xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision 1075b321)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <linux/module.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_ih.h"
29 #include "amdgpu_gfx.h"
30 #include "cikd.h"
31 #include "cik.h"
32 #include "cik_structs.h"
33 #include "atom.h"
34 #include "amdgpu_ucode.h"
35 #include "clearstate_ci.h"
36 
37 #include "dce/dce_8_0_d.h"
38 #include "dce/dce_8_0_sh_mask.h"
39 
40 #include "bif/bif_4_1_d.h"
41 #include "bif/bif_4_1_sh_mask.h"
42 
43 #include "gca/gfx_7_0_d.h"
44 #include "gca/gfx_7_2_enum.h"
45 #include "gca/gfx_7_2_sh_mask.h"
46 
47 #include "gmc/gmc_7_0_d.h"
48 #include "gmc/gmc_7_0_sh_mask.h"
49 
50 #include "oss/oss_2_0_d.h"
51 #include "oss/oss_2_0_sh_mask.h"
52 
53 #define NUM_SIMD_PER_CU 0x4 /* missing from the gfx_7 IP headers */
54 
55 #define GFX7_NUM_GFX_RINGS     1
56 #define GFX7_MEC_HPD_SIZE      2048
57 
58 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev);
59 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev);
60 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev);
61 
62 MODULE_FIRMWARE("amdgpu/bonaire_pfp.bin");
63 MODULE_FIRMWARE("amdgpu/bonaire_me.bin");
64 MODULE_FIRMWARE("amdgpu/bonaire_ce.bin");
65 MODULE_FIRMWARE("amdgpu/bonaire_rlc.bin");
66 MODULE_FIRMWARE("amdgpu/bonaire_mec.bin");
67 
68 MODULE_FIRMWARE("amdgpu/hawaii_pfp.bin");
69 MODULE_FIRMWARE("amdgpu/hawaii_me.bin");
70 MODULE_FIRMWARE("amdgpu/hawaii_ce.bin");
71 MODULE_FIRMWARE("amdgpu/hawaii_rlc.bin");
72 MODULE_FIRMWARE("amdgpu/hawaii_mec.bin");
73 
74 MODULE_FIRMWARE("amdgpu/kaveri_pfp.bin");
75 MODULE_FIRMWARE("amdgpu/kaveri_me.bin");
76 MODULE_FIRMWARE("amdgpu/kaveri_ce.bin");
77 MODULE_FIRMWARE("amdgpu/kaveri_rlc.bin");
78 MODULE_FIRMWARE("amdgpu/kaveri_mec.bin");
79 MODULE_FIRMWARE("amdgpu/kaveri_mec2.bin");
80 
81 MODULE_FIRMWARE("amdgpu/kabini_pfp.bin");
82 MODULE_FIRMWARE("amdgpu/kabini_me.bin");
83 MODULE_FIRMWARE("amdgpu/kabini_ce.bin");
84 MODULE_FIRMWARE("amdgpu/kabini_rlc.bin");
85 MODULE_FIRMWARE("amdgpu/kabini_mec.bin");
86 
87 MODULE_FIRMWARE("amdgpu/mullins_pfp.bin");
88 MODULE_FIRMWARE("amdgpu/mullins_me.bin");
89 MODULE_FIRMWARE("amdgpu/mullins_ce.bin");
90 MODULE_FIRMWARE("amdgpu/mullins_rlc.bin");
91 MODULE_FIRMWARE("amdgpu/mullins_mec.bin");
92 
93 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
94 {
95 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
96 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
97 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
98 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
99 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
100 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
101 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
102 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
103 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
104 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
105 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
106 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
107 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
108 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
109 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
110 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
111 };
112 
113 static const u32 spectre_rlc_save_restore_register_list[] =
114 {
115 	(0x0e00 << 16) | (0xc12c >> 2),
116 	0x00000000,
117 	(0x0e00 << 16) | (0xc140 >> 2),
118 	0x00000000,
119 	(0x0e00 << 16) | (0xc150 >> 2),
120 	0x00000000,
121 	(0x0e00 << 16) | (0xc15c >> 2),
122 	0x00000000,
123 	(0x0e00 << 16) | (0xc168 >> 2),
124 	0x00000000,
125 	(0x0e00 << 16) | (0xc170 >> 2),
126 	0x00000000,
127 	(0x0e00 << 16) | (0xc178 >> 2),
128 	0x00000000,
129 	(0x0e00 << 16) | (0xc204 >> 2),
130 	0x00000000,
131 	(0x0e00 << 16) | (0xc2b4 >> 2),
132 	0x00000000,
133 	(0x0e00 << 16) | (0xc2b8 >> 2),
134 	0x00000000,
135 	(0x0e00 << 16) | (0xc2bc >> 2),
136 	0x00000000,
137 	(0x0e00 << 16) | (0xc2c0 >> 2),
138 	0x00000000,
139 	(0x0e00 << 16) | (0x8228 >> 2),
140 	0x00000000,
141 	(0x0e00 << 16) | (0x829c >> 2),
142 	0x00000000,
143 	(0x0e00 << 16) | (0x869c >> 2),
144 	0x00000000,
145 	(0x0600 << 16) | (0x98f4 >> 2),
146 	0x00000000,
147 	(0x0e00 << 16) | (0x98f8 >> 2),
148 	0x00000000,
149 	(0x0e00 << 16) | (0x9900 >> 2),
150 	0x00000000,
151 	(0x0e00 << 16) | (0xc260 >> 2),
152 	0x00000000,
153 	(0x0e00 << 16) | (0x90e8 >> 2),
154 	0x00000000,
155 	(0x0e00 << 16) | (0x3c000 >> 2),
156 	0x00000000,
157 	(0x0e00 << 16) | (0x3c00c >> 2),
158 	0x00000000,
159 	(0x0e00 << 16) | (0x8c1c >> 2),
160 	0x00000000,
161 	(0x0e00 << 16) | (0x9700 >> 2),
162 	0x00000000,
163 	(0x0e00 << 16) | (0xcd20 >> 2),
164 	0x00000000,
165 	(0x4e00 << 16) | (0xcd20 >> 2),
166 	0x00000000,
167 	(0x5e00 << 16) | (0xcd20 >> 2),
168 	0x00000000,
169 	(0x6e00 << 16) | (0xcd20 >> 2),
170 	0x00000000,
171 	(0x7e00 << 16) | (0xcd20 >> 2),
172 	0x00000000,
173 	(0x8e00 << 16) | (0xcd20 >> 2),
174 	0x00000000,
175 	(0x9e00 << 16) | (0xcd20 >> 2),
176 	0x00000000,
177 	(0xae00 << 16) | (0xcd20 >> 2),
178 	0x00000000,
179 	(0xbe00 << 16) | (0xcd20 >> 2),
180 	0x00000000,
181 	(0x0e00 << 16) | (0x89bc >> 2),
182 	0x00000000,
183 	(0x0e00 << 16) | (0x8900 >> 2),
184 	0x00000000,
185 	0x3,
186 	(0x0e00 << 16) | (0xc130 >> 2),
187 	0x00000000,
188 	(0x0e00 << 16) | (0xc134 >> 2),
189 	0x00000000,
190 	(0x0e00 << 16) | (0xc1fc >> 2),
191 	0x00000000,
192 	(0x0e00 << 16) | (0xc208 >> 2),
193 	0x00000000,
194 	(0x0e00 << 16) | (0xc264 >> 2),
195 	0x00000000,
196 	(0x0e00 << 16) | (0xc268 >> 2),
197 	0x00000000,
198 	(0x0e00 << 16) | (0xc26c >> 2),
199 	0x00000000,
200 	(0x0e00 << 16) | (0xc270 >> 2),
201 	0x00000000,
202 	(0x0e00 << 16) | (0xc274 >> 2),
203 	0x00000000,
204 	(0x0e00 << 16) | (0xc278 >> 2),
205 	0x00000000,
206 	(0x0e00 << 16) | (0xc27c >> 2),
207 	0x00000000,
208 	(0x0e00 << 16) | (0xc280 >> 2),
209 	0x00000000,
210 	(0x0e00 << 16) | (0xc284 >> 2),
211 	0x00000000,
212 	(0x0e00 << 16) | (0xc288 >> 2),
213 	0x00000000,
214 	(0x0e00 << 16) | (0xc28c >> 2),
215 	0x00000000,
216 	(0x0e00 << 16) | (0xc290 >> 2),
217 	0x00000000,
218 	(0x0e00 << 16) | (0xc294 >> 2),
219 	0x00000000,
220 	(0x0e00 << 16) | (0xc298 >> 2),
221 	0x00000000,
222 	(0x0e00 << 16) | (0xc29c >> 2),
223 	0x00000000,
224 	(0x0e00 << 16) | (0xc2a0 >> 2),
225 	0x00000000,
226 	(0x0e00 << 16) | (0xc2a4 >> 2),
227 	0x00000000,
228 	(0x0e00 << 16) | (0xc2a8 >> 2),
229 	0x00000000,
230 	(0x0e00 << 16) | (0xc2ac  >> 2),
231 	0x00000000,
232 	(0x0e00 << 16) | (0xc2b0 >> 2),
233 	0x00000000,
234 	(0x0e00 << 16) | (0x301d0 >> 2),
235 	0x00000000,
236 	(0x0e00 << 16) | (0x30238 >> 2),
237 	0x00000000,
238 	(0x0e00 << 16) | (0x30250 >> 2),
239 	0x00000000,
240 	(0x0e00 << 16) | (0x30254 >> 2),
241 	0x00000000,
242 	(0x0e00 << 16) | (0x30258 >> 2),
243 	0x00000000,
244 	(0x0e00 << 16) | (0x3025c >> 2),
245 	0x00000000,
246 	(0x4e00 << 16) | (0xc900 >> 2),
247 	0x00000000,
248 	(0x5e00 << 16) | (0xc900 >> 2),
249 	0x00000000,
250 	(0x6e00 << 16) | (0xc900 >> 2),
251 	0x00000000,
252 	(0x7e00 << 16) | (0xc900 >> 2),
253 	0x00000000,
254 	(0x8e00 << 16) | (0xc900 >> 2),
255 	0x00000000,
256 	(0x9e00 << 16) | (0xc900 >> 2),
257 	0x00000000,
258 	(0xae00 << 16) | (0xc900 >> 2),
259 	0x00000000,
260 	(0xbe00 << 16) | (0xc900 >> 2),
261 	0x00000000,
262 	(0x4e00 << 16) | (0xc904 >> 2),
263 	0x00000000,
264 	(0x5e00 << 16) | (0xc904 >> 2),
265 	0x00000000,
266 	(0x6e00 << 16) | (0xc904 >> 2),
267 	0x00000000,
268 	(0x7e00 << 16) | (0xc904 >> 2),
269 	0x00000000,
270 	(0x8e00 << 16) | (0xc904 >> 2),
271 	0x00000000,
272 	(0x9e00 << 16) | (0xc904 >> 2),
273 	0x00000000,
274 	(0xae00 << 16) | (0xc904 >> 2),
275 	0x00000000,
276 	(0xbe00 << 16) | (0xc904 >> 2),
277 	0x00000000,
278 	(0x4e00 << 16) | (0xc908 >> 2),
279 	0x00000000,
280 	(0x5e00 << 16) | (0xc908 >> 2),
281 	0x00000000,
282 	(0x6e00 << 16) | (0xc908 >> 2),
283 	0x00000000,
284 	(0x7e00 << 16) | (0xc908 >> 2),
285 	0x00000000,
286 	(0x8e00 << 16) | (0xc908 >> 2),
287 	0x00000000,
288 	(0x9e00 << 16) | (0xc908 >> 2),
289 	0x00000000,
290 	(0xae00 << 16) | (0xc908 >> 2),
291 	0x00000000,
292 	(0xbe00 << 16) | (0xc908 >> 2),
293 	0x00000000,
294 	(0x4e00 << 16) | (0xc90c >> 2),
295 	0x00000000,
296 	(0x5e00 << 16) | (0xc90c >> 2),
297 	0x00000000,
298 	(0x6e00 << 16) | (0xc90c >> 2),
299 	0x00000000,
300 	(0x7e00 << 16) | (0xc90c >> 2),
301 	0x00000000,
302 	(0x8e00 << 16) | (0xc90c >> 2),
303 	0x00000000,
304 	(0x9e00 << 16) | (0xc90c >> 2),
305 	0x00000000,
306 	(0xae00 << 16) | (0xc90c >> 2),
307 	0x00000000,
308 	(0xbe00 << 16) | (0xc90c >> 2),
309 	0x00000000,
310 	(0x4e00 << 16) | (0xc910 >> 2),
311 	0x00000000,
312 	(0x5e00 << 16) | (0xc910 >> 2),
313 	0x00000000,
314 	(0x6e00 << 16) | (0xc910 >> 2),
315 	0x00000000,
316 	(0x7e00 << 16) | (0xc910 >> 2),
317 	0x00000000,
318 	(0x8e00 << 16) | (0xc910 >> 2),
319 	0x00000000,
320 	(0x9e00 << 16) | (0xc910 >> 2),
321 	0x00000000,
322 	(0xae00 << 16) | (0xc910 >> 2),
323 	0x00000000,
324 	(0xbe00 << 16) | (0xc910 >> 2),
325 	0x00000000,
326 	(0x0e00 << 16) | (0xc99c >> 2),
327 	0x00000000,
328 	(0x0e00 << 16) | (0x9834 >> 2),
329 	0x00000000,
330 	(0x0000 << 16) | (0x30f00 >> 2),
331 	0x00000000,
332 	(0x0001 << 16) | (0x30f00 >> 2),
333 	0x00000000,
334 	(0x0000 << 16) | (0x30f04 >> 2),
335 	0x00000000,
336 	(0x0001 << 16) | (0x30f04 >> 2),
337 	0x00000000,
338 	(0x0000 << 16) | (0x30f08 >> 2),
339 	0x00000000,
340 	(0x0001 << 16) | (0x30f08 >> 2),
341 	0x00000000,
342 	(0x0000 << 16) | (0x30f0c >> 2),
343 	0x00000000,
344 	(0x0001 << 16) | (0x30f0c >> 2),
345 	0x00000000,
346 	(0x0600 << 16) | (0x9b7c >> 2),
347 	0x00000000,
348 	(0x0e00 << 16) | (0x8a14 >> 2),
349 	0x00000000,
350 	(0x0e00 << 16) | (0x8a18 >> 2),
351 	0x00000000,
352 	(0x0600 << 16) | (0x30a00 >> 2),
353 	0x00000000,
354 	(0x0e00 << 16) | (0x8bf0 >> 2),
355 	0x00000000,
356 	(0x0e00 << 16) | (0x8bcc >> 2),
357 	0x00000000,
358 	(0x0e00 << 16) | (0x8b24 >> 2),
359 	0x00000000,
360 	(0x0e00 << 16) | (0x30a04 >> 2),
361 	0x00000000,
362 	(0x0600 << 16) | (0x30a10 >> 2),
363 	0x00000000,
364 	(0x0600 << 16) | (0x30a14 >> 2),
365 	0x00000000,
366 	(0x0600 << 16) | (0x30a18 >> 2),
367 	0x00000000,
368 	(0x0600 << 16) | (0x30a2c >> 2),
369 	0x00000000,
370 	(0x0e00 << 16) | (0xc700 >> 2),
371 	0x00000000,
372 	(0x0e00 << 16) | (0xc704 >> 2),
373 	0x00000000,
374 	(0x0e00 << 16) | (0xc708 >> 2),
375 	0x00000000,
376 	(0x0e00 << 16) | (0xc768 >> 2),
377 	0x00000000,
378 	(0x0400 << 16) | (0xc770 >> 2),
379 	0x00000000,
380 	(0x0400 << 16) | (0xc774 >> 2),
381 	0x00000000,
382 	(0x0400 << 16) | (0xc778 >> 2),
383 	0x00000000,
384 	(0x0400 << 16) | (0xc77c >> 2),
385 	0x00000000,
386 	(0x0400 << 16) | (0xc780 >> 2),
387 	0x00000000,
388 	(0x0400 << 16) | (0xc784 >> 2),
389 	0x00000000,
390 	(0x0400 << 16) | (0xc788 >> 2),
391 	0x00000000,
392 	(0x0400 << 16) | (0xc78c >> 2),
393 	0x00000000,
394 	(0x0400 << 16) | (0xc798 >> 2),
395 	0x00000000,
396 	(0x0400 << 16) | (0xc79c >> 2),
397 	0x00000000,
398 	(0x0400 << 16) | (0xc7a0 >> 2),
399 	0x00000000,
400 	(0x0400 << 16) | (0xc7a4 >> 2),
401 	0x00000000,
402 	(0x0400 << 16) | (0xc7a8 >> 2),
403 	0x00000000,
404 	(0x0400 << 16) | (0xc7ac >> 2),
405 	0x00000000,
406 	(0x0400 << 16) | (0xc7b0 >> 2),
407 	0x00000000,
408 	(0x0400 << 16) | (0xc7b4 >> 2),
409 	0x00000000,
410 	(0x0e00 << 16) | (0x9100 >> 2),
411 	0x00000000,
412 	(0x0e00 << 16) | (0x3c010 >> 2),
413 	0x00000000,
414 	(0x0e00 << 16) | (0x92a8 >> 2),
415 	0x00000000,
416 	(0x0e00 << 16) | (0x92ac >> 2),
417 	0x00000000,
418 	(0x0e00 << 16) | (0x92b4 >> 2),
419 	0x00000000,
420 	(0x0e00 << 16) | (0x92b8 >> 2),
421 	0x00000000,
422 	(0x0e00 << 16) | (0x92bc >> 2),
423 	0x00000000,
424 	(0x0e00 << 16) | (0x92c0 >> 2),
425 	0x00000000,
426 	(0x0e00 << 16) | (0x92c4 >> 2),
427 	0x00000000,
428 	(0x0e00 << 16) | (0x92c8 >> 2),
429 	0x00000000,
430 	(0x0e00 << 16) | (0x92cc >> 2),
431 	0x00000000,
432 	(0x0e00 << 16) | (0x92d0 >> 2),
433 	0x00000000,
434 	(0x0e00 << 16) | (0x8c00 >> 2),
435 	0x00000000,
436 	(0x0e00 << 16) | (0x8c04 >> 2),
437 	0x00000000,
438 	(0x0e00 << 16) | (0x8c20 >> 2),
439 	0x00000000,
440 	(0x0e00 << 16) | (0x8c38 >> 2),
441 	0x00000000,
442 	(0x0e00 << 16) | (0x8c3c >> 2),
443 	0x00000000,
444 	(0x0e00 << 16) | (0xae00 >> 2),
445 	0x00000000,
446 	(0x0e00 << 16) | (0x9604 >> 2),
447 	0x00000000,
