xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision 568b9de4)
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  *
1584  * Select which SE, SH combinations to address. Certain
1585  * registers are instanced per SE or SH.  0xffffffff means
1586  * broadcast to all SEs or SHs (CIK).
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  * @se_num: number of SEs (shader engines) for the asic
1783  * @sh_per_se: number of SH blocks per SE for the asic
1784  *
1785  * Configures per-SE/SH RB registers (CIK).
1786  */
1787 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1788 {
1789 	int i, j;
1790 	u32 data;
1791 	u32 raster_config = 0, raster_config_1 = 0;
1792 	u32 active_rbs = 0;
1793 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1794 					adev->gfx.config.max_sh_per_se;
1795 	unsigned num_rb_pipes;
1796 
1797 	mutex_lock(&adev->grbm_idx_mutex);
1798 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1799 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1800 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1801 			data = gfx_v7_0_get_rb_active_bitmap(adev);
1802 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1803 					       rb_bitmap_width_per_sh);
1804 		}
1805 	}
1806 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1807 
1808 	adev->gfx.config.backend_enable_mask = active_rbs;
1809 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1810 
1811 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1812 			     adev->gfx.config.max_shader_engines, 16);
1813 
1814 	gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1815 
1816 	if (!adev->gfx.config.backend_enable_mask ||
1817 			adev->gfx.config.num_rbs >= num_rb_pipes) {
1818 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1819 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1820 	} else {
1821 		gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1822 							adev->gfx.config.backend_enable_mask,
1823 							num_rb_pipes);
1824 	}
1825 
1826 	/* cache the values for userspace */
1827 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1828 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1829 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1830 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
1831 				RREG32(mmCC_RB_BACKEND_DISABLE);
1832 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
1833 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1834 			adev->gfx.config.rb_config[i][j].raster_config =
1835 				RREG32(mmPA_SC_RASTER_CONFIG);
1836 			adev->gfx.config.rb_config[i][j].raster_config_1 =
1837 				RREG32(mmPA_SC_RASTER_CONFIG_1);
1838 		}
1839 	}
1840 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1841 	mutex_unlock(&adev->grbm_idx_mutex);
1842 }
1843 
1844 /**
1845  * gfx_v7_0_init_compute_vmid - gart enable
1846  *
1847  * @adev: amdgpu_device pointer
1848  *
1849  * Initialize compute vmid sh_mem registers
1850  *
1851  */
1852 #define DEFAULT_SH_MEM_BASES	(0x6000)
1853 #define FIRST_COMPUTE_VMID	(8)
1854 #define LAST_COMPUTE_VMID	(16)
1855 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1856 {
1857 	int i;
1858 	uint32_t sh_mem_config;
1859 	uint32_t sh_mem_bases;
1860 
1861 	/*
1862 	 * Configure apertures:
1863 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1864 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1865 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1866 	*/
1867 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1868 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1869 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1870 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1871 	mutex_lock(&adev->srbm_mutex);
1872 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1873 		cik_srbm_select(adev, 0, 0, 0, i);
1874 		/* CP and shaders */
1875 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1876 		WREG32(mmSH_MEM_APE1_BASE, 1);
1877 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1878 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1879 	}
1880 	cik_srbm_select(adev, 0, 0, 0, 0);
1881 	mutex_unlock(&adev->srbm_mutex);
1882 }
1883 
1884 static void gfx_v7_0_config_init(struct amdgpu_device *adev)
1885 {
1886 	adev->gfx.config.double_offchip_lds_buf = 1;
1887 }
1888 
1889 /**
1890  * gfx_v7_0_constants_init - setup the 3D engine
1891  *
1892  * @adev: amdgpu_device pointer
1893  *
1894  * init the gfx constants such as the 3D engine, tiling configuration
1895  * registers, maximum number of quad pipes, render backends...
1896  */
1897 static void gfx_v7_0_constants_init(struct amdgpu_device *adev)
1898 {
1899 	u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base;
1900 	u32 tmp;
1901 	int i;
1902 
1903 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1904 
1905 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1906 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1907 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1908 
1909 	gfx_v7_0_tiling_mode_table_init(adev);
1910 
1911 	gfx_v7_0_setup_rb(adev);
1912 	gfx_v7_0_get_cu_info(adev);
1913 	gfx_v7_0_config_init(adev);
1914 
1915 	/* set HW defaults for 3D engine */
1916 	WREG32(mmCP_MEQ_THRESHOLDS,
1917 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1918 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1919 
1920 	mutex_lock(&adev->grbm_idx_mutex);
1921 	/*
1922 	 * making sure that the following register writes will be broadcasted
1923 	 * to all the shaders
1924 	 */
1925 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1926 
1927 	/* XXX SH_MEM regs */
1928 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1929 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1930 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1931 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE,
1932 				   MTYPE_NC);
1933 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE,
1934 				   MTYPE_UC);
1935 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0);
1936 
1937 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
1938 				   SWIZZLE_ENABLE, 1);
1939 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1940 				   ELEMENT_SIZE, 1);
1941 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1942 				   INDEX_STRIDE, 3);
1943 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
1944 
1945 	mutex_lock(&adev->srbm_mutex);
1946 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
1947 		if (i == 0)
1948 			sh_mem_base = 0;
1949 		else
1950 			sh_mem_base = adev->gmc.shared_aperture_start >> 48;
1951 		cik_srbm_select(adev, 0, 0, 0, i);
1952 		/* CP and shaders */
1953 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1954 		WREG32(mmSH_MEM_APE1_BASE, 1);
1955 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1956 		WREG32(mmSH_MEM_BASES, sh_mem_base);
1957 	}
1958 	cik_srbm_select(adev, 0, 0, 0, 0);
1959 	mutex_unlock(&adev->srbm_mutex);
1960 
1961 	gfx_v7_0_init_compute_vmid(adev);
1962 
1963 	WREG32(mmSX_DEBUG_1, 0x20);
1964 
1965 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1966 
1967 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1968 	tmp |= 0x03000000;
1969 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1970 
1971 	WREG32(mmSQ_CONFIG, 1);
1972 
1973 	WREG32(mmDB_DEBUG, 0);
1974 
1975 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
1976 	tmp |= 0x00000400;
1977 	WREG32(mmDB_DEBUG2, tmp);
1978 
1979 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
1980 	tmp |= 0x00020200;
1981 	WREG32(mmDB_DEBUG3, tmp);
1982 
1983 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
1984 	tmp |= 0x00018208;
1985 	WREG32(mmCB_HW_CONTROL, tmp);
1986 
1987 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
1988 
1989 	WREG32(mmPA_SC_FIFO_SIZE,
1990 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
1991 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
1992 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
1993 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
1994 
1995 	WREG32(mmVGT_NUM_INSTANCES, 1);
1996 
1997 	WREG32(mmCP_PERFMON_CNTL, 0);
1998 
1999 	WREG32(mmSQ_CONFIG, 0);
2000 
2001 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
2002 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
2003 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
2004 
2005 	WREG32(mmVGT_CACHE_INVALIDATION,
2006 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
2007 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
2008 
2009 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
2010 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
2011 
2012 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
2013 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
2014 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
2015 
2016 	tmp = RREG32(mmSPI_ARB_PRIORITY);
2017 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
2018 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
2019 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
2020 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
2021 	WREG32(mmSPI_ARB_PRIORITY, tmp);
2022 
2023 	mutex_unlock(&adev->grbm_idx_mutex);
2024 
2025 	udelay(50);
2026 }
2027 
2028 /*
2029  * GPU scratch registers helpers function.
2030  */
2031 /**
2032  * gfx_v7_0_scratch_init - setup driver info for CP scratch regs
2033  *
2034  * @adev: amdgpu_device pointer
2035  *
2036  * Set up the number and offset of the CP scratch registers.
2037  * NOTE: use of CP scratch registers is a legacy inferface and
2038  * is not used by default on newer asics (r6xx+).  On newer asics,
2039  * memory buffers are used for fences rather than scratch regs.
2040  */
2041 static void gfx_v7_0_scratch_init(struct amdgpu_device *adev)
2042 {
2043 	adev->gfx.scratch.num_reg = 8;
2044 	adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
2045 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
2046 }
2047 
2048 /**
2049  * gfx_v7_0_ring_test_ring - basic gfx ring test
2050  *
2051  * @adev: amdgpu_device pointer
2052  * @ring: amdgpu_ring structure holding ring information
2053  *
2054  * Allocate a scratch register and write to it using the gfx ring (CIK).
2055  * Provides a basic gfx ring test to verify that the ring is working.
2056  * Used by gfx_v7_0_cp_gfx_resume();
2057  * Returns 0 on success, error on failure.
2058  */
2059 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2060 {
2061 	struct amdgpu_device *adev = ring->adev;
2062 	uint32_t scratch;
2063 	uint32_t tmp = 0;
2064 	unsigned i;
2065 	int r;
2066 
2067 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2068 	if (r)
2069 		return r;
2070 
2071 	WREG32(scratch, 0xCAFEDEAD);
2072 	r = amdgpu_ring_alloc(ring, 3);
2073 	if (r)
2074 		goto error_free_scratch;
2075 
2076 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2077 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
2078 	amdgpu_ring_write(ring, 0xDEADBEEF);
2079 	amdgpu_ring_commit(ring);
2080 
2081 	for (i = 0; i < adev->usec_timeout; i++) {
2082 		tmp = RREG32(scratch);
2083 		if (tmp == 0xDEADBEEF)
2084 			break;
2085 		udelay(1);
2086 	}
2087 	if (i >= adev->usec_timeout)
2088 		r = -ETIMEDOUT;
2089 
2090 error_free_scratch:
2091 	amdgpu_gfx_scratch_free(adev, scratch);
2092 	return r;
2093 }
2094 
2095 /**
2096  * gfx_v7_0_ring_emit_hdp - emit an hdp flush on the cp
2097  *
2098  * @adev: amdgpu_device pointer
2099  * @ridx: amdgpu ring index
2100  *
2101  * Emits an hdp flush on the cp.
2102  */
2103 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2104 {
2105 	u32 ref_and_mask;
2106 	int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2107 
2108 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2109 		switch (ring->me) {
2110 		case 1:
2111 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2112 			break;
2113 		case 2:
2114 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2115 			break;
2116 		default:
2117 			return;
2118 		}
2119 	} else {
2120 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2121 	}
2122 
2123 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2124 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2125 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2126 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2127 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2128 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2129 	amdgpu_ring_write(ring, ref_and_mask);
2130 	amdgpu_ring_write(ring, ref_and_mask);
2131 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2132 }
2133 
2134 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2135 {
2136 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2137 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2138 		EVENT_INDEX(4));
2139 
2140 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2141 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2142 		EVENT_INDEX(0));
2143 }
2144 
2145 /**
2146  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2147  *
2148  * @adev: amdgpu_device pointer
2149  * @fence: amdgpu fence object
2150  *
2151  * Emits a fence sequnce number on the gfx ring and flushes
2152  * GPU caches.
2153  */
2154 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2155 					 u64 seq, unsigned flags)
2156 {
2157 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2158 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2159 	/* Workaround for cache flush problems. First send a dummy EOP
2160 	 * event down the pipe with seq one below.
