RREG32(mmCP_CPF_DEBUG);
3079 	tmp |= (1 << 23);
3080 	WREG32(mmCP_CPF_DEBUG, tmp);
3081 
3082 	/* init all pipes (even the ones we don't own) */
3083 	for (i = 0; i < adev->gfx.mec.num_mec; i++)
3084 		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++)
3085 			gfx_v7_0_compute_pipe_init(adev, i, j);
3086 
3087 	/* init the queues */
3088 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3089 		r = gfx_v7_0_compute_queue_init(adev, i);
3090 		if (r) {
3091 			gfx_v7_0_cp_compute_fini(adev);
3092 			return r;
3093 		}
3094 	}
3095 
3096 	gfx_v7_0_cp_compute_enable(adev, true);
3097 
3098 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3099 		ring = &adev->gfx.compute_ring[i];
3100 		amdgpu_ring_test_helper(ring);
3101 	}
3102 
3103 	return 0;
3104 }
3105 
3106 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3107 {
3108 	gfx_v7_0_cp_gfx_enable(adev, enable);
3109 	gfx_v7_0_cp_compute_enable(adev, enable);
3110 }
3111 
3112 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3113 {
3114 	int r;
3115 
3116 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3117 	if (r)
3118 		return r;
3119 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3120 	if (r)
3121 		return r;
3122 
3123 	return 0;
3124 }
3125 
3126 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3127 					       bool enable)
3128 {
3129 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3130 
3131 	if (enable)
3132 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3133 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3134 	else
3135 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3136 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3137 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3138 }
3139 
3140 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3141 {
3142 	int r;
3143 
3144 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3145 
3146 	r = gfx_v7_0_cp_load_microcode(adev);
3147 	if (r)
3148 		return r;
3149 
3150 	r = gfx_v7_0_cp_gfx_resume(adev);
3151 	if (r)
3152 		return r;
3153 	r = gfx_v7_0_cp_compute_resume(adev);
3154 	if (r)
3155 		return r;
3156 
3157 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3158 
3159 	return 0;
3160 }
3161 
3162 /**
3163  * gfx_v7_0_ring_emit_pipeline_sync - cik vm flush using the CP
3164  *
3165  * @ring: the ring to emit the commands to
3166  *
3167  * Sync the command pipeline with the PFP. E.g. wait for everything
3168  * to be completed.
3169  */
3170 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3171 {
3172 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3173 	uint32_t seq = ring->fence_drv.sync_seq;
3174 	uint64_t addr = ring->fence_drv.gpu_addr;
3175 
3176 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3177 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3178 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3179 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3180 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3181 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3182 	amdgpu_ring_write(ring, seq);
3183 	amdgpu_ring_write(ring, 0xffffffff);
3184 	amdgpu_ring_write(ring, 4); /* poll interval */
3185 
3186 	if (usepfp) {
3187 		/* sync CE with ME to prevent CE fetch CEIB before context switch done */
3188 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3189 		amdgpu_ring_write(ring, 0);
3190 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3191 		amdgpu_ring_write(ring, 0);
3192 	}
3193 }
3194 
3195 /*
3196  * vm
3197  * VMID 0 is the physical GPU addresses as used by the kernel.
3198  * VMIDs 1-15 are used for userspace clients and are handled
3199  * by the amdgpu vm/hsa code.
3200  */
3201 /**
3202  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3203  *
3204  * @ring: amdgpu_ring pointer
3205  * @vmid: vmid number to use
3206  * @pd_addr: address
3207  *
3208  * Update the page table base and flush the VM TLB
3209  * using the CP (CIK).
3210  */
3211 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3212 					unsigned vmid, uint64_t pd_addr)
3213 {
3214 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3215 
3216 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3217 
3218 	/* wait for the invalidate to complete */
3219 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3220 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3221 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3222 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3223 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3224 	amdgpu_ring_write(ring, 0);
3225 	amdgpu_ring_write(ring, 0); /* ref */
3226 	amdgpu_ring_write(ring, 0); /* mask */
3227 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3228 
3229 	/* compute doesn't have PFP */
3230 	if (usepfp) {
3231 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3232 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3233 		amdgpu_ring_write(ring, 0x0);
3234 
3235 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3236 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3237 		amdgpu_ring_write(ring, 0);
3238 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3239 		amdgpu_ring_write(ring, 0);
3240 	}
3241 }
3242 
3243 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
3244 				    uint32_t reg, uint32_t val)
3245 {
3246 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3247 
3248 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3249 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3250 				 WRITE_DATA_DST_SEL(0)));
3251 	amdgpu_ring_write(ring, reg);
3252 	amdgpu_ring_write(ring, 0);
3253 	amdgpu_ring_write(ring, val);
3254 }
3255 
3256 /*
3257  * RLC
3258  * The RLC is a multi-purpose microengine that handles a
3259  * variety of functions.
3260  */
3261 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3262 {
3263 	const u32 *src_ptr;
3264 	u32 dws;
3265 	const struct cs_section_def *cs_data;
3266 	int r;
3267 
3268 	/* allocate rlc buffers */
3269 	if (adev->flags & AMD_IS_APU) {
3270 		if (adev->asic_type == CHIP_KAVERI) {
3271 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3272 			adev->gfx.rlc.reg_list_size =
3273 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3274 		} else {
3275 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3276 			adev->gfx.rlc.reg_list_size =
3277 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3278 		}
3279 	}
3280 	adev->gfx.rlc.cs_data = ci_cs_data;
3281 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3282 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3283 
3284 	src_ptr = adev->gfx.rlc.reg_list;
3285 	dws = adev->gfx.rlc.reg_list_size;
3286 	dws += (5 * 16) + 48 + 48 + 64;
3287 
3288 	cs_data = adev->gfx.rlc.cs_data;
3289 
3290 	if (src_ptr) {
3291 		/* init save restore block */
3292 		r = amdgpu_gfx_rlc_init_sr(adev, dws);
3293 		if (r)
3294 			return r;
3295 	}
3296 
3297 	if (cs_data) {
3298 		/* init clear state block */
3299 		r = amdgpu_gfx_rlc_init_csb(adev);
3300 		if (r)
3301 			return r;
3302 	}
3303 
3304 	if (adev->gfx.rlc.cp_table_size) {
3305 		r = amdgpu_gfx_rlc_init_cpt(adev);
3306 		if (r)
3307 			return r;
3308 	}
3309 
3310 	/* init spm vmid with 0xf */
3311 	if (adev->gfx.rlc.funcs->update_spm_vmid)
3312 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
3313 
3314 	return 0;
3315 }
3316 
3317 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3318 {
3319 	u32 tmp;
3320 
3321 	tmp = RREG32(mmRLC_LB_CNTL);
3322 	if (enable)
3323 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3324 	else
3325 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3326 	WREG32(mmRLC_LB_CNTL, tmp);
3327 }
3328 
3329 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3330 {
3331 	u32 i, j, k;
3332 	u32 mask;
3333 
3334 	mutex_lock(&adev->grbm_idx_mutex);
3335 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3336 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3337 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
3338 			for (k = 0; k < adev->usec_timeout; k++) {
3339 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3340 					break;
3341 				udelay(1);
3342 			}
3343 		}
3344 	}
3345 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3346 	mutex_unlock(&adev->grbm_idx_mutex);
3347 
3348 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3349 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3350 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3351 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3352 	for (k = 0; k < adev->usec_timeout; k++) {
3353 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3354 			break;
3355 		udelay(1);
3356 	}
3357 }
3358 
3359 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3360 {
3361 	u32 tmp;
3362 
3363 	tmp = RREG32(mmRLC_CNTL);
3364 	if (tmp != rlc)
3365 		WREG32(mmRLC_CNTL, rlc);
3366 }
3367 
3368 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3369 {
3370 	u32 data, orig;
3371 
3372 	orig = data = RREG32(mmRLC_CNTL);
3373 
3374 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3375 		u32 i;
3376 
3377 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3378 		WREG32(mmRLC_CNTL, data);
3379 
3380 		for (i = 0; i < adev->usec_timeout; i++) {
3381 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3382 				break;
3383 			udelay(1);
3384 		}
3385 
3386 		gfx_v7_0_wait_for_rlc_serdes(adev);
3387 	}
3388 
3389 	return orig;
3390 }
3391 
3392 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev)
3393 {
3394 	return true;
3395 }
3396 
3397 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev)
3398 {
3399 	u32 tmp, i, mask;
3400 
3401 	tmp = 0x1 | (1 << 1);
3402 	WREG32(mmRLC_GPR_REG2, tmp);
3403 
3404 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3405 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3406 	for (i = 0; i < adev->usec_timeout; i++) {
3407 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3408 			break;
3409 		udelay(1);
3410 	}
3411 
3412 	for (i = 0; i < adev->usec_timeout; i++) {
3413 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3414 			break;
3415 		udelay(1);
3416 	}
3417 }
3418 
3419 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev)
3420 {
3421 	u32 tmp;
3422 
3423 	tmp = 0x1 | (0 << 1);
3424 	WREG32(mmRLC_GPR_REG2, tmp);
3425 }
3426 
3427 /**
3428  * gfx_v7_0_rlc_stop - stop the RLC ME
3429  *
3430  * @adev: amdgpu_device pointer
3431  *
3432  * Halt the RLC ME (MicroEngine) (CIK).