448 	(0x0e00 << 16) | (0xac08 >> 2),
449 	0x00000000,
450 	(0x0e00 << 16) | (0xac0c >> 2),
451 	0x00000000,
452 	(0x0e00 << 16) | (0xac10 >> 2),
453 	0x00000000,
454 	(0x0e00 << 16) | (0xac14 >> 2),
455 	0x00000000,
456 	(0x0e00 << 16) | (0xac58 >> 2),
457 	0x00000000,
458 	(0x0e00 << 16) | (0xac68 >> 2),
459 	0x00000000,
460 	(0x0e00 << 16) | (0xac6c >> 2),
461 	0x00000000,
462 	(0x0e00 << 16) | (0xac70 >> 2),
463 	0x00000000,
464 	(0x0e00 << 16) | (0xac74 >> 2),
465 	0x00000000,
466 	(0x0e00 << 16) | (0xac78 >> 2),
467 	0x00000000,
468 	(0x0e00 << 16) | (0xac7c >> 2),
469 	0x00000000,
470 	(0x0e00 << 16) | (0xac80 >> 2),
471 	0x00000000,
472 	(0x0e00 << 16) | (0xac84 >> 2),
473 	0x00000000,
474 	(0x0e00 << 16) | (0xac88 >> 2),
475 	0x00000000,
476 	(0x0e00 << 16) | (0xac8c >> 2),
477 	0x00000000,
478 	(0x0e00 << 16) | (0x970c >> 2),
479 	0x00000000,
480 	(0x0e00 << 16) | (0x9714 >> 2),
481 	0x00000000,
482 	(0x0e00 << 16) | (0x9718 >> 2),
483 	0x00000000,
484 	(0x0e00 << 16) | (0x971c >> 2),
485 	0x00000000,
486 	(0x0e00 << 16) | (0x31068 >> 2),
487 	0x00000000,
488 	(0x4e00 << 16) | (0x31068 >> 2),
489 	0x00000000,
490 	(0x5e00 << 16) | (0x31068 >> 2),
491 	0x00000000,
492 	(0x6e00 << 16) | (0x31068 >> 2),
493 	0x00000000,
494 	(0x7e00 << 16) | (0x31068 >> 2),
495 	0x00000000,
496 	(0x8e00 << 16) | (0x31068 >> 2),
497 	0x00000000,
498 	(0x9e00 << 16) | (0x31068 >> 2),
499 	0x00000000,
500 	(0xae00 << 16) | (0x31068 >> 2),
501 	0x00000000,
502 	(0xbe00 << 16) | (0x31068 >> 2),
503 	0x00000000,
504 	(0x0e00 << 16) | (0xcd10 >> 2),
505 	0x00000000,
506 	(0x0e00 << 16) | (0xcd14 >> 2),
507 	0x00000000,
508 	(0x0e00 << 16) | (0x88b0 >> 2),
509 	0x00000000,
510 	(0x0e00 << 16) | (0x88b4 >> 2),
511 	0x00000000,
512 	(0x0e00 << 16) | (0x88b8 >> 2),
513 	0x00000000,
514 	(0x0e00 << 16) | (0x88bc >> 2),
515 	0x00000000,
516 	(0x0400 << 16) | (0x89c0 >> 2),
517 	0x00000000,
518 	(0x0e00 << 16) | (0x88c4 >> 2),
519 	0x00000000,
520 	(0x0e00 << 16) | (0x88c8 >> 2),
521 	0x00000000,
522 	(0x0e00 << 16) | (0x88d0 >> 2),
523 	0x00000000,
524 	(0x0e00 << 16) | (0x88d4 >> 2),
525 	0x00000000,
526 	(0x0e00 << 16) | (0x88d8 >> 2),
527 	0x00000000,
528 	(0x0e00 << 16) | (0x8980 >> 2),
529 	0x00000000,
530 	(0x0e00 << 16) | (0x30938 >> 2),
531 	0x00000000,
532 	(0x0e00 << 16) | (0x3093c >> 2),
533 	0x00000000,
534 	(0x0e00 << 16) | (0x30940 >> 2),
535 	0x00000000,
536 	(0x0e00 << 16) | (0x89a0 >> 2),
537 	0x00000000,
538 	(0x0e00 << 16) | (0x30900 >> 2),
539 	0x00000000,
540 	(0x0e00 << 16) | (0x30904 >> 2),
541 	0x00000000,
542 	(0x0e00 << 16) | (0x89b4 >> 2),
543 	0x00000000,
544 	(0x0e00 << 16) | (0x3c210 >> 2),
545 	0x00000000,
546 	(0x0e00 << 16) | (0x3c214 >> 2),
547 	0x00000000,
548 	(0x0e00 << 16) | (0x3c218 >> 2),
549 	0x00000000,
550 	(0x0e00 << 16) | (0x8904 >> 2),
551 	0x00000000,
552 	0x5,
553 	(0x0e00 << 16) | (0x8c28 >> 2),
554 	(0x0e00 << 16) | (0x8c2c >> 2),
555 	(0x0e00 << 16) | (0x8c30 >> 2),
556 	(0x0e00 << 16) | (0x8c34 >> 2),
557 	(0x0e00 << 16) | (0x9600 >> 2),
558 };
559 
560 static const u32 kalindi_rlc_save_restore_register_list[] =
561 {
562 	(0x0e00 << 16) | (0xc12c >> 2),
563 	0x00000000,
564 	(0x0e00 << 16) | (0xc140 >> 2),
565 	0x00000000,
566 	(0x0e00 << 16) | (0xc150 >> 2),
567 	0x00000000,
568 	(0x0e00 << 16) | (0xc15c >> 2),
569 	0x00000000,
570 	(0x0e00 << 16) | (0xc168 >> 2),
571 	0x00000000,
572 	(0x0e00 << 16) | (0xc170 >> 2),
573 	0x00000000,
574 	(0x0e00 << 16) | (0xc204 >> 2),
575 	0x00000000,
576 	(0x0e00 << 16) | (0xc2b4 >> 2),
577 	0x00000000,
578 	(0x0e00 << 16) | (0xc2b8 >> 2),
579 	0x00000000,
580 	(0x0e00 << 16) | (0xc2bc >> 2),
581 	0x00000000,
582 	(0x0e00 << 16) | (0xc2c0 >> 2),
583 	0x00000000,
584 	(0x0e00 << 16) | (0x8228 >> 2),
585 	0x00000000,
586 	(0x0e00 << 16) | (0x829c >> 2),
587 	0x00000000,
588 	(0x0e00 << 16) | (0x869c >> 2),
589 	0x00000000,
590 	(0x0600 << 16) | (0x98f4 >> 2),
591 	0x00000000,
592 	(0x0e00 << 16) | (0x98f8 >> 2),
593 	0x00000000,
594 	(0x0e00 << 16) | (0x9900 >> 2),
595 	0x00000000,
596 	(0x0e00 << 16) | (0xc260 >> 2),
597 	0x00000000,
598 	(0x0e00 << 16) | (0x90e8 >> 2),
599 	0x00000000,
600 	(0x0e00 << 16) | (0x3c000 >> 2),
601 	0x00000000,
602 	(0x0e00 << 16) | (0x3c00c >> 2),
603 	0x00000000,
604 	(0x0e00 << 16) | (0x8c1c >> 2),
605 	0x00000000,
606 	(0x0e00 << 16) | (0x9700 >> 2),
607 	0x00000000,
608 	(0x0e00 << 16) | (0xcd20 >> 2),
609 	0x00000000,
610 	(0x4e00 << 16) | (0xcd20 >> 2),
611 	0x00000000,
612 	(0x5e00 << 16) | (0xcd20 >> 2),
613 	0x00000000,
614 	(0x6e00 << 16) | (0xcd20 >> 2),
615 	0x00000000,
616 	(0x7e00 << 16) | (0xcd20 >> 2),
617 	0x00000000,
618 	(0x0e00 << 16) | (0x89bc >> 2),
619 	0x00000000,
620 	(0x0e00 << 16) | (0x8900 >> 2),
621 	0x00000000,
622 	0x3,
623 	(0x0e00 << 16) | (0xc130 >> 2),
624 	0x00000000,
625 	(0x0e00 << 16) | (0xc134 >> 2),
626 	0x00000000,
627 	(0x0e00 << 16) | (0xc1fc >> 2),
628 	0x00000000,
629 	(0x0e00 << 16) | (0xc208 >> 2),
630 	0x00000000,
631 	(0x0e00 << 16) | (0xc264 >> 2),
632 	0x00000000,
633 	(0x0e00 << 16) | (0xc268 >> 2),
634 	0x00000000,
635 	(0x0e00 << 16) | (0xc26c >> 2),
636 	0x00000000,
637 	(0x0e00 << 16) | (0xc270 >> 2),
638 	0x00000000,
639 	(0x0e00 << 16) | (0xc274 >> 2),
640 	0x00000000,
641 	(0x0e00 << 16) | (0xc28c >> 2),
642 	0x00000000,
643 	(0x0e00 << 16) | (0xc290 >> 2),
644 	0x00000000,
645 	(0x0e00 << 16) | (0xc294 >> 2),
646 	0x00000000,
647 	(0x0e00 << 16) | (0xc298 >> 2),
648 	0x00000000,
649 	(0x0e00 << 16) | (0xc2a0 >> 2),
650 	0x00000000,
651 	(0x0e00 << 16) | (0xc2a4 >> 2),
652 	0x00000000,
653 	(0x0e00 << 16) | (0xc2a8 >> 2),
654 	0x00000000,
655 	(0x0e00 << 16) | (0xc2ac >> 2),
656 	0x00000000,
657 	(0x0e00 << 16) | (0x301d0 >> 2),
658 	0x00000000,
659 	(0x0e00 << 16) | (0x30238 >> 2),
660 	0x00000000,
661 	(0x0e00 << 16) | (0x30250 >> 2),
662 	0x00000000,
663 	(0x0e00 << 16) | (0x30254 >> 2),
664 	0x00000000,
665 	(0x0e00 << 16) | (0x30258 >> 2),
666 	0x00000000,
667 	(0x0e00 << 16) | (0x3025c >> 2),
668 	0x00000000,
669 	(0x4e00 << 16) | (0xc900 >> 2),
670 	0x00000000,
671 	(0x5e00 << 16) | (0xc900 >> 2),
672 	0x00000000,
673 	(0x6e00 << 16) | (0xc900 >> 2),
674 	0x00000000,
675 	(0x7e00 << 16) | (0xc900 >> 2),
676 	0x00000000,
677 	(0x4e00 << 16) | (0xc904 >> 2),
678 	0x00000000,
679 	(0x5e00 << 16) | (0xc904 >> 2),
680 	0x00000000,
681 	(0x6e00 << 16) | (0xc904 >> 2),
682 	0x00000000,
683 	(0x7e00 << 16) | (0xc904 >> 2),
684 	0x00000000,
685 	(0x4e00 << 16) | (0xc908 >> 2),
686 	0x00000000,
687 	(0x5e00 << 16) | (0xc908 >> 2),
688 	0x00000000,
689 	(0x6e00 << 16) | (0xc908 >> 2),
690 	0x00000000,
691 	(0x7e00 << 16) | (0xc908 >> 2),
692 	0x00000000,
693 	(0x4e00 << 16) | (0xc90c >> 2),
694 	0x00000000,
695 	(0x5e00 << 16) | (0xc90c >> 2),
696 	0x00000000,
697 	(0x6e00 << 16) | (0xc90c >> 2),
698 	0x00000000,
699 	(0x7e00 << 16) | (0xc90c >> 2),
700 	0x00000000,
701 	(0x4e00 << 16) | (0xc910 >> 2),
702 	0x00000000,
703 	(0x5e00 << 16) | (0xc910 >> 2),
704 	0x00000000,
705 	(0x6e00 << 16) | (0xc910 >> 2),
706 	0x00000000,
707 	(0x7e00 << 16) | (0xc910 >> 2),
708 	0x00000000,
709 	(0x0e00 << 16) | (0xc99c >> 2),
710 	0x00000000,
711 	(0x0e00 << 16) | (0x9834 >> 2),
712 	0x00000000,
713 	(0x0000 << 16) | (0x30f00 >> 2),
714 	0x00000000,
715 	(0x0000 << 16) | (0x30f04 >> 2),
716 	0x00000000,
717 	(0x0000 << 16) | (0x30f08 >> 2),
718 	0x00000000,
719 	(0x0000 << 16) | (0x30f0c >> 2),
720 	0x00000000,
721 	(0x0600 << 16) | (0x9b7c >> 2),
722 	0x00000000,
723 	(0x0e00 << 16) | (0x8a14 >> 2),
724 	0x00000000,
725 	(0x0e00 << 16) | (0x8a18 >> 2),
726 	0x00000000,
727 	(0x0600 << 16) | (0x30a00 >> 2),
728 	0x00000000,
729 	(0x0e00 << 16) | (0x8bf0 >> 2),
730 	0x00000000,
731 	(0x0e00 << 16) | (0x8bcc >> 2),
732 	0x00000000,
733 	(0x0e00 << 16) | (0x8b24 >> 2),
734 	0x00000000,
735 	(0x0e00 << 16) | (0x30a04 >> 2),
736 	0x00000000,
737 	(0x0600 << 16) | (0x30a10 >> 2),
738 	0x00000000,
739 	(0x0600 << 16) | (0x30a14 >> 2),
740 	0x00000000,
741 	(0x0600 << 16) | (0x30a18 >> 2),
742 	0x00000000,
743 	(0x0600 << 16) | (0x30a2c >> 2),
744 	0x00000000,
745 	(0x0e00 << 16) | (0xc700 >> 2),
746 	0x00000000,
747 	(0x0e00 << 16) | (0xc704 >> 2),
748 	0x00000000,
749 	(0x0e00 << 16) | (0xc708 >> 2),
750 	0x00000000,
751 	(0x0e00 << 16) | (0xc768 >> 2),
752 	0x00000000,
753 	(0x0400 << 16) | (0xc770 >> 2),
754 	0x00000000,
755 	(0x0400 << 16) | (0xc774 >> 2),
756 	0x00000000,
757 	(0x0400 << 16) | (0xc798 >> 2),
758 	0x00000000,
759 	(0x0400 << 16) | (0xc79c >> 2),
760 	0x00000000,
761 	(0x0e00 << 16) | (0x9100 >> 2),
762 	0x00000000,
763 	(0x0e00 << 16) | (0x3c010 >> 2),
764 	0x00000000,
765 	(0x0e00 << 16) | (0x8c00 >> 2),
766 	0x00000000,
767 	(0x0e00 << 16) | (0x8c04 >> 2),
768 	0x00000000,
769 	(0x0e00 << 16) | (0x8c20 >> 2),
770 	0x00000000,
771 	(0x0e00 << 16) | (0x8c38 >> 2),
772 	0x00000000,
773 	(0x0e00 << 16) | (0x8c3c >> 2),
774 	0x00000000,
775 	(0x0e00 << 16) | (0xae00 >> 2),
776 	0x00000000,
777 	(0x0e00 << 16) | (0x9604 >> 2),
778 	0x00000000,
779 	(0x0e00 << 16) | (0xac08 >> 2),
780 	0x00000000,
781 	(0x0e00 << 16) | (0xac0c >> 2),
782 	0x00000000,
783 	(0x0e00 << 16) | (0xac10 >> 2),
784 	0x00000000,
785 	(0x0e00 << 16) | (0xac14 >> 2),
786 	0x00000000,
787 	(0x0e00 << 16) | (0xac58 >> 2),
788 	0x00000000,
789 	(0x0e00 << 16) | (0xac68 >> 2),
790 	0x00000000,
791 	(0x0e00 << 16) | (0xac6c >> 2),
792 	0x00000000,
793 	(0x0e00 << 16) | (0xac70 >> 2),
794 	0x00000000,
795 	(0x0e00 << 16) | (0xac74 >> 2),
796 	0x00000000,
797 	(0x0e00 << 16) | (0xac78 >> 2),
798 	0x00000000,
799 	(0x0e00 << 16) | (0xac7c >> 2),
800 	0x00000000,
801 	(0x0e00 << 16) | (0xac80 >> 2),
802 	0x00000000,
803 	(0x0e00 << 16) | (0xac84 >> 2),
804 	0x00000000,
805 	(0x0e00 << 16) | (0xac88 >> 2),
806 	0x00000000,
807 	(0x0e00 << 16) | (0xac8c >> 2),
808 	0x00000000,
809 	(0x0e00 << 16) | (0x970c >> 2),
810 	0x00000000,
811 	(0x0e00 << 16) | (0x9714 >> 2),
812 	0x00000000,
813 	(0x0e00 << 16) | (0x9718 >> 2),
814 	0x00000000,
815 	(0x0e00 << 16) | (0x971c >> 2),
816 	0x00000000,
817 	(0x0e00 << 16) | (0x31068 >> 2),
818 	0x00000000,
819 	(0x4e00 << 16) | (0x31068 >> 2),
820 	0x00000000,
821 	(0x5e00 << 16) | (0x31068 >> 2),
822 	0x00000000,
823 	(0x6e00 << 16) | (0x31068 >> 2),
824 	0x00000000,
825 	(0x7e00 << 16) | (0x31068 >> 2),
826 	0x00000000,
827 	(0x0e00 << 16) | (0xcd10 >> 2),
828 	0x00000000,
829 	(0x0e00 << 16) | (0xcd14 >> 2),
830 	0x00000000,
831 	(0x0e00 << 16) | (0x88b0 >> 2),
832 	0x00000000,
833 	(0x0e00 << 16) | (0x88b4 >> 2),
834 	0x00000000,
835 	(0x0e00 << 16) | (0x88b8 >> 2),
836 	0x00000000,
837 	(0x0e00 << 16) | (0x88bc >> 2),
838 	0x00000000,
839 	(0x0400 << 16) | (0x89c0 >> 2),
840 	0x00000000,
841 	(0x0e00 << 16) | (0x88c4 >> 2),
842 	0x00000000,
843 	(0x0e00 << 16) | (0x88c8 >> 2),
844 	0x00000000,
845 	(0x0e00 << 16) | (0x88d0 >> 2),
846 	0x00000000,
847 	(0x0e00 << 16) | (0x88d4 >> 2),
848 	0x00000000,
849 	(0x0e00 << 16) | (0x88d8 >> 2),
850 	0x00000000,
851 	(0x0e00 << 16) | (0x8980 >> 2),
852 	0x00000000,
853 	(0x0e00 << 16) | (0x30938 >> 2),
854 	0x00000000,
855 	(0x0e00 << 16) | (0x3093c >> 2),
856 	0x00000000,
857 	(0x0e00 << 16) | (0x30940 >> 2),
858 	0x00000000,
859 	(0x0e00 << 16) | (0x89a0 >> 2),
860 	0x00000000,
861 	(0x0e00 << 16) | (0x30900 >> 2),
862 	0x00000000,
863 	(0x0e00 << 16) | (0x30904 >> 2),
864 	0x00000000,
865 	(0x0e00 << 16) | (0x89b4 >> 2),
866 	0x00000000,
867 	(0x0e00 << 16) | (0x3e1fc >> 2),
868 	0x00000000,
869 	(0x0e00 << 16) | (0x3c210 >> 2),
870 	0x00000000,
871 	(0x0e00 << 16) | (0x3c214 >> 2),
872 	0x00000000,
873 	(0x0e00 << 16) | (0x3c218 >> 2),
874 	0x00000000,
875 	(0x0e00 << 16) | (0x8904 >> 2),
876 	0x00000000,
877 	0x5,
878 	(0x0e00 << 16) | (0x8c28 >> 2),
879 	(0x0e00 << 16) | (0x8c2c >> 2),
880 	(0x0e00 << 16) | (0x8c30 >> 2),
881 	(0x0e00 << 16) | (0x8c34 >> 2),
882 	(0x0e00 << 16) | (0x9600 >> 2),
883 };
884 
885 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev);
886 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer);
887 static void gfx_v7_0_init_pg(struct amdgpu_device *adev);
888 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev);
889 
890 /*
891  * Core functions
892  */
893 /**
894  * gfx_v7_0_init_microcode - load ucode images from disk
895  *
896  * @adev: amdgpu_device pointer
897  *
898  * Use the firmware interface to load the ucode images into
899  * the driver (not loaded into hw).