2161 	 */
2162 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2163 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2164 				 EOP_TC_ACTION_EN |
2165 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2166 				 EVENT_INDEX(5)));
2167 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2168 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2169 				DATA_SEL(1) | INT_SEL(0));
2170 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2171 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2172 
2173 	/* Then send the real EOP event down the pipe. */
2174 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2175 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2176 				 EOP_TC_ACTION_EN |
2177 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2178 				 EVENT_INDEX(5)));
2179 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2180 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2181 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2182 	amdgpu_ring_write(ring, lower_32_bits(seq));
2183 	amdgpu_ring_write(ring, upper_32_bits(seq));
2184 }
2185 
2186 /**
2187  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2188  *
2189  * @adev: amdgpu_device pointer
2190  * @fence: amdgpu fence object
2191  *
2192  * Emits a fence sequnce number on the compute ring and flushes
2193  * GPU caches.
2194  */
2195 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2196 					     u64 addr, u64 seq,
2197 					     unsigned flags)
2198 {
2199 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2200 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2201 
2202 	/* RELEASE_MEM - flush caches, send int */
2203 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2204 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2205 				 EOP_TC_ACTION_EN |
2206 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2207 				 EVENT_INDEX(5)));
2208 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2209 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2210 	amdgpu_ring_write(ring, upper_32_bits(addr));
2211 	amdgpu_ring_write(ring, lower_32_bits(seq));
2212 	amdgpu_ring_write(ring, upper_32_bits(seq));
2213 }
2214 
2215 /*
2216  * IB stuff
2217  */
2218 /**
2219  * gfx_v7_0_ring_emit_ib - emit an IB (Indirect Buffer) on the ring
2220  *
2221  * @ring: amdgpu_ring structure holding ring information
2222  * @ib: amdgpu indirect buffer object
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  * sheduling 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  *
2320  * Allocate an IB and execute it on the gfx ring (CIK).
2321  * Provides a basic gfx ring test to verify that IBs are working.
2322  * Returns 0 on success, error on failure.
2323  */
2324 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2325 {
2326 	struct amdgpu_device *adev = ring->adev;
2327 	struct amdgpu_ib ib;
2328 	struct dma_fence *f = NULL;
2329 	uint32_t scratch;
2330 	uint32_t tmp = 0;
2331 	long r;
2332 
2333 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2334 	if (r)
2335 		return r;
2336 
2337 	WREG32(scratch, 0xCAFEDEAD);
2338 	memset(&ib, 0, sizeof(ib));
2339 	r = amdgpu_ib_get(adev, NULL, 256, &ib);
2340 	if (r)
2341 		goto err1;
2342 
2343 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2344 	ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
2345 	ib.ptr[2] = 0xDEADBEEF;
2346 	ib.length_dw = 3;
2347 
2348 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2349 	if (r)
2350 		goto err2;
2351 
2352 	r = dma_fence_wait_timeout(f, false, timeout);
2353 	if (r == 0) {
2354 		r = -ETIMEDOUT;
2355 		goto err2;
2356 	} else if (r < 0) {
2357 		goto err2;
2358 	}
2359 	tmp = RREG32(scratch);
2360 	if (tmp == 0xDEADBEEF)
2361 		r = 0;
2362 	else
2363 		r = -EINVAL;
2364 
2365 err2:
2366 	amdgpu_ib_free(adev, &ib, NULL);
2367 	dma_fence_put(f);
2368 err1:
2369 	amdgpu_gfx_scratch_free(adev, scratch);
2370 	return r;
2371 }
2372 
2373 /*
2374  * CP.
2375  * On CIK, gfx and compute now have independant command processors.
2376  *
2377  * GFX
2378  * Gfx consists of a single ring and can process both gfx jobs and
2379  * compute jobs.  The gfx CP consists of three microengines (ME):
2380  * PFP - Pre-Fetch Parser
2381  * ME - Micro Engine
2382  * CE - Constant Engine
2383  * The PFP and ME make up what is considered the Drawing Engine (DE).
2384  * The CE is an asynchronous engine used for updating buffer desciptors
2385  * used by the DE so that they can be loaded into cache in parallel
2386  * while the DE is processing state update packets.
2387  *
2388  * Compute
2389  * The compute CP consists of two microengines (ME):
2390  * MEC1 - Compute MicroEngine 1
2391  * MEC2 - Compute MicroEngine 2
2392  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2393  * The queues are exposed to userspace and are programmed directly
2394  * by the compute runtime.
2395  */
2396 /**
2397  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2398  *
2399  * @adev: amdgpu_device pointer
2400  * @enable: enable or disable the MEs
2401  *
2402  * Halts or unhalts the gfx MEs.
2403  */
2404 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2405 {
2406 	int i;
2407 
2408 	if (enable) {
2409 		WREG32(mmCP_ME_CNTL, 0);
2410 	} else {
2411 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK));
2412 		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2413 			adev->gfx.gfx_ring[i].sched.ready = false;
2414 	}
2415 	udelay(50);
2416 }
2417 
2418 /**
2419  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2420  *
2421  * @adev: amdgpu_device pointer
2422  *
2423  * Loads the gfx PFP, ME, and CE ucode.
2424  * Returns 0 for success, -EINVAL if the ucode is not available.
2425  */
2426 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2427 {
2428 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2429 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2430 	const struct gfx_firmware_header_v1_0 *me_hdr;
2431 	const __le32 *fw_data;
2432 	unsigned i, fw_size;
2433 
2434 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2435 		return -EINVAL;
2436 
2437 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2438 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2439 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2440 
2441 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2442 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2443 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2444 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2445 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2446 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2447 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2448 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2449 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2450 
2451 	gfx_v7_0_cp_gfx_enable(adev, false);
2452 
2453 	/* PFP */
2454 	fw_data = (const __le32 *)
2455 		(adev->gfx.pfp_fw->data +
2456 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2457 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2458 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2459 	for (i = 0; i < fw_size; i++)
2460 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2461 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2462 
2463 	/* CE */
2464 	fw_data = (const __le32 *)
2465 		(adev->gfx.ce_fw->data +
2466 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2467 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2468 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2469 	for (i = 0; i < fw_size; i++)
2470 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2471 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2472 
2473 	/* ME */
2474 	fw_data = (const __le32 *)
2475 		(adev->gfx.me_fw->data +
2476 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2477 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2478 	WREG32(mmCP_ME_RAM_WADDR, 0);
2479 	for (i = 0; i < fw_size; i++)
2480 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2481 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2482 
2483 	return 0;
2484 }
2485 
2486 /**
2487  * gfx_v7_0_cp_gfx_start - start the gfx ring
2488  *
2489  * @adev: amdgpu_device pointer
2490  *
2491  * Enables the ring and loads the clear state context and other
2492  * packets required to init the ring.
2493  * Returns 0 for success, error for failure.
2494  */
2495 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2496 {
2497 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2498 	const struct cs_section_def *sect = NULL;
2499 	const struct cs_extent_def *ext = NULL;
2500 	int r, i;
2501 
2502 	/* init the CP */
2503 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2504 	WREG32(mmCP_ENDIAN_SWAP, 0);
2505 	WREG32(mmCP_DEVICE_ID, 1);
2506 
2507 	gfx_v7_0_cp_gfx_enable(adev, true);
2508 
2509 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2510 	if (r) {
2511 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2512 		return r;
2513 	}
2514 
2515 	/* init the CE partitions.  CE only used for gfx on CIK */
2516 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2517 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2518 	amdgpu_ring_write(ring, 0x8000);
2519 	amdgpu_ring_write(ring, 0x8000);
2520 
2521 	/* clear state buffer */
2522 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2523 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2524 
2525 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2526 	amdgpu_ring_write(ring, 0x80000000);
2527 	amdgpu_ring_write(ring, 0x80000000);
2528 
2529 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2530 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2531 			if (sect->id == SECT_CONTEXT) {
2532 				amdgpu_ring_write(ring,
2533 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2534 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2535 				for (i = 0; i < ext->reg_count; i++)
2536 					amdgpu_ring_write(ring, ext->extent[i]);
2537 			}
2538 		}
2539 	}
2540 
2541 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2542 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2543 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
2544 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
2545 
2546 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2547 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2548 
2549 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2550 	amdgpu_ring_write(ring, 0);
2551 
2552 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2553 	amdgpu_ring_write(ring, 0x00000316);
2554 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2555 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2556 
2557 	amdgpu_ring_commit(ring);
2558 
2559 	return 0;
2560 }
2561 
2562 /**
2563  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2564  *
2565  * @adev: amdgpu_device pointer
2566  *
2567  * Program the location and size of the gfx ring buffer
2568  * and test it to make sure it's working.
2569  * Returns 0 for success, error for failure.
2570  */
2571 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2572 {
2573 	struct amdgpu_ring *ring;
2574 	u32 tmp;
2575 	u32 rb_bufsz;
2576 	u64 rb_addr, rptr_addr;
2577 	int r;
2578 
2579 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2580 	if (adev->asic_type != CHIP_HAWAII)
2581 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2582 
2583 	/* Set the write pointer delay */
2584 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2585 
2586 	/* set the RB to use vmid 0 */
2587 	WREG32(mmCP_RB_VMID, 0);
2588 
2589 	WREG32(mmSCRATCH_ADDR, 0);
2590 
2591 	/* ring 0 - compute and gfx */
2592 	/* Set ring buffer size */
2593 	ring = &adev->gfx.gfx_ring[0];
2594 	rb_bufsz = order_base_2(ring->ring_size / 8);
2595 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2596 #ifdef __BIG_ENDIAN
2597 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2598 #endif
2599 	WREG32(mmCP_RB0_CNTL, tmp);
2600 
2601 	/* Initialize the ring buffer's read and write pointers */
2602 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2603 	ring->wptr = 0;
2604 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2605 
2606 	/* set the wb address wether it's enabled or not */
2607 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2608 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2609 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2610 
2611 	/* scratch register shadowing is no longer supported */
2612 	WREG32(mmSCRATCH_UMSK, 0);
2613 
2614 	mdelay(1);
2615 	WREG32(mmCP_RB0_CNTL, tmp);
2616 
2617 	rb_addr = ring->gpu_addr >> 8;
2618 	WREG32(mmCP_RB0_BASE, rb_addr);
2619 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2620 
2621 	/* start the ring */
2622 	gfx_v7_0_cp_gfx_start(adev);
2623 	r = amdgpu_ring_test_helper(ring);
2624 	if (r)
2625 		return r;
2626 
2627 	return 0;
2628 }
2629 
2630 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2631 {
2632 	return ring->adev->wb.wb[ring->rptr_offs];
2633 }
2634 
2635 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2636 {
2637 	struct amdgpu_device *adev = ring->adev;
2638 
2639 	return RREG32(mmCP_RB0_WPTR);
2640 }
2641 
2642 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2643 {
2644 	struct amdgpu_device *adev = ring->adev;
2645 
2646 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2647 	(void)RREG32(mmCP_RB0_WPTR);
2648 }
2649 
2650 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2651 {
2652 	/* XXX check if swapping is necessary on BE */
2653 	return ring->adev->wb.wb[ring->wptr_offs];
2654 }
2655 
2656 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2657 {
2658 	struct amdgpu_device *adev = ring->adev;
2659 
2660 	/* XXX check if swapping is necessary on BE */
2661 	adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2662 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2663 }
2664 
2665 /**
2666  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2667  *
2668  * @adev: amdgpu_device pointer
2669  * @enable: enable or disable the MEs
2670  *
2671  * Halts or unhalts the compute MEs.