3433  */
3434 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3435 {
3436 	WREG32(mmRLC_CNTL, 0);
3437 
3438 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3439 
3440 	gfx_v7_0_wait_for_rlc_serdes(adev);
3441 }
3442 
3443 /**
3444  * gfx_v7_0_rlc_start - start the RLC ME
3445  *
3446  * @adev: amdgpu_device pointer
3447  *
3448  * Unhalt the RLC ME (MicroEngine) (CIK).
3449  */
3450 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3451 {
3452 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3453 
3454 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3455 
3456 	udelay(50);
3457 }
3458 
3459 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3460 {
3461 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3462 
3463 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3464 	WREG32(mmGRBM_SOFT_RESET, tmp);
3465 	udelay(50);
3466 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3467 	WREG32(mmGRBM_SOFT_RESET, tmp);
3468 	udelay(50);
3469 }
3470 
3471 /**
3472  * gfx_v7_0_rlc_resume - setup the RLC hw
3473  *
3474  * @adev: amdgpu_device pointer
3475  *
3476  * Initialize the RLC registers, load the ucode,
3477  * and start the RLC (CIK).
3478  * Returns 0 for success, -EINVAL if the ucode is not available.
3479  */
3480 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3481 {
3482 	const struct rlc_firmware_header_v1_0 *hdr;
3483 	const __le32 *fw_data;
3484 	unsigned i, fw_size;
3485 	u32 tmp;
3486 
3487 	if (!adev->gfx.rlc_fw)
3488 		return -EINVAL;
3489 
3490 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3491 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3492 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3493 	adev->gfx.rlc_feature_version = le32_to_cpu(
3494 					hdr->ucode_feature_version);
3495 
3496 	adev->gfx.rlc.funcs->stop(adev);
3497 
3498 	/* disable CG */
3499 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3500 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3501 
3502 	adev->gfx.rlc.funcs->reset(adev);
3503 
3504 	gfx_v7_0_init_pg(adev);
3505 
3506 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3507 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3508 
3509 	mutex_lock(&adev->grbm_idx_mutex);
3510 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3511 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3512 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3513 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3514 	mutex_unlock(&adev->grbm_idx_mutex);
3515 
3516 	WREG32(mmRLC_MC_CNTL, 0);
3517 	WREG32(mmRLC_UCODE_CNTL, 0);
3518 
3519 	fw_data = (const __le32 *)
3520 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3521 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3522 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3523 	for (i = 0; i < fw_size; i++)
3524 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3525 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3526 
3527 	/* XXX - find out what chips support lbpw */
3528 	gfx_v7_0_enable_lbpw(adev, false);
3529 
3530 	if (adev->asic_type == CHIP_BONAIRE)
3531 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3532 
3533 	adev->gfx.rlc.funcs->start(adev);
3534 
3535 	return 0;
3536 }
3537 
3538 static void gfx_v7_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
3539 {
3540 	u32 data;
3541 
3542 	amdgpu_gfx_off_ctrl(adev, false);
3543 
3544 	data = RREG32(mmRLC_SPM_VMID);
3545 
3546 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
3547 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
3548 
3549 	WREG32(mmRLC_SPM_VMID, data);
3550 
3551 	amdgpu_gfx_off_ctrl(adev, true);
3552 }
3553 
3554 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3555 {
3556 	u32 data, orig, tmp, tmp2;
3557 
3558 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3559 
3560 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3561 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3562 
3563 		tmp = gfx_v7_0_halt_rlc(adev);
3564 
3565 		mutex_lock(&adev->grbm_idx_mutex);
3566 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3567 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3568 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3569 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3570 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3571 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3572 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3573 		mutex_unlock(&adev->grbm_idx_mutex);
3574 
3575 		gfx_v7_0_update_rlc(adev, tmp);
3576 
3577 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3578 		if (orig != data)
3579 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3580 
3581 	} else {
3582 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3583 
3584 		RREG32(mmCB_CGTT_SCLK_CTRL);
3585 		RREG32(mmCB_CGTT_SCLK_CTRL);
3586 		RREG32(mmCB_CGTT_SCLK_CTRL);
3587 		RREG32(mmCB_CGTT_SCLK_CTRL);
3588 
3589 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3590 		if (orig != data)
3591 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3592 
3593 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3594 	}
3595 }
3596 
3597 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3598 {
3599 	u32 data, orig, tmp = 0;
3600 
3601 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3602 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3603 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3604 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3605 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3606 				if (orig != data)
3607 					WREG32(mmCP_MEM_SLP_CNTL, data);
3608 			}
3609 		}
3610 
3611 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3612 		data |= 0x00000001;
3613 		data &= 0xfffffffd;
3614 		if (orig != data)
3615 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3616 
3617 		tmp = gfx_v7_0_halt_rlc(adev);
3618 
3619 		mutex_lock(&adev->grbm_idx_mutex);
3620 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3621 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3622 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3623 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3624 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3625 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3626 		mutex_unlock(&adev->grbm_idx_mutex);
3627 
3628 		gfx_v7_0_update_rlc(adev, tmp);
3629 
3630 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3631 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3632 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3633 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3634 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3635 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3636 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3637 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3638 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3639 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3640 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3641 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3642 			if (orig != data)
3643 				WREG32(mmCGTS_SM_CTRL_REG, data);
3644 		}
3645 	} else {
3646 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3647 		data |= 0x00000003;
3648 		if (orig != data)
3649 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3650 
3651 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3652 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3653 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3654 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3655 		}
3656 
3657 		data = RREG32(mmCP_MEM_SLP_CNTL);
3658 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3659 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3660 			WREG32(mmCP_MEM_SLP_CNTL, data);
3661 		}
3662 
3663 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3664 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3665 		if (orig != data)
3666 			WREG32(mmCGTS_SM_CTRL_REG, data);
3667 
3668 		tmp = gfx_v7_0_halt_rlc(adev);
3669 
3670 		mutex_lock(&adev->grbm_idx_mutex);
3671 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3672 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3673 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3674 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3675 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3676 		mutex_unlock(&adev->grbm_idx_mutex);
3677 
3678 		gfx_v7_0_update_rlc(adev, tmp);
3679 	}
3680 }
3681 
3682 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3683 			       bool enable)
3684 {
