xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision 29c37341)
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 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1854 {
1855 	int i;
1856 	uint32_t sh_mem_config;
1857 	uint32_t sh_mem_bases;
1858 
1859 	/*
1860 	 * Configure apertures:
1861 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1862 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1863 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1864 	*/
1865 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1866 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1867 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1868 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1869 	mutex_lock(&adev->srbm_mutex);
1870 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1871 		cik_srbm_select(adev, 0, 0, 0, i);
1872 		/* CP and shaders */
1873 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1874 		WREG32(mmSH_MEM_APE1_BASE, 1);
1875 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1876 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1877 	}
1878 	cik_srbm_select(adev, 0, 0, 0, 0);
1879 	mutex_unlock(&adev->srbm_mutex);
1880 
1881 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1882 	   acccess. These should be enabled by FW for target VMIDs. */
1883 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1884 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
1885 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
1886 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
1887 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
1888 	}
1889 }
1890 
1891 static void gfx_v7_0_init_gds_vmid(struct amdgpu_device *adev)
1892 {
1893 	int vmid;
1894 
1895 	/*
1896 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1897 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1898 	 * the driver can enable them for graphics. VMID0 should maintain
1899 	 * access so that HWS firmware can save/restore entries.
1900 	 */
1901 	for (vmid = 1; vmid < 16; vmid++) {
1902 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
1903 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
1904 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
1905 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
1906 	}
1907 }
1908 
1909 static void gfx_v7_0_config_init(struct amdgpu_device *adev)
1910 {
1911 	adev->gfx.config.double_offchip_lds_buf = 1;
1912 }
1913 
1914 /**
1915  * gfx_v7_0_constants_init - setup the 3D engine
1916  *
1917  * @adev: amdgpu_device pointer
1918  *
1919  * init the gfx constants such as the 3D engine, tiling configuration
1920  * registers, maximum number of quad pipes, render backends...
1921  */
1922 static void gfx_v7_0_constants_init(struct amdgpu_device *adev)
1923 {
1924 	u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base;
1925 	u32 tmp;
1926 	int i;
1927 
1928 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1929 
1930 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1931 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1932 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1933 
1934 	gfx_v7_0_tiling_mode_table_init(adev);
1935 
1936 	gfx_v7_0_setup_rb(adev);
1937 	gfx_v7_0_get_cu_info(adev);
1938 	gfx_v7_0_config_init(adev);
1939 
1940 	/* set HW defaults for 3D engine */
1941 	WREG32(mmCP_MEQ_THRESHOLDS,
1942 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1943 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1944 
1945 	mutex_lock(&adev->grbm_idx_mutex);
1946 	/*
1947 	 * making sure that the following register writes will be broadcasted
1948 	 * to all the shaders
1949 	 */
1950 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1951 
1952 	/* XXX SH_MEM regs */
1953 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1954 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1955 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1956 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE,
1957 				   MTYPE_NC);
1958 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE,
1959 				   MTYPE_UC);
1960 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0);
1961 
1962 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
1963 				   SWIZZLE_ENABLE, 1);
1964 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1965 				   ELEMENT_SIZE, 1);
1966 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1967 				   INDEX_STRIDE, 3);
1968 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
1969 
1970 	mutex_lock(&adev->srbm_mutex);
1971 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
1972 		if (i == 0)
1973 			sh_mem_base = 0;
1974 		else
1975 			sh_mem_base = adev->gmc.shared_aperture_start >> 48;
1976 		cik_srbm_select(adev, 0, 0, 0, i);
1977 		/* CP and shaders */
1978 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1979 		WREG32(mmSH_MEM_APE1_BASE, 1);
1980 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1981 		WREG32(mmSH_MEM_BASES, sh_mem_base);
1982 	}
1983 	cik_srbm_select(adev, 0, 0, 0, 0);
1984 	mutex_unlock(&adev->srbm_mutex);
1985 
1986 	gfx_v7_0_init_compute_vmid(adev);
1987 	gfx_v7_0_init_gds_vmid(adev);
1988 
1989 	WREG32(mmSX_DEBUG_1, 0x20);
1990 
1991 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1992 
1993 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1994 	tmp |= 0x03000000;
1995 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1996 
1997 	WREG32(mmSQ_CONFIG, 1);
1998 
1999 	WREG32(mmDB_DEBUG, 0);
2000 
2001 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
2002 	tmp |= 0x00000400;
2003 	WREG32(mmDB_DEBUG2, tmp);
2004 
2005 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
2006 	tmp |= 0x00020200;
2007 	WREG32(mmDB_DEBUG3, tmp);
2008 
2009 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
2010 	tmp |= 0x00018208;
2011 	WREG32(mmCB_HW_CONTROL, tmp);
2012 
2013 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
2014 
2015 	WREG32(mmPA_SC_FIFO_SIZE,
2016 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
2017 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
2018 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
2019 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
2020 
2021 	WREG32(mmVGT_NUM_INSTANCES, 1);
2022 
2023 	WREG32(mmCP_PERFMON_CNTL, 0);
2024 
2025 	WREG32(mmSQ_CONFIG, 0);
2026 
2027 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
2028 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
2029 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
2030 
2031 	WREG32(mmVGT_CACHE_INVALIDATION,
2032 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
2033 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
2034 
2035 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
2036 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
2037 
2038 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
2039 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
2040 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
2041 
2042 	tmp = RREG32(mmSPI_ARB_PRIORITY);
2043 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
2044 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
2045 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
2046 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
2047 	WREG32(mmSPI_ARB_PRIORITY, tmp);
2048 
2049 	mutex_unlock(&adev->grbm_idx_mutex);
2050 
2051 	udelay(50);
2052 }
2053 
2054 /*
2055  * GPU scratch registers helpers function.
2056  */
2057 /**
2058  * gfx_v7_0_scratch_init - setup driver info for CP scratch regs
2059  *
2060  * @adev: amdgpu_device pointer
2061  *
2062  * Set up the number and offset of the CP scratch registers.
2063  * NOTE: use of CP scratch registers is a legacy inferface and
2064  * is not used by default on newer asics (r6xx+).  On newer asics,
2065  * memory buffers are used for fences rather than scratch regs.
2066  */
2067 static void gfx_v7_0_scratch_init(struct amdgpu_device *adev)
2068 {
2069 	adev->gfx.scratch.num_reg = 8;
2070 	adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
2071 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
2072 }
2073 
2074 /**
2075  * gfx_v7_0_ring_test_ring - basic gfx ring test
2076  *
2077  * @adev: amdgpu_device pointer
2078  * @ring: amdgpu_ring structure holding ring information
2079  *
2080  * Allocate a scratch register and write to it using the gfx ring (CIK).
2081  * Provides a basic gfx ring test to verify that the ring is working.
2082  * Used by gfx_v7_0_cp_gfx_resume();
2083  * Returns 0 on success, error on failure.
2084  */
2085 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2086 {
2087 	struct amdgpu_device *adev = ring->adev;
2088 	uint32_t scratch;
2089 	uint32_t tmp = 0;
2090 	unsigned i;
2091 	int r;
2092 
2093 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2094 	if (r)
2095 		return r;
2096 
2097 	WREG32(scratch, 0xCAFEDEAD);
2098 	r = amdgpu_ring_alloc(ring, 3);
2099 	if (r)
2100 		goto error_free_scratch;
2101 
2102 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2103 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
2104 	amdgpu_ring_write(ring, 0xDEADBEEF);
2105 	amdgpu_ring_commit(ring);
2106 
2107 	for (i = 0; i < adev->usec_timeout; i++) {
2108 		tmp = RREG32(scratch);
2109 		if (tmp == 0xDEADBEEF)
2110 			break;
2111 		udelay(1);
2112 	}
2113 	if (i >= adev->usec_timeout)
2114 		r = -ETIMEDOUT;
2115 
2116 error_free_scratch:
2117 	amdgpu_gfx_scratch_free(adev, scratch);
2118 	return r;
2119 }
2120 
2121 /**
2122  * gfx_v7_0_ring_emit_hdp - emit an hdp flush on the cp
2123  *
2124  * @adev: amdgpu_device pointer
2125  * @ridx: amdgpu ring index
2126  *
2127  * Emits an hdp flush on the cp.
2128  */
2129 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2130 {
2131 	u32 ref_and_mask;
2132 	int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2133 
2134 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2135 		switch (ring->me) {
2136 		case 1:
2137 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2138 			break;
2139 		case 2:
2140 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2141 			break;
2142 		default:
2143 			return;
2144 		}
2145 	} else {
2146 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2147 	}
2148 
2149 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2150 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2151 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2152 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2153 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2154 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2155 	amdgpu_ring_write(ring, ref_and_mask);
2156 	amdgpu_ring_write(ring, ref_and_mask);
2157 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2158 }
2159 
2160 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2161 {
2162 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2163 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2164 		EVENT_INDEX(4));
2165 
2166 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2167 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2168 		EVENT_INDEX(0));
2169 }
2170 
2171 /**
2172  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2173  *
2174  * @adev: amdgpu_device pointer
2175  * @fence: amdgpu fence object
2176  *
2177  * Emits a fence sequnce number on the gfx ring and flushes
2178  * GPU caches.
2179  */
2180 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2181 					 u64 seq, unsigned flags)
2182 {
2183 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2184 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2185 	/* Workaround for cache flush problems. First send a dummy EOP
2186 	 * event down the pipe with seq one below.
2187 	 */
2188 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2189 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2190 				 EOP_TC_ACTION_EN |
2191 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2192 				 EVENT_INDEX(5)));
2193 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2194 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2195 				DATA_SEL(1) | INT_SEL(0));
2196 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2197 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2198 
2199 	/* Then send the real EOP event down the pipe. */
2200 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2201 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2202 				 EOP_TC_ACTION_EN |
2203 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2204 				 EVENT_INDEX(5)));
2205 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2206 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2207 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2208 	amdgpu_ring_write(ring, lower_32_bits(seq));
2209 	amdgpu_ring_write(ring, upper_32_bits(seq));
2210 }
2211 
2212 /**
2213  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2214  *
2215  * @adev: amdgpu_device pointer
2216  * @fence: amdgpu fence object
2217  *
2218  * Emits a fence sequnce number on the compute ring and flushes
2219  * GPU caches.
2220  */
2221 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2222 					     u64 addr, u64 seq,
2223 					     unsigned flags)
2224 {
2225 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2226 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2227 
2228 	/* RELEASE_MEM - flush caches, send int */
2229 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2230 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2231 				 EOP_TC_ACTION_EN |
2232 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2233 				 EVENT_INDEX(5)));
2234 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2235 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2236 	amdgpu_ring_write(ring, upper_32_bits(addr));
2237 	amdgpu_ring_write(ring, lower_32_bits(seq));
2238 	amdgpu_ring_write(ring, upper_32_bits(seq));
2239 }
2240 
2241 /*
2242  * IB stuff
2243  */
2244 /**
2245  * gfx_v7_0_ring_emit_ib - emit an IB (Indirect Buffer) on the ring
2246  *
2247  * @ring: amdgpu_ring structure holding ring information
2248  * @ib: amdgpu indirect buffer object
2249  *
2250  * Emits an DE (drawing engine) or CE (constant engine) IB
2251  * on the gfx ring.  IBs are usually generated by userspace
2252  * acceleration drivers and submitted to the kernel for
2253  * sheduling on the ring.  This function schedules the IB
2254  * on the gfx ring for execution by the GPU.