2672  */
2673 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2674 {
2675 	int i;
2676 
2677 	if (enable) {
2678 		WREG32(mmCP_MEC_CNTL, 0);
2679 	} else {
2680 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2681 		for (i = 0; i < adev->gfx.num_compute_rings; i++)
2682 			adev->gfx.compute_ring[i].sched.ready = false;
2683 	}
2684 	udelay(50);
2685 }
2686 
2687 /**
2688  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2689  *
2690  * @adev: amdgpu_device pointer
2691  *
2692  * Loads the compute MEC1&2 ucode.
2693  * Returns 0 for success, -EINVAL if the ucode is not available.
2694  */
2695 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2696 {
2697 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2698 	const __le32 *fw_data;
2699 	unsigned i, fw_size;
2700 
2701 	if (!adev->gfx.mec_fw)
2702 		return -EINVAL;
2703 
2704 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2705 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2706 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2707 	adev->gfx.mec_feature_version = le32_to_cpu(
2708 					mec_hdr->ucode_feature_version);
2709 
2710 	gfx_v7_0_cp_compute_enable(adev, false);
2711 
2712 	/* MEC1 */
2713 	fw_data = (const __le32 *)
2714 		(adev->gfx.mec_fw->data +
2715 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2716 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2717 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2718 	for (i = 0; i < fw_size; i++)
2719 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2720 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2721 
2722 	if (adev->asic_type == CHIP_KAVERI) {
2723 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2724 
2725 		if (!adev->gfx.mec2_fw)
2726 			return -EINVAL;
2727 
2728 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2729 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2730 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2731 		adev->gfx.mec2_feature_version = le32_to_cpu(
2732 				mec2_hdr->ucode_feature_version);
2733 
2734 		/* MEC2 */
2735 		fw_data = (const __le32 *)
2736 			(adev->gfx.mec2_fw->data +
2737 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2738 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2739 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2740 		for (i = 0; i < fw_size; i++)
2741 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2742 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2743 	}
2744 
2745 	return 0;
2746 }
2747 
2748 /**
2749  * gfx_v7_0_cp_compute_fini - stop the compute queues
2750  *
2751  * @adev: amdgpu_device pointer
2752  *
2753  * Stop the compute queues and tear down the driver queue
2754  * info.
2755  */
2756 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2757 {
2758 	int i;
2759 
2760 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2761 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2762 
2763 		amdgpu_bo_free_kernel(&ring->mqd_obj, NULL, NULL);
2764 	}
2765 }
2766 
2767 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2768 {
2769 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
2770 }
2771 
2772 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2773 {
2774 	int r;
2775 	u32 *hpd;
2776 	size_t mec_hpd_size;
2777 
2778 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
2779 
2780 	/* take ownership of the relevant compute queues */
2781 	amdgpu_gfx_compute_queue_acquire(adev);
2782 
2783 	/* allocate space for ALL pipes (even the ones we don't own) */
2784 	mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec
2785 		* GFX7_MEC_HPD_SIZE * 2;
2786 
2787 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
2788 				      AMDGPU_GEM_DOMAIN_VRAM,
2789 				      &adev->gfx.mec.hpd_eop_obj,
2790 				      &adev->gfx.mec.hpd_eop_gpu_addr,
2791 				      (void **)&hpd);
2792 	if (r) {
2793 		dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r);
2794 		gfx_v7_0_mec_fini(adev);
2795 		return r;
2796 	}
2797 
2798 	/* clear memory.  Not sure if this is required or not */
2799 	memset(hpd, 0, mec_hpd_size);
2800 
2801 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2802 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2803 
2804 	return 0;
2805 }
2806 
2807 struct hqd_registers
2808 {
2809 	u32 cp_mqd_base_addr;
2810 	u32 cp_mqd_base_addr_hi;
2811 	u32 cp_hqd_active;
2812 	u32 cp_hqd_vmid;
2813 	u32 cp_hqd_persistent_state;
2814 	u32 cp_hqd_pipe_priority;
2815 	u32 cp_hqd_queue_priority;
2816 	u32 cp_hqd_quantum;
2817 	u32 cp_hqd_pq_base;
2818 	u32 cp_hqd_pq_base_hi;
2819 	u32 cp_hqd_pq_rptr;
2820 	u32 cp_hqd_pq_rptr_report_addr;
2821 	u32 cp_hqd_pq_rptr_report_addr_hi;
2822 	u32 cp_hqd_pq_wptr_poll_addr;
2823 	u32 cp_hqd_pq_wptr_poll_addr_hi;
2824 	u32 cp_hqd_pq_doorbell_control;
2825 	u32 cp_hqd_pq_wptr;
2826 	u32 cp_hqd_pq_control;
2827 	u32 cp_hqd_ib_base_addr;
2828 	u32 cp_hqd_ib_base_addr_hi;
2829 	u32 cp_hqd_ib_rptr;
2830 	u32 cp_hqd_ib_control;
2831 	u32 cp_hqd_iq_timer;
2832 	u32 cp_hqd_iq_rptr;
2833 	u32 cp_hqd_dequeue_request;
2834 	u32 cp_hqd_dma_offload;
2835 	u32 cp_hqd_sema_cmd;
2836 	u32 cp_hqd_msg_type;
2837 	u32 cp_hqd_atomic0_preop_lo;
2838 	u32 cp_hqd_atomic0_preop_hi;
2839 	u32 cp_hqd_atomic1_preop_lo;
2840 	u32 cp_hqd_atomic1_preop_hi;
2841 	u32 cp_hqd_hq_scheduler0;
2842 	u32 cp_hqd_hq_scheduler1;
2843 	u32 cp_mqd_control;
2844 };
2845 
2846 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev,
2847 				       int mec, int pipe)
2848 {
2849 	u64 eop_gpu_addr;
2850 	u32 tmp;
2851 	size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe)
2852 			    * GFX7_MEC_HPD_SIZE * 2;
2853 
2854 	mutex_lock(&adev->srbm_mutex);
2855 	eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset;
2856 
2857 	cik_srbm_select(adev, mec + 1, pipe, 0, 0);
2858 
2859 	/* write the EOP addr */
2860 	WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2861 	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2862 
2863 	/* set the VMID assigned */
2864 	WREG32(mmCP_HPD_EOP_VMID, 0);
2865 
2866 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2867 	tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2868 	tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2869 	tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8);
2870 	WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2871 
2872 	cik_srbm_select(adev, 0, 0, 0, 0);
2873 	mutex_unlock(&adev->srbm_mutex);
2874 }
2875 
2876 static int gfx_v7_0_mqd_deactivate(struct amdgpu_device *adev)
2877 {
2878 	int i;
2879 
2880 	/* disable the queue if it's active */
2881 	if (RREG32(mmCP_HQD_ACTIVE) & 1) {
2882 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
2883 		for (i = 0; i < adev->usec_timeout; i++) {
2884 			if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
2885 				break;
2886 			udelay(1);
2887 		}
2888 
2889 		if (i == adev->usec_timeout)
2890 			return -ETIMEDOUT;
2891 
2892 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
2893 		WREG32(mmCP_HQD_PQ_RPTR, 0);
2894 		WREG32(mmCP_HQD_PQ_WPTR, 0);
2895 	}
2896 
2897 	return 0;
2898 }
2899 
2900 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev,
2901 			     struct cik_mqd *mqd,
2902 			     uint64_t mqd_gpu_addr,
2903 			     struct amdgpu_ring *ring)
2904 {
2905 	u64 hqd_gpu_addr;
2906 	u64 wb_gpu_addr;
2907 
2908 	/* init the mqd struct */
2909 	memset(mqd, 0, sizeof(struct cik_mqd));
2910 
2911 	mqd->header = 0xC0310800;
2912 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2913 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2914 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2915 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2916 
2917 	/* enable doorbell? */
2918 	mqd->cp_hqd_pq_doorbell_control =
2919 		RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2920 	if (ring->use_doorbell)
2921 		mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2922 	else
2923 		mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2924 
2925 	/* set the pointer to the MQD */
2926 	mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc;
2927 	mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
2928 
2929 	/* set MQD vmid to 0 */
2930 	mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
2931 	mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
2932 
2933 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2934 	hqd_gpu_addr = ring->gpu_addr >> 8;
2935 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2936 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2937 
2938 	/* set up the HQD, this is similar to CP_RB0_CNTL */
2939 	mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
2940 	mqd->cp_hqd_pq_control &=
2941 		~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
2942 				CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
2943 
2944 	mqd->cp_hqd_pq_control |=
2945 		order_base_2(ring->ring_size / 8);
2946 	mqd->cp_hqd_pq_control |=
2947 		(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
2948 #ifdef __BIG_ENDIAN
2949 	mqd->cp_hqd_pq_control |=
2950 		2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
2951 #endif
2952 	mqd->cp_hqd_pq_control &=
2953 		~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
2954 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
2955 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
2956 	mqd->cp_hqd_pq_control |=
2957 		CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
2958 		CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
2959 
2960 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2961 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2962 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2963 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2964 
2965 	/* set the wb address wether it's enabled or not */
2966 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2967 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2968 	mqd->cp_hqd_pq_rptr_report_addr_hi =
2969 		upper_32_bits(wb_gpu_addr) & 0xffff;
2970 
2971 	/* enable the doorbell if requested */
2972 	if (ring->use_doorbell) {
2973 		mqd->cp_hqd_pq_doorbell_control =
2974 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2975 		mqd->cp_hqd_pq_doorbell_control &=
2976 			~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
2977 		mqd->cp_hqd_pq_doorbell_control |=
2978 			(ring->doorbell_index <<
2979 			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
2980 		mqd->cp_hqd_pq_doorbell_control |=
2981 			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2982 		mqd->cp_hqd_pq_doorbell_control &=
2983 			~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
2984 					CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
2985 
2986 	} else {
2987 		mqd->cp_hqd_pq_doorbell_control = 0;
2988 	}
2989 
2990 	/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
2991 	ring->wptr = 0;
2992 	mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr);
2993 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2994 
2995 	/* set the vmid for the queue */
2996 	mqd->cp_hqd_vmid = 0;
2997 
2998 	/* defaults */
2999 	mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL);
3000 	mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR);
3001 	mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI);
3002 	mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR);
3003 	mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE);
3004 	mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD);
3005 	mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE);
3006 	mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO);
3007 	mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI);
3008 	mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO);
3009 	mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI);
3010 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
3011 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
3012 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
3013 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
3014 	mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR);
3015 
3016 	/* activate the queue */
3017 	mqd->cp_hqd_active = 1;
3018 }
3019 
3020 int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd)
3021 {
3022 	uint32_t tmp;
3023 	uint32_t mqd_reg;
3024 	uint32_t *mqd_data;
3025 
3026 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */
3027 	mqd_data = &mqd->cp_mqd_base_addr_lo;
3028 
3029 	/* disable wptr polling */
3030 	tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
3031 	tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3032 	WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
3033 
3034 	/* program all HQD registers */
3035 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++)
3036 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3037 
3038 	/* activate the HQD */
3039 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
3040 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3041 
3042 	return 0;
3043 }
3044 
3045 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id)
3046 {
3047 	int r;
3048 	u64 mqd_gpu_addr;
3049 	struct cik_mqd *mqd;
3050 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
3051 
3052 	r = amdgpu_bo_create_reserved(adev, sizeof(struct cik_mqd), PAGE_SIZE,
3053 				      AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj,
3054 				      &mqd_gpu_addr, (void **)&mqd);
3055 	if (r) {
3056 		dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
3057 		return r;
3058 	}
3059 
3060 	mutex_lock(&adev->srbm_mutex);
3061 	cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3062 
3063 	gfx_v7_0_mqd_init(adev, mqd, mqd_gpu_addr, ring);
3064 	gfx_v7_0_mqd_deactivate(adev);
3065 	gfx_v7_0_mqd_commit(adev, mqd);
3066 
3067 	cik_srbm_select(adev, 0, 0, 0, 0);
3068 	mutex_unlock(&adev->srbm_mutex);
3069 
3070 	amdgpu_bo_kunmap(ring->mqd_obj);
3071 	amdgpu_bo_unreserve(ring->mqd_obj);
3072 	return 0;
3073 }
3074 
3075 /**
3076  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
3077  *
3078  * @adev: amdgpu_device pointer
3079  *
3080  * Program the compute queues and test them to make sure they
3081  * are working.