900  * Returns 0 on success, error on failure.
901  */
902 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev)
903 {
904 	const char *chip_name;
905 	char fw_name[30];
906 	int err;
907 
908 	DRM_DEBUG("\n");
909 
910 	switch (adev->asic_type) {
911 	case CHIP_BONAIRE:
912 		chip_name = "bonaire";
913 		break;
914 	case CHIP_HAWAII:
915 		chip_name = "hawaii";
916 		break;
917 	case CHIP_KAVERI:
918 		chip_name = "kaveri";
919 		break;
920 	case CHIP_KABINI:
921 		chip_name = "kabini";
922 		break;
923 	case CHIP_MULLINS:
924 		chip_name = "mullins";
925 		break;
926 	default: BUG();
927 	}
928 
929 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
930 	err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
931 	if (err)
932 		goto out;
933 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
934 	if (err)
935 		goto out;
936 
937 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
938 	err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
939 	if (err)
940 		goto out;
941 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
942 	if (err)
943 		goto out;
944 
945 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
946 	err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
947 	if (err)
948 		goto out;
949 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
950 	if (err)
951 		goto out;
952 
953 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
954 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
955 	if (err)
956 		goto out;
957 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
958 	if (err)
959 		goto out;
960 
961 	if (adev->asic_type == CHIP_KAVERI) {
962 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
963 		err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
964 		if (err)
965 			goto out;
966 		err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
967 		if (err)
968 			goto out;
969 	}
970 
971 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
972 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
973 	if (err)
974 		goto out;
975 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
976 
977 out:
978 	if (err) {
979 		pr_err("gfx7: Failed to load firmware \"%s\"\n", fw_name);
980 		release_firmware(adev->gfx.pfp_fw);
981 		adev->gfx.pfp_fw = NULL;
982 		release_firmware(adev->gfx.me_fw);
983 		adev->gfx.me_fw = NULL;
984 		release_firmware(adev->gfx.ce_fw);
985 		adev->gfx.ce_fw = NULL;
986 		release_firmware(adev->gfx.mec_fw);
987 		adev->gfx.mec_fw = NULL;
988 		release_firmware(adev->gfx.mec2_fw);
989 		adev->gfx.mec2_fw = NULL;
990 		release_firmware(adev->gfx.rlc_fw);
991 		adev->gfx.rlc_fw = NULL;
992 	}
993 	return err;
994 }
995 
996 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev)
997 {
998 	release_firmware(adev->gfx.pfp_fw);
999 	adev->gfx.pfp_fw = NULL;
1000 	release_firmware(adev->gfx.me_fw);
1001 	adev->gfx.me_fw = NULL;
1002 	release_firmware(adev->gfx.ce_fw);
1003 	adev->gfx.ce_fw = NULL;
1004 	release_firmware(adev->gfx.mec_fw);
1005 	adev->gfx.mec_fw = NULL;
1006 	release_firmware(adev->gfx.mec2_fw);
1007 	adev->gfx.mec2_fw = NULL;
1008 	release_firmware(adev->gfx.rlc_fw);
1009 	adev->gfx.rlc_fw = NULL;
1010 }
1011 
1012 /**
1013  * gfx_v7_0_tiling_mode_table_init - init the hw tiling table
1014  *
1015  * @adev: amdgpu_device pointer
1016  *
1017  * Starting with SI, the tiling setup is done globally in a
1018  * set of 32 tiling modes.  Rather than selecting each set of
1019  * parameters per surface as on older asics, we just select
1020  * which index in the tiling table we want to use, and the
1021  * surface uses those parameters (CIK).
1022  */
1023 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev)
1024 {
1025 	const u32 num_tile_mode_states =
1026 			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
1027 	const u32 num_secondary_tile_mode_states =
1028 			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
1029 	u32 reg_offset, split_equal_to_row_size;
1030 	uint32_t *tile, *macrotile;
1031 
1032 	tile = adev->gfx.config.tile_mode_array;
1033 	macrotile = adev->gfx.config.macrotile_mode_array;
1034 
1035 	switch (adev->gfx.config.mem_row_size_in_kb) {
1036 	case 1:
1037 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1038 		break;
1039 	case 2:
1040 	default:
1041 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1042 		break;
1043 	case 4:
1044 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1045 		break;
1046 	}
1047 
1048 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1049 		tile[reg_offset] = 0;
1050 	for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1051 		macrotile[reg_offset] = 0;
1052 
1053 	switch (adev->asic_type) {
1054 	case CHIP_BONAIRE:
1055 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1056 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1057 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1058 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1059 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1060 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1061 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1062 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1063 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1064 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1065 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1066 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1067 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1068 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1069 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1070 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1071 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1072 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1073 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1074 			   TILE_SPLIT(split_equal_to_row_size));
1075 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1076 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1077 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1078 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1079 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1080 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1081 			   TILE_SPLIT(split_equal_to_row_size));
1082 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1083 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1084 			   PIPE_CONFIG(ADDR_SURF_P4_16x16));
1085 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1086 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1087 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1088 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1089 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1090 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1091 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1092 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1093 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1094 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1095 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1096 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1097 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1098 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1099 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1100 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1101 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1102 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1103 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1104 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1105 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1106 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1107 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1108 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1109 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1110 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1111 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1112 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1113 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1114 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1115 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1116 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1117 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1118 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1119 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1120 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1121 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1122 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1123 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1124 		tile[21] =  (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1125 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1126 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1127 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1128 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1129 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1130 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1131 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1132 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1133 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1134 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1135 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1136 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1137 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1138 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1139 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1140 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1141 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1142 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1143 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1144 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1145 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1146 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1147 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1148 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1149 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1150 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1151 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1152 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1153 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1154 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1155 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1156 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1157 
1158 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1159 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1160 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1161 				NUM_BANKS(ADDR_SURF_16_BANK));
1162 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1163 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1164 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1165 				NUM_BANKS(ADDR_SURF_16_BANK));
1166 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1167 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1168 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1169 				NUM_BANKS(ADDR_SURF_16_BANK));
1170 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1171 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1172 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1173 				NUM_BANKS(ADDR_SURF_16_BANK));
1174 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1175 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1176 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1177 				NUM_BANKS(ADDR_SURF_16_BANK));
1178 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1179 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1180 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1181 				NUM_BANKS(ADDR_SURF_8_BANK));
1182 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1183 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1184 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1185 				NUM_BANKS(ADDR_SURF_4_BANK));
1186 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1187 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1188 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1189 				NUM_BANKS(ADDR_SURF_16_BANK));
1190 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1191 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1192 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1193 				NUM_BANKS(ADDR_SURF_16_BANK));
1194 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1195 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1196 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1197 				NUM_BANKS(ADDR_SURF_16_BANK));
1198 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1199 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1200 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1201 				NUM_BANKS(ADDR_SURF_16_BANK));
1202 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1203 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1204 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1205 				NUM_BANKS(ADDR_SURF_16_BANK));
1206 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1207 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1208 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1209 				NUM_BANKS(ADDR_SURF_8_BANK));
1210 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1211 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1212 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1213 				NUM_BANKS(ADDR_SURF_4_BANK));
1214 
1215 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1216 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1217 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1218 			if (reg_offset != 7)
1219 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1220 		break;
1221 	case CHIP_HAWAII:
1222 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1223 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1224 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1225 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1226 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1227 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1228 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1229 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1230 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1231 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1232 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1233 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1234 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1235 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1236 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1237 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1238 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1239 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1240 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1241 			   TILE_SPLIT(split_equal_to_row_size));
1242 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1243 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1244 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1245 			   TILE_SPLIT(split_equal_to_row_size));
1246 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1247 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1248 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1249 			   TILE_SPLIT(split_equal_to_row_size));
1250 		tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1251 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1252 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1253 			   TILE_SPLIT(split_equal_to_row_size));
1254 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1255 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
1256 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1257 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1258 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1259 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1260 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1261 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1262 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1263 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1264 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1265 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1266 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1267 		tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1268 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1269 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1270 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1271 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1272 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1273 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1274 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1275 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1276 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1277 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1278 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1279 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1280 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1281 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1282 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1283 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1284 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1285 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1286 		tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1287 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1288 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1289 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1290 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1291 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1292 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1293 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1294 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1295 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1296 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1297 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1298 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1299 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1300 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1301 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1302 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1303 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1304 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1305 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1306 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1307 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1308 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1309 		tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1310 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1311 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1312 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1313 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1314 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1315 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1316 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1317 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1318 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1319 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1320 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1321 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1322 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1323 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1324 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1325 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1326 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1327 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1328 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1329 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1330 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1331 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1332 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1333 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1334 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1335 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1336 		tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1337 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1338 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1339 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1340 
1341 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1342 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1343 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1344 				NUM_BANKS(ADDR_SURF_16_BANK));
1345 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1346 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1347 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1348 				NUM_BANKS(ADDR_SURF_16_BANK));
1349 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1350 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1351 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1352 				NUM_BANKS(ADDR_SURF_16_BANK));
1353 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1354 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1355 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1356 				NUM_BANKS(ADDR_SURF_16_BANK));
1357 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1358 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1359 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1360 				NUM_BANKS(ADDR_SURF_8_BANK));
1361 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1362 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1363 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1364 				NUM_BANKS(ADDR_SURF_4_BANK));
1365 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1366 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1367 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1368 				NUM_BANKS(ADDR_SURF_4_BANK));
1369 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1370 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1371 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1372 				NUM_BANKS(ADDR_SURF_16_BANK));
1373 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1374 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1375 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1376 				NUM_BANKS(ADDR_SURF_16_BANK));
1377 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1378 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1379 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1380 				NUM_BANKS(ADDR_SURF_16_BANK));
1381 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1382 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1383 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1384 				NUM_BANKS(ADDR_SURF_8_BANK));
1385 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1386 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1387 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1388 				NUM_BANKS(ADDR_SURF_16_BANK));
1389 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1390 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1391 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1392 				NUM_BANKS(ADDR_SURF_8_BANK));
1393 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1394 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1395 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1396 				NUM_BANKS(ADDR_SURF_4_BANK));
1397 
1398 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1399 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1400 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1401 			if (reg_offset != 7)
1402 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1403 		break;
1404 	case CHIP_KABINI:
1405 	case CHIP_KAVERI:
1406 	case CHIP_MULLINS:
1407 	default:
1408 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1409 			   PIPE_CONFIG(ADDR_SURF_P2) |
1410 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1411 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1412 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1413 			   PIPE_CONFIG(ADDR_SURF_P2) |
1414 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1415 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1416 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1417 			   PIPE_CONFIG(ADDR_SURF_P2) |
1418 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1419 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1420 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1421 			   PIPE_CONFIG(ADDR_SURF_P2) |
1422 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1423 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1424 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1425 			   PIPE_CONFIG(ADDR_SURF_P2) |
1426 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1427 			   TILE_SPLIT(split_equal_to_row_size));
1428 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1429 			   PIPE_CONFIG(ADDR_SURF_P2) |
1430 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1431 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1432 			   PIPE_CONFIG(ADDR_SURF_P2) |
1433 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1434 			   TILE_SPLIT(split_equal_to_row_size));
1435 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1436 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1437 			   PIPE_CONFIG(ADDR_SURF_P2));
1438 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1439 			   PIPE_CONFIG(ADDR_SURF_P2) |
1440 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1441 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1442 			    PIPE_CONFIG(ADDR_SURF_P2) |
1443 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1444 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1445 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1446 			    PIPE_CONFIG(ADDR_SURF_P2) |
1447 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1448 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1449 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1450 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1451 			    PIPE_CONFIG(ADDR_SURF_P2) |
1452 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1453 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1454 			    PIPE_CONFIG(ADDR_SURF_P2) |
1455 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1456 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1457 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1458 			    PIPE_CONFIG(ADDR_SURF_P2) |
1459 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1460 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1461 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1462 			    PIPE_CONFIG(ADDR_SURF_P2) |
1463 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1464 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1465 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1466 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1467 			    PIPE_CONFIG(ADDR_SURF_P2) |
1468 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1469 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1470 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1471 			    PIPE_CONFIG(ADDR_SURF_P2) |
1472 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1473 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1474 			    PIPE_CONFIG(ADDR_SURF_P2) |
1475 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1476 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1477 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1478 			    PIPE_CONFIG(ADDR_SURF_P2) |
1479 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1480 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1481 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1482 			    PIPE_CONFIG(ADDR_SURF_P2) |
1483 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1484 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1485 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1486 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1487 			    PIPE_CONFIG(ADDR_SURF_P2) |
1488 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1489 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1490 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1491 			    PIPE_CONFIG(ADDR_SURF_P2) |
1492 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1493 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1494 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1495 			    PIPE_CONFIG(ADDR_SURF_P2) |
1496 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1497 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1498 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1499 			    PIPE_CONFIG(ADDR_SURF_P2) |
1500 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1501 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1502 			    PIPE_CONFIG(ADDR_SURF_P2) |
1503 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1504 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1505 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1506 			    PIPE_CONFIG(ADDR_SURF_P2) |
1507 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1508 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1509 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1510 
1511 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1512 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1513 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1514 				NUM_BANKS(ADDR_SURF_8_BANK));
1515 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1516 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1517 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1518 				NUM_BANKS(ADDR_SURF_8_BANK));
1519 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1520 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1521 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1522 				NUM_BANKS(ADDR_SURF_8_BANK));
1523 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1524 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1525 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1526 				NUM_BANKS(ADDR_SURF_8_BANK));
1527 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1528 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1529 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1530 				NUM_BANKS(ADDR_SURF_8_BANK));
1531 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1532 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1533 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1534 				NUM_BANKS(ADDR_SURF_8_BANK));
1535 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1536 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1537 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1538 				NUM_BANKS(ADDR_SURF_8_BANK));
1539 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1540 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1541 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1542 				NUM_BANKS(ADDR_SURF_16_BANK));
1543 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1544 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1545 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1546 				NUM_BANKS(ADDR_SURF_16_BANK));
1547 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1548 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1549 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1550 				NUM_BANKS(ADDR_SURF_16_BANK));
1551 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1552 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1553 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1554 				NUM_BANKS(ADDR_SURF_16_BANK));
1555 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1556 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1557 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1558 				NUM_BANKS(ADDR_SURF_16_BANK));
1559 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1560 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1561 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1562 				NUM_BANKS(ADDR_SURF_16_BANK));
1563 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1564 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1565 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1566 				NUM_BANKS(ADDR_SURF_8_BANK));
1567 
1568 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1569 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1570 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1571 			if (reg_offset != 7)
1572 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1573 		break;
1574 	}
1575 }
1576 
1577 /**
1578  * gfx_v7_0_select_se_sh - select which SE, SH to address
1579  *
1580  * @adev: amdgpu_device pointer
1581  * @se_num: shader engine to address
1582  * @sh_num: sh block to address
1583  *
1584  * Select which SE, SH combinations to address. Certain
1585  * registers are instanced per SE or SH.  0xffffffff means
1586  * broadcast to all SEs or SHs (CIK).
1587  */
1588 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev,
1589 				  u32 se_num, u32 sh_num, u32 instance)
1590 {
1591 	u32 data;
1592 
1593 	if (instance == 0xffffffff)
1594 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1595 	else
1596 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1597 
1598 	if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1599 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1600 			GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
1601 	else if (se_num == 0xffffffff)
1602 		data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK |
1603 			(sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT);
1604 	else if (sh_num == 0xffffffff)
1605 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1606 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1607 	else
1608 		data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) |
1609 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1610 	WREG32(mmGRBM_GFX_INDEX, data);
1611 }
1612 
1613 /**
1614  * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs
1615  *
1616  * @adev: amdgpu_device pointer
1617  *
1618  * Calculates the bitmask of enabled RBs (CIK).
1619  * Returns the enabled RB bitmask.
1620  */
1621 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1622 {
1623 	u32 data, mask;
1624 
1625 	data = RREG32(mmCC_RB_BACKEND_DISABLE);
1626 	data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1627 
1628 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1629 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1630 
1631 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1632 					 adev->gfx.config.max_sh_per_se);
1633 
1634 	return (~data) & mask;
1635 }
1636 
1637 static void
1638 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
1639 {
1640 	switch (adev->asic_type) {
1641 	case CHIP_BONAIRE:
1642 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
1643 			  SE_XSEL(1) | SE_YSEL(1);
1644 		*rconf1 |= 0x0;
1645 		break;
1646 	case CHIP_HAWAII:
1647 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
1648 			  RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) |
1649 			  PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) |
1650 			  SE_YSEL(3);
1651 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
1652 			   SE_PAIR_YSEL(2);
1653 		break;
1654 	case CHIP_KAVERI:
1655 		*rconf |= RB_MAP_PKR0(2);
1656 		*rconf1 |= 0x0;
1657 		break;
1658 	case CHIP_KABINI:
1659 	case CHIP_MULLINS:
1660 		*rconf |= 0x0;
1661 		*rconf1 |= 0x0;
1662 		break;
1663 	default:
1664 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
1665 		break;
1666 	}
1667 }
1668 
1669 static void
1670 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev,
1671 					u32 raster_config, u32 raster_config_1,
1672 					unsigned rb_mask, unsigned num_rb)
1673 {
1674 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
1675 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
1676 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
1677 	unsigned rb_per_se = num_rb / num_se;
1678 	unsigned se_mask[4];
1679 	unsigned se;
1680 
1681 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
1682 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
1683 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
1684 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
1685 
1686 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
1687 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
1688 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
1689 
1690 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
1691 			     (!se_mask[2] && !se_mask[3]))) {
1692 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
1693 
1694 		if (!se_mask[0] && !se_mask[1]) {
1695 			raster_config_1 |=
1696 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
1697 		} else {
1698 			raster_config_1 |=
1699 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
1700 		}
1701 	}
1702 
1703 	for (se = 0; se < num_se; se++) {
1704 		unsigned raster_config_se = raster_config;
1705 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
1706 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
1707 		int idx = (se / 2) * 2;
1708 
1709 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
1710 			raster_config_se &= ~SE_MAP_MASK;
1711 
1712 			if (!se_mask[idx]) {
1713 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
1714 			} else {
1715 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
1716 			}
1717 		}
1718 
1719 		pkr0_mask &= rb_mask;
1720 		pkr1_mask &= rb_mask;
1721 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
1722 			raster_config_se &= ~PKR_MAP_MASK;
1723 
1724 			if (!pkr0_mask) {
1725 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
1726 			} else {
1727 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
1728 			}
1729 		}
1730 
1731 		if (rb_per_se >= 2) {
1732 			unsigned rb0_mask = 1 << (se * rb_per_se);
1733 			unsigned rb1_mask = rb0_mask << 1;
1734 
1735 			rb0_mask &= rb_mask;
1736 			rb1_mask &= rb_mask;
1737 			if (!rb0_mask || !rb1_mask) {
1738 				raster_config_se &= ~RB_MAP_PKR0_MASK;
1739 
1740 				if (!rb0_mask) {
1741 					raster_config_se |=
1742 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
1743 				} else {
1744 					raster_config_se |=
1745 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
1746 				}
1747 			}
1748 
1749 			if (rb_per_se > 2) {
1750 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
1751 				rb1_mask = rb0_mask << 1;
1752 				rb0_mask &= rb_mask;
1753 				rb1_mask &= rb_mask;
1754 				if (!rb0_mask || !rb1_mask) {
1755 					raster_config_se &= ~RB_MAP_PKR1_MASK;
1756 
1757 					if (!rb0_mask) {
1758 						raster_config_se |=
1759 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
1760 					} else {
1761 						raster_config_se |=
1762 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
1763 					}
1764 				}
1765 			}
1766 		}
1767 
1768 		/* GRBM_GFX_INDEX has a different offset on CI+ */
1769 		gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
1770 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
1771 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1772 	}
1773 
1774 	/* GRBM_GFX_INDEX has a different offset on CI+ */
1775 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1776 }
1777 
1778 /**
1779  * gfx_v7_0_setup_rb - setup the RBs on the asic
1780  *
1781  * @adev: amdgpu_device pointer
1782  * @se_num: number of SEs (shader engines) for the asic
1783  * @sh_per_se: number of SH blocks per SE for the asic
1784  *
1785  * Configures per-SE/SH RB registers (CIK).