2255  */
2256 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2257 					struct amdgpu_job *job,
2258 					struct amdgpu_ib *ib,
2259 					uint32_t flags)
2260 {
2261 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2262 	u32 header, control = 0;
2263 
2264 	/* insert SWITCH_BUFFER packet before first IB in the ring frame */
2265 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2266 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2267 		amdgpu_ring_write(ring, 0);
2268 	}
2269 
2270 	if (ib->flags & AMDGPU_IB_FLAG_CE)
2271 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2272 	else
2273 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2274 
2275 	control |= ib->length_dw | (vmid << 24);
2276 
2277 	amdgpu_ring_write(ring, header);
2278 	amdgpu_ring_write(ring,
2279 #ifdef __BIG_ENDIAN
2280 			  (2 << 0) |
2281 #endif
2282 			  (ib->gpu_addr & 0xFFFFFFFC));
2283 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2284 	amdgpu_ring_write(ring, control);
2285 }
2286 
2287 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
2288 					  struct amdgpu_job *job,
2289 					  struct amdgpu_ib *ib,
2290 					  uint32_t flags)
2291 {
2292 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2293 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
2294 
2295 	/* Currently, there is a high possibility to get wave ID mismatch
2296 	 * between ME and GDS, leading to a hw deadlock, because ME generates
2297 	 * different wave IDs than the GDS expects. This situation happens
2298 	 * randomly when at least 5 compute pipes use GDS ordered append.
2299 	 * The wave IDs generated by ME are also wrong after suspend/resume.
2300 	 * Those are probably bugs somewhere else in the kernel driver.
2301 	 *
2302 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
2303 	 * GDS to 0 for this ring (me/pipe).
2304 	 */
2305 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
2306 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2307 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
2308 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
2309 	}
2310 
2311 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2312 	amdgpu_ring_write(ring,
2313 #ifdef __BIG_ENDIAN
2314 					  (2 << 0) |
2315 #endif
2316 					  (ib->gpu_addr & 0xFFFFFFFC));
2317 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2318 	amdgpu_ring_write(ring, control);
2319 }
2320 
2321 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2322 {
2323 	uint32_t dw2 = 0;
2324 
2325 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2326 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2327 		gfx_v7_0_ring_emit_vgt_flush(ring);
2328 		/* set load_global_config & load_global_uconfig */
2329 		dw2 |= 0x8001;
2330 		/* set load_cs_sh_regs */
2331 		dw2 |= 0x01000000;
2332 		/* set load_per_context_state & load_gfx_sh_regs */
2333 		dw2 |= 0x10002;
2334 	}
2335 
2336 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2337 	amdgpu_ring_write(ring, dw2);
2338 	amdgpu_ring_write(ring, 0);
2339 }
2340 
2341 /**
2342  * gfx_v7_0_ring_test_ib - basic ring IB test
2343  *
2344  * @ring: amdgpu_ring structure holding ring information
2345  *
2346  * Allocate an IB and execute it on the gfx ring (CIK).
2347  * Provides a basic gfx ring test to verify that IBs are working.
2348  * Returns 0 on success, error on failure.
2349  */
2350 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2351 {
2352 	struct amdgpu_device *adev = ring->adev;
2353 	struct amdgpu_ib ib;
2354 	struct dma_fence *f = NULL;
2355 	uint32_t scratch;
2356 	uint32_t tmp = 0;
2357 	long r;
2358 
2359 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2360 	if (r)
2361 		return r;
2362 
2363 	WREG32(scratch, 0xCAFEDEAD);
2364 	memset(&ib, 0, sizeof(ib));
2365 	r = amdgpu_ib_get(adev, NULL, 256,
2366 					AMDGPU_IB_POOL_DIRECT, &ib);
2367 	if (r)
2368 		goto err1;
2369 
2370 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2371 	ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
2372 	ib.ptr[2] = 0xDEADBEEF;
2373 	ib.length_dw = 3;
2374 
2375 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2376 	if (r)
2377 		goto err2;
2378 
2379 	r = dma_fence_wait_timeout(f, false, timeout);
2380 	if (r == 0) {
2381 		r = -ETIMEDOUT;
2382 		goto err2;
2383 	} else if (r < 0) {
2384 		goto err2;
2385 	}
2386 	tmp = RREG32(scratch);
2387 	if (tmp == 0xDEADBEEF)
2388 		r = 0;
2389 	else
2390 		r = -EINVAL;
2391 
2392 err2:
2393 	amdgpu_ib_free(adev, &ib, NULL);
2394 	dma_fence_put(f);
2395 err1:
2396 	amdgpu_gfx_scratch_free(adev, scratch);
2397 	return r;
2398 }
2399 
2400 /*
2401  * CP.
2402  * On CIK, gfx and compute now have independant command processors.
2403  *
2404  * GFX
2405  * Gfx consists of a single ring and can process both gfx jobs and
2406  * compute jobs.  The gfx CP consists of three microengines (ME):
2407  * PFP - Pre-Fetch Parser
2408  * ME - Micro Engine
2409  * CE - Constant Engine
2410  * The PFP and ME make up what is considered the Drawing Engine (DE).
2411  * The CE is an asynchronous engine used for updating buffer desciptors
2412  * used by the DE so that they can be loaded into cache in parallel
2413  * while the DE is processing state update packets.
2414  *
2415  * Compute
2416  * The compute CP consists of two microengines (ME):
2417  * MEC1 - Compute MicroEngine 1
2418  * MEC2 - Compute MicroEngine 2
2419  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2420  * The queues are exposed to userspace and are programmed directly
2421  * by the compute runtime.
2422  */
2423 /**
2424  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2425  *
2426  * @adev: amdgpu_device pointer
2427  * @enable: enable or disable the MEs
2428  *
2429  * Halts or unhalts the gfx MEs.
2430  */
2431 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2432 {
2433 	if (enable)
2434 		WREG32(mmCP_ME_CNTL, 0);
2435 	else
2436 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK |
2437 				      CP_ME_CNTL__PFP_HALT_MASK |
2438 				      CP_ME_CNTL__CE_HALT_MASK));
2439 	udelay(50);
2440 }
2441 
2442 /**
2443  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2444  *
2445  * @adev: amdgpu_device pointer
2446  *
2447  * Loads the gfx PFP, ME, and CE ucode.
2448  * Returns 0 for success, -EINVAL if the ucode is not available.
2449  */
2450 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2451 {
2452 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2453 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2454 	const struct gfx_firmware_header_v1_0 *me_hdr;
2455 	const __le32 *fw_data;
2456 	unsigned i, fw_size;
2457 
2458 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2459 		return -EINVAL;
2460 
2461 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2462 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2463 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2464 
2465 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2466 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2467 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2468 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2469 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2470 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2471 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2472 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2473 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2474 
2475 	gfx_v7_0_cp_gfx_enable(adev, false);
2476 
2477 	/* PFP */
2478 	fw_data = (const __le32 *)
2479 		(adev->gfx.pfp_fw->data +
2480 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2481 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2482 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2483 	for (i = 0; i < fw_size; i++)
2484 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2485 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2486 
2487 	/* CE */
2488 	fw_data = (const __le32 *)
2489 		(adev->gfx.ce_fw->data +
2490 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2491 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2492 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2493 	for (i = 0; i < fw_size; i++)
2494 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2495 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2496 
2497 	/* ME */
2498 	fw_data = (const __le32 *)
2499 		(adev->gfx.me_fw->data +
2500 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2501 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2502 	WREG32(mmCP_ME_RAM_WADDR, 0);
2503 	for (i = 0; i < fw_size; i++)
2504 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2505 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2506 
2507 	return 0;
2508 }
2509 
2510 /**
2511  * gfx_v7_0_cp_gfx_start - start the gfx ring
2512  *
2513  * @adev: amdgpu_device pointer
2514  *
2515  * Enables the ring and loads the clear state context and other
2516  * packets required to init the ring.
2517  * Returns 0 for success, error for failure.
2518  */
2519 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2520 {
2521 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2522 	const struct cs_section_def *sect = NULL;
2523 	const struct cs_extent_def *ext = NULL;
2524 	int r, i;
2525 
2526 	/* init the CP */
2527 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2528 	WREG32(mmCP_ENDIAN_SWAP, 0);
2529 	WREG32(mmCP_DEVICE_ID, 1);
2530 
2531 	gfx_v7_0_cp_gfx_enable(adev, true);
2532 
2533 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2534 	if (r) {
2535 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2536 		return r;
2537 	}
2538 
2539 	/* init the CE partitions.  CE only used for gfx on CIK */
2540 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2541 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2542 	amdgpu_ring_write(ring, 0x8000);
2543 	amdgpu_ring_write(ring, 0x8000);
2544 
2545 	/* clear state buffer */
2546 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2547 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2548 
2549 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2550 	amdgpu_ring_write(ring, 0x80000000);
2551 	amdgpu_ring_write(ring, 0x80000000);
2552 
2553 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2554 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2555 			if (sect->id == SECT_CONTEXT) {
2556 				amdgpu_ring_write(ring,
2557 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2558 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2559 				for (i = 0; i < ext->reg_count; i++)
2560 					amdgpu_ring_write(ring, ext->extent[i]);
2561 			}
2562 		}
2563 	}
2564 
2565 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2566 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2567 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
2568 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
2569 
2570 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2571 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2572 
2573 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2574 	amdgpu_ring_write(ring, 0);
2575 
2576 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2577 	amdgpu_ring_write(ring, 0x00000316);
2578 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2579 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2580 
2581 	amdgpu_ring_commit(ring);
2582 
2583 	return 0;
2584 }
2585 
2586 /**
2587  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2588  *
2589  * @adev: amdgpu_device pointer
2590  *
2591  * Program the location and size of the gfx ring buffer
2592  * and test it to make sure it's working.
2593  * Returns 0 for success, error for failure.
2594  */
2595 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2596 {
2597 	struct amdgpu_ring *ring;
2598 	u32 tmp;
2599 	u32 rb_bufsz;
2600 	u64 rb_addr, rptr_addr;
2601 	int r;
2602 
2603 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2604 	if (adev->asic_type != CHIP_HAWAII)
2605 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2606 
2607 	/* Set the write pointer delay */
2608 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2609 
2610 	/* set the RB to use vmid 0 */
2611 	WREG32(mmCP_RB_VMID, 0);
2612 
2613 	WREG32(mmSCRATCH_ADDR, 0);
2614 
2615 	/* ring 0 - compute and gfx */
2616 	/* Set ring buffer size */
2617 	ring = &adev->gfx.gfx_ring[0];
2618 	rb_bufsz = order_base_2(ring->ring_size / 8);
2619 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2620 #ifdef __BIG_ENDIAN
2621 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2622 #endif
2623 	WREG32(mmCP_RB0_CNTL, tmp);
2624 
2625 	/* Initialize the ring buffer's read and write pointers */
2626 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2627 	ring->wptr = 0;
2628 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2629 
2630 	/* set the wb address wether it's enabled or not */
2631 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2632 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2633 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2634 
2635 	/* scratch register shadowing is no longer supported */
2636 	WREG32(mmSCRATCH_UMSK, 0);
2637 
2638 	mdelay(1);
2639 	WREG32(mmCP_RB0_CNTL, tmp);
2640 
2641 	rb_addr = ring->gpu_addr >> 8;
2642 	WREG32(mmCP_RB0_BASE, rb_addr);
2643 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2644 
2645 	/* start the ring */
2646 	gfx_v7_0_cp_gfx_start(adev);
2647 	r = amdgpu_ring_test_helper(ring);
2648 	if (r)
2649 		return r;
2650 
2651 	return 0;
2652 }
2653 
2654 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2655 {
2656 	return ring->adev->wb.wb[ring->rptr_offs];
2657 }
2658 
2659 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2660 {
2661 	struct amdgpu_device *adev = ring->adev;
2662 
2663 	return RREG32(mmCP_RB0_WPTR);
2664 }
2665 
2666 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2667 {
2668 	struct amdgpu_device *adev = ring->adev;
2669 
2670 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2671 	(void)RREG32(mmCP_RB0_WPTR);
2672 }
2673 
2674 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2675 {
2676 	/* XXX check if swapping is necessary on BE */
2677 	return ring->adev->wb.wb[ring->wptr_offs];
2678 }
2679 
2680 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2681 {
2682 	struct amdgpu_device *adev = ring->adev;
2683 
2684 	/* XXX check if swapping is necessary on BE */
2685 	adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2686 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2687 }
2688 
2689 /**
2690  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2691  *
2692  * @adev: amdgpu_device pointer
2693  * @enable: enable or disable the MEs
2694  *
2695  * Halts or unhalts the compute MEs.
