xref: /linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision 71dfa617ea9f18e4585fe78364217cd32b1fc382)
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 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
95 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
96 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
97 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
98 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
99 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
100 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
101 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
102 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
103 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
104 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
105 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
106 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
107 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
108 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
109 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
110 };
111 
112 static const u32 spectre_rlc_save_restore_register_list[] = {
113 	(0x0e00 << 16) | (0xc12c >> 2),
114 	0x00000000,
115 	(0x0e00 << 16) | (0xc140 >> 2),
116 	0x00000000,
117 	(0x0e00 << 16) | (0xc150 >> 2),
118 	0x00000000,
119 	(0x0e00 << 16) | (0xc15c >> 2),
120 	0x00000000,
121 	(0x0e00 << 16) | (0xc168 >> 2),
122 	0x00000000,
123 	(0x0e00 << 16) | (0xc170 >> 2),
124 	0x00000000,
125 	(0x0e00 << 16) | (0xc178 >> 2),
126 	0x00000000,
127 	(0x0e00 << 16) | (0xc204 >> 2),
128 	0x00000000,
129 	(0x0e00 << 16) | (0xc2b4 >> 2),
130 	0x00000000,
131 	(0x0e00 << 16) | (0xc2b8 >> 2),
132 	0x00000000,
133 	(0x0e00 << 16) | (0xc2bc >> 2),
134 	0x00000000,
135 	(0x0e00 << 16) | (0xc2c0 >> 2),
136 	0x00000000,
137 	(0x0e00 << 16) | (0x8228 >> 2),
138 	0x00000000,
139 	(0x0e00 << 16) | (0x829c >> 2),
140 	0x00000000,
141 	(0x0e00 << 16) | (0x869c >> 2),
142 	0x00000000,
143 	(0x0600 << 16) | (0x98f4 >> 2),
144 	0x00000000,
145 	(0x0e00 << 16) | (0x98f8 >> 2),
146 	0x00000000,
147 	(0x0e00 << 16) | (0x9900 >> 2),
148 	0x00000000,
149 	(0x0e00 << 16) | (0xc260 >> 2),
150 	0x00000000,
151 	(0x0e00 << 16) | (0x90e8 >> 2),
152 	0x00000000,
153 	(0x0e00 << 16) | (0x3c000 >> 2),
154 	0x00000000,
155 	(0x0e00 << 16) | (0x3c00c >> 2),
156 	0x00000000,
157 	(0x0e00 << 16) | (0x8c1c >> 2),
158 	0x00000000,
159 	(0x0e00 << 16) | (0x9700 >> 2),
160 	0x00000000,
161 	(0x0e00 << 16) | (0xcd20 >> 2),
162 	0x00000000,
163 	(0x4e00 << 16) | (0xcd20 >> 2),
164 	0x00000000,
165 	(0x5e00 << 16) | (0xcd20 >> 2),
166 	0x00000000,
167 	(0x6e00 << 16) | (0xcd20 >> 2),
168 	0x00000000,
169 	(0x7e00 << 16) | (0xcd20 >> 2),
170 	0x00000000,
171 	(0x8e00 << 16) | (0xcd20 >> 2),
172 	0x00000000,
173 	(0x9e00 << 16) | (0xcd20 >> 2),
174 	0x00000000,
175 	(0xae00 << 16) | (0xcd20 >> 2),
176 	0x00000000,
177 	(0xbe00 << 16) | (0xcd20 >> 2),
178 	0x00000000,
179 	(0x0e00 << 16) | (0x89bc >> 2),
180 	0x00000000,
181 	(0x0e00 << 16) | (0x8900 >> 2),
182 	0x00000000,
183 	0x3,
184 	(0x0e00 << 16) | (0xc130 >> 2),
185 	0x00000000,
186 	(0x0e00 << 16) | (0xc134 >> 2),
187 	0x00000000,
188 	(0x0e00 << 16) | (0xc1fc >> 2),
189 	0x00000000,
190 	(0x0e00 << 16) | (0xc208 >> 2),
191 	0x00000000,
192 	(0x0e00 << 16) | (0xc264 >> 2),
193 	0x00000000,
194 	(0x0e00 << 16) | (0xc268 >> 2),
195 	0x00000000,
196 	(0x0e00 << 16) | (0xc26c >> 2),
197 	0x00000000,
198 	(0x0e00 << 16) | (0xc270 >> 2),
199 	0x00000000,
200 	(0x0e00 << 16) | (0xc274 >> 2),
201 	0x00000000,
202 	(0x0e00 << 16) | (0xc278 >> 2),
203 	0x00000000,
204 	(0x0e00 << 16) | (0xc27c >> 2),
205 	0x00000000,
206 	(0x0e00 << 16) | (0xc280 >> 2),