3685 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3686 	/* order matters! */
3687 	if (enable) {
3688 		gfx_v7_0_enable_mgcg(adev, true);
3689 		gfx_v7_0_enable_cgcg(adev, true);
3690 	} else {
3691 		gfx_v7_0_enable_cgcg(adev, false);
3692 		gfx_v7_0_enable_mgcg(adev, false);
3693 	}
3694 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3695 }
3696 
3697 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3698 						bool enable)
3699 {
3700 	u32 data, orig;
3701 
3702 	orig = data = RREG32(mmRLC_PG_CNTL);
3703 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3704 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3705 	else
3706 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3707 	if (orig != data)
3708 		WREG32(mmRLC_PG_CNTL, data);
3709 }
3710 
3711 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3712 						bool enable)
3713 {
3714 	u32 data, orig;
3715 
3716 	orig = data = RREG32(mmRLC_PG_CNTL);
3717 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3718 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3719 	else
3720 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3721 	if (orig != data)
3722 		WREG32(mmRLC_PG_CNTL, data);
3723 }
3724 
3725 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3726 {
3727 	u32 data, orig;
3728 
3729 	orig = data = RREG32(mmRLC_PG_CNTL);
3730 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3731 		data &= ~0x8000;
3732 	else
3733 		data |= 0x8000;
3734 	if (orig != data)
3735 		WREG32(mmRLC_PG_CNTL, data);
3736 }
3737 
3738 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3739 {
3740 	u32 data, orig;
3741 
3742 	orig = data = RREG32(mmRLC_PG_CNTL);
3743 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3744 		data &= ~0x2000;
3745 	else
3746 		data |= 0x2000;
3747 	if (orig != data)
3748 		WREG32(mmRLC_PG_CNTL, data);
3749 }
3750 
3751 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev)
3752 {
3753 	if (adev->asic_type == CHIP_KAVERI)
3754 		return 5;
3755 	else
3756 		return 4;
3757 }
3758 
3759 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
3760 				     bool enable)
3761 {
3762 	u32 data, orig;
3763 
3764 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
3765 		orig = data = RREG32(mmRLC_PG_CNTL);
3766 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3767 		if (orig != data)
3768 			WREG32(mmRLC_PG_CNTL, data);
3769 
3770 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3771 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3772 		if (orig != data)
3773 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3774 	} else {
3775 		orig = data = RREG32(mmRLC_PG_CNTL);
3776 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3777 		if (orig != data)
3778 			WREG32(mmRLC_PG_CNTL, data);
3779 
3780 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3781 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3782 		if (orig != data)
3783 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3784 
3785 		data = RREG32(mmDB_RENDER_CONTROL);
3786 	}
3787 }
3788 
3789 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
3790 						 u32 bitmap)
3791 {
3792 	u32 data;
3793 
3794 	if (!bitmap)
3795 		return;
3796 
3797 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3798 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3799 
3800 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
3801 }
3802 
3803 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
3804 {
3805 	u32 data, mask;
3806 
3807 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
3808 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
3809 
3810 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3811 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3812 
3813 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
3814 
3815 	return (~data) & mask;
3816 }
3817 
3818 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
3819 {
3820 	u32 tmp;
3821 
3822 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
3823 
3824 	tmp = RREG32(mmRLC_MAX_PG_CU);
3825 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
3826 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
3827 	WREG32(mmRLC_MAX_PG_CU, tmp);
3828 }
3829 
3830 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
3831 					    bool enable)
3832 {
3833 	u32 data, orig;
3834 
3835 	orig = data = RREG32(mmRLC_PG_CNTL);
3836 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
3837 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3838 	else
3839 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3840 	if (orig != data)
3841 		WREG32(mmRLC_PG_CNTL, data);
3842 }
3843 
3844 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
3845 					     bool enable)
3846 {
3847 	u32 data, orig;
3848 
3849 	orig = data = RREG32(mmRLC_PG_CNTL);
3850 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
3851 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3852 	else
3853 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3854 	if (orig != data)
3855 		WREG32(mmRLC_PG_CNTL, data);
3856 }
3857 
3858 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
3859 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
3860 
3861 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
3862 {
3863 	u32 data, orig;
3864 	u32 i;
3865 
3866 	if (adev->gfx.rlc.cs_data) {
3867 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3868 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3869 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3870 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
3871 	} else {
3872 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3873 		for (i = 0; i < 3; i++)
3874 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
3875 	}
3876 	if (adev->gfx.rlc.reg_list) {
3877 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
3878 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3879 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
3880 	}
3881 
3882 	orig = data = RREG32(mmRLC_PG_CNTL);
3883 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
3884 	if (orig != data)
3885 		WREG32(mmRLC_PG_CNTL, data);
3886 
3887 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
3888 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
3889 
3890 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
3891 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
3892 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3893 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
3894 
3895 	data = 0x10101010;
3896 	WREG32(mmRLC_PG_DELAY, data);
3897 
3898 	data = RREG32(mmRLC_PG_DELAY_2);
3899 	data &= ~0xff;
3900 	data |= 0x3;
3901 	WREG32(mmRLC_PG_DELAY_2, data);
3902 
3903 	data = RREG32(mmRLC_AUTO_PG_CTRL);
3904 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
3905 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
3906 	WREG32(mmRLC_AUTO_PG_CTRL, data);
3907 
3908 }
3909 
3910 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
3911 {
3912 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
3913 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
3914 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
3915 }
3916 
3917 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
3918 {
3919 	u32 count = 0;
3920 	const struct cs_section_def *sect = NULL;
3921 	const struct cs_extent_def *ext = NULL;
3922 
3923 	if (adev->gfx.rlc.cs_data == NULL)
3924 		return 0;
3925 
3926 	/* begin clear state */
3927 	count += 2;
3928 	/* context control state */
3929 	count += 3;
3930 
3931 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3932 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3933 			if (sect->id == SECT_CONTEXT)
3934 				count += 2 + ext->reg_count;
3935 			else
3936 				return 0;
3937 		}
3938 	}
3939 	/* pa_sc_raster_config/pa_sc_raster_config1 */
3940 	count += 4;
3941 	/* end clear state */
3942 	count += 2;
3943 	/* clear state */
3944 	count += 2;
3945 
3946 	return count;
3947 }
3948 
3949 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
3950 				    volatile u32 *buffer)
3951 {
3952 	u32 count = 0, i;
3953 	const struct cs_section_def *sect = NULL;
3954 	const struct cs_extent_def *ext = NULL;
3955 
3956 	if (adev->gfx.rlc.cs_data == NULL)
3957 		return;
3958 	if (buffer == NULL)
3959 		return;
3960 
3961 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3962 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3963 
3964 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3965 	buffer[count++] = cpu_to_le32(0x80000000);
3966 	buffer[count++] = cpu_to_le32(0x80000000);
3967 