1786  */
1787 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1788 {
1789 	int i, j;
1790 	u32 data;
1791 	u32 raster_config = 0, raster_config_1 = 0;
1792 	u32 active_rbs = 0;
1793 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1794 					adev->gfx.config.max_sh_per_se;
1795 	unsigned num_rb_pipes;
1796 
1797 	mutex_lock(&adev->grbm_idx_mutex);
1798 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1799 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1800 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1801 			data = gfx_v7_0_get_rb_active_bitmap(adev);
1802 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1803 					       rb_bitmap_width_per_sh);
1804 		}
1805 	}
1806 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1807 
1808 	adev->gfx.config.backend_enable_mask = active_rbs;
1809 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1810 
1811 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1812 			     adev->gfx.config.max_shader_engines, 16);
1813 
1814 	gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1815 
1816 	if (!adev->gfx.config.backend_enable_mask ||
1817 			adev->gfx.config.num_rbs >= num_rb_pipes) {
1818 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1819 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1820 	} else {
1821 		gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1822 							adev->gfx.config.backend_enable_mask,
1823 							num_rb_pipes);
1824 	}
1825 
1826 	/* cache the values for userspace */
1827 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1828 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1829 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1830 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
1831 				RREG32(mmCC_RB_BACKEND_DISABLE);
1832 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
1833 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1834 			adev->gfx.config.rb_config[i][j].raster_config =
1835 				RREG32(mmPA_SC_RASTER_CONFIG);
1836 			adev->gfx.config.rb_config[i][j].raster_config_1 =
1837 				RREG32(mmPA_SC_RASTER_CONFIG_1);
1838 		}
1839 	}
1840 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1841 	mutex_unlock(&adev->grbm_idx_mutex);
1842 }
1843 
1844 /**
1845  * gfx_v7_0_init_compute_vmid - gart enable
1846  *
1847  * @adev: amdgpu_device pointer
1848  *
1849  * Initialize compute vmid sh_mem registers
1850  *
1851  */
1852 #define DEFAULT_SH_MEM_BASES	(0x6000)
1853 #define FIRST_COMPUTE_VMID	(8)
1854 #define LAST_COMPUTE_VMID	(16)
1855 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1856 {
1857 	int i;
1858 	uint32_t sh_mem_config;
1859 	uint32_t sh_mem_bases;
1860 
1861 	/*
1862 	 * Configure apertures:
1863 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1864 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1865 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1866 	*/
1867 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1868 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1869 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1870 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1871 	mutex_lock(&adev->srbm_mutex);
1872 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1873 		cik_srbm_select(adev, 0, 0, 0, i);
1874 		/* CP and shaders */
1875 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1876 		WREG32(mmSH_MEM_APE1_BASE, 1);
1877 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1878 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1879 	}
1880 	cik_srbm_select(adev, 0, 0, 0, 0);
1881 	mutex_unlock(&adev->srbm_mutex);
1882 
1883 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1884 	   acccess. These should be enabled by FW for target VMIDs. */
1885 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1886 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
1887 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
1888 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
1889 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
1890 	}
1891 }
1892 
1893 static void gfx_v7_0_init_gds_vmid(struct amdgpu_device *adev)
1894 {
1895 	int vmid;
1896 
1897 	/*
1898 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1899 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1900 	 * the driver can enable them for graphics. VMID0 should maintain
1901 	 * access so that HWS firmware can save/restore entries.
1902 	 */
1903 	for (vmid = 1; vmid < 16; vmid++) {
1904 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
1905 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
1906 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
1907 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
1908 	}
1909 }
1910 
1911 static void gfx_v7_0_config_init(struct amdgpu_device *adev)
1912 {
1913 	adev->gfx.config.double_offchip_lds_buf = 1;
1914 }
1915 
1916 /**
1917  * gfx_v7_0_constants_init - setup the 3D engine
1918  *
1919  * @adev: amdgpu_device pointer
1920  *
1921  * init the gfx constants such as the 3D engine, tiling configuration
1922  * registers, maximum number of quad pipes, render backends...
1923  */
1924 static void gfx_v7_0_constants_init(struct amdgpu_device *adev)
1925 {
1926 	u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base;
1927 	u32 tmp;
1928 	int i;
1929 
1930 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1931 
1932 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1933 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1934 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1935 
1936 	gfx_v7_0_tiling_mode_table_init(adev);
1937 
1938 	gfx_v7_0_setup_rb(adev);
1939 	gfx_v7_0_get_cu_info(adev);
1940 	gfx_v7_0_config_init(adev);
1941 
1942 	/* set HW defaults for 3D engine */
1943 	WREG32(mmCP_MEQ_THRESHOLDS,
1944 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1945 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1946 
1947 	mutex_lock(&adev->grbm_idx_mutex);
1948 	/*
1949 	 * making sure that the following register writes will be broadcasted
1950 	 * to all the shaders
1951 	 */
1952 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1953 
1954 	/* XXX SH_MEM regs */
1955 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1956 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1957 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1958 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE,
1959 				   MTYPE_NC);
1960 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE,
1961 				   MTYPE_UC);
1962 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0);
1963 
1964 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
1965 				   SWIZZLE_ENABLE, 1);
1966 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1967 				   ELEMENT_SIZE, 1);
1968 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1969 				   INDEX_STRIDE, 3);
1970 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
1971 
1972 	mutex_lock(&adev->srbm_mutex);
1973 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
1974 		if (i == 0)
1975 			sh_mem_base = 0;
1976 		else
1977 			sh_mem_base = adev->gmc.shared_aperture_start >> 48;
1978 		cik_srbm_select(adev, 0, 0, 0, i);
1979 		/* CP and shaders */
1980 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1981 		WREG32(mmSH_MEM_APE1_BASE, 1);
1982 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1983 		WREG32(mmSH_MEM_BASES, sh_mem_base);
1984 	}
1985 	cik_srbm_select(adev, 0, 0, 0, 0);
1986 	mutex_unlock(&adev->srbm_mutex);
1987 
1988 	gfx_v7_0_init_compute_vmid(adev);
1989 	gfx_v7_0_init_gds_vmid(adev);
1990 
1991 	WREG32(mmSX_DEBUG_1, 0x20);
1992 
1993 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1994 
1995 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1996 	tmp |= 0x03000000;
1997 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1998 
1999 	WREG32(mmSQ_CONFIG, 1);
2000 
2001 	WREG32(mmDB_DEBUG, 0);
2002 
2003 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
2004 	tmp |= 0x00000400;
2005 	WREG32(mmDB_DEBUG2, tmp);
2006 
2007 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
2008 	tmp |= 0x00020200;
2009 	WREG32(mmDB_DEBUG3, tmp);
2010 
2011 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
2012 	tmp |= 0x00018208;
2013 	WREG32(mmCB_HW_CONTROL, tmp);
2014 
2015 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
2016 
2017 	WREG32(mmPA_SC_FIFO_SIZE,
2018 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
2019 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
2020 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
2021 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
2022 
2023 	WREG32(mmVGT_NUM_INSTANCES, 1);
2024 
2025 	WREG32(mmCP_PERFMON_CNTL, 0);
2026 
2027 	WREG32(mmSQ_CONFIG, 0);
2028 
2029 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
2030 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
2031 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
2032 
2033 	WREG32(mmVGT_CACHE_INVALIDATION,
2034 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
2035 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
2036 
2037 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
2038 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
2039 
2040 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
2041 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
2042 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
2043 
2044 	tmp = RREG32(mmSPI_ARB_PRIORITY);
2045 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
2046 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
2047 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
2048 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
2049 	WREG32(mmSPI_ARB_PRIORITY, tmp);
2050 
2051 	mutex_unlock(&adev->grbm_idx_mutex);
2052 
2053 	udelay(50);
2054 }
2055 
2056 /*
2057  * GPU scratch registers helpers function.
2058  */
2059 /**
2060  * gfx_v7_0_scratch_init - setup driver info for CP scratch regs
2061  *
2062  * @adev: amdgpu_device pointer
2063  *
2064  * Set up the number and offset of the CP scratch registers.
2065  * NOTE: use of CP scratch registers is a legacy inferface and
2066  * is not used by default on newer asics (r6xx+).  On newer asics,
2067  * memory buffers are used for fences rather than scratch regs.
2068  */
2069 static void gfx_v7_0_scratch_init(struct amdgpu_device *adev)
2070 {
2071 	adev->gfx.scratch.num_reg = 8;
2072 	adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
2073 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
2074 }
2075 
2076 /**
2077  * gfx_v7_0_ring_test_ring - basic gfx ring test
2078  *
2079  * @adev: amdgpu_device pointer
2080  * @ring: amdgpu_ring structure holding ring information
2081  *
2082  * Allocate a scratch register and write to it using the gfx ring (CIK).
2083  * Provides a basic gfx ring test to verify that the ring is working.
2084  * Used by gfx_v7_0_cp_gfx_resume();
2085  * Returns 0 on success, error on failure.
2086  */
2087 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2088 {
2089 	struct amdgpu_device *adev = ring->adev;
2090 	uint32_t scratch;
2091 	uint32_t tmp = 0;
2092 	unsigned i;
2093 	int r;
2094 
2095 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2096 	if (r)
2097 		return r;
2098 
2099 	WREG32(scratch, 0xCAFEDEAD);
2100 	r = amdgpu_ring_alloc(ring, 3);
2101 	if (r)
2102 		goto error_free_scratch;
2103 
2104 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2105 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
2106 	amdgpu_ring_write(ring, 0xDEADBEEF);
2107 	amdgpu_ring_commit(ring);
2108 
2109 	for (i = 0; i < adev->usec_timeout; i++) {
2110 		tmp = RREG32(scratch);
2111 		if (tmp == 0xDEADBEEF)
2112 			break;
2113 		udelay(1);
2114 	}
2115 	if (i >= adev->usec_timeout)
2116 		r = -ETIMEDOUT;
2117 
2118 error_free_scratch:
2119 	amdgpu_gfx_scratch_free(adev, scratch);
2120 	return r;
2121 }
2122 
2123 /**
2124  * gfx_v7_0_ring_emit_hdp - emit an hdp flush on the cp
2125  *
2126  * @adev: amdgpu_device pointer
2127  * @ridx: amdgpu ring index
2128  *
2129  * Emits an hdp flush on the cp.
2130  */
2131 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2132 {
2133 	u32 ref_and_mask;
2134 	int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2135 
2136 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2137 		switch (ring->me) {
2138 		case 1:
2139 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2140 			break;
2141 		case 2:
2142 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2143 			break;
2144 		default:
2145 			return;
2146 		}
2147 	} else {
2148 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2149 	}
2150 
2151 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2152 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2153 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2154 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2155 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2156 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2157 	amdgpu_ring_write(ring, ref_and_mask);
2158 	amdgpu_ring_write(ring, ref_and_mask);
2159 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2160 }
2161 
2162 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2163 {
2164 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2165 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2166 		EVENT_INDEX(4));
2167 
2168 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2169 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2170 		EVENT_INDEX(0));
2171 }
2172 
2173 /**
2174  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2175  *
2176  * @adev: amdgpu_device pointer
2177  * @fence: amdgpu fence object
2178  *
2179  * Emits a fence sequnce number on the gfx ring and flushes
2180  * GPU caches.
2181  */
2182 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2183 					 u64 seq, unsigned flags)
2184 {
2185 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2186 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2187 	/* Workaround for cache flush problems. First send a dummy EOP
2188 	 * event down the pipe with seq one below.
2189 	 */
2190 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2191 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2192 				 EOP_TC_ACTION_EN |
2193 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2194 				 EVENT_INDEX(5)));
2195 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2196 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2197 				DATA_SEL(1) | INT_SEL(0));
2198 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2199 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2200 
2201 	/* Then send the real EOP event down the pipe. */
2202 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2203 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2204 				 EOP_TC_ACTION_EN |
2205 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2206 				 EVENT_INDEX(5)));
2207 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2208 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2209 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2210 	amdgpu_ring_write(ring, lower_32_bits(seq));
2211 	amdgpu_ring_write(ring, upper_32_bits(seq));
2212 }
2213 
2214 /**
2215  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2216  *
2217  * @adev: amdgpu_device pointer
2218  * @fence: amdgpu fence object
2219  *
2220  * Emits a fence sequnce number on the compute ring and flushes
2221  * GPU caches.
2222  */
2223 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2224 					     u64 addr, u64 seq,
2225 					     unsigned flags)
2226 {
2227 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2228 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2229 
2230 	/* RELEASE_MEM - flush caches, send int */
2231 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2232 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2233 				 EOP_TC_ACTION_EN |
2234 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2235 				 EVENT_INDEX(5)));
2236 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2237 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2238 	amdgpu_ring_write(ring, upper_32_bits(addr));
2239 	amdgpu_ring_write(ring, lower_32_bits(seq));
2240 	amdgpu_ring_write(ring, upper_32_bits(seq));
2241 }
2242 
2243 /*
2244  * IB stuff
2245  */
2246 /**
2247  * gfx_v7_0_ring_emit_ib - emit an IB (Indirect Buffer) on the ring
2248  *
2249  * @ring: amdgpu_ring structure holding ring information
2250  * @ib: amdgpu indirect buffer object
2251  *
2252  * Emits an DE (drawing engine) or CE (constant engine) IB
2253  * on the gfx ring.  IBs are usually generated by userspace
2254  * acceleration drivers and submitted to the kernel for
2255  * sheduling on the ring.  This function schedules the IB
2256  * on the gfx ring for execution by the GPU.
2257  */
2258 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2259 					struct amdgpu_job *job,
2260 					struct amdgpu_ib *ib,
2261 					uint32_t flags)
2262 {
2263 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2264 	u32 header, control = 0;
2265 
2266 	/* insert SWITCH_BUFFER packet before first IB in the ring frame */
2267 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2268 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2269 		amdgpu_ring_write(ring, 0);
2270 	}
2271 
2272 	if (ib->flags & AMDGPU_IB_FLAG_CE)
2273 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2274 	else
2275 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2276 
2277 	control |= ib->length_dw | (vmid << 24);
2278 
2279 	amdgpu_ring_write(ring, header);
2280 	amdgpu_ring_write(ring,
2281 #ifdef __BIG_ENDIAN
2282 			  (2 << 0) |
2283 #endif
2284 			  (ib->gpu_addr & 0xFFFFFFFC));
2285 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2286 	amdgpu_ring_write(ring, control);
2287 }
2288 
2289 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
2290 					  struct amdgpu_job *job,
2291 					  struct amdgpu_ib *ib,
2292 					  uint32_t flags)
2293 {
2294 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2295 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
2296 
2297 	/* Currently, there is a high possibility to get wave ID mismatch
2298 	 * between ME and GDS, leading to a hw deadlock, because ME generates
2299 	 * different wave IDs than the GDS expects. This situation happens
2300 	 * randomly when at least 5 compute pipes use GDS ordered append.
2301 	 * The wave IDs generated by ME are also wrong after suspend/resume.
2302 	 * Those are probably bugs somewhere else in the kernel driver.
2303 	 *
2304 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
2305 	 * GDS to 0 for this ring (me/pipe).
2306 	 */
2307 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
2308 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2309 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
2310 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
2311 	}
2312 
2313 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2314 	amdgpu_ring_write(ring,
2315 #ifdef __BIG_ENDIAN
2316 					  (2 << 0) |
2317 #endif
2318 					  (ib->gpu_addr & 0xFFFFFFFC));
2319 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2320 	amdgpu_ring_write(ring, control);
2321 }
2322 
2323 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2324 {
2325 	uint32_t dw2 = 0;
2326 
2327 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2328 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2329 		gfx_v7_0_ring_emit_vgt_flush(ring);
2330 		/* set load_global_config & load_global_uconfig */
2331 		dw2 |= 0x8001;
2332 		/* set load_cs_sh_regs */
2333 		dw2 |= 0x01000000;
2334 		/* set load_per_context_state & load_gfx_sh_regs */
2335 		dw2 |= 0x10002;
2336 	}
2337 
2338 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2339 	amdgpu_ring_write(ring, dw2);
2340 	amdgpu_ring_write(ring, 0);
2341 }
2342 
2343 /**
2344  * gfx_v7_0_ring_test_ib - basic ring IB test
2345  *
2346  * @ring: amdgpu_ring structure holding ring information
2347  *
2348  * Allocate an IB and execute it on the gfx ring (CIK).
2349  * Provides a basic gfx ring test to verify that IBs are working.
2350  * Returns 0 on success, error on failure.
2351  */
2352 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2353 {
2354 	struct amdgpu_device *adev = ring->adev;
2355 	struct amdgpu_ib ib;
2356 	struct dma_fence *f = NULL;
2357 	uint32_t scratch;
2358 	uint32_t tmp = 0;
2359 	long r;
2360 
2361 	r = amdgpu_gfx_scratch_get(adev, &scratch);
2362 	if (r)
2363 		return r;
2364 
2365 	WREG32(scratch, 0xCAFEDEAD);
2366 	memset(&ib, 0, sizeof(ib));
2367 	r = amdgpu_ib_get(adev, NULL, 256, &ib);
2368 	if (r)
2369 		goto err1;
2370 
2371 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2372 	ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
2373 	ib.ptr[2] = 0xDEADBEEF;
2374 	ib.length_dw = 3;
2375 
2376 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2377 	if (r)
2378 		goto err2;
2379 
2380 	r = dma_fence_wait_timeout(f, false, timeout);
2381 	if (r == 0) {
2382 		r = -ETIMEDOUT;
2383 		goto err2;
2384 	} else if (r < 0) {
2385 		goto err2;
2386 	}
2387 	tmp = RREG32(scratch);
2388 	if (tmp == 0xDEADBEEF)
2389 		r = 0;
2390 	else
2391 		r = -EINVAL;
2392 
2393 err2:
2394 	amdgpu_ib_free(adev, &ib, NULL);
2395 	dma_fence_put(f);
2396 err1:
2397 	amdgpu_gfx_scratch_free(adev, scratch);
2398 	return r;
2399 }
2400 
2401 /*
2402  * CP.