3082  * Returns 0 for success, error for failure.
3083  */
3084 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
3085 {
3086 	int r, i, j;
3087 	u32 tmp;
3088 	struct amdgpu_ring *ring;
3089 
3090 	/* fix up chicken bits */
3091 	tmp = RREG32(mmCP_CPF_DEBUG);
3092 	tmp |= (1 << 23);
3093 	WREG32(mmCP_CPF_DEBUG, tmp);
3094 
3095 	/* init all pipes (even the ones we don't own) */
3096 	for (i = 0; i < adev->gfx.mec.num_mec; i++)
3097 		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++)
3098 			gfx_v7_0_compute_pipe_init(adev, i, j);
3099 
3100 	/* init the queues */
3101 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3102 		r = gfx_v7_0_compute_queue_init(adev, i);
3103 		if (r) {
3104 			gfx_v7_0_cp_compute_fini(adev);
3105 			return r;
3106 		}
3107 	}
3108 
3109 	gfx_v7_0_cp_compute_enable(adev, true);
3110 
3111 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3112 		ring = &adev->gfx.compute_ring[i];
3113 		amdgpu_ring_test_helper(ring);
3114 	}
3115 
3116 	return 0;
3117 }
3118 
3119 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3120 {
3121 	gfx_v7_0_cp_gfx_enable(adev, enable);
3122 	gfx_v7_0_cp_compute_enable(adev, enable);
3123 }
3124 
3125 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3126 {
3127 	int r;
3128 
3129 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3130 	if (r)
3131 		return r;
3132 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3133 	if (r)
3134 		return r;
3135 
3136 	return 0;
3137 }
3138 
3139 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3140 					       bool enable)
3141 {
3142 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3143 
3144 	if (enable)
3145 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3146 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3147 	else
3148 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3149 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3150 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3151 }
3152 
3153 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3154 {
3155 	int r;
3156 
3157 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3158 
3159 	r = gfx_v7_0_cp_load_microcode(adev);
3160 	if (r)
3161 		return r;
3162 
3163 	r = gfx_v7_0_cp_gfx_resume(adev);
3164 	if (r)
3165 		return r;
3166 	r = gfx_v7_0_cp_compute_resume(adev);
3167 	if (r)
3168 		return r;
3169 
3170 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3171 
3172 	return 0;
3173 }
3174 
3175 /**
3176  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3177  *
3178  * @ring: the ring to emmit the commands to
3179  *
3180  * Sync the command pipeline with the PFP. E.g. wait for everything
3181  * to be completed.
3182  */
3183 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3184 {
3185 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3186 	uint32_t seq = ring->fence_drv.sync_seq;
3187 	uint64_t addr = ring->fence_drv.gpu_addr;
3188 
3189 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3190 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3191 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3192 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3193 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3194 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3195 	amdgpu_ring_write(ring, seq);
3196 	amdgpu_ring_write(ring, 0xffffffff);
3197 	amdgpu_ring_write(ring, 4); /* poll interval */
3198 
3199 	if (usepfp) {
3200 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3201 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3202 		amdgpu_ring_write(ring, 0);
3203 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3204 		amdgpu_ring_write(ring, 0);
3205 	}
3206 }
3207 
3208 /*
3209  * vm
3210  * VMID 0 is the physical GPU addresses as used by the kernel.
3211  * VMIDs 1-15 are used for userspace clients and are handled
3212  * by the amdgpu vm/hsa code.
3213  */
3214 /**
3215  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3216  *
3217  * @adev: amdgpu_device pointer
3218  *
3219  * Update the page table base and flush the VM TLB
3220  * using the CP (CIK).
3221  */
3222 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3223 					unsigned vmid, uint64_t pd_addr)
3224 {
3225 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3226 
3227 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3228 
3229 	/* wait for the invalidate to complete */
3230 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3231 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3232 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3233 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3234 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3235 	amdgpu_ring_write(ring, 0);
3236 	amdgpu_ring_write(ring, 0); /* ref */
3237 	amdgpu_ring_write(ring, 0); /* mask */
3238 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3239 
3240 	/* compute doesn't have PFP */
3241 	if (usepfp) {
3242 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3243 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3244 		amdgpu_ring_write(ring, 0x0);
3245 
3246 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3247 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3248 		amdgpu_ring_write(ring, 0);
3249 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3250 		amdgpu_ring_write(ring, 0);
3251 	}
3252 }
3253 
3254 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
3255 				    uint32_t reg, uint32_t val)
3256 {
3257 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3258 
3259 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3260 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3261 				 WRITE_DATA_DST_SEL(0)));
3262 	amdgpu_ring_write(ring, reg);
3263 	amdgpu_ring_write(ring, 0);
3264 	amdgpu_ring_write(ring, val);
3265 }
3266 
3267 /*
3268  * RLC
3269  * The RLC is a multi-purpose microengine that handles a
3270  * variety of functions.
3271  */
3272 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3273 {
3274 	const u32 *src_ptr;
3275 	u32 dws;
3276 	const struct cs_section_def *cs_data;
3277 	int r;
3278 
3279 	/* allocate rlc buffers */
3280 	if (adev->flags & AMD_IS_APU) {
3281 		if (adev->asic_type == CHIP_KAVERI) {
3282 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3283 			adev->gfx.rlc.reg_list_size =
3284 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3285 		} else {
3286 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3287 			adev->gfx.rlc.reg_list_size =
3288 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3289 		}
3290 	}
3291 	adev->gfx.rlc.cs_data = ci_cs_data;
3292 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3293 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3294 
3295 	src_ptr = adev->gfx.rlc.reg_list;
3296 	dws = adev->gfx.rlc.reg_list_size;
3297 	dws += (5 * 16) + 48 + 48 + 64;
3298 
3299 	cs_data = adev->gfx.rlc.cs_data;
3300 
3301 	if (src_ptr) {
3302 		/* init save restore block */
3303 		r = amdgpu_gfx_rlc_init_sr(adev, dws);
3304 		if (r)
3305 			return r;
3306 	}
3307 
3308 	if (cs_data) {
3309 		/* init clear state block */
3310 		r = amdgpu_gfx_rlc_init_csb(adev);
3311 		if (r)
3312 			return r;
3313 	}
3314 
3315 	if (adev->gfx.rlc.cp_table_size) {
3316 		r = amdgpu_gfx_rlc_init_cpt(adev);
3317 		if (r)
3318 			return r;
3319 	}
3320 
3321 	return 0;
3322 }
3323 
3324 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3325 {
3326 	u32 tmp;
3327 
3328 	tmp = RREG32(mmRLC_LB_CNTL);
3329 	if (enable)
3330 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3331 	else
3332 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3333 	WREG32(mmRLC_LB_CNTL, tmp);
3334 }
3335 
3336 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3337 {
3338 	u32 i, j, k;
3339 	u32 mask;
3340 
3341 	mutex_lock(&adev->grbm_idx_mutex);
3342 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3343 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3344 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
3345 			for (k = 0; k < adev->usec_timeout; k++) {
3346 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3347 					break;
3348 				udelay(1);
3349 			}
3350 		}
3351 	}
3352 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3353 	mutex_unlock(&adev->grbm_idx_mutex);
3354 
3355 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3356 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3357 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3358 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3359 	for (k = 0; k < adev->usec_timeout; k++) {
3360 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3361 			break;
3362 		udelay(1);
3363 	}
3364 }
3365 
3366 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3367 {
3368 	u32 tmp;
3369 
3370 	tmp = RREG32(mmRLC_CNTL);
3371 	if (tmp != rlc)
3372 		WREG32(mmRLC_CNTL, rlc);
3373 }
3374 
3375 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3376 {
3377 	u32 data, orig;
3378 
3379 	orig = data = RREG32(mmRLC_CNTL);
3380 
3381 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3382 		u32 i;
3383 
3384 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3385 		WREG32(mmRLC_CNTL, data);
3386 
3387 		for (i = 0; i < adev->usec_timeout; i++) {
3388 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3389 				break;
3390 			udelay(1);
3391 		}
3392 
3393 		gfx_v7_0_wait_for_rlc_serdes(adev);
3394 	}
3395 
3396 	return orig;
3397 }
3398 
3399 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev)
3400 {
3401 	return true;
3402 }
3403 
3404 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev)
3405 {
3406 	u32 tmp, i, mask;
3407 
3408 	tmp = 0x1 | (1 << 1);
3409 	WREG32(mmRLC_GPR_REG2, tmp);
3410 
3411 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3412 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3413 	for (i = 0; i < adev->usec_timeout; i++) {
3414 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3415 			break;
3416 		udelay(1);
3417 	}
3418 
3419 	for (i = 0; i < adev->usec_timeout; i++) {
3420 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3421 			break;
3422 		udelay(1);
3423 	}
3424 }
3425 
3426 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev)
3427 {
3428 	u32 tmp;
3429 
3430 	tmp = 0x1 | (0 << 1);
3431 	WREG32(mmRLC_GPR_REG2, tmp);
3432 }
3433 
3434 /**
3435  * gfx_v7_0_rlc_stop - stop the RLC ME
3436  *
3437  * @adev: amdgpu_device pointer
3438  *
3439  * Halt the RLC ME (MicroEngine) (CIK).
3440  */
3441 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3442 {
3443 	WREG32(mmRLC_CNTL, 0);
3444 
3445 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3446 
3447 	gfx_v7_0_wait_for_rlc_serdes(adev);
3448 }
3449 
3450 /**
3451  * gfx_v7_0_rlc_start - start the RLC ME
3452  *
3453  * @adev: amdgpu_device pointer
3454  *
3455  * Unhalt the RLC ME (MicroEngine) (CIK).
3456  */
3457 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3458 {
3459 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3460 
3461 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3462 
3463 	udelay(50);
3464 }
3465 
3466 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3467 {
3468 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3469 
3470 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3471 	WREG32(mmGRBM_SOFT_RESET, tmp);
3472 	udelay(50);
3473 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3474 	WREG32(mmGRBM_SOFT_RESET, tmp);
3475 	udelay(50);
3476 }
3477 
3478 /**
3479  * gfx_v7_0_rlc_resume - setup the RLC hw
3480  *
3481  * @adev: amdgpu_device pointer
3482  *
3483  * Initialize the RLC registers, load the ucode,
3484  * and start the RLC (CIK).
3485  * Returns 0 for success, -EINVAL if the ucode is not available.