2696  */
2697 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2698 {
2699 	if (enable)
2700 		WREG32(mmCP_MEC_CNTL, 0);
2701 	else
2702 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK |
2703 				       CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2704 	udelay(50);
2705 }
2706 
2707 /**
2708  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2709  *
2710  * @adev: amdgpu_device pointer
2711  *
2712  * Loads the compute MEC1&2 ucode.
2713  * Returns 0 for success, -EINVAL if the ucode is not available.
2714  */
2715 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2716 {
2717 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2718 	const __le32 *fw_data;
2719 	unsigned i, fw_size;
2720 
2721 	if (!adev->gfx.mec_fw)
2722 		return -EINVAL;
2723 
2724 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2725 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2726 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2727 	adev->gfx.mec_feature_version = le32_to_cpu(
2728 					mec_hdr->ucode_feature_version);
2729 
2730 	gfx_v7_0_cp_compute_enable(adev, false);
2731 
2732 	/* MEC1 */
2733 	fw_data = (const __le32 *)
2734 		(adev->gfx.mec_fw->data +
2735 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2736 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2737 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2738 	for (i = 0; i < fw_size; i++)
2739 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2740 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2741 
2742 	if (adev->asic_type == CHIP_KAVERI) {
2743 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2744 
2745 		if (!adev->gfx.mec2_fw)
2746 			return -EINVAL;
2747 
2748 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2749 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2750 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2751 		adev->gfx.mec2_feature_version = le32_to_cpu(
2752 				mec2_hdr->ucode_feature_version);
2753 
2754 		/* MEC2 */
2755 		fw_data = (const __le32 *)
2756 			(adev->gfx.mec2_fw->data +
2757 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2758 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2759 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2760 		for (i = 0; i < fw_size; i++)
2761 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2762 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2763 	}
2764 
2765 	return 0;
2766 }
2767 
2768 /**
2769  * gfx_v7_0_cp_compute_fini - stop the compute queues
2770  *
2771  * @adev: amdgpu_device pointer
2772  *
2773  * Stop the compute queues and tear down the driver queue
2774  * info.
2775  */
2776 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2777 {
2778 	int i;
2779 
2780 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2781 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2782 
2783 		amdgpu_bo_free_kernel(&ring->mqd_obj, NULL, NULL);
2784 	}
2785 }
2786 
2787 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2788 {
2789 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
2790 }
2791 
2792 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2793 {
2794 	int r;
2795 	u32 *hpd;
2796 	size_t mec_hpd_size;
2797 
2798 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
2799 
2800 	/* take ownership of the relevant compute queues */
2801 	amdgpu_gfx_compute_queue_acquire(adev);
2802 
2803 	/* allocate space for ALL pipes (even the ones we don't own) */
2804 	mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec
2805 		* GFX7_MEC_HPD_SIZE * 2;
2806 
2807 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
2808 				      AMDGPU_GEM_DOMAIN_VRAM,
2809 				      &adev->gfx.mec.hpd_eop_obj,
2810 				      &adev->gfx.mec.hpd_eop_gpu_addr,
2811 				      (void **)&hpd);
2812 	if (r) {
2813 		dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r);
2814 		gfx_v7_0_mec_fini(adev);
2815 		return r;
2816 	}
2817 
2818 	/* clear memory.  Not sure if this is required or not */
2819 	memset(hpd, 0, mec_hpd_size);
2820 
2821 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2822 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2823 
2824 	return 0;
2825 }
2826 
2827 struct hqd_registers
2828 {
2829 	u32 cp_mqd_base_addr;
2830 	u32 cp_mqd_base_addr_hi;
2831 	u32 cp_hqd_active;
2832 	u32 cp_hqd_vmid;
2833 	u32 cp_hqd_persistent_state;
2834 	u32 cp_hqd_pipe_priority;
2835 	u32 cp_hqd_queue_priority;
2836 	u32 cp_hqd_quantum;
2837 	u32 cp_hqd_pq_base;
2838 	u32 cp_hqd_pq_base_hi;
2839 	u32 cp_hqd_pq_rptr;
2840 	u32 cp_hqd_pq_rptr_report_addr;
2841 	u32 cp_hqd_pq_rptr_report_addr_hi;
2842 	u32 cp_hqd_pq_wptr_poll_addr;
2843 	u32 cp_hqd_pq_wptr_poll_addr_hi;
2844 	u32 cp_hqd_pq_doorbell_control;
2845 	u32 cp_hqd_pq_wptr;
2846 	u32 cp_hqd_pq_control;
2847 	u32 cp_hqd_ib_base_addr;
2848 	u32 cp_hqd_ib_base_addr_hi;
2849 	u32 cp_hqd_ib_rptr;
2850 	u32 cp_hqd_ib_control;
2851 	u32 cp_hqd_iq_timer;
2852 	u32 cp_hqd_iq_rptr;
2853 	u32 cp_hqd_dequeue_request;
2854 	u32 cp_hqd_dma_offload;
2855 	u32 cp_hqd_sema_cmd;
2856 	u32 cp_hqd_msg_type;
2857 	u32 cp_hqd_atomic0_preop_lo;
2858 	u32 cp_hqd_atomic0_preop_hi;
2859 	u32 cp_hqd_atomic1_preop_lo;
2860 	u32 cp_hqd_atomic1_preop_hi;
2861 	u32 cp_hqd_hq_scheduler0;
2862 	u32 cp_hqd_hq_scheduler1;
2863 	u32 cp_mqd_control;
2864 };
2865 
2866 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev,
2867 				       int mec, int pipe)
2868 {
2869 	u64 eop_gpu_addr;
2870 	u32 tmp;
2871 	size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe)
2872 			    * GFX7_MEC_HPD_SIZE * 2;
2873 
2874 	mutex_lock(&adev->srbm_mutex);
2875 	eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset;
2876 
2877 	cik_srbm_select(adev, mec + 1, pipe, 0, 0);
2878 
2879 	/* write the EOP addr */
2880 	WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2881 	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2882 
2883 	/* set the VMID assigned */
2884 	WREG32(mmCP_HPD_EOP_VMID, 0);
2885 
2886 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2887 	tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2888 	tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2889 	tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8);
2890 	WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2891 
2892 	cik_srbm_select(adev, 0, 0, 0, 0);
2893 	mutex_unlock(&adev->srbm_mutex);
2894 }
2895 
2896 static int gfx_v7_0_mqd_deactivate(struct amdgpu_device *adev)
2897 {
2898 	int i;
2899 
2900 	/* disable the queue if it's active */
2901 	if (RREG32(mmCP_HQD_ACTIVE) & 1) {
2902 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
2903 		for (i = 0; i < adev->usec_timeout; i++) {
2904 			if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
2905 				break;
2906 			udelay(1);
2907 		}
2908 
2909 		if (i == adev->usec_timeout)
2910 			return -ETIMEDOUT;
2911 
2912 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
2913 		WREG32(mmCP_HQD_PQ_RPTR, 0);
2914 		WREG32(mmCP_HQD_PQ_WPTR, 0);
2915 	}
2916 
2917 	return 0;
2918 }
2919 
2920 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev,
2921 			     struct cik_mqd *mqd,
2922 			     uint64_t mqd_gpu_addr,
2923 			     struct amdgpu_ring *ring)
2924 {
2925 	u64 hqd_gpu_addr;
2926 	u64 wb_gpu_addr;
2927 
2928 	/* init the mqd struct */
2929 	memset(mqd, 0, sizeof(struct cik_mqd));
2930 
2931 	mqd->header = 0xC0310800;
2932 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2933 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2934 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2935 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2936 
2937 	/* enable doorbell? */
2938 	mqd->cp_hqd_pq_doorbell_control =
2939 		RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2940 	if (ring->use_doorbell)
2941 		mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2942 	else
2943 		mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2944 
2945 	/* set the pointer to the MQD */
2946 	mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc;
2947 	mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
2948 
2949 	/* set MQD vmid to 0 */
2950 	mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
2951 	mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
2952 
2953 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2954 	hqd_gpu_addr = ring->gpu_addr >> 8;
2955 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2956 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2957 
2958 	/* set up the HQD, this is similar to CP_RB0_CNTL */
2959 	mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
2960 	mqd->cp_hqd_pq_control &=
2961 		~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
2962 				CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
2963 
2964 	mqd->cp_hqd_pq_control |=
2965 		order_base_2(ring->ring_size / 8);
2966 	mqd->cp_hqd_pq_control |=
2967 		(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
2968 #ifdef __BIG_ENDIAN
2969 	mqd->cp_hqd_pq_control |=
2970 		2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
2971 #endif
2972 	mqd->cp_hqd_pq_control &=
2973 		~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
2974 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
2975 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
2976 	mqd->cp_hqd_pq_control |=
2977 		CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
2978 		CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
2979 
2980 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2981 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2982 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2983 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2984 
2985 	/* set the wb address wether it's enabled or not */
2986 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2987 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2988 	mqd->cp_hqd_pq_rptr_report_addr_hi =
2989 		upper_32_bits(wb_gpu_addr) & 0xffff;
2990 
2991 	/* enable the doorbell if requested */
2992 	if (ring->use_doorbell) {
2993 		mqd->cp_hqd_pq_doorbell_control =
2994 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2995 		mqd->cp_hqd_pq_doorbell_control &=
2996 			~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
2997 		mqd->cp_hqd_pq_doorbell_control |=
2998 			(ring->doorbell_index <<
2999 			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
3000 		mqd->cp_hqd_pq_doorbell_control |=
3001 			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3002 		mqd->cp_hqd_pq_doorbell_control &=
3003 			~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
3004 					CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
3005 
3006 	} else {
3007 		mqd->cp_hqd_pq_doorbell_control = 0;
3008 	}
3009 
3010 	/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3011 	ring->wptr = 0;
3012 	mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr);
3013 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
3014 
3015 	/* set the vmid for the queue */
3016 	mqd->cp_hqd_vmid = 0;
3017 
3018 	/* defaults */
3019 	mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL);
3020 	mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR);
3021 	mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI);
3022 	mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR);
3023 	mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE);
3024 	mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD);
3025 	mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE);
3026 	mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO);
3027 	mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI);
3028 	mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO);
3029 	mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI);
3030 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
3031 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
3032 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
3033 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
3034 	mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR);
3035 
3036 	/* activate the queue */
3037 	mqd->cp_hqd_active = 1;
3038 }
3039 
3040 static int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd)
3041 {
3042 	uint32_t tmp;
3043 	uint32_t mqd_reg;
3044 	uint32_t *mqd_data;
3045 
3046 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */
3047 	mqd_data = &mqd->cp_mqd_base_addr_lo;
3048 
3049 	/* disable wptr polling */
3050 	tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
3051 	tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3052 	WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
3053 
3054 	/* program all HQD registers */
3055 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++)
3056 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3057 
3058 	/* activate the HQD */
3059 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
3060 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3061 
3062 	return 0;
3063 }
3064 
3065 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id)
3066 {
3067 	int r;
3068 	u64 mqd_gpu_addr;
3069 	struct cik_mqd *mqd;
3070 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
3071 
3072 	r = amdgpu_bo_create_reserved(adev, sizeof(struct cik_mqd), PAGE_SIZE,
3073 				      AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj,
3074 				      &mqd_gpu_addr, (void **)&mqd);
3075 	if (r) {
3076 		dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
3077 		return r;
3078 	}
3079 
3080 	mutex_lock(&adev->srbm_mutex);
3081 	cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3082 
3083 	gfx_v7_0_mqd_init(adev, mqd, mqd_gpu_addr, ring);
3084 	gfx_v7_0_mqd_deactivate(adev);
3085 	gfx_v7_0_mqd_commit(adev, mqd);
3086 
3087 	cik_srbm_select(adev, 0, 0, 0, 0);
3088 	mutex_unlock(&adev->srbm_mutex);
3089 
3090 	amdgpu_bo_kunmap(ring->mqd_obj);
3091 	amdgpu_bo_unreserve(ring->mqd_obj);
3092 	return 0;
3093 }
3094 
3095 /**
3096  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
3097  *
3098  * @adev: amdgpu_device pointer
3099  *
3100  * Program the compute queues and test them to make sure they
3101  * are working.