2403  * On CIK, gfx and compute now have independant command processors.
2404  *
2405  * GFX
2406  * Gfx consists of a single ring and can process both gfx jobs and
2407  * compute jobs.  The gfx CP consists of three microengines (ME):
2408  * PFP - Pre-Fetch Parser
2409  * ME - Micro Engine
2410  * CE - Constant Engine
2411  * The PFP and ME make up what is considered the Drawing Engine (DE).
2412  * The CE is an asynchronous engine used for updating buffer desciptors
2413  * used by the DE so that they can be loaded into cache in parallel
2414  * while the DE is processing state update packets.
2415  *
2416  * Compute
2417  * The compute CP consists of two microengines (ME):
2418  * MEC1 - Compute MicroEngine 1
2419  * MEC2 - Compute MicroEngine 2
2420  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2421  * The queues are exposed to userspace and are programmed directly
2422  * by the compute runtime.
2423  */
2424 /**
2425  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2426  *
2427  * @adev: amdgpu_device pointer
2428  * @enable: enable or disable the MEs
2429  *
2430  * Halts or unhalts the gfx MEs.
2431  */
2432 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2433 {
2434 	int i;
2435 
2436 	if (enable) {
2437 		WREG32(mmCP_ME_CNTL, 0);
2438 	} else {
2439 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK));
2440 		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2441 			adev->gfx.gfx_ring[i].sched.ready = false;
2442 	}
2443 	udelay(50);
2444 }
2445 
2446 /**
2447  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2448  *
2449  * @adev: amdgpu_device pointer
2450  *
2451  * Loads the gfx PFP, ME, and CE ucode.
2452  * Returns 0 for success, -EINVAL if the ucode is not available.
2453  */
2454 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2455 {
2456 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2457 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2458 	const struct gfx_firmware_header_v1_0 *me_hdr;
2459 	const __le32 *fw_data;
2460 	unsigned i, fw_size;
2461 
2462 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2463 		return -EINVAL;
2464 
2465 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2466 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2467 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2468 
2469 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2470 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2471 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2472 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2473 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2474 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2475 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2476 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2477 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2478 
2479 	gfx_v7_0_cp_gfx_enable(adev, false);
2480 
2481 	/* PFP */
2482 	fw_data = (const __le32 *)
2483 		(adev->gfx.pfp_fw->data +
2484 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2485 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2486 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2487 	for (i = 0; i < fw_size; i++)
2488 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2489 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2490 
2491 	/* CE */
2492 	fw_data = (const __le32 *)
2493 		(adev->gfx.ce_fw->data +
2494 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2495 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2496 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2497 	for (i = 0; i < fw_size; i++)
2498 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2499 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2500 
2501 	/* ME */
2502 	fw_data = (const __le32 *)
2503 		(adev->gfx.me_fw->data +
2504 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2505 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2506 	WREG32(mmCP_ME_RAM_WADDR, 0);
2507 	for (i = 0; i < fw_size; i++)
2508 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2509 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2510 
2511 	return 0;
2512 }
2513 
2514 /**
2515  * gfx_v7_0_cp_gfx_start - start the gfx ring
2516  *
2517  * @adev: amdgpu_device pointer
2518  *
2519  * Enables the ring and loads the clear state context and other
2520  * packets required to init the ring.
2521  * Returns 0 for success, error for failure.
2522  */
2523 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2524 {
2525 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2526 	const struct cs_section_def *sect = NULL;
2527 	const struct cs_extent_def *ext = NULL;
2528 	int r, i;
2529 
2530 	/* init the CP */
2531 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2532 	WREG32(mmCP_ENDIAN_SWAP, 0);
2533 	WREG32(mmCP_DEVICE_ID, 1);
2534 
2535 	gfx_v7_0_cp_gfx_enable(adev, true);
2536 
2537 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2538 	if (r) {
2539 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2540 		return r;
2541 	}
2542 
2543 	/* init the CE partitions.  CE only used for gfx on CIK */
2544 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2545 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2546 	amdgpu_ring_write(ring, 0x8000);
2547 	amdgpu_ring_write(ring, 0x8000);
2548 
2549 	/* clear state buffer */
2550 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2551 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2552 
2553 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2554 	amdgpu_ring_write(ring, 0x80000000);
2555 	amdgpu_ring_write(ring, 0x80000000);
2556 
2557 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2558 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2559 			if (sect->id == SECT_CONTEXT) {
2560 				amdgpu_ring_write(ring,
2561 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2562 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2563 				for (i = 0; i < ext->reg_count; i++)
2564 					amdgpu_ring_write(ring, ext->extent[i]);
2565 			}
2566 		}
2567 	}
2568 
2569 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2570 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2571 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
2572 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
2573 
2574 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2575 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2576 
2577 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2578 	amdgpu_ring_write(ring, 0);
2579 
2580 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2581 	amdgpu_ring_write(ring, 0x00000316);
2582 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2583 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2584 
2585 	amdgpu_ring_commit(ring);
2586 
2587 	return 0;
2588 }
2589 
2590 /**
2591  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2592  *
2593  * @adev: amdgpu_device pointer
2594  *
2595  * Program the location and size of the gfx ring buffer
2596  * and test it to make sure it's working.
2597  * Returns 0 for success, error for failure.
2598  */
2599 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2600 {
2601 	struct amdgpu_ring *ring;
2602 	u32 tmp;
2603 	u32 rb_bufsz;
2604 	u64 rb_addr, rptr_addr;
2605 	int r;
2606 
2607 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2608 	if (adev->asic_type != CHIP_HAWAII)
2609 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2610 
2611 	/* Set the write pointer delay */
2612 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2613 
2614 	/* set the RB to use vmid 0 */
2615 	WREG32(mmCP_RB_VMID, 0);
2616 
2617 	WREG32(mmSCRATCH_ADDR, 0);
2618 
2619 	/* ring 0 - compute and gfx */
2620 	/* Set ring buffer size */
2621 	ring = &adev->gfx.gfx_ring[0];
2622 	rb_bufsz = order_base_2(ring->ring_size / 8);
2623 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2624 #ifdef __BIG_ENDIAN
2625 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2626 #endif
2627 	WREG32(mmCP_RB0_CNTL, tmp);
2628 
2629 	/* Initialize the ring buffer's read and write pointers */
2630 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2631 	ring->wptr = 0;
2632 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2633 
2634 	/* set the wb address wether it's enabled or not */
2635 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2636 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2637 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2638 
2639 	/* scratch register shadowing is no longer supported */
2640 	WREG32(mmSCRATCH_UMSK, 0);
2641 
2642 	mdelay(1);
2643 	WREG32(mmCP_RB0_CNTL, tmp);
2644 
2645 	rb_addr = ring->gpu_addr >> 8;
2646 	WREG32(mmCP_RB0_BASE, rb_addr);
2647 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2648 
2649 	/* start the ring */
2650 	gfx_v7_0_cp_gfx_start(adev);
2651 	r = amdgpu_ring_test_helper(ring);
2652 	if (r)
2653 		return r;
2654 
2655 	return 0;
2656 }
2657 
2658 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2659 {
2660 	return ring->adev->wb.wb[ring->rptr_offs];
2661 }
2662 
2663 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2664 {
2665 	struct amdgpu_device *adev = ring->adev;
2666 
2667 	return RREG32(mmCP_RB0_WPTR);
2668 }
2669 
2670 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2671 {
2672 	struct amdgpu_device *adev = ring->adev;
2673 
2674 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2675 	(void)RREG32(mmCP_RB0_WPTR);
2676 }
2677 
2678 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2679 {
2680 	/* XXX check if swapping is necessary on BE */
2681 	return ring->adev->wb.wb[ring->wptr_offs];
2682 }
2683 
2684 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2685 {
2686 	struct amdgpu_device *adev = ring->adev;
2687 
2688 	/* XXX check if swapping is necessary on BE */
2689 	adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2690 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2691 }
2692 
2693 /**
2694  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2695  *
2696  * @adev: amdgpu_device pointer
2697  * @enable: enable or disable the MEs
2698  *
2699  * Halts or unhalts the compute MEs.
2700  */
2701 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2702 {
2703 	int i;
2704 
2705 	if (enable) {
2706 		WREG32(mmCP_MEC_CNTL, 0);
2707 	} else {
2708 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2709 		for (i = 0; i < adev->gfx.num_compute_rings; i++)
2710 			adev->gfx.compute_ring[i].sched.ready = false;
2711 	}
2712 	udelay(50);
2713 }
2714 
2715 /**
2716  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2717  *
2718  * @adev: amdgpu_device pointer
2719  *
2720  * Loads the compute MEC1&2 ucode.
2721  * Returns 0 for success, -EINVAL if the ucode is not available.
2722  */
2723 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2724 {
2725 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2726 	const __le32 *fw_data;
2727 	unsigned i, fw_size;
2728 
2729 	if (!adev->gfx.mec_fw)
2730 		return -EINVAL;
2731 
2732 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2733 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2734 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2735 	adev->gfx.mec_feature_version = le32_to_cpu(
2736 					mec_hdr->ucode_feature_version);
2737 
2738 	gfx_v7_0_cp_compute_enable(adev, false);
2739 
2740 	/* MEC1 */
2741 	fw_data = (const __le32 *)
2742 		(adev->gfx.mec_fw->data +
2743 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2744 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2745 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2746 	for (i = 0; i < fw_size; i++)
2747 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2748 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2749 
2750 	if (adev->asic_type == CHIP_KAVERI) {
2751 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2752 
2753 		if (!adev->gfx.mec2_fw)
2754 			return -EINVAL;
2755 
2756 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2757 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2758 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2759 		adev->gfx.mec2_feature_version = le32_to_cpu(
2760 				mec2_hdr->ucode_feature_version);
2761 
2762 		/* MEC2 */
2763 		fw_data = (const __le32 *)
2764 			(adev->gfx.mec2_fw->data +
2765 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2766 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2767 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2768 		for (i = 0; i < fw_size; i++)
2769 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2770 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2771 	}
2772 
2773 	return 0;
2774 }
2775 
2776 /**
2777  * gfx_v7_0_cp_compute_fini - stop the compute queues
2778  *
2779  * @adev: amdgpu_device pointer
2780  *
2781  * Stop the compute queues and tear down the driver queue
2782  * info.
2783  */
2784 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2785 {
2786 	int i;
2787 
2788 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2789 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2790 
2791 		amdgpu_bo_free_kernel(&ring->mqd_obj, NULL, NULL);
2792 	}
2793 }
2794 
2795 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2796 {
2797 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
2798 }
2799 
2800 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2801 {
2802 	int r;
2803 	u32 *hpd;
2804 	size_t mec_hpd_size;
2805 
2806 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
2807 
2808 	/* take ownership of the relevant compute queues */
2809 	amdgpu_gfx_compute_queue_acquire(adev);
2810 
2811 	/* allocate space for ALL pipes (even the ones we don't own) */
2812 	mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec
2813 		* GFX7_MEC_HPD_SIZE * 2;
2814 
2815 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
2816 				      AMDGPU_GEM_DOMAIN_VRAM,
2817 				      &adev->gfx.mec.hpd_eop_obj,
2818 				      &adev->gfx.mec.hpd_eop_gpu_addr,
2819 				      (void **)&hpd);
2820 	if (r) {
2821 		dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r);
2822 		gfx_v7_0_mec_fini(adev);
2823 		return r;
2824 	}
2825 
2826 	/* clear memory.  Not sure if this is required or not */
2827 	memset(hpd, 0, mec_hpd_size);
2828 
2829 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2830 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2831 
2832 	return 0;
2833 }
2834 
2835 struct hqd_registers
2836 {
2837 	u32 cp_mqd_base_addr;
2838 	u32 cp_mqd_base_addr_hi;
2839 	u32 cp_hqd_active;
2840 	u32 cp_hqd_vmid;
2841 	u32 cp_hqd_persistent_state;
2842 	u32 cp_hqd_pipe_priority;
2843 	u32 cp_hqd_queue_priority;
2844 	u32 cp_hqd_quantum;
2845 	u32 cp_hqd_pq_base;
2846 	u32 cp_hqd_pq_base_hi;
2847 	u32 cp_hqd_pq_rptr;
2848 	u32 cp_hqd_pq_rptr_report_addr;
2849 	u32 cp_hqd_pq_rptr_report_addr_hi;
2850 	u32 cp_hqd_pq_wptr_poll_addr;
2851 	u32 cp_hqd_pq_wptr_poll_addr_hi;
2852 	u32 cp_hqd_pq_doorbell_control;
2853 	u32 cp_hqd_pq_wptr;
2854 	u32 cp_hqd_pq_control;
2855 	u32 cp_hqd_ib_base_addr;
2856 	u32 cp_hqd_ib_base_addr_hi;
2857 	u32 cp_hqd_ib_rptr;
2858 	u32 cp_hqd_ib_control;
2859 	u32 cp_hqd_iq_timer;
2860 	u32 cp_hqd_iq_rptr;
2861 	u32 cp_hqd_dequeue_request;
2862 	u32 cp_hqd_dma_offload;
2863 	u32 cp_hqd_sema_cmd;
2864 	u32 cp_hqd_msg_type;
2865 	u32 cp_hqd_atomic0_preop_lo;
2866 	u32 cp_hqd_atomic0_preop_hi;
2867 	u32 cp_hqd_atomic1_preop_lo;
2868 	u32 cp_hqd_atomic1_preop_hi;
2869 	u32 cp_hqd_hq_scheduler0;
2870 	u32 cp_hqd_hq_scheduler1;
2871 	u32 cp_mqd_control;
2872 };
2873 
2874 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev,
2875 				       int mec, int pipe)
2876 {
2877 	u64 eop_gpu_addr;
2878 	u32 tmp;
2879 	size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe)
2880 			    * GFX7_MEC_HPD_SIZE * 2;
2881 
2882 	mutex_lock(&adev->srbm_mutex);
2883 	eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset;
2884 
2885 	cik_srbm_select(adev, mec + 1, pipe, 0, 0);
2886 
2887 	/* write the EOP addr */
2888 	WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2889 	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2890 
2891 	/* set the VMID assigned */
2892 	WREG32(mmCP_HPD_EOP_VMID, 0);
2893 
2894 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2895 	tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2896 	tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2897 	tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8);
2898 	WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2899 
2900 	cik_srbm_select(adev, 0, 0, 0, 0);
2901 	mutex_unlock(&adev->srbm_mutex);
2902 }
2903 
2904 static int gfx_v7_0_mqd_deactivate(struct amdgpu_device *adev)
2905 {
2906 	int i;
2907 
2908 	/* disable the queue if it's active */
2909 	if (RREG32(mmCP_HQD_ACTIVE) & 1) {
2910 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
2911 		for (i = 0; i < adev->usec_timeout; i++) {
2912 			if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
2913 				break;
2914 			udelay(1);
2915 		}
2916 
2917 		if (i == adev->usec_timeout)
2918 			return -ETIMEDOUT;
2919 
2920 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
2921 		WREG32(mmCP_HQD_PQ_RPTR, 0);
2922 		WREG32(mmCP_HQD_PQ_WPTR, 0);
2923 	}
2924 
2925 	return 0;
2926 }
2927 
2928 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev,
2929 			     struct cik_mqd *mqd,
2930 			     uint64_t mqd_gpu_addr,
2931 			     struct amdgpu_ring *ring)
2932 {
2933 	u64 hqd_gpu_addr;
2934 	u64 wb_gpu_addr;
2935 
2936 	/* init the mqd struct */
2937 	memset(mqd, 0, sizeof(struct cik_mqd));
2938 
2939 	mqd->header = 0xC0310800;
2940 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2941 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2942 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2943 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2944 
2945 	/* enable doorbell? */
2946 	mqd->cp_hqd_pq_doorbell_control =
2947 		RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2948 	if (ring->use_doorbell)
2949 		mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2950 	else
2951 		mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2952 
2953 	/* set the pointer to the MQD */
2954 	mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc;
2955 	mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
2956 
2957 	/* set MQD vmid to 0 */
2958 	mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
2959 	mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
2960 
2961 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2962 	hqd_gpu_addr = ring->gpu_addr >> 8;
2963 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2964 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2965 
2966 	/* set up the HQD, this is similar to CP_RB0_CNTL */
2967 	mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
2968 	mqd->cp_hqd_pq_control &=
2969 		~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
2970 				CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
2971 
2972 	mqd->cp_hqd_pq_control |=
2973 		order_base_2(ring->ring_size / 8);
2974 	mqd->cp_hqd_pq_control |=
2975 		(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
2976 #ifdef __BIG_ENDIAN
2977 	mqd->cp_hqd_pq_control |=
2978 		2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
2979 #endif
2980 	mqd->cp_hqd_pq_control &=
2981 		~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
2982 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
2983 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
2984 	mqd->cp_hqd_pq_control |=
2985 		CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
2986 		CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
2987 
2988 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2989 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2990 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2991 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2992 
2993 	/* set the wb address wether it's enabled or not */
2994 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2995 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2996 	mqd->cp_hqd_pq_rptr_report_addr_hi =
2997 		upper_32_bits(wb_gpu_addr) & 0xffff;
2998 
2999 	/* enable the doorbell if requested */
3000 	if (ring->use_doorbell) {
3001 		mqd->cp_hqd_pq_doorbell_control =
3002 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
3003 		mqd->cp_hqd_pq_doorbell_control &=
3004 			~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
3005 		mqd->cp_hqd_pq_doorbell_control |=
3006 			(ring->doorbell_index <<
3007 			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
3008 		mqd->cp_hqd_pq_doorbell_control |=
3009 			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
3010 		mqd->cp_hqd_pq_doorbell_control &=
3011 			~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
3012 					CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
3013 
3014 	} else {
3015 		mqd->cp_hqd_pq_doorbell_control = 0;
3016 	}
3017 
3018 	/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3019 	ring->wptr = 0;
3020 	mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr);
3021 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
3022 
3023 	/* set the vmid for the queue */
3024 	mqd->cp_hqd_vmid = 0;
3025 
3026 	/* defaults */
3027 	mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL);
3028 	mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR);
3029 	mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI);
3030 	mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR);
3031 	mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE);
3032 	mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD);
3033 	mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE);
3034 	mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO);
3035 	mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI);
3036 	mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO);
3037 	mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI);
3038 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
3039 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
3040 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
3041 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
3042 	mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR);
3043 
3044 	/* activate the queue */
3045 	mqd->cp_hqd_active = 1;
3046 }
3047 
3048 int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd)
3049 {
3050 	uint32_t tmp;
3051 	uint32_t mqd_reg;
3052 	uint32_t *mqd_data;
3053 
3054 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */
3055 	mqd_data = &mqd->cp_mqd_base_addr_lo;
3056 
3057 	/* disable wptr polling */
3058 	tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
3059 	tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3060 	WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
3061 
3062 	/* program all HQD registers */
3063 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++)
3064 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3065 
3066 	/* activate the HQD */
3067 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
3068 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
3069 
3070 	return 0;
3071 }
3072 
3073 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id)
3074 {
3075 	int r;
3076 	u64 mqd_gpu_addr;
3077 	struct cik_mqd *mqd;
3078 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
3079 
3080 	r = amdgpu_bo_create_reserved(adev, sizeof(struct cik_mqd), PAGE_SIZE,
3081 				      AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj,
3082 				      &mqd_gpu_addr, (void **)&mqd);
3083 	if (r) {
3084 		dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
3085 		return r;
3086 	}
3087 
3088 	mutex_lock(&adev->srbm_mutex);
3089 	cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3090 
3091 	gfx_v7_0_mqd_init(adev, mqd, mqd_gpu_addr, ring);
3092 	gfx_v7_0_mqd_deactivate(adev);
3093 	gfx_v7_0_mqd_commit(adev, mqd);
3094 
3095 	cik_srbm_select(adev, 0, 0, 0, 0);
3096 	mutex_unlock(&adev->srbm_mutex);
3097 
3098 	amdgpu_bo_kunmap(ring->mqd_obj);
3099 	amdgpu_bo_unreserve(ring->mqd_obj);
3100 	return 0;
3101 }
3102 
3103 /**
3104  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
3105  *
3106  * @adev: amdgpu_device pointer
3107  *
3108  * Program the compute queues and test them to make sure they
3109  * are working.