3486  */
3487 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3488 {
3489 	const struct rlc_firmware_header_v1_0 *hdr;
3490 	const __le32 *fw_data;
3491 	unsigned i, fw_size;
3492 	u32 tmp;
3493 
3494 	if (!adev->gfx.rlc_fw)
3495 		return -EINVAL;
3496 
3497 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3498 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3499 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3500 	adev->gfx.rlc_feature_version = le32_to_cpu(
3501 					hdr->ucode_feature_version);
3502 
3503 	adev->gfx.rlc.funcs->stop(adev);
3504 
3505 	/* disable CG */
3506 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3507 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3508 
3509 	adev->gfx.rlc.funcs->reset(adev);
3510 
3511 	gfx_v7_0_init_pg(adev);
3512 
3513 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3514 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3515 
3516 	mutex_lock(&adev->grbm_idx_mutex);
3517 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3518 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3519 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3520 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3521 	mutex_unlock(&adev->grbm_idx_mutex);
3522 
3523 	WREG32(mmRLC_MC_CNTL, 0);
3524 	WREG32(mmRLC_UCODE_CNTL, 0);
3525 
3526 	fw_data = (const __le32 *)
3527 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3528 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3529 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3530 	for (i = 0; i < fw_size; i++)
3531 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3532 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3533 
3534 	/* XXX - find out what chips support lbpw */
3535 	gfx_v7_0_enable_lbpw(adev, false);
3536 
3537 	if (adev->asic_type == CHIP_BONAIRE)
3538 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3539 
3540 	adev->gfx.rlc.funcs->start(adev);
3541 
3542 	return 0;
3543 }
3544 
3545 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3546 {
3547 	u32 data, orig, tmp, tmp2;
3548 
3549 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3550 
3551 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3552 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3553 
3554 		tmp = gfx_v7_0_halt_rlc(adev);
3555 
3556 		mutex_lock(&adev->grbm_idx_mutex);
3557 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3558 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3559 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3560 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3561 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3562 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3563 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3564 		mutex_unlock(&adev->grbm_idx_mutex);
3565 
3566 		gfx_v7_0_update_rlc(adev, tmp);
3567 
3568 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3569 		if (orig != data)
3570 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3571 
3572 	} else {
3573 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3574 
3575 		RREG32(mmCB_CGTT_SCLK_CTRL);
3576 		RREG32(mmCB_CGTT_SCLK_CTRL);
3577 		RREG32(mmCB_CGTT_SCLK_CTRL);
3578 		RREG32(mmCB_CGTT_SCLK_CTRL);
3579 
3580 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3581 		if (orig != data)
3582 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3583 
3584 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3585 	}
3586 }
3587 
3588 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3589 {
3590 	u32 data, orig, tmp = 0;
3591 
3592 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3593 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3594 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3595 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3596 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3597 				if (orig != data)
3598 					WREG32(mmCP_MEM_SLP_CNTL, data);
3599 			}
3600 		}
3601 
3602 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3603 		data |= 0x00000001;
3604 		data &= 0xfffffffd;
3605 		if (orig != data)
3606 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3607 
3608 		tmp = gfx_v7_0_halt_rlc(adev);
3609 
3610 		mutex_lock(&adev->grbm_idx_mutex);
3611 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3612 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3613 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3614 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3615 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3616 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3617 		mutex_unlock(&adev->grbm_idx_mutex);
3618 
3619 		gfx_v7_0_update_rlc(adev, tmp);
3620 
3621 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3622 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3623 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3624 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3625 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3626 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3627 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3628 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3629 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3630 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3631 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3632 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3633 			if (orig != data)
3634 				WREG32(mmCGTS_SM_CTRL_REG, data);
3635 		}
3636 	} else {
3637 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3638 		data |= 0x00000003;
3639 		if (orig != data)
3640 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3641 
3642 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3643 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3644 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3645 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3646 		}
3647 
3648 		data = RREG32(mmCP_MEM_SLP_CNTL);
3649 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3650 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3651 			WREG32(mmCP_MEM_SLP_CNTL, data);
3652 		}
3653 
3654 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3655 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3656 		if (orig != data)
3657 			WREG32(mmCGTS_SM_CTRL_REG, data);
3658 
3659 		tmp = gfx_v7_0_halt_rlc(adev);
3660 
3661 		mutex_lock(&adev->grbm_idx_mutex);
3662 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3663 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3664 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3665 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3666 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3667 		mutex_unlock(&adev->grbm_idx_mutex);
3668 
3669 		gfx_v7_0_update_rlc(adev, tmp);
3670 	}
3671 }
3672 
3673 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3674 			       bool enable)
3675 {
3676 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3677 	/* order matters! */
3678 	if (enable) {
3679 		gfx_v7_0_enable_mgcg(adev, true);
3680 		gfx_v7_0_enable_cgcg(adev, true);
3681 	} else {
3682 		gfx_v7_0_enable_cgcg(adev, false);
3683 		gfx_v7_0_enable_mgcg(adev, false);
3684 	}
3685 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3686 }
3687 
3688 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3689 						bool enable)
3690 {
3691 	u32 data, orig;
3692 
3693 	orig = data = RREG32(mmRLC_PG_CNTL);
3694 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3695 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3696 	else
3697 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3698 	if (orig != data)
3699 		WREG32(mmRLC_PG_CNTL, data);
3700 }
3701 
3702 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3703 						bool enable)
3704 {
3705 	u32 data, orig;
3706 
3707 	orig = data = RREG32(mmRLC_PG_CNTL);
3708 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3709 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3710 	else
3711 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3712 	if (orig != data)
3713 		WREG32(mmRLC_PG_CNTL, data);
3714 }
3715 
3716 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3717 {
3718 	u32 data, orig;
3719 
3720 	orig = data = RREG32(mmRLC_PG_CNTL);
3721 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3722 		data &= ~0x8000;
3723 	else
3724 		data |= 0x8000;
3725 	if (orig != data)
3726 		WREG32(mmRLC_PG_CNTL, data);
3727 }
3728 
3729 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3730 {
3731 	u32 data, orig;
3732 
3733 	orig = data = RREG32(mmRLC_PG_CNTL);
3734 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3735 		data &= ~0x2000;
3736 	else
3737 		data |= 0x2000;
3738 	if (orig != data)
3739 		WREG32(mmRLC_PG_CNTL, data);
3740 }
3741 
3742 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev)
3743 {
3744 	if (adev->asic_type == CHIP_KAVERI)
3745 		return 5;
3746 	else
3747 		return 4;
3748 }
3749 
3750 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
3751 				     bool enable)
3752 {
3753 	u32 data, orig;
3754 
3755 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
3756 		orig = data = RREG32(mmRLC_PG_CNTL);
3757 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3758 		if (orig != data)
3759 			WREG32(mmRLC_PG_CNTL, data);
3760 
3761 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3762 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3763 		if (orig != data)
3764 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3765 	} else {
3766 		orig = data = RREG32(mmRLC_PG_CNTL);
3767 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3768 		if (orig != data)
3769 			WREG32(mmRLC_PG_CNTL, data);
3770 
3771 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3772 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3773 		if (orig != data)
3774 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3775 
3776 		data = RREG32(mmDB_RENDER_CONTROL);
3777 	}
3778 }
3779 
3780 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
3781 						 u32 bitmap)
3782 {
3783 	u32 data;
3784 
3785 	if (!bitmap)
3786 		return;
3787 
3788 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3789 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3790 
3791 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
3792 }
3793 
3794 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
3795 {
3796 	u32 data, mask;
3797 
3798 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
3799 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
3800 
3801 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3802 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3803 
3804 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
3805 
3806 	return (~data) & mask;
3807 }
3808 
3809 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
3810 {
3811 	u32 tmp;
3812 
3813 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
3814 
3815 	tmp = RREG32(mmRLC_MAX_PG_CU);
3816 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
3817 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
3818 	WREG32(mmRLC_MAX_PG_CU, tmp);
3819 }
3820 
3821 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
3822 					    bool enable)
3823 {
3824 	u32 data, orig;
3825 
3826 	orig = data = RREG32(mmRLC_PG_CNTL);
3827 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
3828 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3829 	else
3830 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3831 	if (orig != data)
3832 		WREG32(mmRLC_PG_CNTL, data);
3833 }
3834 
3835 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
3836 					     bool enable)
3837 {
3838 	u32 data, orig;
3839 
3840 	orig = data = RREG32(mmRLC_PG_CNTL);
3841 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
3842 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3843 	else
3844 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3845 	if (orig != data)
3846 		WREG32(mmRLC_PG_CNTL, data);
3847 }
3848 
3849 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
3850 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
3851 
3852 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
3853 {
3854 	u32 data, orig;
3855 	u32 i;
3856 
3857 	if (adev->gfx.rlc.cs_data) {
3858 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3859 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3860 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3861 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
3862 	} else {
3863 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3864 		for (i = 0; i < 3; i++)
3865 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
3866 	}
3867 	if (adev->gfx.rlc.reg_list) {
3868 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
3869 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3870 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
3871 	}
3872 
3873 	orig = data = RREG32(mmRLC_PG_CNTL);
3874 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
3875 	if (orig != data)
3876 		WREG32(mmRLC_PG_CNTL, data);
3877 
3878 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
3879 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
3880 
3881 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
3882 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
3883 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3884 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
3885 
3886 	data = 0x10101010;
3887 	WREG32(mmRLC_PG_DELAY, data);
3888 
3889 	data = RREG32(mmRLC_PG_DELAY_2);
3890 	data &= ~0xff;
3891 	data |= 0x3;
3892 	WREG32(mmRLC_PG_DELAY_2, data);
3893 
3894 	data = RREG32(mmRLC_AUTO_PG_CTRL);
3895 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
3896 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
3897 	WREG32(mmRLC_AUTO_PG_CTRL, data);
3898 
3899 }
3900 
3901 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
3902 {
3903 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
3904 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
3905 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
3906 }
3907 
3908 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
3909 {
3910 	u32 count = 0;
3911 	const struct cs_section_def *sect = NULL;
3912 	const struct cs_extent_def *ext = NULL;
3913 
3914 	if (adev->gfx.rlc.cs_data == NULL)
3915 		return 0;
3916 
3917 	/* begin clear state */
3918 	count += 2;
3919 	/* context control state */
3920 	count += 3;
3921 
3922 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3923 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3924 			if (sect->id == SECT_CONTEXT)
3925 				count += 2 + ext->reg_count;
3926 			else
3927 				return 0;
3928 		}
3929 	}
3930 	/* pa_sc_raster_config/pa_sc_raster_config1 */
3931 	count += 4;
3932 	/* end clear state */
3933 	count += 2;
3934 	/* clear state */
3935 	count += 2;
3936 
3937 	return count;
3938 }
3939 
3940 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
3941 				    volatile u32 *buffer)
3942 {
3943 	u32 count = 0, i;
3944 	const struct cs_section_def *sect = NULL;
3945 	const struct cs_extent_def *ext = NULL;
3946 
3947 	if (adev->gfx.rlc.cs_data == NULL)
3948 		return;
3949 	if (buffer == NULL)
3950 		return;
3951 
3952 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3953 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3954 
3955 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3956 	buffer[count++] = cpu_to_le32(0x80000000);
3957 	buffer[count++] = cpu_to_le32(0x80000000);
3958 
3959 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3960 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3961 			if (sect->id == SECT_CONTEXT) {
3962 				buffer[count++] =
3963 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
3964 				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3965 				for (i = 0; i < ext->reg_count; i++)
3966 					buffer[count++] = cpu_to_le32(ext->extent[i]);
3967 			} else {
3968 				return;
3969 			}
3970 		}
3971 	}
3972 
3973 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
3974 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
3975 	switch (adev->asic_type) {
3976 	case CHIP_BONAIRE:
3977 		buffer[count++] = cpu_to_le32(0x16000012);
3978 		buffer[count++] = cpu_to_le32(0x00000000);
3979 		break;
3980 	case CHIP_KAVERI:
3981 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3982 		buffer[count++] = cpu_to_le32(0x00000000);
3983 		break;
3984 	case CHIP_KABINI:
3985 	case CHIP_MULLINS:
3986 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3987 		buffer[count++] = cpu_to_le32(0x00000000);
3988 		break;
3989 	case CHIP_HAWAII:
3990 		buffer[count++] = cpu_to_le32(0x3a00161a);
3991 		buffer[count++] = cpu_to_le32(0x0000002e);
3992 		break;
3993 	default:
3994 		buffer[count++] = cpu_to_le32(0x00000000);
3995 		buffer[count++] = cpu_to_le32(0x00000000);
3996 		break;
3997 	}
3998 
3999 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4000 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
4001 
4002 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
4003 	buffer[count++] = cpu_to_le32(0);
4004 }
4005 
4006 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
4007 {
4008 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4009 			      AMD_PG_SUPPORT_GFX_SMG |
4010 			      AMD_PG_SUPPORT_GFX_DMG |
4011 			      AMD_PG_SUPPORT_CP |
4012 			      AMD_PG_SUPPORT_GDS |
4013 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4014 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
4015 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
4016 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4017 			gfx_v7_0_init_gfx_cgpg(adev);
4018 			gfx_v7_0_enable_cp_pg(adev, true);
4019 			gfx_v7_0_enable_gds_pg(adev, true);
4020 		}
4021 		gfx_v7_0_init_ao_cu_mask(adev);
4022 		gfx_v7_0_update_gfx_pg(adev, true);
4023 	}
4024 }
4025 
4026 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4027 {
4028 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4029 			      AMD_PG_SUPPORT_GFX_SMG |
4030 			      AMD_PG_SUPPORT_GFX_DMG |
4031 			      AMD_PG_SUPPORT_CP |
4032 			      AMD_PG_SUPPORT_GDS |
4033 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4034 		gfx_v7_0_update_gfx_pg(adev, false);
4035 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4036 			gfx_v7_0_enable_cp_pg(adev, false);
4037 			gfx_v7_0_enable_gds_pg(adev, false);
4038 		}
4039 	}
4040 }
4041 
4042 /**
4043  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4044  *
4045  * @adev: amdgpu_device pointer
4046  *
4047  * Fetches a GPU clock counter snapshot (SI).