3102  * Returns 0 for success, error for failure.
3103  */
3104 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
3105 {
3106 	int r, i, j;
3107 	u32 tmp;
3108 	struct amdgpu_ring *ring;
3109 
3110 	/* fix up chicken bits */
3111 	tmp = RREG32(mmCP_CPF_DEBUG);
3112 	tmp |= (1 << 23);
3113 	WREG32(mmCP_CPF_DEBUG, tmp);
3114 
3115 	/* init all pipes (even the ones we don't own) */
3116 	for (i = 0; i < adev->gfx.mec.num_mec; i++)
3117 		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++)
3118 			gfx_v7_0_compute_pipe_init(adev, i, j);
3119 
3120 	/* init the queues */
3121 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3122 		r = gfx_v7_0_compute_queue_init(adev, i);
3123 		if (r) {
3124 			gfx_v7_0_cp_compute_fini(adev);
3125 			return r;
3126 		}
3127 	}
3128 
3129 	gfx_v7_0_cp_compute_enable(adev, true);
3130 
3131 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3132 		ring = &adev->gfx.compute_ring[i];
3133 		amdgpu_ring_test_helper(ring);
3134 	}
3135 
3136 	return 0;
3137 }
3138 
3139 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3140 {
3141 	gfx_v7_0_cp_gfx_enable(adev, enable);
3142 	gfx_v7_0_cp_compute_enable(adev, enable);
3143 }
3144 
3145 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3146 {
3147 	int r;
3148 
3149 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3150 	if (r)
3151 		return r;
3152 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3153 	if (r)
3154 		return r;
3155 
3156 	return 0;
3157 }
3158 
3159 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3160 					       bool enable)
3161 {
3162 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3163 
3164 	if (enable)
3165 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3166 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3167 	else
3168 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3169 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3170 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3171 }
3172 
3173 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3174 {
3175 	int r;
3176 
3177 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3178 
3179 	r = gfx_v7_0_cp_load_microcode(adev);
3180 	if (r)
3181 		return r;
3182 
3183 	r = gfx_v7_0_cp_gfx_resume(adev);
3184 	if (r)
3185 		return r;
3186 	r = gfx_v7_0_cp_compute_resume(adev);
3187 	if (r)
3188 		return r;
3189 
3190 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3191 
3192 	return 0;
3193 }
3194 
3195 /**
3196  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3197  *
3198  * @ring: the ring to emmit the commands to
3199  *
3200  * Sync the command pipeline with the PFP. E.g. wait for everything
3201  * to be completed.
3202  */
3203 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3204 {
3205 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3206 	uint32_t seq = ring->fence_drv.sync_seq;
3207 	uint64_t addr = ring->fence_drv.gpu_addr;
3208 
3209 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3210 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3211 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3212 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3213 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3214 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3215 	amdgpu_ring_write(ring, seq);
3216 	amdgpu_ring_write(ring, 0xffffffff);
3217 	amdgpu_ring_write(ring, 4); /* poll interval */
3218 
3219 	if (usepfp) {
3220 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3221 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3222 		amdgpu_ring_write(ring, 0);
3223 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3224 		amdgpu_ring_write(ring, 0);
3225 	}
3226 }
3227 
3228 /*
3229  * vm
3230  * VMID 0 is the physical GPU addresses as used by the kernel.
3231  * VMIDs 1-15 are used for userspace clients and are handled
3232  * by the amdgpu vm/hsa code.
3233  */
3234 /**
3235  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3236  *
3237  * @adev: amdgpu_device pointer
3238  *
3239  * Update the page table base and flush the VM TLB
3240  * using the CP (CIK).
3241  */
3242 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3243 					unsigned vmid, uint64_t pd_addr)
3244 {
3245 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3246 
3247 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3248 
3249 	/* wait for the invalidate to complete */
3250 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3251 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3252 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3253 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3254 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3255 	amdgpu_ring_write(ring, 0);
3256 	amdgpu_ring_write(ring, 0); /* ref */
3257 	amdgpu_ring_write(ring, 0); /* mask */
3258 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3259 
3260 	/* compute doesn't have PFP */
3261 	if (usepfp) {
3262 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3263 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3264 		amdgpu_ring_write(ring, 0x0);
3265 
3266 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3267 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3268 		amdgpu_ring_write(ring, 0);
3269 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3270 		amdgpu_ring_write(ring, 0);
3271 	}
3272 }
3273 
3274 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
3275 				    uint32_t reg, uint32_t val)
3276 {
3277 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3278 
3279 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3280 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3281 				 WRITE_DATA_DST_SEL(0)));
3282 	amdgpu_ring_write(ring, reg);
3283 	amdgpu_ring_write(ring, 0);
3284 	amdgpu_ring_write(ring, val);
3285 }
3286 
3287 /*
3288  * RLC
3289  * The RLC is a multi-purpose microengine that handles a
3290  * variety of functions.
3291  */
3292 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3293 {
3294 	const u32 *src_ptr;
3295 	u32 dws;
3296 	const struct cs_section_def *cs_data;
3297 	int r;
3298 
3299 	/* allocate rlc buffers */
3300 	if (adev->flags & AMD_IS_APU) {
3301 		if (adev->asic_type == CHIP_KAVERI) {
3302 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3303 			adev->gfx.rlc.reg_list_size =
3304 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3305 		} else {
3306 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3307 			adev->gfx.rlc.reg_list_size =
3308 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3309 		}
3310 	}
3311 	adev->gfx.rlc.cs_data = ci_cs_data;
3312 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3313 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3314 
3315 	src_ptr = adev->gfx.rlc.reg_list;
3316 	dws = adev->gfx.rlc.reg_list_size;
3317 	dws += (5 * 16) + 48 + 48 + 64;
3318 
3319 	cs_data = adev->gfx.rlc.cs_data;
3320 
3321 	if (src_ptr) {
3322 		/* init save restore block */
3323 		r = amdgpu_gfx_rlc_init_sr(adev, dws);
3324 		if (r)
3325 			return r;
3326 	}
3327 
3328 	if (cs_data) {
3329 		/* init clear state block */
3330 		r = amdgpu_gfx_rlc_init_csb(adev);
3331 		if (r)
3332 			return r;
3333 	}
3334 
3335 	if (adev->gfx.rlc.cp_table_size) {
3336 		r = amdgpu_gfx_rlc_init_cpt(adev);
3337 		if (r)
3338 			return r;
3339 	}
3340 
3341 	/* init spm vmid with 0xf */
3342 	if (adev->gfx.rlc.funcs->update_spm_vmid)
3343 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
3344 
3345 	return 0;
3346 }
3347 
3348 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3349 {
3350 	u32 tmp;
3351 
3352 	tmp = RREG32(mmRLC_LB_CNTL);
3353 	if (enable)
3354 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3355 	else
3356 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3357 	WREG32(mmRLC_LB_CNTL, tmp);
3358 }
3359 
3360 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3361 {
3362 	u32 i, j, k;
3363 	u32 mask;
3364 
3365 	mutex_lock(&adev->grbm_idx_mutex);
3366 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3367 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3368 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
3369 			for (k = 0; k < adev->usec_timeout; k++) {
3370 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3371 					break;
3372 				udelay(1);
3373 			}
3374 		}
3375 	}
3376 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3377 	mutex_unlock(&adev->grbm_idx_mutex);
3378 
3379 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3380 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3381 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3382 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3383 	for (k = 0; k < adev->usec_timeout; k++) {
3384 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3385 			break;
3386 		udelay(1);
3387 	}
3388 }
3389 
3390 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3391 {
3392 	u32 tmp;
3393 
3394 	tmp = RREG32(mmRLC_CNTL);
3395 	if (tmp != rlc)
3396 		WREG32(mmRLC_CNTL, rlc);
3397 }
3398 
3399 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3400 {
3401 	u32 data, orig;
3402 
3403 	orig = data = RREG32(mmRLC_CNTL);
3404 
3405 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3406 		u32 i;
3407 
3408 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3409 		WREG32(mmRLC_CNTL, data);
3410 
3411 		for (i = 0; i < adev->usec_timeout; i++) {
3412 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3413 				break;
3414 			udelay(1);
3415 		}
3416 
3417 		gfx_v7_0_wait_for_rlc_serdes(adev);
3418 	}
3419 
3420 	return orig;
3421 }
3422 
3423 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev)
3424 {
3425 	return true;
3426 }
3427 
3428 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev)
3429 {
3430 	u32 tmp, i, mask;
3431 
3432 	tmp = 0x1 | (1 << 1);
3433 	WREG32(mmRLC_GPR_REG2, tmp);
3434 
3435 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3436 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3437 	for (i = 0; i < adev->usec_timeout; i++) {
3438 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3439 			break;
3440 		udelay(1);
3441 	}
3442 
3443 	for (i = 0; i < adev->usec_timeout; i++) {
3444 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3445 			break;
3446 		udelay(1);
3447 	}
3448 }
3449 
3450 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev)
3451 {
3452 	u32 tmp;
3453 
3454 	tmp = 0x1 | (0 << 1);
3455 	WREG32(mmRLC_GPR_REG2, tmp);
3456 }
3457 
3458 /**
3459  * gfx_v7_0_rlc_stop - stop the RLC ME
3460  *
3461  * @adev: amdgpu_device pointer
3462  *
3463  * Halt the RLC ME (MicroEngine) (CIK).
3464  */
3465 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3466 {
3467 	WREG32(mmRLC_CNTL, 0);
3468 
3469 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3470 
3471 	gfx_v7_0_wait_for_rlc_serdes(adev);
3472 }
3473 
3474 /**
3475  * gfx_v7_0_rlc_start - start the RLC ME
3476  *
3477  * @adev: amdgpu_device pointer
3478  *
3479  * Unhalt the RLC ME (MicroEngine) (CIK).
3480  */
3481 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3482 {
3483 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3484 
3485 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3486 
3487 	udelay(50);
3488 }
3489 
3490 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3491 {
3492 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3493 
3494 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3495 	WREG32(mmGRBM_SOFT_RESET, tmp);
3496 	udelay(50);
3497 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3498 	WREG32(mmGRBM_SOFT_RESET, tmp);
3499 	udelay(50);
3500 }
3501 
3502 /**
3503  * gfx_v7_0_rlc_resume - setup the RLC hw
3504  *
3505  * @adev: amdgpu_device pointer
3506  *
3507  * Initialize the RLC registers, load the ucode,
3508  * and start the RLC (CIK).
3509  * Returns 0 for success, -EINVAL if the ucode is not available.