3968 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3969 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3970 			if (sect->id == SECT_CONTEXT) {
3971 				buffer[count++] =
3972 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
3973 				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3974 				for (i = 0; i < ext->reg_count; i++)
3975 					buffer[count++] = cpu_to_le32(ext->extent[i]);
3976 			} else {
3977 				return;
3978 			}
3979 		}
3980 	}
3981 
3982 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
3983 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
3984 	switch (adev->asic_type) {
3985 	case CHIP_BONAIRE:
3986 		buffer[count++] = cpu_to_le32(0x16000012);
3987 		buffer[count++] = cpu_to_le32(0x00000000);
3988 		break;
3989 	case CHIP_KAVERI:
3990 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3991 		buffer[count++] = cpu_to_le32(0x00000000);
3992 		break;
3993 	case CHIP_KABINI:
3994 	case CHIP_MULLINS:
3995 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3996 		buffer[count++] = cpu_to_le32(0x00000000);
3997 		break;
3998 	case CHIP_HAWAII:
3999 		buffer[count++] = cpu_to_le32(0x3a00161a);
4000 		buffer[count++] = cpu_to_le32(0x0000002e);
4001 		break;
4002 	default:
4003 		buffer[count++] = cpu_to_le32(0x00000000);
4004 		buffer[count++] = cpu_to_le32(0x00000000);
4005 		break;
4006 	}
4007 
4008 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4009 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
4010 
4011 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
4012 	buffer[count++] = cpu_to_le32(0);
4013 }
4014 
4015 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
4016 {
4017 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4018 			      AMD_PG_SUPPORT_GFX_SMG |
4019 			      AMD_PG_SUPPORT_GFX_DMG |
4020 			      AMD_PG_SUPPORT_CP |
4021 			      AMD_PG_SUPPORT_GDS |
4022 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4023 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
4024 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
4025 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4026 			gfx_v7_0_init_gfx_cgpg(adev);
4027 			gfx_v7_0_enable_cp_pg(adev, true);
4028 			gfx_v7_0_enable_gds_pg(adev, true);
4029 		}
4030 		gfx_v7_0_init_ao_cu_mask(adev);
4031 		gfx_v7_0_update_gfx_pg(adev, true);
4032 	}
4033 }
4034 
4035 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4036 {
4037 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4038 			      AMD_PG_SUPPORT_GFX_SMG |
4039 			      AMD_PG_SUPPORT_GFX_DMG |
4040 			      AMD_PG_SUPPORT_CP |
4041 			      AMD_PG_SUPPORT_GDS |
4042 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4043 		gfx_v7_0_update_gfx_pg(adev, false);
4044 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4045 			gfx_v7_0_enable_cp_pg(adev, false);
4046 			gfx_v7_0_enable_gds_pg(adev, false);
4047 		}
4048 	}
4049 }
4050 
4051 /**
4052  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4053  *
4054  * @adev: amdgpu_device pointer
4055  *
4056  * Fetches a GPU clock counter snapshot (SI).
4057  * Returns the 64 bit clock counter snapshot.
4058  */
4059 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4060 {
4061 	uint64_t clock;
4062 
4063 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4064 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4065 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4066 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4067 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4068 	return clock;
4069 }
4070 
4071 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4072 					  uint32_t vmid,
4073 					  uint32_t gds_base, uint32_t gds_size,
4074 					  uint32_t gws_base, uint32_t gws_size,
4075 					  uint32_t oa_base, uint32_t oa_size)
4076 {
4077 	/* GDS Base */
4078 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4079 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4080 				WRITE_DATA_DST_SEL(0)));
4081 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4082 	amdgpu_ring_write(ring, 0);
4083 	amdgpu_ring_write(ring, gds_base);
4084 
4085 	/* GDS Size */
4086 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4087 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4088 				WRITE_DATA_DST_SEL(0)));
4089 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
4090 	amdgpu_ring_write(ring, 0);
4091 	amdgpu_ring_write(ring, gds_size);
4092 
4093 	/* GWS */
4094 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4095 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4096 				WRITE_DATA_DST_SEL(0)));
4097 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
4098 	amdgpu_ring_write(ring, 0);
4099 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4100 
4101 	/* OA */
4102 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4103 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4104 				WRITE_DATA_DST_SEL(0)));
4105 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
4106 	amdgpu_ring_write(ring, 0);
4107 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4108 }
4109 
4110 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
4111 {
4112 	struct amdgpu_device *adev = ring->adev;
4113 	uint32_t value = 0;
4114 
4115 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
4116 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
4117 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
4118 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
4119 	WREG32(mmSQ_CMD, value);
4120 }
4121 
4122 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4123 {
4124 	WREG32(mmSQ_IND_INDEX,
4125 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4126 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4127 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4128 		(SQ_IND_INDEX__FORCE_READ_MASK));
4129 	return RREG32(mmSQ_IND_DATA);
4130 }
4131 
4132 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4133 			   uint32_t wave, uint32_t thread,
4134 			   uint32_t regno, uint32_t num, uint32_t *out)
4135 {
4136 	WREG32(mmSQ_IND_INDEX,
4137 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4138 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4139 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4140 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4141 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4142 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4143 	while (num--)
4144 		*(out++) = RREG32(mmSQ_IND_DATA);
4145 }
4146 
4147 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4148 {
4149 	/* type 0 wave data */
4150 	dst[(*no_fields)++] = 0;
4151 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4152 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4153 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4154 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4155 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4156 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4157 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4158 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4159 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4160 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4161 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4162 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4163 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4164 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4165 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4166 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4167 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4168 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4169 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
4170 }
4171 
4172 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
4173 				     uint32_t wave, uint32_t start,
4174 				     uint32_t size, uint32_t *dst)
4175 {
4176 	wave_read_regs(
4177 		adev, simd, wave, 0,
4178 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4179 }
4180 
4181 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
4182 				  u32 me, u32 pipe, u32 q, u32 vm)
4183 {
4184 	cik_srbm_select(adev, me, pipe, q, vm);
4185 }
4186 
4187 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4188 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4189 	.select_se_sh = &gfx_v7_0_select_se_sh,
4190 	.read_wave_data = &gfx_v7_0_read_wave_data,
4191 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4192 	.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q
4193 };
4194 
4195 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4196 	.is_rlc_enabled = gfx_v7_0_is_rlc_enabled,