4048  * Returns the 64 bit clock counter snapshot.
4049  */
4050 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4051 {
4052 	uint64_t clock;
4053 
4054 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4055 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4056 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4057 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4058 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4059 	return clock;
4060 }
4061 
4062 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4063 					  uint32_t vmid,
4064 					  uint32_t gds_base, uint32_t gds_size,
4065 					  uint32_t gws_base, uint32_t gws_size,
4066 					  uint32_t oa_base, uint32_t oa_size)
4067 {
4068 	/* GDS Base */
4069 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4070 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4071 				WRITE_DATA_DST_SEL(0)));
4072 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4073 	amdgpu_ring_write(ring, 0);
4074 	amdgpu_ring_write(ring, gds_base);
4075 
4076 	/* GDS Size */
4077 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4078 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4079 				WRITE_DATA_DST_SEL(0)));
4080 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
4081 	amdgpu_ring_write(ring, 0);
4082 	amdgpu_ring_write(ring, gds_size);
4083 
4084 	/* GWS */
4085 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4086 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4087 				WRITE_DATA_DST_SEL(0)));
4088 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
4089 	amdgpu_ring_write(ring, 0);
4090 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4091 
4092 	/* OA */
4093 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4094 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4095 				WRITE_DATA_DST_SEL(0)));
4096 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
4097 	amdgpu_ring_write(ring, 0);
4098 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4099 }
4100 
4101 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
4102 {
4103 	struct amdgpu_device *adev = ring->adev;
4104 	uint32_t value = 0;
4105 
4106 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
4107 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
4108 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
4109 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
4110 	WREG32(mmSQ_CMD, value);
4111 }
4112 
4113 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4114 {
4115 	WREG32(mmSQ_IND_INDEX,
4116 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4117 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4118 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4119 		(SQ_IND_INDEX__FORCE_READ_MASK));
4120 	return RREG32(mmSQ_IND_DATA);
4121 }
4122 
4123 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4124 			   uint32_t wave, uint32_t thread,
4125 			   uint32_t regno, uint32_t num, uint32_t *out)
4126 {
4127 	WREG32(mmSQ_IND_INDEX,
4128 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4129 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4130 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4131 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4132 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4133 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4134 	while (num--)
4135 		*(out++) = RREG32(mmSQ_IND_DATA);
4136 }
4137 
4138 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4139 {
4140 	/* type 0 wave data */
4141 	dst[(*no_fields)++] = 0;
4142 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4143 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4144 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4145 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4146 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4147 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4148 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4149 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4150 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4151 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4152 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4153 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4154 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4155 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4156 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4157 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4158 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4159 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4160 }
4161 
4162 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
4163 				     uint32_t wave, uint32_t start,
4164 				     uint32_t size, uint32_t *dst)
4165 {
4166 	wave_read_regs(
4167 		adev, simd, wave, 0,
4168 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4169 }
4170 
4171 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
4172 				  u32 me, u32 pipe, u32 q, u32 vm)
4173 {
4174 	cik_srbm_select(adev, me, pipe, q, vm);
4175 }
4176 
4177 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4178 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4179 	.select_se_sh = &gfx_v7_0_select_se_sh,
4180 	.read_wave_data = &gfx_v7_0_read_wave_data,
4181 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4182 	.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q
4183 };
4184 
4185 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4186 	.is_rlc_enabled = gfx_v7_0_is_rlc_enabled,
4187 	.set_safe_mode = gfx_v7_0_set_safe_mode,
4188 	.unset_safe_mode = gfx_v7_0_unset_safe_mode,
4189 	.init = gfx_v7_0_rlc_init,
4190 	.get_csb_size = gfx_v7_0_get_csb_size,
4191 	.get_csb_buffer = gfx_v7_0_get_csb_buffer,
4192 	.get_cp_table_num = gfx_v7_0_cp_pg_table_num,
4193 	.resume = gfx_v7_0_rlc_resume,
4194 	.stop = gfx_v7_0_rlc_stop,
4195 	.reset = gfx_v7_0_rlc_reset,
4196 	.start = gfx_v7_0_rlc_start
4197 };
4198 
4199 static int gfx_v7_0_early_init(void *handle)
4200 {
4201 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4202 
4203 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4204 	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
4205 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4206 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4207 	gfx_v7_0_set_ring_funcs(adev);
4208 	gfx_v7_0_set_irq_funcs(adev);
4209 	gfx_v7_0_set_gds_init(adev);
4210 
4211 	return 0;
4212 }
4213 
4214 static int gfx_v7_0_late_init(void *handle)
4215 {
4216 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4217 	int r;
4218 
4219 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4220 	if (r)
4221 		return r;
4222 
4223 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4224 	if (r)
4225 		return r;
4226 
4227 	return 0;
4228 }
4229 
4230 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4231 {
4232 	u32 gb_addr_config;
4233 	u32 mc_shared_chmap, mc_arb_ramcfg;
4234 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4235 	u32 tmp;
4236 
4237 	switch (adev->asic_type) {
4238 	case CHIP_BONAIRE:
4239 		adev->gfx.config.max_shader_engines = 2;
4240 		adev->gfx.config.max_tile_pipes = 4;
4241 		adev->gfx.config.max_cu_per_sh = 7;
4242 		adev->gfx.config.max_sh_per_se = 1;
4243 		adev->gfx.config.max_backends_per_se = 2;
4244 		adev->gfx.config.max_texture_channel_caches = 4;
4245 		adev->gfx.config.max_gprs = 256;
4246 		adev->gfx.config.max_gs_threads = 32;
4247 		adev->gfx.config.max_hw_contexts = 8;
4248 
4249 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4250 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4251 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4252 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4253 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4254 		break;
4255 	case CHIP_HAWAII:
4256 		adev->gfx.config.max_shader_engines = 4;
4257 		adev->gfx.config.max_tile_pipes = 16;
4258 		adev->gfx.config.max_cu_per_sh = 11;
4259 		adev->gfx.config.max_sh_per_se = 1;
4260 		adev->gfx.config.max_backends_per_se = 4;
4261 		adev->gfx.config.max_texture_channel_caches = 16;
4262 		adev->gfx.config.max_gprs = 256;
4263 		adev->gfx.config.max_gs_threads = 32;
4264 		adev->gfx.config.max_hw_contexts = 8;
4265 
4266 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4267 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4268 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4269 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4270 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4271 		break;
4272 	case CHIP_KAVERI:
4273 		adev->gfx.config.max_shader_engines = 1;
4274 		adev->gfx.config.max_tile_pipes = 4;
4275 		adev->gfx.config.max_cu_per_sh = 8;
4276 		adev->gfx.config.max_backends_per_se = 2;
4277 		adev->gfx.config.max_sh_per_se = 1;
4278 		adev->gfx.config.max_texture_channel_caches = 4;
4279 		adev->gfx.config.max_gprs = 256;
4280 		adev->gfx.config.max_gs_threads = 16;
4281 		adev->gfx.config.max_hw_contexts = 8;
4282 
4283 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4284 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4285 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4286 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4287 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4288 		break;
4289 	case CHIP_KABINI:
4290 	case CHIP_MULLINS:
4291 	default:
4292 		adev->gfx.config.max_shader_engines = 1;
4293 		adev->gfx.config.max_tile_pipes = 2;
4294 		adev->gfx.config.max_cu_per_sh = 2;
4295 		adev->gfx.config.max_sh_per_se = 1;
4296 		adev->gfx.config.max_backends_per_se = 1;
4297 		adev->gfx.config.max_texture_channel_caches = 2;
4298 		adev->gfx.config.max_gprs = 256;
4299 		adev->gfx.config.max_gs_threads = 16;
4300 		adev->gfx.config.max_hw_contexts = 8;
4301 
4302 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4303 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4304 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4305 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4306 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4307 		break;
4308 	}
4309 
4310 	mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP);
4311 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4312 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4313 
4314 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4315 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4316 	if (adev->flags & AMD_IS_APU) {
4317 		/* Get memory bank mapping mode. */
4318 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4319 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4320 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4321 
4322 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4323 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4324 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4325 
4326 		/* Validate settings in case only one DIMM installed. */
4327 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4328 			dimm00_addr_map = 0;
4329 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4330 			dimm01_addr_map = 0;