3110  * Returns 0 for success, error for failure.
3111  */
3112 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
3113 {
3114 	int r, i, j;
3115 	u32 tmp;
3116 	struct amdgpu_ring *ring;
3117 
3118 	/* fix up chicken bits */
3119 	tmp = RREG32(mmCP_CPF_DEBUG);
3120 	tmp |= (1 << 23);
3121 	WREG32(mmCP_CPF_DEBUG, tmp);
3122 
3123 	/* init all pipes (even the ones we don't own) */
3124 	for (i = 0; i < adev->gfx.mec.num_mec; i++)
3125 		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++)
3126 			gfx_v7_0_compute_pipe_init(adev, i, j);
3127 
3128 	/* init the queues */
3129 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3130 		r = gfx_v7_0_compute_queue_init(adev, i);
3131 		if (r) {
3132 			gfx_v7_0_cp_compute_fini(adev);
3133 			return r;
3134 		}
3135 	}
3136 
3137 	gfx_v7_0_cp_compute_enable(adev, true);
3138 
3139 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3140 		ring = &adev->gfx.compute_ring[i];
3141 		amdgpu_ring_test_helper(ring);
3142 	}
3143 
3144 	return 0;
3145 }
3146 
3147 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3148 {
3149 	gfx_v7_0_cp_gfx_enable(adev, enable);
3150 	gfx_v7_0_cp_compute_enable(adev, enable);
3151 }
3152 
3153 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3154 {
3155 	int r;
3156 
3157 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3158 	if (r)
3159 		return r;
3160 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3161 	if (r)
3162 		return r;
3163 
3164 	return 0;
3165 }
3166 
3167 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3168 					       bool enable)
3169 {
3170 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3171 
3172 	if (enable)
3173 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3174 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3175 	else
3176 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3177 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3178 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3179 }
3180 
3181 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3182 {
3183 	int r;
3184 
3185 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3186 
3187 	r = gfx_v7_0_cp_load_microcode(adev);
3188 	if (r)
3189 		return r;
3190 
3191 	r = gfx_v7_0_cp_gfx_resume(adev);
3192 	if (r)
3193 		return r;
3194 	r = gfx_v7_0_cp_compute_resume(adev);
3195 	if (r)
3196 		return r;
3197 
3198 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3199 
3200 	return 0;
3201 }
3202 
3203 /**
3204  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3205  *
3206  * @ring: the ring to emmit the commands to
3207  *
3208  * Sync the command pipeline with the PFP. E.g. wait for everything
3209  * to be completed.
3210  */
3211 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3212 {
3213 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3214 	uint32_t seq = ring->fence_drv.sync_seq;
3215 	uint64_t addr = ring->fence_drv.gpu_addr;
3216 
3217 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3218 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3219 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3220 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3221 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3222 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3223 	amdgpu_ring_write(ring, seq);
3224 	amdgpu_ring_write(ring, 0xffffffff);
3225 	amdgpu_ring_write(ring, 4); /* poll interval */
3226 
3227 	if (usepfp) {
3228 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3229 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3230 		amdgpu_ring_write(ring, 0);
3231 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3232 		amdgpu_ring_write(ring, 0);
3233 	}
3234 }
3235 
3236 /*
3237  * vm
3238  * VMID 0 is the physical GPU addresses as used by the kernel.
3239  * VMIDs 1-15 are used for userspace clients and are handled
3240  * by the amdgpu vm/hsa code.
3241  */
3242 /**
3243  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3244  *
3245  * @adev: amdgpu_device pointer
3246  *
3247  * Update the page table base and flush the VM TLB
3248  * using the CP (CIK).
3249  */
3250 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3251 					unsigned vmid, uint64_t pd_addr)
3252 {
3253 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3254 
3255 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3256 
3257 	/* wait for the invalidate to complete */
3258 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3259 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3260 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3261 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3262 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3263 	amdgpu_ring_write(ring, 0);
3264 	amdgpu_ring_write(ring, 0); /* ref */
3265 	amdgpu_ring_write(ring, 0); /* mask */
3266 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3267 
3268 	/* compute doesn't have PFP */
3269 	if (usepfp) {
3270 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3271 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3272 		amdgpu_ring_write(ring, 0x0);
3273 
3274 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3275 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3276 		amdgpu_ring_write(ring, 0);
3277 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3278 		amdgpu_ring_write(ring, 0);
3279 	}
3280 }
3281 
3282 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
3283 				    uint32_t reg, uint32_t val)
3284 {
3285 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3286 
3287 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3288 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3289 				 WRITE_DATA_DST_SEL(0)));
3290 	amdgpu_ring_write(ring, reg);
3291 	amdgpu_ring_write(ring, 0);
3292 	amdgpu_ring_write(ring, val);
3293 }
3294 
3295 /*
3296  * RLC
3297  * The RLC is a multi-purpose microengine that handles a
3298  * variety of functions.
3299  */
3300 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3301 {
3302 	const u32 *src_ptr;
3303 	u32 dws;
3304 	const struct cs_section_def *cs_data;
3305 	int r;
3306 
3307 	/* allocate rlc buffers */
3308 	if (adev->flags & AMD_IS_APU) {
3309 		if (adev->asic_type == CHIP_KAVERI) {
3310 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3311 			adev->gfx.rlc.reg_list_size =
3312 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3313 		} else {
3314 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3315 			adev->gfx.rlc.reg_list_size =
3316 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3317 		}
3318 	}
3319 	adev->gfx.rlc.cs_data = ci_cs_data;
3320 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3321 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3322 
3323 	src_ptr = adev->gfx.rlc.reg_list;
3324 	dws = adev->gfx.rlc.reg_list_size;
3325 	dws += (5 * 16) + 48 + 48 + 64;
3326 
3327 	cs_data = adev->gfx.rlc.cs_data;
3328 
3329 	if (src_ptr) {
3330 		/* init save restore block */
3331 		r = amdgpu_gfx_rlc_init_sr(adev, dws);
3332 		if (r)
3333 			return r;
3334 	}
3335 
3336 	if (cs_data) {
3337 		/* init clear state block */
3338 		r = amdgpu_gfx_rlc_init_csb(adev);
3339 		if (r)
3340 			return r;
3341 	}
3342 
3343 	if (adev->gfx.rlc.cp_table_size) {
3344 		r = amdgpu_gfx_rlc_init_cpt(adev);
3345 		if (r)
3346 			return r;
3347 	}
3348 
3349 	return 0;
3350 }
3351 
3352 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3353 {
3354 	u32 tmp;
3355 
3356 	tmp = RREG32(mmRLC_LB_CNTL);
3357 	if (enable)
3358 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3359 	else
3360 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3361 	WREG32(mmRLC_LB_CNTL, tmp);
3362 }
3363 
3364 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3365 {
3366 	u32 i, j, k;
3367 	u32 mask;
3368 
3369 	mutex_lock(&adev->grbm_idx_mutex);
3370 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3371 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3372 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
3373 			for (k = 0; k < adev->usec_timeout; k++) {
3374 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3375 					break;
3376 				udelay(1);
3377 			}
3378 		}
3379 	}
3380 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3381 	mutex_unlock(&adev->grbm_idx_mutex);
3382 
3383 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3384 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3385 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3386 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3387 	for (k = 0; k < adev->usec_timeout; k++) {
3388 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3389 			break;
3390 		udelay(1);
3391 	}
3392 }
3393 
3394 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3395 {
3396 	u32 tmp;
3397 
3398 	tmp = RREG32(mmRLC_CNTL);
3399 	if (tmp != rlc)
3400 		WREG32(mmRLC_CNTL, rlc);
3401 }
3402 
3403 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3404 {
3405 	u32 data, orig;
3406 
3407 	orig = data = RREG32(mmRLC_CNTL);
3408 
3409 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3410 		u32 i;
3411 
3412 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3413 		WREG32(mmRLC_CNTL, data);
3414 
3415 		for (i = 0; i < adev->usec_timeout; i++) {
3416 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3417 				break;
3418 			udelay(1);
3419 		}
3420 
3421 		gfx_v7_0_wait_for_rlc_serdes(adev);
3422 	}
3423 
3424 	return orig;
3425 }
3426 
3427 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev)
3428 {
3429 	return true;
3430 }
3431 
3432 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev)
3433 {
3434 	u32 tmp, i, mask;
3435 
3436 	tmp = 0x1 | (1 << 1);
3437 	WREG32(mmRLC_GPR_REG2, tmp);
3438 
3439 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3440 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3441 	for (i = 0; i < adev->usec_timeout; i++) {
3442 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3443 			break;
3444 		udelay(1);
3445 	}
3446 
3447 	for (i = 0; i < adev->usec_timeout; i++) {
3448 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3449 			break;
3450 		udelay(1);
3451 	}
3452 }
3453 
3454 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev)
3455 {
3456 	u32 tmp;
3457 
3458 	tmp = 0x1 | (0 << 1);
3459 	WREG32(mmRLC_GPR_REG2, tmp);
3460 }
3461 
3462 /**
3463  * gfx_v7_0_rlc_stop - stop the RLC ME
3464  *
3465  * @adev: amdgpu_device pointer
3466  *
3467  * Halt the RLC ME (MicroEngine) (CIK).
3468  */
3469 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3470 {
3471 	WREG32(mmRLC_CNTL, 0);
3472 
3473 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3474 
3475 	gfx_v7_0_wait_for_rlc_serdes(adev);
3476 }
3477 
3478 /**
3479  * gfx_v7_0_rlc_start - start the RLC ME
3480  *
3481  * @adev: amdgpu_device pointer
3482  *
3483  * Unhalt the RLC ME (MicroEngine) (CIK).
3484  */
3485 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3486 {
3487 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3488 
3489 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3490 
3491 	udelay(50);
3492 }
3493 
3494 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3495 {
3496 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3497 
3498 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3499 	WREG32(mmGRBM_SOFT_RESET, tmp);
3500 	udelay(50);
3501 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3502 	WREG32(mmGRBM_SOFT_RESET, tmp);
3503 	udelay(50);
3504 }
3505 
3506 /**
3507  * gfx_v7_0_rlc_resume - setup the RLC hw
3508  *
3509  * @adev: amdgpu_device pointer
3510  *
3511  * Initialize the RLC registers, load the ucode,
3512  * and start the RLC (CIK).
3513  * Returns 0 for success, -EINVAL if the ucode is not available.
3514  */
3515 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3516 {
3517 	const struct rlc_firmware_header_v1_0 *hdr;
3518 	const __le32 *fw_data;
3519 	unsigned i, fw_size;
3520 	u32 tmp;
3521 
3522 	if (!adev->gfx.rlc_fw)
3523 		return -EINVAL;
3524 
3525 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3526 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3527 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3528 	adev->gfx.rlc_feature_version = le32_to_cpu(
3529 					hdr->ucode_feature_version);
3530 
3531 	adev->gfx.rlc.funcs->stop(adev);
3532 
3533 	/* disable CG */
3534 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3535 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3536 
3537 	adev->gfx.rlc.funcs->reset(adev);
3538 
3539 	gfx_v7_0_init_pg(adev);
3540 
3541 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3542 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3543 
3544 	mutex_lock(&adev->grbm_idx_mutex);
3545 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3546 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3547 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3548 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3549 	mutex_unlock(&adev->grbm_idx_mutex);
3550 
3551 	WREG32(mmRLC_MC_CNTL, 0);
3552 	WREG32(mmRLC_UCODE_CNTL, 0);
3553 
3554 	fw_data = (const __le32 *)
3555 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3556 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3557 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3558 	for (i = 0; i < fw_size; i++)
3559 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3560 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3561 
3562 	/* XXX - find out what chips support lbpw */
3563 	gfx_v7_0_enable_lbpw(adev, false);
3564 
3565 	if (adev->asic_type == CHIP_BONAIRE)
3566 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3567 
3568 	adev->gfx.rlc.funcs->start(adev);
3569 
3570 	return 0;
3571 }
3572 
3573 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3574 {
3575 	u32 data, orig, tmp, tmp2;
3576 
3577 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3578 
3579 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3580 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3581 
3582 		tmp = gfx_v7_0_halt_rlc(adev);
3583 
3584 		mutex_lock(&adev->grbm_idx_mutex);
3585 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3586 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3587 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3588 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3589 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3590 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3591 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3592 		mutex_unlock(&adev->grbm_idx_mutex);
3593 
3594 		gfx_v7_0_update_rlc(adev, tmp);
3595 
3596 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3597 		if (orig != data)
3598 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3599 
3600 	} else {
3601 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3602 
3603 		RREG32(mmCB_CGTT_SCLK_CTRL);
3604 		RREG32(mmCB_CGTT_SCLK_CTRL);
3605 		RREG32(mmCB_CGTT_SCLK_CTRL);
3606 		RREG32(mmCB_CGTT_SCLK_CTRL);
3607 
3608 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3609 		if (orig != data)
3610 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3611 
3612 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3613 	}
3614 }
3615 
3616 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3617 {
3618 	u32 data, orig, tmp = 0;
3619 
3620 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3621 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3622 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3623 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3624 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3625 				if (orig != data)
3626 					WREG32(mmCP_MEM_SLP_CNTL, data);
3627 			}
3628 		}
3629 
3630 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3631 		data |= 0x00000001;
3632 		data &= 0xfffffffd;
3633 		if (orig != data)
3634 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3635 
3636 		tmp = gfx_v7_0_halt_rlc(adev);
3637 
3638 		mutex_lock(&adev->grbm_idx_mutex);
3639 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3640 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3641 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3642 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3643 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3644 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3645 		mutex_unlock(&adev->grbm_idx_mutex);
3646 
3647 		gfx_v7_0_update_rlc(adev, tmp);
3648 
3649 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3650 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3651 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3652 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3653 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3654 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3655 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3656 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3657 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3658 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3659 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3660 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3661 			if (orig != data)
3662 				WREG32(mmCGTS_SM_CTRL_REG, data);
3663 		}
3664 	} else {
3665 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3666 		data |= 0x00000003;
3667 		if (orig != data)
3668 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3669 
3670 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3671 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3672 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3673 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3674 		}
3675 
3676 		data = RREG32(mmCP_MEM_SLP_CNTL);
3677 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3678 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3679 			WREG32(mmCP_MEM_SLP_CNTL, data);
3680 		}
3681 
3682 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3683 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3684 		if (orig != data)
3685 			WREG32(mmCGTS_SM_CTRL_REG, data);
3686 
3687 		tmp = gfx_v7_0_halt_rlc(adev);
3688 
3689 		mutex_lock(&adev->grbm_idx_mutex);
3690 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3691 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3692 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3693 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3694 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3695 		mutex_unlock(&adev->grbm_idx_mutex);
3696 
3697 		gfx_v7_0_update_rlc(adev, tmp);
3698 	}
3699 }
3700 
3701 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3702 			       bool enable)
3703 {
3704 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3705 	/* order matters! */
3706 	if (enable) {
3707 		gfx_v7_0_enable_mgcg(adev, true);
3708 		gfx_v7_0_enable_cgcg(adev, true);
3709 	} else {
3710 		gfx_v7_0_enable_cgcg(adev, false);
3711 		gfx_v7_0_enable_mgcg(adev, false);
3712 	}
3713 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3714 }
3715 
3716 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3717 						bool enable)
3718 {
3719 	u32 data, orig;
3720 
3721 	orig = data = RREG32(mmRLC_PG_CNTL);
3722 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3723 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3724 	else
3725 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3726 	if (orig != data)
3727 		WREG32(mmRLC_PG_CNTL, data);
3728 }
3729 
3730 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3731 						bool enable)
3732 {
3733 	u32 data, orig;
3734 
3735 	orig = data = RREG32(mmRLC_PG_CNTL);
3736 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3737 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3738 	else
3739 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3740 	if (orig != data)
3741 		WREG32(mmRLC_PG_CNTL, data);
3742 }
3743 
3744 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3745 {
3746 	u32 data, orig;
3747 
3748 	orig = data = RREG32(mmRLC_PG_CNTL);
3749 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3750 		data &= ~0x8000;
3751 	else
3752 		data |= 0x8000;
3753 	if (orig != data)
3754 		WREG32(mmRLC_PG_CNTL, data);
3755 }
3756 
3757 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3758 {
3759 	u32 data, orig;
3760 
3761 	orig = data = RREG32(mmRLC_PG_CNTL);
3762 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3763 		data &= ~0x2000;
3764 	else
3765 		data |= 0x2000;
3766 	if (orig != data)
3767 		WREG32(mmRLC_PG_CNTL, data);
3768 }
3769 
3770 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev)
3771 {
3772 	if (adev->asic_type == CHIP_KAVERI)
3773 		return 5;
3774 	else
3775 		return 4;
3776 }
3777 
3778 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
3779 				     bool enable)
3780 {
3781 	u32 data, orig;
3782 
3783 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
3784 		orig = data = RREG32(mmRLC_PG_CNTL);
3785 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3786 		if (orig != data)
3787 			WREG32(mmRLC_PG_CNTL, data);
3788 
3789 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3790 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3791 		if (orig != data)
3792 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3793 	} else {
3794 		orig = data = RREG32(mmRLC_PG_CNTL);
3795 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3796 		if (orig != data)
3797 			WREG32(mmRLC_PG_CNTL, data);
3798 
3799 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3800 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3801 		if (orig != data)
3802 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3803 
3804 		data = RREG32(mmDB_RENDER_CONTROL);
3805 	}
3806 }
3807 
3808 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
3809 						 u32 bitmap)
3810 {
3811 	u32 data;
3812 
3813 	if (!bitmap)
3814 		return;
3815 
3816 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3817 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3818 
3819 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
3820 }
3821 
3822 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
3823 {
3824 	u32 data, mask;
3825 
3826 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
3827 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
3828 
3829 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3830 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3831 
3832 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
3833 
3834 	return (~data) & mask;
3835 }
3836 
3837 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