4197 	.set_safe_mode = gfx_v7_0_set_safe_mode,
4198 	.unset_safe_mode = gfx_v7_0_unset_safe_mode,
4199 	.init = gfx_v7_0_rlc_init,
4200 	.get_csb_size = gfx_v7_0_get_csb_size,
4201 	.get_csb_buffer = gfx_v7_0_get_csb_buffer,
4202 	.get_cp_table_num = gfx_v7_0_cp_pg_table_num,
4203 	.resume = gfx_v7_0_rlc_resume,
4204 	.stop = gfx_v7_0_rlc_stop,
4205 	.reset = gfx_v7_0_rlc_reset,
4206 	.start = gfx_v7_0_rlc_start,
4207 	.update_spm_vmid = gfx_v7_0_update_spm_vmid
4208 };
4209 
4210 static int gfx_v7_0_early_init(void *handle)
4211 {
4212 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4213 
4214 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4215 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4216 					  AMDGPU_MAX_COMPUTE_RINGS);
4217 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4218 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4219 	gfx_v7_0_set_ring_funcs(adev);
4220 	gfx_v7_0_set_irq_funcs(adev);
4221 	gfx_v7_0_set_gds_init(adev);
4222 
4223 	return 0;
4224 }
4225 
4226 static int gfx_v7_0_late_init(void *handle)
4227 {
4228 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4229 	int r;
4230 
4231 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4232 	if (r)
4233 		return r;
4234 
4235 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4236 	if (r)
4237 		return r;
4238 
4239 	return 0;
4240 }
4241 
4242 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4243 {
4244 	u32 gb_addr_config;
4245 	u32 mc_arb_ramcfg;
4246 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4247 	u32 tmp;
4248 
4249 	switch (adev->asic_type) {
4250 	case CHIP_BONAIRE:
4251 		adev->gfx.config.max_shader_engines = 2;
4252 		adev->gfx.config.max_tile_pipes = 4;
4253 		adev->gfx.config.max_cu_per_sh = 7;
4254 		adev->gfx.config.max_sh_per_se = 1;
4255 		adev->gfx.config.max_backends_per_se = 2;
4256 		adev->gfx.config.max_texture_channel_caches = 4;
4257 		adev->gfx.config.max_gprs = 256;
4258 		adev->gfx.config.max_gs_threads = 32;
4259 		adev->gfx.config.max_hw_contexts = 8;
4260 
4261 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4262 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4263 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4264 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4265 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4266 		break;
4267 	case CHIP_HAWAII:
4268 		adev->gfx.config.max_shader_engines = 4;
4269 		adev->gfx.config.max_tile_pipes = 16;
4270 		adev->gfx.config.max_cu_per_sh = 11;
4271 		adev->gfx.config.max_sh_per_se = 1;
4272 		adev->gfx.config.max_backends_per_se = 4;
4273 		adev->gfx.config.max_texture_channel_caches = 16;
4274 		adev->gfx.config.max_gprs = 256;
4275 		adev->gfx.config.max_gs_threads = 32;
4276 		adev->gfx.config.max_hw_contexts = 8;
4277 
4278 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4279 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4280 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4281 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4282 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4283 		break;
4284 	case CHIP_KAVERI:
4285 		adev->gfx.config.max_shader_engines = 1;
4286 		adev->gfx.config.max_tile_pipes = 4;
4287 		adev->gfx.config.max_cu_per_sh = 8;
4288 		adev->gfx.config.max_backends_per_se = 2;
4289 		adev->gfx.config.max_sh_per_se = 1;
4290 		adev->gfx.config.max_texture_channel_caches = 4;
4291 		adev->gfx.config.max_gprs = 256;
4292 		adev->gfx.config.max_gs_threads = 16;
4293 		adev->gfx.config.max_hw_contexts = 8;
4294 
4295 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4296 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4297 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4298 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4299 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4300 		break;
4301 	case CHIP_KABINI:
4302 	case CHIP_MULLINS:
4303 	default:
4304 		adev->gfx.config.max_shader_engines = 1;
4305 		adev->gfx.config.max_tile_pipes = 2;
4306 		adev->gfx.config.max_cu_per_sh = 2;
4307 		adev->gfx.config.max_sh_per_se = 1;
4308 		adev->gfx.config.max_backends_per_se = 1;
4309 		adev->gfx.config.max_texture_channel_caches = 2;
4310 		adev->gfx.config.max_gprs = 256;
4311 		adev->gfx.config.max_gs_threads = 16;
4312 		adev->gfx.config.max_hw_contexts = 8;
4313 
4314 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4315 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4316 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4317 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4318 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4319 		break;
4320 	}
4321 
4322 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4323 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4324 
4325 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
4326 				MC_ARB_RAMCFG, NOOFBANK);
4327 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
4328 				MC_ARB_RAMCFG, NOOFRANKS);
4329 
4330 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4331 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4332 	if (adev->flags & AMD_IS_APU) {
4333 		/* Get memory bank mapping mode. */
4334 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4335 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4336 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4337 
4338 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4339 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4340 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4341 
4342 		/* Validate settings in case only one DIMM installed. */
4343 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4344 			dimm00_addr_map = 0;
4345 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4346 			dimm01_addr_map = 0;
4347 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4348 			dimm10_addr_map = 0;
4349 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4350 			dimm11_addr_map = 0;
4351 
4352 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4353 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4354 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4355 			adev->gfx.config.mem_row_size_in_kb = 2;
4356 		else
4357 			adev->gfx.config.mem_row_size_in_kb = 1;
4358 	} else {
4359 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4360 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4361 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4362 			adev->gfx.config.mem_row_size_in_kb = 4;
4363 	}
4364 	/* XXX use MC settings? */
4365 	adev->gfx.config.shader_engine_tile_size = 32;
4366 	adev->gfx.config.num_gpus = 1;
4367 	adev->gfx.config.multi_gpu_tile_size = 64;
4368 
4369 	/* fix up row size */
4370 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4371 	switch (adev->gfx.config.mem_row_size_in_kb) {
4372 	case 1:
4373 	default:
4374 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4375 		break;
4376 	case 2:
4377 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4378 		break;
4379 	case 4:
4380 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4381 		break;
4382 	}
4383 	adev->gfx.config.gb_addr_config = gb_addr_config;
4384 }
4385 
4386 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
4387 					int mec, int pipe, int queue)
4388 {
4389 	int r;
4390 	unsigned irq_type;
4391 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
4392 
4393 	/* mec0 is me1 */
4394 	ring->me = mec + 1;
4395 	ring->pipe = pipe;
4396 	ring->queue = queue;
4397 
4398 	ring->ring_obj = NULL;
4399 	ring->use_doorbell = true;
4400 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
4401 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4402 
4403 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
4404 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
4405 		+ ring->pipe;
4406 
4407 	/* type-2 packets are deprecated on MEC, use type-3 instead */
4408 	r = amdgpu_ring_init(adev, ring, 1024,
4409 			     &adev->gfx.eop_irq, irq_type,
4410 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
4411 	if (r)
4412 		return r;
4413 
4414 
4415 	return 0;
4416 }
4417 
4418 static int gfx_v7_0_sw_init(void *handle)
4419 {
4420 	struct amdgpu_ring *ring;
4421 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4422 	int i, j, k, r, ring_id;
4423 
4424 	switch (adev->asic_type) {
4425 	case CHIP_KAVERI:
4426 		adev->gfx.mec.num_mec = 2;
4427 		break;
4428 	case CHIP_BONAIRE:
4429 	case CHIP_HAWAII:
4430 	case CHIP_KABINI:
4431 	case CHIP_MULLINS:
4432 	default:
4433 		adev->gfx.mec.num_mec = 1;
4434 		break;
4435 	}
4436 	adev->gfx.mec.num_pipe_per_mec = 4;
4437 	adev->gfx.mec.num_queue_per_pipe = 8;
4438 
4439 	/* EOP Event */
4440 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq);
4441 	if (r)
4442 		return r;
4443 
4444 	/* Privileged reg */
4445 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184,
4446 			      &adev->gfx.priv_reg_irq);
4447 	if (r)
4448 		return r;
4449 
4450 	/* Privileged inst */
4451 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185,
4452 			      &adev->gfx.priv_inst_irq);
4453 	if (r)
4454 		return r;
4455 
4456 	r = gfx_v7_0_init_microcode(adev);
4457 	if (r) {
4458 		DRM_ERROR("Failed to load gfx firmware!\n");
4459 		return r;
4460 	}
4461 
4462 	r = adev->gfx.rlc.funcs->init(adev);
4463 	if (r) {
4464 		DRM_ERROR("Failed to init rlc BOs!\n");
4465 		return r;
4466 	}
4467 
4468 	/* allocate mec buffers */
4469 	r = gfx_v7_0_mec_init(adev);
4470 	if (r) {
4471 		DRM_ERROR("Failed to init MEC BOs!\n");
4472 		return r;
4473 	}
4474 
4475 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4476 		ring = &adev->gfx.gfx_ring[i];
4477 		ring->ring_obj = NULL;
4478 		sprintf(ring->name, "gfx");
4479 		r = amdgpu_ring_init(adev, ring, 1024,
4480 				     &adev->gfx.eop_irq,
4481 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
4482 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
4483 		if (r)
4484 			return r;
4485 	}
4486 
4487 	/* set up the compute queues - allocate horizontally across pipes */
4488 	ring_id = 0;
4489 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4490 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4491 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4492 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
4493 					continue;
4494 
4495 				r = gfx_v7_0_compute_ring_init(adev,
4496 								ring_id,
4497 								i, k, j);
4498 				if (r)
4499 					return r;
4500 
4501 				ring_id++;
4502 			}
4503 		}
4504 	}
4505 
4506 	adev->gfx.ce_ram_size = 0x8000;
4507 
4508 	gfx_v7_0_gpu_early_init(adev);
4509 
4510 	return r;
4511 }
4512 
4513 static int gfx_v7_0_sw_fini(void *handle)
4514 {
4515 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4516 	int i;
4517 
4518 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4519 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4520 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
4521 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4522 
4523 	gfx_v7_0_cp_compute_fini(adev);
4524 	amdgpu_gfx_rlc_fini(adev);
4525 	gfx_v7_0_mec_fini(adev);
4526 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
4527 				&adev->gfx.rlc.clear_state_gpu_addr,
4528 				(void **)&adev->gfx.rlc.cs_ptr);
4529 	if (adev->gfx.rlc.cp_table_size) {
4530 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
4531 				&adev->gfx.rlc.cp_table_gpu_addr,
4532 				(void **)&adev->gfx.rlc.cp_table_ptr);
4533 	}
4534 	gfx_v7_0_free_microcode(adev);
4535 
4536 	return 0;
4537 }
4538 
4539 static int gfx_v7_0_hw_init(void *handle)
4540 {
4541 	int r;
4542 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4543 
4544 	gfx_v7_0_constants_init(adev);
4545 
4546 	/* init CSB */
4547 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
4548 	/* init rlc */
4549 	r = adev->gfx.rlc.funcs->resume(adev);
4550 	if (r)
4551 		return r;
4552 
4553 	r = gfx_v7_0_cp_resume(adev);
4554 	if (r)
4555 		return r;
4556 
4557 	return r;
4558 }
4559 
4560 static int gfx_v7_0_hw_fini(void *handle)
4561 {
4562 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4563 
4564 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4565 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4566 	gfx_v7_0_cp_enable(adev, false);
4567 	adev->gfx.rlc.funcs->stop(adev);
4568 	gfx_v7_0_fini_pg(adev);
4569 
4570 	return 0;
4571 }
4572 
4573 static int gfx_v7_0_suspend(void *handle)
4574 {
4575 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4576 
4577 	return gfx_v7_0_hw_fini(adev);
4578 }
4579 
4580 static int gfx_v7_0_resume(void *handle)
4581 {
4582 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4583 
4584 	return gfx_v7_0_hw_init(adev);
4585 }
4586 
4587 static bool gfx_v7_0_is_idle(void *handle)
4588 {
4589 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4590 
4591 	if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK)
4592 		return false;
4593 	else
4594 		return true;
4595 }
4596 
4597 static int gfx_v7_0_wait_for_idle(void *handle)
4598 {
4599 	unsigned i;
4600 	u32 tmp;
4601 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4602 
4603 	for (i = 0; i < adev->usec_timeout; i++) {
4604 		/* read MC_STATUS */
4605 		tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4606 
4607 		if (!tmp)
4608 			return 0;
4609 		udelay(1);
4610 	}
4611 	return -ETIMEDOUT;
4612 }
4613 
4614 static int gfx_v7_0_soft_reset(void *handle)
4615 {
4616 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4617 	u32 tmp;
4618 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4619 
4620 	/* GRBM_STATUS */
4621 	tmp = RREG32(mmGRBM_STATUS);
4622 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4623 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4624 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4625 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4626 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4627 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK))
4628 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK |
4629 			GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK;
4630 
4631 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4632 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK;
4633 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4634 	}
4635 
4636 	/* GRBM_STATUS2 */
4637 	tmp = RREG32(mmGRBM_STATUS2);
4638 	if (tmp & GRBM_STATUS2__RLC_BUSY_MASK)
4639 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
4640 
4641 	/* SRBM_STATUS */
4642 	tmp = RREG32(mmSRBM_STATUS);
4643 	if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK)
4644 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4645 
4646 	if (grbm_soft_reset || srbm_soft_reset) {
4647 		/* disable CG/PG */
4648 		gfx_v7_0_fini_pg(adev);
4649 		gfx_v7_0_update_cg(adev, false);
4650 
4651 		/* stop the rlc */
4652 		adev->gfx.rlc.funcs->stop(adev);
4653 
4654 		/* Disable GFX parsing/prefetching */
4655 		WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK);
4656 
4657 		/* Disable MEC parsing/prefetching */
4658 		WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK);
4659 
4660 		if (grbm_soft_reset) {
4661 			tmp = RREG32(mmGRBM_SOFT_RESET);
4662 			tmp |= grbm_soft_reset;
4663 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4664 			WREG32(mmGRBM_SOFT_RESET, tmp);
4665 			tmp = RREG32(mmGRBM_SOFT_RESET);
4666 
4667 			udelay(50);
4668 
4669 			tmp &= ~grbm_soft_reset;
4670 			WREG32(mmGRBM_SOFT_RESET, tmp);
4671 			tmp = RREG32(mmGRBM_SOFT_RESET);
4672 		}
4673 
4674 		if (srbm_soft_reset) {
4675 			tmp = RREG32(mmSRBM_SOFT_RESET);
4676 			tmp |= srbm_soft_reset;
4677 			dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4678 			WREG32(mmSRBM_SOFT_RESET, tmp);
4679 			tmp = RREG32(mmSRBM_SOFT_RESET);
4680 
4681 			udelay(50);
4682 
4683 			tmp &= ~srbm_soft_reset;
4684 			WREG32(mmSRBM_SOFT_RESET, tmp);
4685 			tmp = RREG32(mmSRBM_SOFT_RESET);
4686 		}
4687 		/* Wait a little for things to settle down */
4688 		udelay(50);
4689 	}
4690 	return 0;
4691 }
4692 
4693 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4694 						 enum amdgpu_interrupt_state state)
4695 {
4696 	u32 cp_int_cntl;
4697 
4698 	switch (state) {
4699 	case AMDGPU_IRQ_STATE_DISABLE:
4700 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4701 		cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4702 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4703 		break;
4704 	case AMDGPU_IRQ_STATE_ENABLE:
4705 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4706 		cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4707 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4708 		break;
4709 	default:
4710 		break;
4711 	}
4712 }
4713 
4714 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4715 						     int me, int pipe,
4716 						     enum amdgpu_interrupt_state state)
4717 {
4718 	u32 mec_int_cntl, mec_int_cntl_reg;
4719 
4720 	/*
4721 	 * amdgpu controls only the first MEC. That's why this function only
4722 	 * handles the setting of interrupts for this specific MEC. All other
4723 	 * pipes' interrupts are set by amdkfd.