4331 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4332 			dimm10_addr_map = 0;
4333 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4334 			dimm11_addr_map = 0;
4335 
4336 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4337 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4338 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4339 			adev->gfx.config.mem_row_size_in_kb = 2;
4340 		else
4341 			adev->gfx.config.mem_row_size_in_kb = 1;
4342 	} else {
4343 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4344 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4345 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4346 			adev->gfx.config.mem_row_size_in_kb = 4;
4347 	}
4348 	/* XXX use MC settings? */
4349 	adev->gfx.config.shader_engine_tile_size = 32;
4350 	adev->gfx.config.num_gpus = 1;
4351 	adev->gfx.config.multi_gpu_tile_size = 64;
4352 
4353 	/* fix up row size */
4354 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4355 	switch (adev->gfx.config.mem_row_size_in_kb) {
4356 	case 1:
4357 	default:
4358 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4359 		break;
4360 	case 2:
4361 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4362 		break;
4363 	case 4:
4364 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4365 		break;
4366 	}
4367 	adev->gfx.config.gb_addr_config = gb_addr_config;
4368 }
4369 
4370 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
4371 					int mec, int pipe, int queue)
4372 {
4373 	int r;
4374 	unsigned irq_type;
4375 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
4376 
4377 	/* mec0 is me1 */
4378 	ring->me = mec + 1;
4379 	ring->pipe = pipe;
4380 	ring->queue = queue;
4381 
4382 	ring->ring_obj = NULL;
4383 	ring->use_doorbell = true;
4384 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
4385 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4386 
4387 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
4388 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
4389 		+ ring->pipe;
4390 
4391 	/* type-2 packets are deprecated on MEC, use type-3 instead */
4392 	r = amdgpu_ring_init(adev, ring, 1024,
4393 			&adev->gfx.eop_irq, irq_type);
4394 	if (r)
4395 		return r;
4396 
4397 
4398 	return 0;
4399 }
4400 
4401 static int gfx_v7_0_sw_init(void *handle)
4402 {
4403 	struct amdgpu_ring *ring;
4404 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4405 	int i, j, k, r, ring_id;
4406 
4407 	switch (adev->asic_type) {
4408 	case CHIP_KAVERI:
4409 		adev->gfx.mec.num_mec = 2;
4410 		break;
4411 	case CHIP_BONAIRE:
4412 	case CHIP_HAWAII:
4413 	case CHIP_KABINI:
4414 	case CHIP_MULLINS:
4415 	default:
4416 		adev->gfx.mec.num_mec = 1;
4417 		break;
4418 	}
4419 	adev->gfx.mec.num_pipe_per_mec = 4;
4420 	adev->gfx.mec.num_queue_per_pipe = 8;
4421 
4422 	/* EOP Event */
4423 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq);
4424 	if (r)
4425 		return r;
4426 
4427 	/* Privileged reg */
4428 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184,
4429 			      &adev->gfx.priv_reg_irq);
4430 	if (r)
4431 		return r;
4432 
4433 	/* Privileged inst */
4434 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185,
4435 			      &adev->gfx.priv_inst_irq);
4436 	if (r)
4437 		return r;
4438 
4439 	gfx_v7_0_scratch_init(adev);
4440 
4441 	r = gfx_v7_0_init_microcode(adev);
4442 	if (r) {
4443 		DRM_ERROR("Failed to load gfx firmware!\n");
4444 		return r;
4445 	}
4446 
4447 	r = adev->gfx.rlc.funcs->init(adev);
4448 	if (r) {
4449 		DRM_ERROR("Failed to init rlc BOs!\n");
4450 		return r;
4451 	}
4452 
4453 	/* allocate mec buffers */
4454 	r = gfx_v7_0_mec_init(adev);
4455 	if (r) {
4456 		DRM_ERROR("Failed to init MEC BOs!\n");
4457 		return r;
4458 	}
4459 
4460 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4461 		ring = &adev->gfx.gfx_ring[i];
4462 		ring->ring_obj = NULL;
4463 		sprintf(ring->name, "gfx");
4464 		r = amdgpu_ring_init(adev, ring, 1024,
4465 				     &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP);
4466 		if (r)
4467 			return r;
4468 	}
4469 
4470 	/* set up the compute queues - allocate horizontally across pipes */
4471 	ring_id = 0;
4472 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4473 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4474 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4475 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
4476 					continue;
4477 
4478 				r = gfx_v7_0_compute_ring_init(adev,
4479 								ring_id,
4480 								i, k, j);
4481 				if (r)
4482 					return r;
4483 
4484 				ring_id++;
4485 			}
4486 		}
4487 	}
4488 
4489 	adev->gfx.ce_ram_size = 0x8000;
4490 
4491 	gfx_v7_0_gpu_early_init(adev);
4492 
4493 	return r;
4494 }
4495 
4496 static int gfx_v7_0_sw_fini(void *handle)
4497 {
4498 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4499 	int i;
4500 
4501 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4502 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4503 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
4504 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4505 
4506 	gfx_v7_0_cp_compute_fini(adev);
4507 	amdgpu_gfx_rlc_fini(adev);
4508 	gfx_v7_0_mec_fini(adev);
4509 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
4510 				&adev->gfx.rlc.clear_state_gpu_addr,
4511 				(void **)&adev->gfx.rlc.cs_ptr);
4512 	if (adev->gfx.rlc.cp_table_size) {
4513 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
4514 				&adev->gfx.rlc.cp_table_gpu_addr,
4515 				(void **)&adev->gfx.rlc.cp_table_ptr);
4516 	}
4517 	gfx_v7_0_free_microcode(adev);
4518 
4519 	return 0;
4520 }
4521 
4522 static int gfx_v7_0_hw_init(void *handle)
4523 {
4524 	int r;
4525 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4526 
4527 	gfx_v7_0_constants_init(adev);
4528 
4529 	/* init rlc */
4530 	r = adev->gfx.rlc.funcs->resume(adev);
4531 	if (r)
4532 		return r;
4533 
4534 	r = gfx_v7_0_cp_resume(adev);
4535 	if (r)
4536 		return r;
4537 
4538 	return r;
4539 }
4540 
4541 static int gfx_v7_0_hw_fini(void *handle)
4542 {
4543 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4544 
4545 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4546 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4547 	gfx_v7_0_cp_enable(adev, false);
4548 	adev->gfx.rlc.funcs->stop(adev);
4549 	gfx_v7_0_fini_pg(adev);
4550 
4551 	return 0;
4552 }
4553 
4554 static int gfx_v7_0_suspend(void *handle)
4555 {
4556 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4557 
4558 	return gfx_v7_0_hw_fini(adev);
4559 }
4560 
4561 static int gfx_v7_0_resume(void *handle)
4562 {
4563 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4564 
4565 	return gfx_v7_0_hw_init(adev);
4566 }
4567 
4568 static bool gfx_v7_0_is_idle(void *handle)
4569 {
4570 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4571 
4572 	if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK)
4573 		return false;
4574 	else
4575 		return true;
4576 }
4577 
4578 static int gfx_v7_0_wait_for_idle(void *handle)
4579 {
4580 	unsigned i;
4581 	u32 tmp;
4582 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4583 
4584 	for (i = 0; i < adev->usec_timeout; i++) {
4585 		/* read MC_STATUS */
4586 		tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4587 
4588 		if (!tmp)
4589 			return 0;
4590 		udelay(1);
4591 	}
4592 	return -ETIMEDOUT;
4593 }
4594 
4595 static int gfx_v7_0_soft_reset(void *handle)
4596 {
4597 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4598 	u32 tmp;
4599 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4600 
4601 	/* GRBM_STATUS */
4602 	tmp = RREG32(mmGRBM_STATUS);
4603 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4604 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4605 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4606 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4607 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4608 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK))
4609 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK |
4610 			GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK;
4611 
4612 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4613 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK;
4614 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4615 	}
4616 
4617 	/* GRBM_STATUS2 */
4618 	tmp = RREG32(mmGRBM_STATUS2);
4619 	if (tmp & GRBM_STATUS2__RLC_BUSY_MASK)
4620 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
4621 
4622 	/* SRBM_STATUS */
4623 	tmp = RREG32(mmSRBM_STATUS);
4624 	if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK)
4625 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4626 
4627 	if (grbm_soft_reset || srbm_soft_reset) {
4628 		/* disable CG/PG */
4629 		gfx_v7_0_fini_pg(adev);
4630 		gfx_v7_0_update_cg(adev, false);
4631 
4632 		/* stop the rlc */
4633 		adev->gfx.rlc.funcs->stop(adev);
4634 
4635 		/* Disable GFX parsing/prefetching */
4636 		WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK);
4637 
4638 		/* Disable MEC parsing/prefetching */
4639 		WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK);
4640 
4641 		if (grbm_soft_reset) {
4642 			tmp = RREG32(mmGRBM_SOFT_RESET);
4643 			tmp |= grbm_soft_reset;
4644 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4645 			WREG32(mmGRBM_SOFT_RESET, tmp);
4646 			tmp = RREG32(mmGRBM_SOFT_RESET);
4647 
4648 			udelay(50);
4649 
4650 			tmp &= ~grbm_soft_reset;
4651 			WREG32(mmGRBM_SOFT_RESET, tmp);
4652 			tmp = RREG32(mmGRBM_SOFT_RESET);
4653 		}
4654 
4655 		if (srbm_soft_reset) {
4656 			tmp = RREG32(mmSRBM_SOFT_RESET);
4657 			tmp |= srbm_soft_reset;
4658 			dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4659 			WREG32(mmSRBM_SOFT_RESET, tmp);
4660 			tmp = RREG32(mmSRBM_SOFT_RESET);
4661 
4662 			udelay(50);
4663 
4664 			tmp &= ~srbm_soft_reset;
4665 			WREG32(mmSRBM_SOFT_RESET, tmp);
4666 			tmp = RREG32(mmSRBM_SOFT_RESET);
4667 		}
4668 		/* Wait a little for things to settle down */
4669 		udelay(50);
4670 	}
4671 	return 0;
4672 }
4673 
4674 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4675 						 enum amdgpu_interrupt_state state)
4676 {
4677 	u32 cp_int_cntl;
4678 
4679 	switch (state) {
4680 	case AMDGPU_IRQ_STATE_DISABLE:
4681 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4682 		cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4683 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4684 		break;
4685 	case AMDGPU_IRQ_STATE_ENABLE:
4686 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4687 		cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4688 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4689 		break;
4690 	default:
4691 		break;
4692 	}
4693 }
4694 
4695 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4696 						     int me, int pipe,
4697 						     enum amdgpu_interrupt_state state)
4698 {
4699 	u32 mec_int_cntl, mec_int_cntl_reg;
4700 
4701 	/*
4702 	 * amdgpu controls only the first MEC. That's why this function only
4703 	 * handles the setting of interrupts for this specific MEC. All other
4704 	 * pipes' interrupts are set by amdkfd.