3838 {
3839 	u32 tmp;
3840 
3841 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
3842 
3843 	tmp = RREG32(mmRLC_MAX_PG_CU);
3844 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
3845 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
3846 	WREG32(mmRLC_MAX_PG_CU, tmp);
3847 }
3848 
3849 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
3850 					    bool enable)
3851 {
3852 	u32 data, orig;
3853 
3854 	orig = data = RREG32(mmRLC_PG_CNTL);
3855 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
3856 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3857 	else
3858 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3859 	if (orig != data)
3860 		WREG32(mmRLC_PG_CNTL, data);
3861 }
3862 
3863 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
3864 					     bool enable)
3865 {
3866 	u32 data, orig;
3867 
3868 	orig = data = RREG32(mmRLC_PG_CNTL);
3869 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
3870 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3871 	else
3872 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3873 	if (orig != data)
3874 		WREG32(mmRLC_PG_CNTL, data);
3875 }
3876 
3877 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
3878 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
3879 
3880 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
3881 {
3882 	u32 data, orig;
3883 	u32 i;
3884 
3885 	if (adev->gfx.rlc.cs_data) {
3886 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3887 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3888 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3889 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
3890 	} else {
3891 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3892 		for (i = 0; i < 3; i++)
3893 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
3894 	}
3895 	if (adev->gfx.rlc.reg_list) {
3896 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
3897 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3898 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
3899 	}
3900 
3901 	orig = data = RREG32(mmRLC_PG_CNTL);
3902 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
3903 	if (orig != data)
3904 		WREG32(mmRLC_PG_CNTL, data);
3905 
3906 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
3907 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
3908 
3909 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
3910 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
3911 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3912 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
3913 
3914 	data = 0x10101010;
3915 	WREG32(mmRLC_PG_DELAY, data);
3916 
3917 	data = RREG32(mmRLC_PG_DELAY_2);
3918 	data &= ~0xff;
3919 	data |= 0x3;
3920 	WREG32(mmRLC_PG_DELAY_2, data);
3921 
3922 	data = RREG32(mmRLC_AUTO_PG_CTRL);
3923 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
3924 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
3925 	WREG32(mmRLC_AUTO_PG_CTRL, data);
3926 
3927 }
3928 
3929 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
3930 {
3931 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
3932 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
3933 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
3934 }
3935 
3936 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
3937 {
3938 	u32 count = 0;
3939 	const struct cs_section_def *sect = NULL;
3940 	const struct cs_extent_def *ext = NULL;
3941 
3942 	if (adev->gfx.rlc.cs_data == NULL)
3943 		return 0;
3944 
3945 	/* begin clear state */
3946 	count += 2;
3947 	/* context control state */
3948 	count += 3;
3949 
3950 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3951 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3952 			if (sect->id == SECT_CONTEXT)
3953 				count += 2 + ext->reg_count;
3954 			else
3955 				return 0;
3956 		}
3957 	}
3958 	/* pa_sc_raster_config/pa_sc_raster_config1 */
3959 	count += 4;
3960 	/* end clear state */
3961 	count += 2;
3962 	/* clear state */
3963 	count += 2;
3964 
3965 	return count;
3966 }
3967 
3968 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
3969 				    volatile u32 *buffer)
3970 {
3971 	u32 count = 0, i;
3972 	const struct cs_section_def *sect = NULL;
3973 	const struct cs_extent_def *ext = NULL;
3974 
3975 	if (adev->gfx.rlc.cs_data == NULL)
3976 		return;
3977 	if (buffer == NULL)
3978 		return;
3979 
3980 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3981 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3982 
3983 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3984 	buffer[count++] = cpu_to_le32(0x80000000);
3985 	buffer[count++] = cpu_to_le32(0x80000000);
3986 
3987 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3988 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3989 			if (sect->id == SECT_CONTEXT) {
3990 				buffer[count++] =
3991 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
3992 				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3993 				for (i = 0; i < ext->reg_count; i++)
3994 					buffer[count++] = cpu_to_le32(ext->extent[i]);
3995 			} else {
3996 				return;
3997 			}
3998 		}
3999 	}
4000 
4001 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4002 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4003 	switch (adev->asic_type) {
4004 	case CHIP_BONAIRE:
4005 		buffer[count++] = cpu_to_le32(0x16000012);
4006 		buffer[count++] = cpu_to_le32(0x00000000);
4007 		break;
4008 	case CHIP_KAVERI:
4009 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4010 		buffer[count++] = cpu_to_le32(0x00000000);
4011 		break;
4012 	case CHIP_KABINI:
4013 	case CHIP_MULLINS:
4014 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
4015 		buffer[count++] = cpu_to_le32(0x00000000);
4016 		break;
4017 	case CHIP_HAWAII:
4018 		buffer[count++] = cpu_to_le32(0x3a00161a);
4019 		buffer[count++] = cpu_to_le32(0x0000002e);
4020 		break;
4021 	default:
4022 		buffer[count++] = cpu_to_le32(0x00000000);
4023 		buffer[count++] = cpu_to_le32(0x00000000);
4024 		break;
4025 	}
4026 
4027 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4028 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
4029 
4030 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
4031 	buffer[count++] = cpu_to_le32(0);
4032 }
4033 
4034 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
4035 {
4036 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4037 			      AMD_PG_SUPPORT_GFX_SMG |
4038 			      AMD_PG_SUPPORT_GFX_DMG |
4039 			      AMD_PG_SUPPORT_CP |
4040 			      AMD_PG_SUPPORT_GDS |
4041 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4042 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
4043 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
4044 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4045 			gfx_v7_0_init_gfx_cgpg(adev);
4046 			gfx_v7_0_enable_cp_pg(adev, true);
4047 			gfx_v7_0_enable_gds_pg(adev, true);
4048 		}
4049 		gfx_v7_0_init_ao_cu_mask(adev);
4050 		gfx_v7_0_update_gfx_pg(adev, true);
4051 	}
4052 }
4053 
4054 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4055 {
4056 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4057 			      AMD_PG_SUPPORT_GFX_SMG |
4058 			      AMD_PG_SUPPORT_GFX_DMG |
4059 			      AMD_PG_SUPPORT_CP |
4060 			      AMD_PG_SUPPORT_GDS |
4061 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4062 		gfx_v7_0_update_gfx_pg(adev, false);
4063 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4064 			gfx_v7_0_enable_cp_pg(adev, false);
4065 			gfx_v7_0_enable_gds_pg(adev, false);
4066 		}
4067 	}
4068 }
4069 
4070 /**
4071  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4072  *
4073  * @adev: amdgpu_device pointer
4074  *
4075  * Fetches a GPU clock counter snapshot (SI).
4076  * Returns the 64 bit clock counter snapshot.
4077  */
4078 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4079 {
4080 	uint64_t clock;
4081 
4082 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4083 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4084 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4085 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4086 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4087 	return clock;
4088 }
4089 
4090 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4091 					  uint32_t vmid,
4092 					  uint32_t gds_base, uint32_t gds_size,
4093 					  uint32_t gws_base, uint32_t gws_size,
4094 					  uint32_t oa_base, uint32_t oa_size)
4095 {
4096 	/* GDS Base */
4097 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4098 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4099 				WRITE_DATA_DST_SEL(0)));
4100 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4101 	amdgpu_ring_write(ring, 0);
4102 	amdgpu_ring_write(ring, gds_base);
4103 
4104 	/* GDS Size */
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_size);
4109 	amdgpu_ring_write(ring, 0);
4110 	amdgpu_ring_write(ring, gds_size);
4111 
4112 	/* GWS */
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].gws);
4117 	amdgpu_ring_write(ring, 0);
4118 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4119 
4120 	/* OA */
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].oa);
4125 	amdgpu_ring_write(ring, 0);
4126 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4127 }
4128 
4129 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
4130 {
4131 	struct amdgpu_device *adev = ring->adev;
4132 	uint32_t value = 0;
4133 
4134 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
4135 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
4136 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
4137 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
4138 	WREG32(mmSQ_CMD, value);
4139 }
4140 
4141 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4142 {
4143 	WREG32(mmSQ_IND_INDEX,
4144 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4145 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4146 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4147 		(SQ_IND_INDEX__FORCE_READ_MASK));
4148 	return RREG32(mmSQ_IND_DATA);
4149 }
4150 
4151 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4152 			   uint32_t wave, uint32_t thread,
4153 			   uint32_t regno, uint32_t num, uint32_t *out)
4154 {
4155 	WREG32(mmSQ_IND_INDEX,
4156 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4157 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4158 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4159 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4160 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4161 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4162 	while (num--)
4163 		*(out++) = RREG32(mmSQ_IND_DATA);
4164 }
4165 
4166 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4167 {
4168 	/* type 0 wave data */
4169 	dst[(*no_fields)++] = 0;
4170 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4171 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4172 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4173 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4174 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4175 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4176 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4177 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4178 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4179 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4180 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4181 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4182 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4183 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4184 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4185 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4186 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4187 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4188 }
4189 
4190 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
4191 				     uint32_t wave, uint32_t start,
4192 				     uint32_t size, uint32_t *dst)
4193 {
4194 	wave_read_regs(
4195 		adev, simd, wave, 0,
4196 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4197 }
4198 
4199 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
4200 				  u32 me, u32 pipe, u32 q, u32 vm)
4201 {
4202 	cik_srbm_select(adev, me, pipe, q, vm);
4203 }
4204 
4205 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4206 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4207 	.select_se_sh = &gfx_v7_0_select_se_sh,
4208 	.read_wave_data = &gfx_v7_0_read_wave_data,
4209 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4210 	.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q
4211 };
4212 
4213 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4214 	.is_rlc_enabled = gfx_v7_0_is_rlc_enabled,
4215 	.set_safe_mode = gfx_v7_0_set_safe_mode,
4216 	.unset_safe_mode = gfx_v7_0_unset_safe_mode,
4217 	.init = gfx_v7_0_rlc_init,
4218 	.get_csb_size = gfx_v7_0_get_csb_size,
4219 	.get_csb_buffer = gfx_v7_0_get_csb_buffer,
4220 	.get_cp_table_num = gfx_v7_0_cp_pg_table_num,
4221 	.resume = gfx_v7_0_rlc_resume,
4222 	.stop = gfx_v7_0_rlc_stop,
4223 	.reset = gfx_v7_0_rlc_reset,
4224 	.start = gfx_v7_0_rlc_start
4225 };
4226 
4227 static int gfx_v7_0_early_init(void *handle)
4228 {
4229 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4230 
4231 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4232 	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
4233 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4234 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4235 	gfx_v7_0_set_ring_funcs(adev);
4236 	gfx_v7_0_set_irq_funcs(adev);
4237 	gfx_v7_0_set_gds_init(adev);
4238 
4239 	return 0;
4240 }
4241 
4242 static int gfx_v7_0_late_init(void *handle)
4243 {
4244 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4245 	int r;
4246 
4247 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4248 	if (r)
4249 		return r;
4250 
4251 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4252 	if (r)
4253 		return r;
4254 
4255 	return 0;
4256 }
4257 
4258 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4259 {
4260 	u32 gb_addr_config;
4261 	u32 mc_shared_chmap, mc_arb_ramcfg;
4262 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4263 	u32 tmp;
4264 
4265 	switch (adev->asic_type) {
4266 	case CHIP_BONAIRE:
4267 		adev->gfx.config.max_shader_engines = 2;
4268 		adev->gfx.config.max_tile_pipes = 4;
4269 		adev->gfx.config.max_cu_per_sh = 7;
4270 		adev->gfx.config.max_sh_per_se = 1;
4271 		adev->gfx.config.max_backends_per_se = 2;
4272 		adev->gfx.config.max_texture_channel_caches = 4;
4273 		adev->gfx.config.max_gprs = 256;
4274 		adev->gfx.config.max_gs_threads = 32;
4275 		adev->gfx.config.max_hw_contexts = 8;
4276 
4277 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4278 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4279 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4280 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4281 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4282 		break;
4283 	case CHIP_HAWAII:
4284 		adev->gfx.config.max_shader_engines = 4;
4285 		adev->gfx.config.max_tile_pipes = 16;
4286 		adev->gfx.config.max_cu_per_sh = 11;
4287 		adev->gfx.config.max_sh_per_se = 1;
4288 		adev->gfx.config.max_backends_per_se = 4;
4289 		adev->gfx.config.max_texture_channel_caches = 16;
4290 		adev->gfx.config.max_gprs = 256;
4291 		adev->gfx.config.max_gs_threads = 32;
4292 		adev->gfx.config.max_hw_contexts = 8;
4293 
4294 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4295 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4296 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4297 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4298 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4299 		break;
4300 	case CHIP_KAVERI:
4301 		adev->gfx.config.max_shader_engines = 1;
4302 		adev->gfx.config.max_tile_pipes = 4;
4303 		adev->gfx.config.max_cu_per_sh = 8;
4304 		adev->gfx.config.max_backends_per_se = 2;
4305 		adev->gfx.config.max_sh_per_se = 1;
4306 		adev->gfx.config.max_texture_channel_caches = 4;
4307 		adev->gfx.config.max_gprs = 256;
4308 		adev->gfx.config.max_gs_threads = 16;
4309 		adev->gfx.config.max_hw_contexts = 8;
4310 
4311 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4312 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4313 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4314 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4315 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4316 		break;
4317 	case CHIP_KABINI:
4318 	case CHIP_MULLINS:
4319 	default:
4320 		adev->gfx.config.max_shader_engines = 1;
4321 		adev->gfx.config.max_tile_pipes = 2;
4322 		adev->gfx.config.max_cu_per_sh = 2;
4323 		adev->gfx.config.max_sh_per_se = 1;
4324 		adev->gfx.config.max_backends_per_se = 1;
4325 		adev->gfx.config.max_texture_channel_caches = 2;
4326 		adev->gfx.config.max_gprs = 256;
4327 		adev->gfx.config.max_gs_threads = 16;
4328 		adev->gfx.config.max_hw_contexts = 8;
4329 
4330 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4331 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4332 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4333 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4334 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4335 		break;
4336 	}
4337 
4338 	mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP);
4339 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4340 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4341 
4342 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4343 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4344 	if (adev->flags & AMD_IS_APU) {
4345 		/* Get memory bank mapping mode. */
4346 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4347 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4348 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4349 
4350 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4351 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4352 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4353 
4354 		/* Validate settings in case only one DIMM installed. */
4355 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4356 			dimm00_addr_map = 0;
4357 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4358 			dimm01_addr_map = 0;
4359 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4360 			dimm10_addr_map = 0;
4361 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4362 			dimm11_addr_map = 0;
4363 
4364 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4365 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4366 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4367 			adev->gfx.config.mem_row_size_in_kb = 2;
4368 		else
4369 			adev->gfx.config.mem_row_size_in_kb = 1;
4370 	} else {
4371 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4372 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4373 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4374 			adev->gfx.config.mem_row_size_in_kb = 4;
4375 	}
4376 	/* XXX use MC settings? */
4377 	adev->gfx.config.shader_engine_tile_size = 32;
4378 	adev->gfx.config.num_gpus = 1;
4379 	adev->gfx.config.multi_gpu_tile_size = 64;
4380 
4381 	/* fix up row size */
4382 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4383 	switch (adev->gfx.config.mem_row_size_in_kb) {
4384 	case 1:
4385 	default:
4386 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4387 		break;
4388 	case 2:
4389 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4390 		break;
4391 	case 4:
4392 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4393 		break;
4394 	}
4395 	adev->gfx.config.gb_addr_config = gb_addr_config;
4396 }
4397 
4398 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
4399 					int mec, int pipe, int queue)
4400 {
4401 	int r;
4402 	unsigned irq_type;
4403 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
4404 
4405 	/* mec0 is me1 */
4406 	ring->me = mec + 1;
4407 	ring->pipe = pipe;
4408 	ring->queue = queue;
4409 
4410 	ring->ring_obj = NULL;
4411 	ring->use_doorbell = true;
4412 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
4413 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4414 
4415 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
4416 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
4417 		+ ring->pipe;
4418 
4419 	/* type-2 packets are deprecated on MEC, use type-3 instead */
4420 	r = amdgpu_ring_init(adev, ring, 1024,
4421 			&adev->gfx.eop_irq, irq_type);
4422 	if (r)
4423 		return r;
4424 
4425 
4426 	return 0;
4427 }
4428 
4429 static int gfx_v7_0_sw_init(void *handle)
4430 {
4431 	struct amdgpu_ring *ring;
4432 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4433 	int i, j, k, r, ring_id;
4434 
4435 	switch (adev->asic_type) {
4436 	case CHIP_KAVERI:
4437 		adev->gfx.mec.num_mec = 2;
4438 		break;
4439 	case CHIP_BONAIRE:
4440 	case CHIP_HAWAII:
4441 	case CHIP_KABINI:
4442 	case CHIP_MULLINS:
4443 	default:
4444 		adev->gfx.mec.num_mec = 1;
4445 		break;
4446 	}
4447 	adev->gfx.mec.num_pipe_per_mec = 4;
4448 	adev->gfx.mec.num_queue_per_pipe = 8;
4449 
4450 	/* EOP Event */
4451 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq);
4452 	if (r)
4453 		return r;
4454 
4455 	/* Privileged reg */
4456 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184,
4457 			      &adev->gfx.priv_reg_irq);
4458 	if (r)
4459 		return r;
4460 
4461 	/* Privileged inst */
4462 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185,
4463 			      &adev->gfx.priv_inst_irq);
4464 	if (r)
4465 		return r;
4466 
4467 	gfx_v7_0_scratch_init(adev);
4468 
4469 	r = gfx_v7_0_init_microcode(adev);
4470 	if (r) {
4471 		DRM_ERROR("Failed to load gfx firmware!\n");
4472 		return r;
4473 	}
4474 
4475 	r = adev->gfx.rlc.funcs->init(adev);
4476 	if (r) {
4477 		DRM_ERROR("Failed to init rlc BOs!\n");
4478 		return r;
4479 	}
4480 
4481 	/* allocate mec buffers */
4482 	r = gfx_v7_0_mec_init(adev);
4483 	if (r) {
4484 		DRM_ERROR("Failed to init MEC BOs!\n");
4485 		return r;
4486 	}
4487 
4488 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4489 		ring = &adev->gfx.gfx_ring[i];
4490 		ring->ring_obj = NULL;
4491 		sprintf(ring->name, "gfx");
4492 		r = amdgpu_ring_init(adev, ring, 1024,
4493 				     &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP);
4494 		if (r)
4495 			return r;
4496 	}
4497 