3510  */
3511 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3512 {
3513 	const struct rlc_firmware_header_v1_0 *hdr;
3514 	const __le32 *fw_data;
3515 	unsigned i, fw_size;
3516 	u32 tmp;
3517 
3518 	if (!adev->gfx.rlc_fw)
3519 		return -EINVAL;
3520 
3521 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3522 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3523 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3524 	adev->gfx.rlc_feature_version = le32_to_cpu(
3525 					hdr->ucode_feature_version);
3526 
3527 	adev->gfx.rlc.funcs->stop(adev);
3528 
3529 	/* disable CG */
3530 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3531 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3532 
3533 	adev->gfx.rlc.funcs->reset(adev);
3534 
3535 	gfx_v7_0_init_pg(adev);
3536 
3537 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3538 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3539 
3540 	mutex_lock(&adev->grbm_idx_mutex);
3541 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3542 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3543 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3544 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3545 	mutex_unlock(&adev->grbm_idx_mutex);
3546 
3547 	WREG32(mmRLC_MC_CNTL, 0);
3548 	WREG32(mmRLC_UCODE_CNTL, 0);
3549 
3550 	fw_data = (const __le32 *)
3551 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3552 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3553 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3554 	for (i = 0; i < fw_size; i++)
3555 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3556 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3557 
3558 	/* XXX - find out what chips support lbpw */
3559 	gfx_v7_0_enable_lbpw(adev, false);
3560 
3561 	if (adev->asic_type == CHIP_BONAIRE)
3562 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3563 
3564 	adev->gfx.rlc.funcs->start(adev);
3565 
3566 	return 0;
3567 }
3568 
3569 static void gfx_v7_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
3570 {
3571 	u32 data;
3572 
3573 	data = RREG32(mmRLC_SPM_VMID);
3574 
3575 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
3576 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
3577 
3578 	WREG32(mmRLC_SPM_VMID, data);
3579 }
3580 
3581 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3582 {
3583 	u32 data, orig, tmp, tmp2;
3584 
3585 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3586 
3587 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3588 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3589 
3590 		tmp = gfx_v7_0_halt_rlc(adev);
3591 
3592 		mutex_lock(&adev->grbm_idx_mutex);
3593 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3594 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3595 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3596 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3597 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3598 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3599 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3600 		mutex_unlock(&adev->grbm_idx_mutex);
3601 
3602 		gfx_v7_0_update_rlc(adev, tmp);
3603 
3604 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3605 		if (orig != data)
3606 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3607 
3608 	} else {
3609 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3610 
3611 		RREG32(mmCB_CGTT_SCLK_CTRL);
3612 		RREG32(mmCB_CGTT_SCLK_CTRL);
3613 		RREG32(mmCB_CGTT_SCLK_CTRL);
3614 		RREG32(mmCB_CGTT_SCLK_CTRL);
3615 
3616 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3617 		if (orig != data)
3618 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3619 
3620 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3621 	}
3622 }
3623 
3624 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3625 {
3626 	u32 data, orig, tmp = 0;
3627 
3628 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3629 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3630 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3631 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3632 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3633 				if (orig != data)
3634 					WREG32(mmCP_MEM_SLP_CNTL, data);
3635 			}
3636 		}
3637 
3638 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3639 		data |= 0x00000001;
3640 		data &= 0xfffffffd;
3641 		if (orig != data)
3642 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3643 
3644 		tmp = gfx_v7_0_halt_rlc(adev);
3645 
3646 		mutex_lock(&adev->grbm_idx_mutex);
3647 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3648 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3649 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3650 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3651 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3652 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3653 		mutex_unlock(&adev->grbm_idx_mutex);
3654 
3655 		gfx_v7_0_update_rlc(adev, tmp);
3656 
3657 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3658 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3659 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3660 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3661 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3662 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3663 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3664 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3665 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3666 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3667 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3668 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3669 			if (orig != data)
3670 				WREG32(mmCGTS_SM_CTRL_REG, data);
3671 		}
3672 	} else {
3673 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3674 		data |= 0x00000003;
3675 		if (orig != data)
3676 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3677 
3678 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3679 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3680 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3681 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3682 		}
3683 
3684 		data = RREG32(mmCP_MEM_SLP_CNTL);
3685 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3686 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3687 			WREG32(mmCP_MEM_SLP_CNTL, data);
3688 		}
3689 
3690 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3691 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3692 		if (orig != data)
3693 			WREG32(mmCGTS_SM_CTRL_REG, data);
3694 
3695 		tmp = gfx_v7_0_halt_rlc(adev);
3696 
3697 		mutex_lock(&adev->grbm_idx_mutex);
3698 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3699 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3700 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3701 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3702 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3703 		mutex_unlock(&adev->grbm_idx_mutex);
3704 
3705 		gfx_v7_0_update_rlc(adev, tmp);
3706 	}
3707 }
3708 
3709 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3710 			       bool enable)
3711 {
3712 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3713 	/* order matters! */
3714 	if (enable) {
3715 		gfx_v7_0_enable_mgcg(adev, true);
3716 		gfx_v7_0_enable_cgcg(adev, true);
3717 	} else {
3718 		gfx_v7_0_enable_cgcg(adev, false);
3719 		gfx_v7_0_enable_mgcg(adev, false);
3720 	}
3721 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3722 }
3723 
3724 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3725 						bool enable)
3726 {
3727 	u32 data, orig;
3728 
3729 	orig = data = RREG32(mmRLC_PG_CNTL);
3730 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3731 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3732 	else
3733 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3734 	if (orig != data)
3735 		WREG32(mmRLC_PG_CNTL, data);
3736 }
3737 
3738 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3739 						bool enable)
3740 {
3741 	u32 data, orig;
3742 
3743 	orig = data = RREG32(mmRLC_PG_CNTL);
3744 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3745 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3746 	else
3747 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3748 	if (orig != data)
3749 		WREG32(mmRLC_PG_CNTL, data);
3750 }
3751 
3752 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3753 {
3754 	u32 data, orig;
3755 
3756 	orig = data = RREG32(mmRLC_PG_CNTL);
3757 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3758 		data &= ~0x8000;
3759 	else
3760 		data |= 0x8000;
3761 	if (orig != data)
3762 		WREG32(mmRLC_PG_CNTL, data);
3763 }
3764 
3765 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3766 {
3767 	u32 data, orig;
3768 
3769 	orig = data = RREG32(mmRLC_PG_CNTL);
3770 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3771 		data &= ~0x2000;
3772 	else
3773 		data |= 0x2000;
3774 	if (orig != data)
3775 		WREG32(mmRLC_PG_CNTL, data);
3776 }
3777 
3778 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev)
3779 {
3780 	if (adev->asic_type == CHIP_KAVERI)
3781 		return 5;
3782 	else
3783 		return 4;
3784 }
3785 
3786 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
3787 				     bool enable)
3788 {
3789 	u32 data, orig;
3790 
3791 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
3792 		orig = data = RREG32(mmRLC_PG_CNTL);
3793 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3794 		if (orig != data)
3795 			WREG32(mmRLC_PG_CNTL, data);
3796 
3797 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3798 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3799 		if (orig != data)
3800 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3801 	} else {
3802 		orig = data = RREG32(mmRLC_PG_CNTL);
3803 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3804 		if (orig != data)
3805 			WREG32(mmRLC_PG_CNTL, data);
3806 
3807 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3808 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3809 		if (orig != data)
3810 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3811 
3812 		data = RREG32(mmDB_RENDER_CONTROL);
3813 	}
3814 }
3815 
3816 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
3817 						 u32 bitmap)
3818 {
3819 	u32 data;
3820 
3821 	if (!bitmap)
3822 		return;
3823 
3824 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3825 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3826 
3827 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
3828 }
3829 
3830 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
3831 {
3832 	u32 data, mask;
3833 
3834 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
3835 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
3836 
3837 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3838 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3839 
3840 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
3841 
3842 	return (~data) & mask;
3843 }
3844 
3845 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
3846 {
3847 	u32 tmp;
3848 
3849 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
3850 
3851 	tmp = RREG32(mmRLC_MAX_PG_CU);
3852 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
3853 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
3854 	WREG32(mmRLC_MAX_PG_CU, tmp);
3855 }
3856 
3857 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
3858 					    bool enable)
3859 {
3860 	u32 data, orig;
3861 
3862 	orig = data = RREG32(mmRLC_PG_CNTL);
3863 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
3864 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3865 	else
3866 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3867 	if (orig != data)
3868 		WREG32(mmRLC_PG_CNTL, data);
3869 }
3870 
3871 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
3872 					     bool enable)
3873 {
3874 	u32 data, orig;
3875 
3876 	orig = data = RREG32(mmRLC_PG_CNTL);
3877 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
3878 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3879 	else
3880 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3881 	if (orig != data)
3882 		WREG32(mmRLC_PG_CNTL, data);
3883 }
3884 
3885 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
3886 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
3887 
3888 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
3889 {
3890 	u32 data, orig;
3891 	u32 i;
3892 
3893 	if (adev->gfx.rlc.cs_data) {
3894 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3895 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3896 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3897 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
3898 	} else {
3899 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3900 		for (i = 0; i < 3; i++)
3901 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
3902 	}
3903 	if (adev->gfx.rlc.reg_list) {
3904 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
3905 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3906 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
3907 	}
3908 
3909 	orig = data = RREG32(mmRLC_PG_CNTL);
3910 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
3911 	if (orig != data)
3912 		WREG32(mmRLC_PG_CNTL, data);
3913 
3914 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
3915 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
3916 
3917 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
3918 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
3919 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3920 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
3921 
3922 	data = 0x10101010;
3923 	WREG32(mmRLC_PG_DELAY, data);
3924 
3925 	data = RREG32(mmRLC_PG_DELAY_2);
3926 	data &= ~0xff;
3927 	data |= 0x3;
3928 	WREG32(mmRLC_PG_DELAY_2, data);
3929 
3930 	data = RREG32(mmRLC_AUTO_PG_CTRL);
3931 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
3932 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
3933 	WREG32(mmRLC_AUTO_PG_CTRL, data);
3934 
3935 }
3936 
3937 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
3938 {
3939 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
3940 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
3941 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
3942 }
3943 
3944 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
3945 {
3946 	u32 count = 0;
3947 	const struct cs_section_def *sect = NULL;
3948 	const struct cs_extent_def *ext = NULL;
3949 
3950 	if (adev->gfx.rlc.cs_data == NULL)
3951 		return 0;
3952 
3953 	/* begin clear state */
3954 	count += 2;
3955 	/* context control state */
3956 	count += 3;
3957 
3958 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3959 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3960 			if (sect->id == SECT_CONTEXT)
3961 				count += 2 + ext->reg_count;
3962 			else
3963 				return 0;
3964 		}
3965 	}
3966 	/* pa_sc_raster_config/pa_sc_raster_config1 */
3967 	count += 4;
3968 	/* end clear state */
3969 	count += 2;
3970 	/* clear state */
3971 	count += 2;
3972 
3973 	return count;
3974 }
3975 
3976 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
3977 				    volatile u32 *buffer)
3978 {
3979 	u32 count = 0, i;
3980 	const struct cs_section_def *sect = NULL;
3981 	const struct cs_extent_def *ext = NULL;
3982 
3983 	if (adev->gfx.rlc.cs_data == NULL)
3984 		return;
3985 	if (buffer == NULL)
3986 		return;
3987 
3988 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3989 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3990 
3991 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3992 	buffer[count++] = cpu_to_le32(0x80000000);
3993 	buffer[count++] = cpu_to_le32(0x80000000);
3994 
3995 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3996 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3997 			if (sect->id == SECT_CONTEXT) {
3998 				buffer[count++] =
3999 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
4000 				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4001 				for (i = 0; i < ext->reg_count; i++)
4002 					buffer[count++] = cpu_to_le32(ext->extent[i]);
4003 			} else {
4004 				return;
4005 			}
4006 		}
4007 	}
4008 
4009 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4010 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4011 	switch (adev->asic_type) {
4012 	case CHIP_BONAIRE:
4013 		buffer[count++] = cpu_to_le32(0x16000012);
4014 		buffer[count++] = cpu_to_le32(0x00000000);
4015 		break;
4016 	case CHIP_KAVERI:
4017 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4018 		buffer[count++] = cpu_to_le32(0x00000000);
4019 		break;
4020 	case CHIP_KABINI:
4021 	case CHIP_MULLINS:
4022 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4023 		buffer[count++] = cpu_to_le32(0x00000000);
4024 		break;
4025 	case CHIP_HAWAII:
4026 		buffer[count++] = cpu_to_le32(0x3a00161a);
4027 		buffer[count++] = cpu_to_le32(0x0000002e);
4028 		break;
4029 	default:
4030 		buffer[count++] = cpu_to_le32(0x00000000);
4031 		buffer[count++] = cpu_to_le32(0x00000000);
4032 		break;
4033 	}
4034 
4035 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4036 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
4037 
4038 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
4039 	buffer[count++] = cpu_to_le32(0);
4040 }
4041 
4042 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
4043 {
4044 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4045 			      AMD_PG_SUPPORT_GFX_SMG |
4046 			      AMD_PG_SUPPORT_GFX_DMG |
4047 			      AMD_PG_SUPPORT_CP |
4048 			      AMD_PG_SUPPORT_GDS |
4049 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4050 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
4051 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
4052 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4053 			gfx_v7_0_init_gfx_cgpg(adev);
4054 			gfx_v7_0_enable_cp_pg(adev, true);
4055 			gfx_v7_0_enable_gds_pg(adev, true);
4056 		}
4057 		gfx_v7_0_init_ao_cu_mask(adev);
4058 		gfx_v7_0_update_gfx_pg(adev, true);
4059 	}
4060 }
4061 
4062 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4063 {
4064 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4065 			      AMD_PG_SUPPORT_GFX_SMG |
4066 			      AMD_PG_SUPPORT_GFX_DMG |
4067 			      AMD_PG_SUPPORT_CP |
4068 			      AMD_PG_SUPPORT_GDS |
4069 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4070 		gfx_v7_0_update_gfx_pg(adev, false);
4071 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4072 			gfx_v7_0_enable_cp_pg(adev, false);
4073 			gfx_v7_0_enable_gds_pg(adev, false);
4074 		}
4075 	}
4076 }
4077 
4078 /**
4079  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4080  *
4081  * @adev: amdgpu_device pointer
4082  *
4083  * Fetches a GPU clock counter snapshot (SI).