4705 	 */
4706 
4707 	if (me == 1) {
4708 		switch (pipe) {
4709 		case 0:
4710 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4711 			break;
4712 		case 1:
4713 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
4714 			break;
4715 		case 2:
4716 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
4717 			break;
4718 		case 3:
4719 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
4720 			break;
4721 		default:
4722 			DRM_DEBUG("invalid pipe %d\n", pipe);
4723 			return;
4724 		}
4725 	} else {
4726 		DRM_DEBUG("invalid me %d\n", me);
4727 		return;
4728 	}
4729 
4730 	switch (state) {
4731 	case AMDGPU_IRQ_STATE_DISABLE:
4732 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4733 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4734 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4735 		break;
4736 	case AMDGPU_IRQ_STATE_ENABLE:
4737 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4738 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4739 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4740 		break;
4741 	default:
4742 		break;
4743 	}
4744 }
4745 
4746 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4747 					     struct amdgpu_irq_src *src,
4748 					     unsigned type,
4749 					     enum amdgpu_interrupt_state state)
4750 {
4751 	u32 cp_int_cntl;
4752 
4753 	switch (state) {
4754 	case AMDGPU_IRQ_STATE_DISABLE:
4755 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4756 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4757 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4758 		break;
4759 	case AMDGPU_IRQ_STATE_ENABLE:
4760 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4761 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4762 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4763 		break;
4764 	default:
4765 		break;
4766 	}
4767 
4768 	return 0;
4769 }
4770 
4771 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4772 					      struct amdgpu_irq_src *src,
4773 					      unsigned type,
4774 					      enum amdgpu_interrupt_state state)
4775 {
4776 	u32 cp_int_cntl;
4777 
4778 	switch (state) {
4779 	case AMDGPU_IRQ_STATE_DISABLE:
4780 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4781 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4782 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4783 		break;
4784 	case AMDGPU_IRQ_STATE_ENABLE:
4785 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4786 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4787 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4788 		break;
4789 	default:
4790 		break;
4791 	}
4792 
4793 	return 0;
4794 }
4795 
4796 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4797 					    struct amdgpu_irq_src *src,
4798 					    unsigned type,
4799 					    enum amdgpu_interrupt_state state)
4800 {
4801 	switch (type) {
4802 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
4803 		gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
4804 		break;
4805 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4806 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4807 		break;
4808 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4809 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4810 		break;
4811 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4812 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4813 		break;
4814 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4815 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4816 		break;
4817 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4818 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4819 		break;
4820 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4821 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4822 		break;
4823 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4824 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4825 		break;
4826 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4827 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4828 		break;
4829 	default:
4830 		break;
4831 	}
4832 	return 0;
4833 }
4834 
4835 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
4836 			    struct amdgpu_irq_src *source,
4837 			    struct amdgpu_iv_entry *entry)
4838 {
4839 	u8 me_id, pipe_id;
4840 	struct amdgpu_ring *ring;
4841 	int i;
4842 
4843 	DRM_DEBUG("IH: CP EOP\n");
4844 	me_id = (entry->ring_id & 0x0c) >> 2;
4845 	pipe_id = (entry->ring_id & 0x03) >> 0;
4846 	switch (me_id) {
4847 	case 0:
4848 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4849 		break;
4850 	case 1:
4851 	case 2:
4852 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4853 			ring = &adev->gfx.compute_ring[i];
4854 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4855 				amdgpu_fence_process(ring);
4856 		}
4857 		break;
4858 	}
4859 	return 0;
4860 }
4861 
4862 static void gfx_v7_0_fault(struct amdgpu_device *adev,
4863 			   struct amdgpu_iv_entry *entry)
4864 {
4865 	struct amdgpu_ring *ring;
4866 	u8 me_id, pipe_id;
4867 	int i;
4868 
4869 	me_id = (entry->ring_id & 0x0c) >> 2;
4870 	pipe_id = (entry->ring_id & 0x03) >> 0;
4871 	switch (me_id) {
4872 	case 0:
4873 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
4874 		break;
4875 	case 1:
4876 	case 2:
4877 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4878 			ring = &adev->gfx.compute_ring[i];
4879 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4880 				drm_sched_fault(&ring->sched);
4881 		}
4882 		break;
4883 	}
4884 }
4885 
4886 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
4887 				 struct amdgpu_irq_src *source,
4888 				 struct amdgpu_iv_entry *entry)
4889 {
4890 	DRM_ERROR("Illegal register access in command stream\n");
4891 	gfx_v7_0_fault(adev, entry);
4892 	return 0;
4893 }
4894 
4895 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
4896 				  struct amdgpu_irq_src *source,
4897 				  struct amdgpu_iv_entry *entry)
4898 {
4899 	DRM_ERROR("Illegal instruction in command stream\n");
4900 	// XXX soft reset the gfx block only
4901 	gfx_v7_0_fault(adev, entry);
4902 	return 0;
4903 }
4904 
4905 static int gfx_v7_0_set_clockgating_state(void *handle,
4906 					  enum amd_clockgating_state state)
4907 {
4908 	bool gate = false;
4909 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4910 
4911 	if (state == AMD_CG_STATE_GATE)
4912 		gate = true;
4913 
4914 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
4915 	/* order matters! */
4916 	if (gate) {
4917 		gfx_v7_0_enable_mgcg(adev, true);
4918 		gfx_v7_0_enable_cgcg(adev, true);
4919 	} else {
4920 		gfx_v7_0_enable_cgcg(adev, false);
4921 		gfx_v7_0_enable_mgcg(adev, false);
4922 	}
4923 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
4924 
4925 	return 0;
4926 }
4927 
4928 static int gfx_v7_0_set_powergating_state(void *handle,
4929 					  enum amd_powergating_state state)
4930 {
4931 	bool gate = false;
4932 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4933 
4934 	if (state == AMD_PG_STATE_GATE)
4935 		gate = true;
4936 
4937 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4938 			      AMD_PG_SUPPORT_GFX_SMG |
4939 			      AMD_PG_SUPPORT_GFX_DMG |
4940 			      AMD_PG_SUPPORT_CP |
4941 			      AMD_PG_SUPPORT_GDS |
4942 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4943 		gfx_v7_0_update_gfx_pg(adev, gate);
4944 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4945 			gfx_v7_0_enable_cp_pg(adev, gate);
4946 			gfx_v7_0_enable_gds_pg(adev, gate);
4947 		}
4948 	}
4949 
4950 	return 0;
4951 }
4952 
4953 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
4954 	.name = "gfx_v7_0",
4955 	.early_init = gfx_v7_0_early_init,
4956 	.late_init = gfx_v7_0_late_init,
4957 	.sw_init = gfx_v7_0_sw_init,
4958 	.sw_fini = gfx_v7_0_sw_fini,
4959 	.hw_init = gfx_v7_0_hw_init,
4960 	.hw_fini = gfx_v7_0_hw_fini,
4961 	.suspend = gfx_v7_0_suspend,
4962 	.resume = gfx_v7_0_resume,
4963 	.is_idle = gfx_v7_0_is_idle,
4964 	.wait_for_idle = gfx_v7_0_wait_for_idle,
4965 	.soft_reset = gfx_v7_0_soft_reset,
4966 	.set_clockgating_state = gfx_v7_0_set_clockgating_state,
4967 	.set_powergating_state = gfx_v7_0_set_powergating_state,
4968 };
4969 
4970 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
4971 	.type = AMDGPU_RING_TYPE_GFX,
4972 	.align_mask = 0xff,
4973 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
4974 	.support_64bit_ptrs = false,
4975 	.get_rptr = gfx_v7_0_ring_get_rptr,
4976 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
4977 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
4978 	.emit_frame_size =
4979 		20 + /* gfx_v7_0_ring_emit_gds_switch */
4980 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
4981 		5 + /* hdp invalidate */
4982 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
4983 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
4984 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
4985 		3 + 4, /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
4986 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
4987 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
4988 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
4989 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
4990 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
4991 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
4992 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
4993 	.test_ring = gfx_v7_0_ring_test_ring,
4994 	.test_ib = gfx_v7_0_ring_test_ib,
4995 	.insert_nop = amdgpu_ring_insert_nop,
4996 	.pad_ib = amdgpu_ring_generic_pad_ib,
4997 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
4998 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
4999 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
5000 };
5001 
5002 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5003 	.type = AMDGPU_RING_TYPE_COMPUTE,
5004 	.align_mask = 0xff,
5005 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5006 	.support_64bit_ptrs = false,
5007 	.get_rptr = gfx_v7_0_ring_get_rptr,
5008 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5009 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5010 	.emit_frame_size =
5011 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5012 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5013 		5 + /* hdp invalidate */
5014 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5015 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
5016 		7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5017 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
5018 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5019 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5020 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5021 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5022 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5023 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5024 	.test_ring = gfx_v7_0_ring_test_ring,
5025 	.test_ib = gfx_v7_0_ring_test_ib,
5026 	.insert_nop = amdgpu_ring_insert_nop,
5027 	.pad_ib = amdgpu_ring_generic_pad_ib,
5028 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5029 };
5030 
5031 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5032 {
5033 	int i;
5034 
5035 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5036 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5037 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5038 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5039 }
5040 
5041 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5042 	.set = gfx_v7_0_set_eop_interrupt_state,
5043 	.process = gfx_v7_0_eop_irq,
5044 };
5045 
5046 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5047 	.set = gfx_v7_0_set_priv_reg_fault_state,
5048 	.process = gfx_v7_0_priv_reg_irq,
5049 };
5050 
5051 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5052 	.set = gfx_v7_0_set_priv_inst_fault_state,
5053 	.process = gfx_v7_0_priv_inst_irq,
5054 };
5055 
5056 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5057 {
5058 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5059 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5060 
5061 	adev->gfx.priv_reg_irq.num_types = 1;
5062 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5063 
5064 	adev->gfx.priv_inst_irq.num_types = 1;
5065 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5066 }
5067 
5068 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5069 {
5070 	/* init asci gds info */
5071 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
5072 	adev->gds.gws_size = 64;
5073 	adev->gds.oa_size = 16;
5074 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
5075 }
5076 
5077 
5078 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5079 {
5080 	int i, j, k, counter, active_cu_number = 0;
5081 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5082 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5083 	unsigned disable_masks[4 * 2];
5084 	u32 ao_cu_num;
5085 
5086 	if (adev->flags & AMD_IS_APU)
5087 		ao_cu_num = 2;
5088 	else
5089 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
5090 
5091 	memset(cu_info, 0, sizeof(*cu_info));
5092 
5093 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5094 
5095 	mutex_lock(&adev->grbm_idx_mutex);
5096 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5097 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5098 			mask = 1;
5099 			ao_bitmap = 0;
5100 			counter = 0;
5101 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
5102 			if (i < 4 && j < 2)
5103 				gfx_v7_0_set_user_cu_inactive_bitmap(
5104 					adev, disable_masks[i * 2 + j]);
5105 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5106 			cu_info->bitmap[i][j] = bitmap;
5107 
5108 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
5109 				if (bitmap & mask) {
5110 					if (counter < ao_cu_num)
5111 						ao_bitmap |= mask;
5112 					counter ++;
5113 				}
5114 				mask <<= 1;
5115 			}
5116 			active_cu_number += counter;
5117 			if (i < 2 && j < 2)
5118 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5119 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5120 		}
5121 	}
5122 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5123 	mutex_unlock(&adev->grbm_idx_mutex);
5124 
5125 	cu_info->number = active_cu_number;
5126 	cu_info->ao_cu_mask = ao_cu_mask;
5127 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5128 	cu_info->max_waves_per_simd = 10;
5129 	cu_info->max_scratch_slots_per_cu = 32;
5130 	cu_info->wave_front_size = 64;
5131 	cu_info->lds_size = 64;
5132 }
5133 
5134 const struct amdgpu_ip_block_version gfx_v7_0_ip_block =
5135 {
5136 	.type = AMD_IP_BLOCK_TYPE_GFX,
5137 	.major = 7,
5138 	.minor = 0,
5139 	.rev = 0,
5140 	.funcs = &gfx_v7_0_ip_funcs,
5141 };
5142 
5143 const struct amdgpu_ip_block_version gfx_v7_1_ip_block =
5144 {
5145 	.type = AMD_IP_BLOCK_TYPE_GFX,
5146 	.major = 7,
5147 	.minor = 1,
5148 	.rev = 0,
5149 	.funcs = &gfx_v7_0_ip_funcs,
5150 };
5151 
5152 const struct amdgpu_ip_block_version gfx_v7_2_ip_block =
5153 {
5154 	.type = AMD_IP_BLOCK_TYPE_GFX,
5155 	.major = 7,
5156 	.minor = 2,
5157 	.rev = 0,
5158 	.funcs = &gfx_v7_0_ip_funcs,
5159 };
5160 
5161 const struct amdgpu_ip_block_version gfx_v7_3_ip_block =
5162 {
5163 	.type = AMD_IP_BLOCK_TYPE_GFX,
5164 	.major = 7,
5165 	.minor = 3,
5166 	.rev = 0,
5167 	.funcs = &gfx_v7_0_ip_funcs,
5168 };
5169