4498 	/* set up the compute queues - allocate horizontally across pipes */
4499 	ring_id = 0;
4500 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4501 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4502 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4503 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
4504 					continue;
4505 
4506 				r = gfx_v7_0_compute_ring_init(adev,
4507 								ring_id,
4508 								i, k, j);
4509 				if (r)
4510 					return r;
4511 
4512 				ring_id++;
4513 			}
4514 		}
4515 	}
4516 
4517 	adev->gfx.ce_ram_size = 0x8000;
4518 
4519 	gfx_v7_0_gpu_early_init(adev);
4520 
4521 	return r;
4522 }
4523 
4524 static int gfx_v7_0_sw_fini(void *handle)
4525 {
4526 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4527 	int i;
4528 
4529 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4530 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4531 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
4532 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4533 
4534 	gfx_v7_0_cp_compute_fini(adev);
4535 	amdgpu_gfx_rlc_fini(adev);
4536 	gfx_v7_0_mec_fini(adev);
4537 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
4538 				&adev->gfx.rlc.clear_state_gpu_addr,
4539 				(void **)&adev->gfx.rlc.cs_ptr);
4540 	if (adev->gfx.rlc.cp_table_size) {
4541 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
4542 				&adev->gfx.rlc.cp_table_gpu_addr,
4543 				(void **)&adev->gfx.rlc.cp_table_ptr);
4544 	}
4545 	gfx_v7_0_free_microcode(adev);
4546 
4547 	return 0;
4548 }
4549 
4550 static int gfx_v7_0_hw_init(void *handle)
4551 {
4552 	int r;
4553 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4554 
4555 	gfx_v7_0_constants_init(adev);
4556 
4557 	/* init CSB */
4558 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
4559 	/* init rlc */
4560 	r = adev->gfx.rlc.funcs->resume(adev);
4561 	if (r)
4562 		return r;
4563 
4564 	r = gfx_v7_0_cp_resume(adev);
4565 	if (r)
4566 		return r;
4567 
4568 	return r;
4569 }
4570 
4571 static int gfx_v7_0_hw_fini(void *handle)
4572 {
4573 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4574 
4575 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4576 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4577 	gfx_v7_0_cp_enable(adev, false);
4578 	adev->gfx.rlc.funcs->stop(adev);
4579 	gfx_v7_0_fini_pg(adev);
4580 
4581 	return 0;
4582 }
4583 
4584 static int gfx_v7_0_suspend(void *handle)
4585 {
4586 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4587 
4588 	return gfx_v7_0_hw_fini(adev);
4589 }
4590 
4591 static int gfx_v7_0_resume(void *handle)
4592 {
4593 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4594 
4595 	return gfx_v7_0_hw_init(adev);
4596 }
4597 
4598 static bool gfx_v7_0_is_idle(void *handle)
4599 {
4600 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4601 
4602 	if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK)
4603 		return false;
4604 	else
4605 		return true;
4606 }
4607 
4608 static int gfx_v7_0_wait_for_idle(void *handle)
4609 {
4610 	unsigned i;
4611 	u32 tmp;
4612 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4613 
4614 	for (i = 0; i < adev->usec_timeout; i++) {
4615 		/* read MC_STATUS */
4616 		tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4617 
4618 		if (!tmp)
4619 			return 0;
4620 		udelay(1);
4621 	}
4622 	return -ETIMEDOUT;
4623 }
4624 
4625 static int gfx_v7_0_soft_reset(void *handle)
4626 {
4627 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4628 	u32 tmp;
4629 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4630 
4631 	/* GRBM_STATUS */
4632 	tmp = RREG32(mmGRBM_STATUS);
4633 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4634 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4635 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4636 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4637 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4638 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK))
4639 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK |
4640 			GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK;
4641 
4642 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4643 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK;
4644 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4645 	}
4646 
4647 	/* GRBM_STATUS2 */
4648 	tmp = RREG32(mmGRBM_STATUS2);
4649 	if (tmp & GRBM_STATUS2__RLC_BUSY_MASK)
4650 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
4651 
4652 	/* SRBM_STATUS */
4653 	tmp = RREG32(mmSRBM_STATUS);
4654 	if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK)
4655 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4656 
4657 	if (grbm_soft_reset || srbm_soft_reset) {
4658 		/* disable CG/PG */
4659 		gfx_v7_0_fini_pg(adev);
4660 		gfx_v7_0_update_cg(adev, false);
4661 
4662 		/* stop the rlc */
4663 		adev->gfx.rlc.funcs->stop(adev);
4664 
4665 		/* Disable GFX parsing/prefetching */
4666 		WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK);
4667 
4668 		/* Disable MEC parsing/prefetching */
4669 		WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK);
4670 
4671 		if (grbm_soft_reset) {
4672 			tmp = RREG32(mmGRBM_SOFT_RESET);
4673 			tmp |= grbm_soft_reset;
4674 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4675 			WREG32(mmGRBM_SOFT_RESET, tmp);
4676 			tmp = RREG32(mmGRBM_SOFT_RESET);
4677 
4678 			udelay(50);
4679 
4680 			tmp &= ~grbm_soft_reset;
4681 			WREG32(mmGRBM_SOFT_RESET, tmp);
4682 			tmp = RREG32(mmGRBM_SOFT_RESET);
4683 		}
4684 
4685 		if (srbm_soft_reset) {
4686 			tmp = RREG32(mmSRBM_SOFT_RESET);
4687 			tmp |= srbm_soft_reset;
4688 			dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4689 			WREG32(mmSRBM_SOFT_RESET, tmp);
4690 			tmp = RREG32(mmSRBM_SOFT_RESET);
4691 
4692 			udelay(50);
4693 
4694 			tmp &= ~srbm_soft_reset;
4695 			WREG32(mmSRBM_SOFT_RESET, tmp);
4696 			tmp = RREG32(mmSRBM_SOFT_RESET);
4697 		}
4698 		/* Wait a little for things to settle down */
4699 		udelay(50);
4700 	}
4701 	return 0;
4702 }
4703 
4704 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4705 						 enum amdgpu_interrupt_state state)
4706 {
4707 	u32 cp_int_cntl;
4708 
4709 	switch (state) {
4710 	case AMDGPU_IRQ_STATE_DISABLE:
4711 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4712 		cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4713 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4714 		break;
4715 	case AMDGPU_IRQ_STATE_ENABLE:
4716 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4717 		cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4718 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4719 		break;
4720 	default:
4721 		break;
4722 	}
4723 }
4724 
4725 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4726 						     int me, int pipe,
4727 						     enum amdgpu_interrupt_state state)
4728 {
4729 	u32 mec_int_cntl, mec_int_cntl_reg;
4730 
4731 	/*
4732 	 * amdgpu controls only the first MEC. That's why this function only
4733 	 * handles the setting of interrupts for this specific MEC. All other
4734 	 * pipes' interrupts are set by amdkfd.
4735 	 */
4736 
4737 	if (me == 1) {
4738 		switch (pipe) {
4739 		case 0:
4740 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4741 			break;
4742 		case 1:
4743 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
4744 			break;
4745 		case 2:
4746 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
4747 			break;
4748 		case 3:
4749 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
4750 			break;
4751 		default:
4752 			DRM_DEBUG("invalid pipe %d\n", pipe);
4753 			return;
4754 		}
4755 	} else {
4756 		DRM_DEBUG("invalid me %d\n", me);
4757 		return;
4758 	}
4759 
4760 	switch (state) {
4761 	case AMDGPU_IRQ_STATE_DISABLE:
4762 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4763 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4764 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4765 		break;
4766 	case AMDGPU_IRQ_STATE_ENABLE:
4767 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4768 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4769 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4770 		break;
4771 	default:
4772 		break;
4773 	}
4774 }
4775 
4776 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4777 					     struct amdgpu_irq_src *src,
4778 					     unsigned type,
4779 					     enum amdgpu_interrupt_state state)
4780 {
4781 	u32 cp_int_cntl;
4782 
4783 	switch (state) {
4784 	case AMDGPU_IRQ_STATE_DISABLE:
4785 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4786 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4787 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4788 		break;
4789 	case AMDGPU_IRQ_STATE_ENABLE:
4790 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4791 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4792 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4793 		break;
4794 	default:
4795 		break;
4796 	}
4797 
4798 	return 0;
4799 }
4800 
4801 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4802 					      struct amdgpu_irq_src *src,
4803 					      unsigned type,
4804 					      enum amdgpu_interrupt_state state)
4805 {
4806 	u32 cp_int_cntl;
4807 
4808 	switch (state) {
4809 	case AMDGPU_IRQ_STATE_DISABLE:
4810 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4811 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4812 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4813 		break;
4814 	case AMDGPU_IRQ_STATE_ENABLE:
4815 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4816 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4817 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4818 		break;
4819 	default:
4820 		break;
4821 	}
4822 
4823 	return 0;
4824 }
4825 
4826 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4827 					    struct amdgpu_irq_src *src,
4828 					    unsigned type,
4829 					    enum amdgpu_interrupt_state state)
4830 {
4831 	switch (type) {
4832 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
4833 		gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
4834 		break;
4835 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4836 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4837 		break;
4838 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4839 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4840 		break;
4841 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4842 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4843 		break;
4844 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4845 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4846 		break;
4847 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4848 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4849 		break;
4850 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4851 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4852 		break;
4853 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4854 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4855 		break;
4856 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4857 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4858 		break;
4859 	default:
4860 		break;
4861 	}
4862 	return 0;
4863 }
4864 
4865 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
4866 			    struct amdgpu_irq_src *source,
4867 			    struct amdgpu_iv_entry *entry)
4868 {
4869 	u8 me_id, pipe_id;
4870 	struct amdgpu_ring *ring;
4871 	int i;
4872 
4873 	DRM_DEBUG("IH: CP EOP\n");
4874 	me_id = (entry->ring_id & 0x0c) >> 2;
4875 	pipe_id = (entry->ring_id & 0x03) >> 0;
4876 	switch (me_id) {
4877 	case 0:
4878 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4879 		break;
4880 	case 1:
4881 	case 2:
4882 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4883 			ring = &adev->gfx.compute_ring[i];
4884 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4885 				amdgpu_fence_process(ring);
4886 		}
4887 		break;
4888 	}
4889 	return 0;
4890 }
4891 
4892 static void gfx_v7_0_fault(struct amdgpu_device *adev,
4893 			   struct amdgpu_iv_entry *entry)
4894 {
4895 	struct amdgpu_ring *ring;
4896 	u8 me_id, pipe_id;
4897 	int i;
4898 
4899 	me_id = (entry->ring_id & 0x0c) >> 2;
4900 	pipe_id = (entry->ring_id & 0x03) >> 0;
4901 	switch (me_id) {
4902 	case 0:
4903 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
4904 		break;
4905 	case 1:
4906 	case 2:
4907 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4908 			ring = &adev->gfx.compute_ring[i];
4909 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4910 				drm_sched_fault(&ring->sched);
4911 		}
4912 		break;
4913 	}
4914 }
4915 
4916 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
4917 				 struct amdgpu_irq_src *source,
4918 				 struct amdgpu_iv_entry *entry)
4919 {
4920 	DRM_ERROR("Illegal register access in command stream\n");
4921 	gfx_v7_0_fault(adev, entry);
4922 	return 0;
4923 }
4924 
4925 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
4926 				  struct amdgpu_irq_src *source,
4927 				  struct amdgpu_iv_entry *entry)
4928 {
4929 	DRM_ERROR("Illegal instruction in command stream\n");
4930 	// XXX soft reset the gfx block only
4931 	gfx_v7_0_fault(adev, entry);
4932 	return 0;
4933 }
4934 
4935 static int gfx_v7_0_set_clockgating_state(void *handle,
4936 					  enum amd_clockgating_state state)
4937 {
4938 	bool gate = false;
4939 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4940 
4941 	if (state == AMD_CG_STATE_GATE)
4942 		gate = true;
4943 
4944 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
4945 	/* order matters! */
4946 	if (gate) {
4947 		gfx_v7_0_enable_mgcg(adev, true);
4948 		gfx_v7_0_enable_cgcg(adev, true);
4949 	} else {
4950 		gfx_v7_0_enable_cgcg(adev, false);
4951 		gfx_v7_0_enable_mgcg(adev, false);
4952 	}
4953 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
4954 
4955 	return 0;
4956 }
4957 
4958 static int gfx_v7_0_set_powergating_state(void *handle,
4959 					  enum amd_powergating_state state)
4960 {
4961 	bool gate = false;
4962 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4963 
4964 	if (state == AMD_PG_STATE_GATE)
4965 		gate = true;
4966 
4967 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4968 			      AMD_PG_SUPPORT_GFX_SMG |
4969 			      AMD_PG_SUPPORT_GFX_DMG |
4970 			      AMD_PG_SUPPORT_CP |
4971 			      AMD_PG_SUPPORT_GDS |
4972 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4973 		gfx_v7_0_update_gfx_pg(adev, gate);
4974 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4975 			gfx_v7_0_enable_cp_pg(adev, gate);
4976 			gfx_v7_0_enable_gds_pg(adev, gate);
4977 		}
4978 	}
4979 
4980 	return 0;
4981 }
4982 
4983 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
4984 	.name = "gfx_v7_0",
4985 	.early_init = gfx_v7_0_early_init,
4986 	.late_init = gfx_v7_0_late_init,
4987 	.sw_init = gfx_v7_0_sw_init,
4988 	.sw_fini = gfx_v7_0_sw_fini,
4989 	.hw_init = gfx_v7_0_hw_init,
4990 	.hw_fini = gfx_v7_0_hw_fini,
4991 	.suspend = gfx_v7_0_suspend,
4992 	.resume = gfx_v7_0_resume,
4993 	.is_idle = gfx_v7_0_is_idle,
4994 	.wait_for_idle = gfx_v7_0_wait_for_idle,
4995 	.soft_reset = gfx_v7_0_soft_reset,
4996 	.set_clockgating_state = gfx_v7_0_set_clockgating_state,
4997 	.set_powergating_state = gfx_v7_0_set_powergating_state,
4998 };
4999 
5000 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
5001 	.type = AMDGPU_RING_TYPE_GFX,
5002 	.align_mask = 0xff,
5003 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5004 	.support_64bit_ptrs = false,
5005 	.get_rptr = gfx_v7_0_ring_get_rptr,
5006 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
5007 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
5008 	.emit_frame_size =
5009 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5010 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5011 		5 + /* hdp invalidate */
5012 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
5013 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
5014 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
5015 		3 + 4, /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
5016 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
5017 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
5018 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
5019 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5020 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5021 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5022 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5023 	.test_ring = gfx_v7_0_ring_test_ring,
5024 	.test_ib = gfx_v7_0_ring_test_ib,
5025 	.insert_nop = amdgpu_ring_insert_nop,
5026 	.pad_ib = amdgpu_ring_generic_pad_ib,
5027 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
5028 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5029 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
5030 };
5031 
5032 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5033 	.type = AMDGPU_RING_TYPE_COMPUTE,
5034 	.align_mask = 0xff,
5035 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5036 	.support_64bit_ptrs = false,
5037 	.get_rptr = gfx_v7_0_ring_get_rptr,
5038 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5039 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5040 	.emit_frame_size =
5041 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5042 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5043 		5 + /* hdp invalidate */
5044 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5045 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
5046 		7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5047 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
5048 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5049 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5050 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5051 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5052 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5053 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5054 	.test_ring = gfx_v7_0_ring_test_ring,
5055 	.test_ib = gfx_v7_0_ring_test_ib,
5056 	.insert_nop = amdgpu_ring_insert_nop,
5057 	.pad_ib = amdgpu_ring_generic_pad_ib,
5058 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5059 };
5060 
5061 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5062 {
5063 	int i;
5064 
5065 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5066 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5067 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5068 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5069 }
5070 
5071 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5072 	.set = gfx_v7_0_set_eop_interrupt_state,
5073 	.process = gfx_v7_0_eop_irq,
5074 };
5075 
5076 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5077 	.set = gfx_v7_0_set_priv_reg_fault_state,
5078 	.process = gfx_v7_0_priv_reg_irq,
5079 };
5080 
5081 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5082 	.set = gfx_v7_0_set_priv_inst_fault_state,
5083 	.process = gfx_v7_0_priv_inst_irq,
5084 };
5085 
5086 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5087 {
5088 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5089 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5090 
5091 	adev->gfx.priv_reg_irq.num_types = 1;
5092 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5093 
5094 	adev->gfx.priv_inst_irq.num_types = 1;
5095 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5096 }
5097 
5098 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5099 {
5100 	/* init asci gds info */
5101 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
5102 	adev->gds.gws_size = 64;
5103 	adev->gds.oa_size = 16;
5104 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
5105 }
5106 
5107 
5108 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5109 {
5110 	int i, j, k, counter, active_cu_number = 0;
5111 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5112 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5113 	unsigned disable_masks[4 * 2];
5114 	u32 ao_cu_num;
5115 
5116 	if (adev->flags & AMD_IS_APU)
5117 		ao_cu_num = 2;
5118 	else
5119 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
5120 
5121 	memset(cu_info, 0, sizeof(*cu_info));
5122 
5123 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5124 
5125 	mutex_lock(&adev->grbm_idx_mutex);
5126 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5127 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5128 			mask = 1;
5129 			ao_bitmap = 0;
5130 			counter = 0;
5131 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
5132 			if (i < 4 && j < 2)
5133 				gfx_v7_0_set_user_cu_inactive_bitmap(
5134 					adev, disable_masks[i * 2 + j]);
5135 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5136 			cu_info->bitmap[i][j] = bitmap;
5137 
5138 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
5139 				if (bitmap & mask) {
5140 					if (counter < ao_cu_num)
5141 						ao_bitmap |= mask;
5142 					counter ++;
5143 				}
5144 				mask <<= 1;
5145 			}
5146 			active_cu_number += counter;
5147 			if (i < 2 && j < 2)
5148 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5149 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5150 		}
5151 	}
5152 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5153 	mutex_unlock(&adev->grbm_idx_mutex);
5154 
5155 	cu_info->number = active_cu_number;
5156 	cu_info->ao_cu_mask = ao_cu_mask;
5157 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5158 	cu_info->max_waves_per_simd = 10;
5159 	cu_info->max_scratch_slots_per_cu = 32;
5160 	cu_info->wave_front_size = 64;
5161 	cu_info->lds_size = 64;
5162 }
5163 
5164 const struct amdgpu_ip_block_version gfx_v7_0_ip_block =
5165 {
5166 	.type = AMD_IP_BLOCK_TYPE_GFX,
5167 	.major = 7,
5168 	.minor = 0,
5169 	.rev = 0,
5170 	.funcs = &gfx_v7_0_ip_funcs,
5171 };
5172 
5173 const struct amdgpu_ip_block_version gfx_v7_1_ip_block =
5174 {
5175 	.type = AMD_IP_BLOCK_TYPE_GFX,
5176 	.major = 7,
5177 	.minor = 1,
5178 	.rev = 0,
5179 	.funcs = &gfx_v7_0_ip_funcs,
5180 };
5181 
5182 const struct amdgpu_ip_block_version gfx_v7_2_ip_block =
5183 {
5184 	.type = AMD_IP_BLOCK_TYPE_GFX,
5185 	.major = 7,
5186 	.minor = 2,
5187 	.rev = 0,
5188 	.funcs = &gfx_v7_0_ip_funcs,
5189 };
5190 
5191 const struct amdgpu_ip_block_version gfx_v7_3_ip_block =
5192 {
5193 	.type = AMD_IP_BLOCK_TYPE_GFX,
5194 	.major = 7,
5195 	.minor = 3,
5196 	.rev = 0,
5197 	.funcs = &gfx_v7_0_ip_funcs,
5198 };
5199