4084  * Returns the 64 bit clock counter snapshot.
4085  */
4086 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4087 {
4088 	uint64_t clock;
4089 
4090 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4091 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4092 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4093 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4094 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4095 	return clock;
4096 }
4097 
4098 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4099 					  uint32_t vmid,
4100 					  uint32_t gds_base, uint32_t gds_size,
4101 					  uint32_t gws_base, uint32_t gws_size,
4102 					  uint32_t oa_base, uint32_t oa_size)
4103 {
4104 	/* GDS Base */
4105 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4106 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4107 				WRITE_DATA_DST_SEL(0)));
4108 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4109 	amdgpu_ring_write(ring, 0);
4110 	amdgpu_ring_write(ring, gds_base);
4111 
4112 	/* GDS Size */
4113 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4114 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4115 				WRITE_DATA_DST_SEL(0)));
4116 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
4117 	amdgpu_ring_write(ring, 0);
4118 	amdgpu_ring_write(ring, gds_size);
4119 
4120 	/* GWS */
4121 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4122 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4123 				WRITE_DATA_DST_SEL(0)));
4124 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
4125 	amdgpu_ring_write(ring, 0);
4126 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4127 
4128 	/* OA */
4129 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4130 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4131 				WRITE_DATA_DST_SEL(0)));
4132 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
4133 	amdgpu_ring_write(ring, 0);
4134 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4135 }
4136 
4137 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
4138 {
4139 	struct amdgpu_device *adev = ring->adev;
4140 	uint32_t value = 0;
4141 
4142 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
4143 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
4144 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
4145 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
4146 	WREG32(mmSQ_CMD, value);
4147 }
4148 
4149 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4150 {
4151 	WREG32(mmSQ_IND_INDEX,
4152 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4153 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4154 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4155 		(SQ_IND_INDEX__FORCE_READ_MASK));
4156 	return RREG32(mmSQ_IND_DATA);
4157 }
4158 
4159 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4160 			   uint32_t wave, uint32_t thread,
4161 			   uint32_t regno, uint32_t num, uint32_t *out)
4162 {
4163 	WREG32(mmSQ_IND_INDEX,
4164 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4165 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4166 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4167 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4168 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4169 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4170 	while (num--)
4171 		*(out++) = RREG32(mmSQ_IND_DATA);
4172 }
4173 
4174 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4175 {
4176 	/* type 0 wave data */
4177 	dst[(*no_fields)++] = 0;
4178 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4179 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4180 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4181 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4182 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4183 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4184 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4185 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4186 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4187 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4188 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4189 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4190 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4191 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4192 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4193 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4194 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4195 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4196 }
4197 
4198 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
4199 				     uint32_t wave, uint32_t start,
4200 				     uint32_t size, uint32_t *dst)
4201 {
4202 	wave_read_regs(
4203 		adev, simd, wave, 0,
4204 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4205 }
4206 
4207 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
4208 				  u32 me, u32 pipe, u32 q, u32 vm)
4209 {
4210 	cik_srbm_select(adev, me, pipe, q, vm);
4211 }
4212 
4213 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4214 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4215 	.select_se_sh = &gfx_v7_0_select_se_sh,
4216 	.read_wave_data = &gfx_v7_0_read_wave_data,
4217 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4218 	.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q
4219 };
4220 
4221 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4222 	.is_rlc_enabled = gfx_v7_0_is_rlc_enabled,
4223 	.set_safe_mode = gfx_v7_0_set_safe_mode,
4224 	.unset_safe_mode = gfx_v7_0_unset_safe_mode,
4225 	.init = gfx_v7_0_rlc_init,
4226 	.get_csb_size = gfx_v7_0_get_csb_size,
4227 	.get_csb_buffer = gfx_v7_0_get_csb_buffer,
4228 	.get_cp_table_num = gfx_v7_0_cp_pg_table_num,
4229 	.resume = gfx_v7_0_rlc_resume,
4230 	.stop = gfx_v7_0_rlc_stop,
4231 	.reset = gfx_v7_0_rlc_reset,
4232 	.start = gfx_v7_0_rlc_start,
4233 	.update_spm_vmid = gfx_v7_0_update_spm_vmid
4234 };
4235 
4236 static int gfx_v7_0_early_init(void *handle)
4237 {
4238 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4239 
4240 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4241 	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
4242 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4243 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4244 	gfx_v7_0_set_ring_funcs(adev);
4245 	gfx_v7_0_set_irq_funcs(adev);
4246 	gfx_v7_0_set_gds_init(adev);
4247 
4248 	return 0;
4249 }
4250 
4251 static int gfx_v7_0_late_init(void *handle)
4252 {
4253 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4254 	int r;
4255 
4256 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4257 	if (r)
4258 		return r;
4259 
4260 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4261 	if (r)
4262 		return r;
4263 
4264 	return 0;
4265 }
4266 
4267 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4268 {
4269 	u32 gb_addr_config;
4270 	u32 mc_arb_ramcfg;
4271 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4272 	u32 tmp;
4273 
4274 	switch (adev->asic_type) {
4275 	case CHIP_BONAIRE:
4276 		adev->gfx.config.max_shader_engines = 2;
4277 		adev->gfx.config.max_tile_pipes = 4;
4278 		adev->gfx.config.max_cu_per_sh = 7;
4279 		adev->gfx.config.max_sh_per_se = 1;
4280 		adev->gfx.config.max_backends_per_se = 2;
4281 		adev->gfx.config.max_texture_channel_caches = 4;
4282 		adev->gfx.config.max_gprs = 256;
4283 		adev->gfx.config.max_gs_threads = 32;
4284 		adev->gfx.config.max_hw_contexts = 8;
4285 
4286 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4287 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4288 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4289 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4290 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4291 		break;
4292 	case CHIP_HAWAII:
4293 		adev->gfx.config.max_shader_engines = 4;
4294 		adev->gfx.config.max_tile_pipes = 16;
4295 		adev->gfx.config.max_cu_per_sh = 11;
4296 		adev->gfx.config.max_sh_per_se = 1;
4297 		adev->gfx.config.max_backends_per_se = 4;
4298 		adev->gfx.config.max_texture_channel_caches = 16;
4299 		adev->gfx.config.max_gprs = 256;
4300 		adev->gfx.config.max_gs_threads = 32;
4301 		adev->gfx.config.max_hw_contexts = 8;
4302 
4303 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4304 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4305 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4306 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4307 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4308 		break;
4309 	case CHIP_KAVERI:
4310 		adev->gfx.config.max_shader_engines = 1;
4311 		adev->gfx.config.max_tile_pipes = 4;
4312 		adev->gfx.config.max_cu_per_sh = 8;
4313 		adev->gfx.config.max_backends_per_se = 2;
4314 		adev->gfx.config.max_sh_per_se = 1;
4315 		adev->gfx.config.max_texture_channel_caches = 4;
4316 		adev->gfx.config.max_gprs = 256;
4317 		adev->gfx.config.max_gs_threads = 16;
4318 		adev->gfx.config.max_hw_contexts = 8;
4319 
4320 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4321 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4322 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4323 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4324 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4325 		break;
4326 	case CHIP_KABINI:
4327 	case CHIP_MULLINS:
4328 	default:
4329 		adev->gfx.config.max_shader_engines = 1;
4330 		adev->gfx.config.max_tile_pipes = 2;
4331 		adev->gfx.config.max_cu_per_sh = 2;
4332 		adev->gfx.config.max_sh_per_se = 1;
4333 		adev->gfx.config.max_backends_per_se = 1;
4334 		adev->gfx.config.max_texture_channel_caches = 2;
4335 		adev->gfx.config.max_gprs = 256;
4336 		adev->gfx.config.max_gs_threads = 16;
4337 		adev->gfx.config.max_hw_contexts = 8;
4338 
4339 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4340 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4341 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4342 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4343 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4344 		break;
4345 	}
4346 
4347 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4348 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4349 
4350 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
4351 				MC_ARB_RAMCFG, NOOFBANK);
4352 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
4353 				MC_ARB_RAMCFG, NOOFRANKS);
4354 
4355 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4356 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4357 	if (adev->flags & AMD_IS_APU) {
4358 		/* Get memory bank mapping mode. */
4359 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4360 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4361 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4362 
4363 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4364 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4365 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4366 
4367 		/* Validate settings in case only one DIMM installed. */
4368 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4369 			dimm00_addr_map = 0;
4370 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4371 			dimm01_addr_map = 0;