4724 	 */
4725 
4726 	if (me == 1) {
4727 		switch (pipe) {
4728 		case 0:
4729 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4730 			break;
4731 		case 1:
4732 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
4733 			break;
4734 		case 2:
4735 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
4736 			break;
4737 		case 3:
4738 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
4739 			break;
4740 		default:
4741 			DRM_DEBUG("invalid pipe %d\n", pipe);
4742 			return;
4743 		}
4744 	} else {
4745 		DRM_DEBUG("invalid me %d\n", me);
4746 		return;
4747 	}
4748 
4749 	switch (state) {
4750 	case AMDGPU_IRQ_STATE_DISABLE:
4751 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4752 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4753 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4754 		break;
4755 	case AMDGPU_IRQ_STATE_ENABLE:
4756 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4757 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4758 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4759 		break;
4760 	default:
4761 		break;
4762 	}
4763 }
4764 
4765 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4766 					     struct amdgpu_irq_src *src,
4767 					     unsigned type,
4768 					     enum amdgpu_interrupt_state state)
4769 {
4770 	u32 cp_int_cntl;
4771 
4772 	switch (state) {
4773 	case AMDGPU_IRQ_STATE_DISABLE:
4774 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4775 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4776 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4777 		break;
4778 	case AMDGPU_IRQ_STATE_ENABLE:
4779 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4780 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4781 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4782 		break;
4783 	default:
4784 		break;
4785 	}
4786 
4787 	return 0;
4788 }
4789 
4790 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4791 					      struct amdgpu_irq_src *src,
4792 					      unsigned type,
4793 					      enum amdgpu_interrupt_state state)
4794 {
4795 	u32 cp_int_cntl;
4796 
4797 	switch (state) {
4798 	case AMDGPU_IRQ_STATE_DISABLE:
4799 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4800 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4801 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4802 		break;
4803 	case AMDGPU_IRQ_STATE_ENABLE:
4804 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4805 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4806 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4807 		break;
4808 	default:
4809 		break;
4810 	}
4811 
4812 	return 0;
4813 }
4814 
4815 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4816 					    struct amdgpu_irq_src *src,
4817 					    unsigned type,
4818 					    enum amdgpu_interrupt_state state)
4819 {
4820 	switch (type) {
4821 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
4822 		gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
4823 		break;
4824 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4825 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4826 		break;
4827 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4828 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4829 		break;
4830 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4831 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4832 		break;
4833 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4834 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4835 		break;
4836 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4837 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4838 		break;
4839 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4840 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4841 		break;
4842 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4843 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4844 		break;
4845 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4846 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4847 		break;
4848 	default:
4849 		break;
4850 	}
4851 	return 0;
4852 }
4853 
4854 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
4855 			    struct amdgpu_irq_src *source,
4856 			    struct amdgpu_iv_entry *entry)
4857 {
4858 	u8 me_id, pipe_id;
4859 	struct amdgpu_ring *ring;
4860 	int i;
4861 
4862 	DRM_DEBUG("IH: CP EOP\n");
4863 	me_id = (entry->ring_id & 0x0c) >> 2;
4864 	pipe_id = (entry->ring_id & 0x03) >> 0;
4865 	switch (me_id) {
4866 	case 0:
4867 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4868 		break;
4869 	case 1:
4870 	case 2:
4871 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4872 			ring = &adev->gfx.compute_ring[i];
4873 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4874 				amdgpu_fence_process(ring);
4875 		}
4876 		break;
4877 	}
4878 	return 0;
4879 }
4880 
4881 static void gfx_v7_0_fault(struct amdgpu_device *adev,
4882 			   struct amdgpu_iv_entry *entry)
4883 {
4884 	struct amdgpu_ring *ring;
4885 	u8 me_id, pipe_id;
4886 	int i;
4887 
4888 	me_id = (entry->ring_id & 0x0c) >> 2;
4889 	pipe_id = (entry->ring_id & 0x03) >> 0;
4890 	switch (me_id) {
4891 	case 0:
4892 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
4893 		break;
4894 	case 1:
4895 	case 2:
4896 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4897 			ring = &adev->gfx.compute_ring[i];
4898 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4899 				drm_sched_fault(&ring->sched);
4900 		}
4901 		break;
4902 	}
4903 }
4904 
4905 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
4906 				 struct amdgpu_irq_src *source,
4907 				 struct amdgpu_iv_entry *entry)
4908 {
4909 	DRM_ERROR("Illegal register access in command stream\n");
4910 	gfx_v7_0_fault(adev, entry);
4911 	return 0;
4912 }
4913 
4914 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
4915 				  struct amdgpu_irq_src *source,
4916 				  struct amdgpu_iv_entry *entry)
4917 {
4918 	DRM_ERROR("Illegal instruction in command stream\n");
4919 	// XXX soft reset the gfx block only
4920 	gfx_v7_0_fault(adev, entry);
4921 	return 0;
4922 }
4923 
4924 static int gfx_v7_0_set_clockgating_state(void *handle,
4925 					  enum amd_clockgating_state state)
4926 {
4927 	bool gate = false;
4928 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4929 
4930 	if (state == AMD_CG_STATE_GATE)
4931 		gate = true;
4932 
4933 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
4934 	/* order matters! */
4935 	if (gate) {
4936 		gfx_v7_0_enable_mgcg(adev, true);
4937 		gfx_v7_0_enable_cgcg(adev, true);
4938 	} else {
4939 		gfx_v7_0_enable_cgcg(adev, false);
4940 		gfx_v7_0_enable_mgcg(adev, false);
4941 	}
4942 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
4943 
4944 	return 0;
4945 }
4946 
4947 static int gfx_v7_0_set_powergating_state(void *handle,
4948 					  enum amd_powergating_state state)
4949 {
4950 	bool gate = false;
4951 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4952 
4953 	if (state == AMD_PG_STATE_GATE)
4954 		gate = true;
4955 
4956 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4957 			      AMD_PG_SUPPORT_GFX_SMG |
4958 			      AMD_PG_SUPPORT_GFX_DMG |
4959 			      AMD_PG_SUPPORT_CP |
4960 			      AMD_PG_SUPPORT_GDS |
4961 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4962 		gfx_v7_0_update_gfx_pg(adev, gate);
4963 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4964 			gfx_v7_0_enable_cp_pg(adev, gate);
4965 			gfx_v7_0_enable_gds_pg(adev, gate);
4966 		}
4967 	}
4968 
4969 	return 0;
4970 }
4971 
4972 static void gfx_v7_0_emit_mem_sync(struct amdgpu_ring *ring)
4973 {
4974 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
4975 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
4976 			  PACKET3_TC_ACTION_ENA |
4977 			  PACKET3_SH_KCACHE_ACTION_ENA |
4978 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
4979 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
4980 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
4981 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
4982 }
4983 
4984 static void gfx_v7_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
4985 {
4986 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
4987 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