4372 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4373 			dimm10_addr_map = 0;
4374 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4375 			dimm11_addr_map = 0;
4376 
4377 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4378 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4379 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4380 			adev->gfx.config.mem_row_size_in_kb = 2;
4381 		else
4382 			adev->gfx.config.mem_row_size_in_kb = 1;
4383 	} else {
4384 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4385 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4386 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4387 			adev->gfx.config.mem_row_size_in_kb = 4;
4388 	}
4389 	/* XXX use MC settings? */
4390 	adev->gfx.config.shader_engine_tile_size = 32;
4391 	adev->gfx.config.num_gpus = 1;
4392 	adev->gfx.config.multi_gpu_tile_size = 64;
4393 
4394 	/* fix up row size */
4395 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4396 	switch (adev->gfx.config.mem_row_size_in_kb) {
4397 	case 1:
4398 	default:
4399 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4400 		break;
4401 	case 2:
4402 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4403 		break;
4404 	case 4:
4405 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4406 		break;
4407 	}
4408 	adev->gfx.config.gb_addr_config = gb_addr_config;
4409 }
4410 
4411 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
4412 					int mec, int pipe, int queue)
4413 {
4414 	int r;
4415 	unsigned irq_type;
4416 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
4417 
4418 	/* mec0 is me1 */
4419 	ring->me = mec + 1;
4420 	ring->pipe = pipe;
4421 	ring->queue = queue;
4422 
4423 	ring->ring_obj = NULL;
4424 	ring->use_doorbell = true;
4425 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
4426 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4427 
4428 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
4429 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
4430 		+ ring->pipe;
4431 
4432 	/* type-2 packets are deprecated on MEC, use type-3 instead */
4433 	r = amdgpu_ring_init(adev, ring, 1024,
4434 			     &adev->gfx.eop_irq, irq_type,
4435 			     AMDGPU_RING_PRIO_DEFAULT);
4436 	if (r)
4437 		return r;
4438 
4439 
4440 	return 0;
4441 }
4442 
4443 static int gfx_v7_0_sw_init(void *handle)
4444 {
4445 	struct amdgpu_ring *ring;
4446 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4447 	int i, j, k, r, ring_id;
4448 
4449 	switch (adev->asic_type) {
4450 	case CHIP_KAVERI:
4451 		adev->gfx.mec.num_mec = 2;
4452 		break;
4453 	case CHIP_BONAIRE:
4454 	case CHIP_HAWAII:
4455 	case CHIP_KABINI:
4456 	case CHIP_MULLINS:
4457 	default:
4458 		adev->gfx.mec.num_mec = 1;
4459 		break;
4460 	}
4461 	adev->gfx.mec.num_pipe_per_mec = 4;
4462 	adev->gfx.mec.num_queue_per_pipe = 8;
4463 
4464 	/* EOP Event */
4465 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq);
4466 	if (r)
4467 		return r;
4468 
4469 	/* Privileged reg */
4470 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184,
4471 			      &adev->gfx.priv_reg_irq);
4472 	if (r)
4473 		return r;
4474 
4475 	/* Privileged inst */
4476 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185,
4477 			      &adev->gfx.priv_inst_irq);
4478 	if (r)
4479 		return r;
4480 
4481 	gfx_v7_0_scratch_init(adev);
4482 
4483 	r = gfx_v7_0_init_microcode(adev);
4484 	if (r) {
4485 		DRM_ERROR("Failed to load gfx firmware!\n");
4486 		return r;
4487 	}
4488 
4489 	r = adev->gfx.rlc.funcs->init(adev);
4490 	if (r) {
4491 		DRM_ERROR("Failed to init rlc BOs!\n");
4492 		return r;
4493 	}
4494 
4495 	/* allocate mec buffers */
4496 	r = gfx_v7_0_mec_init(adev);
4497 	if (r) {
4498 		DRM_ERROR("Failed to init MEC BOs!\n");
4499 		return r;
4500 	}
4501 
4502 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4503 		ring = &adev->gfx.gfx_ring[i];
4504 		ring->ring_obj = NULL;
4505 		sprintf(ring->name, "gfx");
4506 		r = amdgpu_ring_init(adev, ring, 1024,
4507 				     &adev->gfx.eop_irq,
4508 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
4509 				     AMDGPU_RING_PRIO_DEFAULT);
4510 		if (r)
4511 			return r;
4512 	}
4513 
4514 	/* set up the compute queues - allocate horizontally across pipes */
4515 	ring_id = 0;
4516 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4517 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4518 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4519 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
4520 					continue;
4521 
4522 				r = gfx_v7_0_compute_ring_init(adev,
4523 								ring_id,
4524 								i, k, j);
4525 				if (r)
4526 					return r;
4527 
4528 				ring_id++;
4529 			}
4530 		}
4531 	}
4532 
4533 	adev->gfx.ce_ram_size = 0x8000;
4534 
4535 	gfx_v7_0_gpu_early_init(adev);
4536 
4537 	return r;
4538 }
4539 
4540 static int gfx_v7_0_sw_fini(void *handle)
4541 {
4542 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4543 	int i;
4544 
4545 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4546 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4547 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
4548 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4549 
4550 	gfx_v7_0_cp_compute_fini(adev);
4551 	amdgpu_gfx_rlc_fini(adev);
4552 	gfx_v7_0_mec_fini(adev);
4553 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
4554 				&adev->gfx.rlc.clear_state_gpu_addr,
4555 				(void **)&adev->gfx.rlc.cs_ptr);
4556 	if (adev->gfx.rlc.cp_table_size) {
4557 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
4558 				&adev->gfx.rlc.cp_table_gpu_addr,
4559 				(void **)&adev->gfx.rlc.cp_table_ptr);
4560 	}
4561 	gfx_v7_0_free_microcode(adev);
4562 
4563 	return 0;
4564 }
4565 
4566 static int gfx_v7_0_hw_init(void *handle)
4567 {
4568 	int r;
4569 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4570 
4571 	gfx_v7_0_constants_init(adev);
4572 
4573 	/* init CSB */
4574 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
4575 	/* init rlc */
4576 	r = adev->gfx.rlc.funcs->resume(adev);
4577 	if (r)
4578 		return r;
4579 
4580 	r = gfx_v7_0_cp_resume(adev);
4581 	if (r)
4582 		return r;
4583 
4584 	return r;
4585 }
4586 
4587 static int gfx_v7_0_hw_fini(void *handle)
4588 {
4589 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4590 
4591 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4592 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4593 	gfx_v7_0_cp_enable(adev, false);
4594 	adev->gfx.rlc.funcs->stop(adev);
4595 	gfx_v7_0_fini_pg(adev);
4596 
4597 	return 0;
4598 }
4599 
4600 static int gfx_v7_0_suspend(void *handle)
4601 {
4602 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4603 
4604 	return gfx_v7_0_hw_fini(adev);
4605 }
4606 
4607 static int gfx_v7_0_resume(void *handle)
4608 {
4609 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4610 
4611 	return gfx_v7_0_hw_init(adev);
4612 }
4613 
4614 static bool gfx_v7_0_is_idle(void *handle)
4615 {
4616 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4617 
4618 	if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK)
4619 		return false;
4620 	else
4621 		return true;
4622 }
4623 
4624 static int gfx_v7_0_wait_for_idle(void *handle)
4625 {
4626 	unsigned i;
4627 	u32 tmp;
4628 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4629 
4630 	for (i = 0; i < adev->usec_timeout; i++) {
4631 		/* read MC_STATUS */
4632 		tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4633 
4634 		if (!tmp)
4635 			return 0;
4636 		udelay(1);
4637 	}
4638 	return -ETIMEDOUT;
4639 }
4640 
4641 static int gfx_v7_0_soft_reset(void *handle)
4642 {
4643 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4644 	u32 tmp;
4645 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4646 
4647 	/* GRBM_STATUS */
4648 	tmp = RREG32(mmGRBM_STATUS);
4649 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4650 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4651 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4652 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4653 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4654 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK))
4655 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK |
4656 			GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK;
4657 
4658 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4659 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK;
4660 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4661 	}
4662 
4663 	/* GRBM_STATUS2 */
4664 	tmp = RREG32(mmGRBM_STATUS2);
4665 	if (tmp & GRBM_STATUS2__RLC_BUSY_MASK)
4666 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
4667 
4668 	/* SRBM_STATUS */
4669 	tmp = RREG32(mmSRBM_STATUS);
4670 	if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK)
4671 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4672 
4673 	if (grbm_soft_reset || srbm_soft_reset) {
4674 		/* disable CG/PG */
4675 		gfx_v7_0_fini_pg(adev);
4676 		gfx_v7_0_update_cg(adev, false);
4677 
4678 		/* stop the rlc */
4679 		adev->gfx.rlc.funcs->stop(adev);
4680 
4681 		/* Disable GFX parsing/prefetching */
4682 		WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK);
4683 
4684 		/* Disable MEC parsing/prefetching */
4685 		WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK);
4686 
4687 		if (grbm_soft_reset) {
4688 			tmp = RREG32(mmGRBM_SOFT_RESET);
4689 			tmp |= grbm_soft_reset;
4690 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4691 			WREG32(mmGRBM_SOFT_RESET, tmp);
4692 			tmp = RREG32(mmGRBM_SOFT_RESET);
4693 
4694 			udelay(50);
4695 
4696 			tmp &= ~grbm_soft_reset;
4697 			WREG32(mmGRBM_SOFT_RESET, tmp);
4698 			tmp = RREG32(mmGRBM_SOFT_RESET);
4699 		}
4700 
4701 		if (srbm_soft_reset) {
4702 			tmp = RREG32(mmSRBM_SOFT_RESET);
4703 			tmp |= srbm_soft_reset;
4704 			dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4705 			WREG32(mmSRBM_SOFT_RESET, tmp);
4706 			tmp = RREG32(mmSRBM_SOFT_RESET);
4707 
4708 			udelay(50);
4709 
4710 			tmp &= ~srbm_soft_reset;
4711 			WREG32(mmSRBM_SOFT_RESET, tmp);
4712 			tmp = RREG32(mmSRBM_SOFT_RESET);
4713 		}
4714 		/* Wait a little for things to settle down */
4715 		udelay(50);
4716 	}
4717 	return 0;
4718 }
4719 
4720 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4721 						 enum amdgpu_interrupt_state state)
4722 {
4723 	u32 cp_int_cntl;
4724 
4725 	switch (state) {
4726 	case AMDGPU_IRQ_STATE_DISABLE:
4727 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4728 		cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4729 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4730 		break;
4731 	case AMDGPU_IRQ_STATE_ENABLE:
4732 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4733 		cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4734 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4735 		break;
4736 	default:
4737 		break;
4738 	}
4739 }
4740 
4741 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4742 						     int me, int pipe,
4743 						     enum amdgpu_interrupt_state state)
4744 {
4745 	u32 mec_int_cntl, mec_int_cntl_reg;
4746 
4747 	/*
4748 	 * amdgpu controls only the first MEC. That's why this function only
4749 	 * handles the setting of interrupts for this specific MEC. All other
4750 	 * pipes' interrupts are set by amdkfd.