4988 			  PACKET3_TC_ACTION_ENA |
4989 			  PACKET3_SH_KCACHE_ACTION_ENA |
4990 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
4991 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
4992 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
4993 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
4994 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
4995 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
4996 }
4997 
4998 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
4999 	.name = "gfx_v7_0",
5000 	.early_init = gfx_v7_0_early_init,
5001 	.late_init = gfx_v7_0_late_init,
5002 	.sw_init = gfx_v7_0_sw_init,
5003 	.sw_fini = gfx_v7_0_sw_fini,
5004 	.hw_init = gfx_v7_0_hw_init,
5005 	.hw_fini = gfx_v7_0_hw_fini,
5006 	.suspend = gfx_v7_0_suspend,
5007 	.resume = gfx_v7_0_resume,
5008 	.is_idle = gfx_v7_0_is_idle,
5009 	.wait_for_idle = gfx_v7_0_wait_for_idle,
5010 	.soft_reset = gfx_v7_0_soft_reset,
5011 	.set_clockgating_state = gfx_v7_0_set_clockgating_state,
5012 	.set_powergating_state = gfx_v7_0_set_powergating_state,
5013 };
5014 
5015 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
5016 	.type = AMDGPU_RING_TYPE_GFX,
5017 	.align_mask = 0xff,
5018 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5019 	.support_64bit_ptrs = false,
5020 	.get_rptr = gfx_v7_0_ring_get_rptr,
5021 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
5022 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
5023 	.emit_frame_size =
5024 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5025 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5026 		5 + /* hdp invalidate */
5027 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
5028 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
5029 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
5030 		3 + 4 + /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
5031 		5, /* SURFACE_SYNC */
5032 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
5033 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
5034 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
5035 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5036 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5037 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5038 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5039 	.test_ring = gfx_v7_0_ring_test_ring,
5040 	.test_ib = gfx_v7_0_ring_test_ib,
5041 	.insert_nop = amdgpu_ring_insert_nop,
5042 	.pad_ib = amdgpu_ring_generic_pad_ib,
5043 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
5044 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5045 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
5046 	.emit_mem_sync = gfx_v7_0_emit_mem_sync,
5047 };
5048 
5049 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5050 	.type = AMDGPU_RING_TYPE_COMPUTE,
5051 	.align_mask = 0xff,
5052 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5053 	.support_64bit_ptrs = false,
5054 	.get_rptr = gfx_v7_0_ring_get_rptr,
5055 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5056 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5057 	.emit_frame_size =
5058 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5059 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5060 		5 + /* hdp invalidate */
5061 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5062 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
5063 		7 + 7 + 7 + /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5064 		7, /* gfx_v7_0_emit_mem_sync_compute */
5065 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
5066 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5067 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5068 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5069 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5070 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5071 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5072 	.test_ring = gfx_v7_0_ring_test_ring,
5073 	.test_ib = gfx_v7_0_ring_test_ib,
5074 	.insert_nop = amdgpu_ring_insert_nop,
5075 	.pad_ib = amdgpu_ring_generic_pad_ib,
5076 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5077 	.emit_mem_sync = gfx_v7_0_emit_mem_sync_compute,
5078 };
5079 
5080 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5081 {
5082 	int i;
5083 
5084 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5085 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5086 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5087 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5088 }
5089 
5090 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5091 	.set = gfx_v7_0_set_eop_interrupt_state,
5092 	.process = gfx_v7_0_eop_irq,
5093 };
5094 
5095 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5096 	.set = gfx_v7_0_set_priv_reg_fault_state,
5097 	.process = gfx_v7_0_priv_reg_irq,
5098 };
5099 
5100 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5101 	.set = gfx_v7_0_set_priv_inst_fault_state,
5102 	.process = gfx_v7_0_priv_inst_irq,
5103 };
5104 
5105 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5106 {
5107 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5108 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5109 
5110 	adev->gfx.priv_reg_irq.num_types = 1;
5111 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5112 
5113 	adev->gfx.priv_inst_irq.num_types = 1;
5114 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5115 }
5116 
5117 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5118 {
5119 	/* init asci gds info */
5120 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
5121 	adev->gds.gws_size = 64;
5122 	adev->gds.oa_size = 16;
5123 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
5124 }
5125 
5126 
5127 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5128 {
5129 	int i, j, k, counter, active_cu_number = 0;
5130 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5131 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5132 	unsigned disable_masks[4 * 2];
5133 	u32 ao_cu_num;
5134 
5135 	if (adev->flags & AMD_IS_APU)
5136 		ao_cu_num = 2;
5137 	else
5138 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
5139 
5140 	memset(cu_info, 0, sizeof(*cu_info));
5141 
5142 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5143 
5144 	mutex_lock(&adev->grbm_idx_mutex);
5145 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5146 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5147 			mask = 1;
5148 			ao_bitmap = 0;
5149 			counter = 0;
5150 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
5151 			if (i < 4 && j < 2)
5152 				gfx_v7_0_set_user_cu_inactive_bitmap(
5153 					adev, disable_masks[i * 2 + j]);
5154 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5155 			cu_info->bitmap[i][j] = bitmap;
5156 
5157 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
5158 				if (bitmap & mask) {
5159 					if (counter < ao_cu_num)
5160 						ao_bitmap |= mask;
5161 					counter ++;
5162 				}
5163 				mask <<= 1;
5164 			}
5165 			active_cu_number += counter;
5166 			if (i < 2 && j < 2)
5167 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5168 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5169 		}
5170 	}
5171 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5172 	mutex_unlock(&adev->grbm_idx_mutex);
5173 
5174 	cu_info->number = active_cu_number;
5175 	cu_info->ao_cu_mask = ao_cu_mask;
5176 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5177 	cu_info->max_waves_per_simd = 10;
5178 	cu_info->max_scratch_slots_per_cu = 32;
5179 	cu_info->wave_front_size = 64;
5180 	cu_info->lds_size = 64;
5181 }
5182 
5183 const struct amdgpu_ip_block_version gfx_v7_1_ip_block =
5184 {
5185 	.type = AMD_IP_BLOCK_TYPE_GFX,
5186 	.major = 7,
5187 	.minor = 1,
5188 	.rev = 0,
5189 	.funcs = &gfx_v7_0_ip_funcs,
5190 };
5191 
5192 const struct amdgpu_ip_block_version gfx_v7_2_ip_block =
5193 {
5194 	.type = AMD_IP_BLOCK_TYPE_GFX,
5195 	.major = 7,
5196 	.minor = 2,
5197 	.rev = 0,
5198 	.funcs = &gfx_v7_0_ip_funcs,
5199 };
5200 
5201 const struct amdgpu_ip_block_version gfx_v7_3_ip_block =
5202 {
5203 	.type = AMD_IP_BLOCK_TYPE_GFX,
5204 	.major = 7,
5205 	.minor = 3,
5206 	.rev = 0,
5207 	.funcs = &gfx_v7_0_ip_funcs,
5208 };
5209