4751 	 */
4752 
4753 	if (me == 1) {
4754 		switch (pipe) {
4755 		case 0:
4756 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4757 			break;
4758 		case 1:
4759 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
4760 			break;
4761 		case 2:
4762 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
4763 			break;
4764 		case 3:
4765 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
4766 			break;
4767 		default:
4768 			DRM_DEBUG("invalid pipe %d\n", pipe);
4769 			return;
4770 		}
4771 	} else {
4772 		DRM_DEBUG("invalid me %d\n", me);
4773 		return;
4774 	}
4775 
4776 	switch (state) {
4777 	case AMDGPU_IRQ_STATE_DISABLE:
4778 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4779 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4780 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4781 		break;
4782 	case AMDGPU_IRQ_STATE_ENABLE:
4783 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4784 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4785 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4786 		break;
4787 	default:
4788 		break;
4789 	}
4790 }
4791 
4792 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4793 					     struct amdgpu_irq_src *src,
4794 					     unsigned type,
4795 					     enum amdgpu_interrupt_state state)
4796 {
4797 	u32 cp_int_cntl;
4798 
4799 	switch (state) {
4800 	case AMDGPU_IRQ_STATE_DISABLE:
4801 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4802 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4803 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4804 		break;
4805 	case AMDGPU_IRQ_STATE_ENABLE:
4806 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4807 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4808 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4809 		break;
4810 	default:
4811 		break;
4812 	}
4813 
4814 	return 0;
4815 }
4816 
4817 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4818 					      struct amdgpu_irq_src *src,
4819 					      unsigned type,
4820 					      enum amdgpu_interrupt_state state)
4821 {
4822 	u32 cp_int_cntl;
4823 
4824 	switch (state) {
4825 	case AMDGPU_IRQ_STATE_DISABLE:
4826 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4827 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4828 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4829 		break;
4830 	case AMDGPU_IRQ_STATE_ENABLE:
4831 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4832 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4833 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4834 		break;
4835 	default:
4836 		break;
4837 	}
4838 
4839 	return 0;
4840 }
4841 
4842 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4843 					    struct amdgpu_irq_src *src,
4844 					    unsigned type,
4845 					    enum amdgpu_interrupt_state state)
4846 {
4847 	switch (type) {
4848 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
4849 		gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
4850 		break;
4851 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4852 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4853 		break;
4854 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4855 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4856 		break;
4857 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4858 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4859 		break;
4860 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4861 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4862 		break;
4863 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4864 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4865 		break;
4866 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4867 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4868 		break;
4869 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4870 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4871 		break;
4872 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4873 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4874 		break;
4875 	default:
4876 		break;
4877 	}
4878 	return 0;
4879 }
4880 
4881 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
4882 			    struct amdgpu_irq_src *source,
4883 			    struct amdgpu_iv_entry *entry)
4884 {
4885 	u8 me_id, pipe_id;
4886 	struct amdgpu_ring *ring;
4887 	int i;
4888 
4889 	DRM_DEBUG("IH: CP EOP\n");
4890 	me_id = (entry->ring_id & 0x0c) >> 2;
4891 	pipe_id = (entry->ring_id & 0x03) >> 0;
4892 	switch (me_id) {
4893 	case 0:
4894 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4895 		break;
4896 	case 1:
4897 	case 2:
4898 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4899 			ring = &adev->gfx.compute_ring[i];
4900 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4901 				amdgpu_fence_process(ring);
4902 		}
4903 		break;
4904 	}
4905 	return 0;
4906 }
4907 
4908 static void gfx_v7_0_fault(struct amdgpu_device *adev,
4909 			   struct amdgpu_iv_entry *entry)
4910 {
4911 	struct amdgpu_ring *ring;
4912 	u8 me_id, pipe_id;
4913 	int i;
4914 
4915 	me_id = (entry->ring_id & 0x0c) >> 2;
4916 	pipe_id = (entry->ring_id & 0x03) >> 0;
4917 	switch (me_id) {
4918 	case 0:
4919 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
4920 		break;
4921 	case 1:
4922 	case 2:
4923 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4924 			ring = &adev->gfx.compute_ring[i];
4925 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4926 				drm_sched_fault(&ring->sched);
4927 		}
4928 		break;
4929 	}
4930 }
4931 
4932 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
4933 				 struct amdgpu_irq_src *source,
4934 				 struct amdgpu_iv_entry *entry)
4935 {
4936 	DRM_ERROR("Illegal register access in command stream\n");
4937 	gfx_v7_0_fault(adev, entry);
4938 	return 0;
4939 }
4940 
4941 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
4942 				  struct amdgpu_irq_src *source,
4943 				  struct amdgpu_iv_entry *entry)
4944 {
4945 	DRM_ERROR("Illegal instruction in command stream\n");
4946 	// XXX soft reset the gfx block only
4947 	gfx_v7_0_fault(adev, entry);
4948 	return 0;
4949 }
4950 
4951 static int gfx_v7_0_set_clockgating_state(void *handle,
4952 					  enum amd_clockgating_state state)
4953 {
4954 	bool gate = false;
4955 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4956 
4957 	if (state == AMD_CG_STATE_GATE)
4958 		gate = true;
4959 
4960 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
4961 	/* order matters! */
4962 	if (gate) {
4963 		gfx_v7_0_enable_mgcg(adev, true);
4964 		gfx_v7_0_enable_cgcg(adev, true);
4965 	} else {
4966 		gfx_v7_0_enable_cgcg(adev, false);
4967 		gfx_v7_0_enable_mgcg(adev, false);
4968 	}
4969 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
4970 
4971 	return 0;
4972 }
4973 
4974 static int gfx_v7_0_set_powergating_state(void *handle,
4975 					  enum amd_powergating_state state)
4976 {
4977 	bool gate = false;
4978 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4979 
4980 	if (state == AMD_PG_STATE_GATE)
4981 		gate = true;
4982 
4983 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4984 			      AMD_PG_SUPPORT_GFX_SMG |
4985 			      AMD_PG_SUPPORT_GFX_DMG |
4986 			      AMD_PG_SUPPORT_CP |
4987 			      AMD_PG_SUPPORT_GDS |
4988 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4989 		gfx_v7_0_update_gfx_pg(adev, gate);
4990 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4991 			gfx_v7_0_enable_cp_pg(adev, gate);
4992 			gfx_v7_0_enable_gds_pg(adev, gate);
4993 		}
4994 	}
4995 
4996 	return 0;
4997 }
4998 
4999 static void gfx_v7_0_emit_mem_sync(struct amdgpu_ring *ring)
5000 {
5001 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
5002 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
5003 			  PACKET3_TC_ACTION_ENA |
5004 			  PACKET3_SH_KCACHE_ACTION_ENA |
5005 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
5006 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
5007 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
5008 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
5009 }
5010 
5011 static void gfx_v7_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
5012 {
5013 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
5014 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
5015 			  PACKET3_TC_ACTION_ENA |
5016 			  PACKET3_SH_KCACHE_ACTION_ENA |
5017 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
5018 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
5019 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
5020 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
5021 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
5022 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
5023 }
5024 
5025 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
5026 	.name = "gfx_v7_0",
5027 	.early_init = gfx_v7_0_early_init,
5028 	.late_init = gfx_v7_0_late_init,
5029 	.sw_init = gfx_v7_0_sw_init,
5030 	.sw_fini = gfx_v7_0_sw_fini,
5031 	.hw_init = gfx_v7_0_hw_init,
5032 	.hw_fini = gfx_v7_0_hw_fini,
5033 	.suspend = gfx_v7_0_suspend,
5034 	.resume = gfx_v7_0_resume,
5035 	.is_idle = gfx_v7_0_is_idle,
5036 	.wait_for_idle = gfx_v7_0_wait_for_idle,
5037 	.soft_reset = gfx_v7_0_soft_reset,
5038 	.set_clockgating_state = gfx_v7_0_set_clockgating_state,
5039 	.set_powergating_state = gfx_v7_0_set_powergating_state,
5040 };
5041 
5042 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
5043 	.type = AMDGPU_RING_TYPE_GFX,
5044 	.align_mask = 0xff,
5045 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5046 	.support_64bit_ptrs = false,
5047 	.get_rptr = gfx_v7_0_ring_get_rptr,
5048 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
5049 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
5050 	.emit_frame_size =
5051 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5052 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5053 		5 + /* hdp invalidate */
5054 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
5055 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
5056 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
5057 		3 + 4 + /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
5058 		5, /* SURFACE_SYNC */
5059 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
5060 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
5061 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
5062 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5063 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5064 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5065 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5066 	.test_ring = gfx_v7_0_ring_test_ring,
5067 	.test_ib = gfx_v7_0_ring_test_ib,
5068 	.insert_nop = amdgpu_ring_insert_nop,
5069 	.pad_ib = amdgpu_ring_generic_pad_ib,
5070 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
5071 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5072 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
5073 	.emit_mem_sync = gfx_v7_0_emit_mem_sync,
5074 };
5075 
5076 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5077 	.type = AMDGPU_RING_TYPE_COMPUTE,
5078 	.align_mask = 0xff,
5079 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5080 	.support_64bit_ptrs = false,
5081 	.get_rptr = gfx_v7_0_ring_get_rptr,
5082 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5083 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5084 	.emit_frame_size =
5085 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5086 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5087 		5 + /* hdp invalidate */
5088 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5089 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
5090 		7 + 7 + 7 + /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5091 		7, /* gfx_v7_0_emit_mem_sync_compute */
5092 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
5093 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5094 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5095 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5096 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5097 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5098 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5099 	.test_ring = gfx_v7_0_ring_test_ring,
5100 	.test_ib = gfx_v7_0_ring_test_ib,
5101 	.insert_nop = amdgpu_ring_insert_nop,
5102 	.pad_ib = amdgpu_ring_generic_pad_ib,
5103 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5104 	.emit_mem_sync = gfx_v7_0_emit_mem_sync_compute,
5105 };
5106 
5107 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5108 {
5109 	int i;
5110 
5111 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5112 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5113 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5114 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5115 }
5116 
5117 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5118 	.set = gfx_v7_0_set_eop_interrupt_state,
5119 	.process = gfx_v7_0_eop_irq,
5120 };
5121 
5122 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5123 	.set = gfx_v7_0_set_priv_reg_fault_state,
5124 	.process = gfx_v7_0_priv_reg_irq,
5125 };
5126 
5127 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5128 	.set = gfx_v7_0_set_priv_inst_fault_state,
5129 	.process = gfx_v7_0_priv_inst_irq,
5130 };
5131 
5132 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5133 {
5134 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5135 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5136 
5137 	adev->gfx.priv_reg_irq.num_types = 1;
5138 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5139 
5140 	adev->gfx.priv_inst_irq.num_types = 1;
5141 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5142 }
5143 
5144 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5145 {
5146 	/* init asci gds info */
5147 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
5148 	adev->gds.gws_size = 64;
5149 	adev->gds.oa_size = 16;
5150 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
5151 }
5152 
5153 
5154 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5155 {
5156 	int i, j, k, counter, active_cu_number = 0;
5157 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5158 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5159 	unsigned disable_masks[4 * 2];
5160 	u32 ao_cu_num;
5161 
5162 	if (adev->flags & AMD_IS_APU)
5163 		ao_cu_num = 2;
5164 	else
5165 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
5166 
5167 	memset(cu_info, 0, sizeof(*cu_info));
5168 
5169 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5170 
5171 	mutex_lock(&adev->grbm_idx_mutex);
5172 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5173 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5174 			mask = 1;
5175 			ao_bitmap = 0;
5176 			counter = 0;
5177 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
5178 			if (i < 4 && j < 2)
5179 				gfx_v7_0_set_user_cu_inactive_bitmap(
5180 					adev, disable_masks[i * 2 + j]);
5181 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5182 			cu_info->bitmap[i][j] = bitmap;
5183 
5184 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
5185 				if (bitmap & mask) {
5186 					if (counter < ao_cu_num)
5187 						ao_bitmap |= mask;
5188 					counter ++;
5189 				}
5190 				mask <<= 1;
5191 			}
5192 			active_cu_number += counter;
5193 			if (i < 2 && j < 2)
5194 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5195 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5196 		}
5197 	}
5198 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5199 	mutex_unlock(&adev->grbm_idx_mutex);
5200 
5201 	cu_info->number = active_cu_number;
5202 	cu_info->ao_cu_mask = ao_cu_mask;
5203 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5204 	cu_info->max_waves_per_simd = 10;
5205 	cu_info->max_scratch_slots_per_cu = 32;
5206 	cu_info->wave_front_size = 64;
5207 	cu_info->lds_size = 64;
5208 }
5209 
5210 static const struct amdgpu_ip_block_version gfx_v7_0_ip_block =
5211 {
5212 	.type = AMD_IP_BLOCK_TYPE_GFX,
5213 	.major = 7,
5214 	.minor = 0,
5215 	.rev = 0,
5216 	.funcs = &gfx_v7_0_ip_funcs,
5217 };
5218 
5219 const struct amdgpu_ip_block_version gfx_v7_1_ip_block =
5220 {
5221 	.type = AMD_IP_BLOCK_TYPE_GFX,
5222 	.major = 7,
5223 	.minor = 1,
5224 	.rev = 0,
5225 	.funcs = &gfx_v7_0_ip_funcs,
5226 };
5227 
5228 const struct amdgpu_ip_block_version gfx_v7_2_ip_block =
5229 {
5230 	.type = AMD_IP_BLOCK_TYPE_GFX,
5231 	.major = 7,
5232 	.minor = 2,
5233 	.rev = 0,
5234 	.funcs = &gfx_v7_0_ip_funcs,
5235 };
5236 
5237 const struct amdgpu_ip_block_version gfx_v7_3_ip_block =
5238 {
5239 	.type = AMD_IP_BLOCK_TYPE_GFX,
5240 	.major = 7,
5241 	.minor = 3,
5242 	.rev = 0,
5243 	.funcs = &gfx_v7_0_ip_funcs,
5244 };
5245