207 	0x00000000,
208 	(0x0e00 << 16) | (0xc284 >> 2),
209 	0x00000000,
210 	(0x0e00 << 16) | (0xc288 >> 2),
211 	0x00000000,
212 	(0x0e00 << 16) | (0xc28c >> 2),
213 	0x00000000,
214 	(0x0e00 << 16) | (0xc290 >> 2),
215 	0x00000000,
216 	(0x0e00 << 16) | (0xc294 >> 2),
217 	0x00000000,
218 	(0x0e00 << 16) | (0xc298 >> 2),
219 	0x00000000,
220 	(0x0e00 << 16) | (0xc29c >> 2),
221 	0x00000000,
222 	(0x0e00 << 16) | (0xc2a0 >> 2),
223 	0x00000000,
224 	(0x0e00 << 16) | (0xc2a4 >> 2),
225 	0x00000000,
226 	(0x0e00 << 16) | (0xc2a8 >> 2),
227 	0x00000000,
228 	(0x0e00 << 16) | (0xc2ac  >> 2),
229 	0x00000000,
230 	(0x0e00 << 16) | (0xc2b0 >> 2),
231 	0x00000000,
232 	(0x0e00 << 16) | (0x301d0 >> 2),
233 	0x00000000,
234 	(0x0e00 << 16) | (0x30238 >> 2),
235 	0x00000000,
236 	(0x0e00 << 16) | (0x30250 >> 2),
237 	0x00000000,
238 	(0x0e00 << 16) | (0x30254 >> 2),
239 	0x00000000,
240 	(0x0e00 << 16) | (0x30258 >> 2),
241 	0x00000000,
242 	(0x0e00 << 16) | (0x3025c >> 2),
243 	0x00000000,
244 	(0x4e00 << 16) | (0xc900 >> 2),
245 	0x00000000,
246 	(0x5e00 << 16) | (0xc900 >> 2),
247 	0x00000000,
248 	(0x6e00 << 16) | (0xc900 >> 2),
249 	0x00000000,
250 	(0x7e00 << 16) | (0xc900 >> 2),
251 	0x00000000,
252 	(0x8e00 << 16) | (0xc900 >> 2),
253 	0x00000000,
254 	(0x9e00 << 16) | (0xc900 >> 2),
255 	0x00000000,
256 	(0xae00 << 16) | (0xc900 >> 2),
257 	0x00000000,
258 	(0xbe00 << 16) | (0xc900 >> 2),
259 	0x00000000,
260 	(0x4e00 << 16) | (0xc904 >> 2),
261 	0x00000000,
262 	(0x5e00 << 16) | (0xc904 >> 2),
263 	0x00000000,
264 	(0x6e00 << 16) | (0xc904 >> 2),
265 	0x00000000,
266 	(0x7e00 << 16) | (0xc904 >> 2),
267 	0x00000000,
268 	(0x8e00 << 16) | (0xc904 >> 2),
269 	0x00000000,
270 	(0x9e00 << 16) | (0xc904 >> 2),
271 	0x00000000,
272 	(0xae00 << 16) | (0xc904 >> 2),
273 	0x00000000,
274 	(0xbe00 << 16) | (0xc904 >> 2),
275 	0x00000000,
276 	(0x4e00 << 16) | (0xc908 >> 2),
277 	0x00000000,
278 	(0x5e00 << 16) | (0xc908 >> 2),
279 	0x00000000,
280 	(0x6e00 << 16) | (0xc908 >> 2),
281 	0x00000000,
282 	(0x7e00 << 16) | (0xc908 >> 2),
283 	0x00000000,
284 	(0x8e00 << 16) | (0xc908 >> 2),
285 	0x00000000,
286 	(0x9e00 << 16) | (0xc908 >> 2),
287 	0x00000000,
288 	(0xae00 << 16) | (0xc908 >> 2),
289 	0x00000000,
290 	(0xbe00 << 16) | (0xc908 >> 2),
291 	0x00000000,
292 	(0x4e00 << 16) | (0xc90c >> 2),
293 	0x00000000,
294 	(0x5e00 << 16) | (0xc90c >> 2),
295 	0x00000000,
296 	(0x6e00 << 16) | (0xc90c >> 2),
297 	0x00000000,
298 	(0x7e00 << 16) | (0xc90c >> 2),
299 	0x00000000,
300 	(0x8e00 << 16) | (0xc90c >> 2),
301 	0x00000000,
302 	(0x9e00 << 16) | (0xc90c >> 2),
303 	0x00000000,
304 	(0xae00 << 16) | (0xc90c >> 2),
305 	0x00000000,
306 	(0xbe00 << 16) | (0xc90c >> 2),
307 	0x00000000,
308 	(0x4e00 << 16) | (0xc910 >> 2),
309 	0x00000000,
310 	(0x5e00 << 16) | (0xc910 >> 2),
311 	0x00000000,
312 	(0x6e00 << 16) | (0xc910 >> 2),
313 	0x00000000,
314 	(0x7e00 << 16) | (0xc910 >> 2),
315 	0x00000000,
316 	(0x8e00 << 16) | (0xc910 >> 2),
317 	0x00000000,
318 	(0x9e00 << 16) | (0xc910 >> 2),
319 	0x00000000,
320 	(0xae00 << 16) | (0xc910 >> 2),
321 	0x00000000,
322 	(0xbe00 << 16) | (0xc910 >> 2),
323 	0x00000000,
324 	(0x0e00 << 16) | (0xc99c >> 2),
325 	0x00000000,
326 	(0x0e00 << 16) | (0x9834 >> 2),
327 	0x00000000,
328 	(0x0000 << 16) | (0x30f00 >> 2),
329 	0x00000000,
330 	(0x0001 << 16) | (0x30f00 >> 2),
331 	0x00000000,
332 	(0x0000 << 16) | (0x30f04 >> 2),
333 	0x00000000,
334 	(0x0001 << 16) | (0x30f04 >> 2),
335 	0x00000000,
336 	(0x0000 << 16) | (0x30f08 >> 2),
337 	0x00000000,
338 	(0x0001 << 16) | (0x30f08 >> 2),
339 	0x00000000,
340 	(0x0000 << 16) | (0x30f0c >> 2),
341 	0x00000000,
342 	(0x0001 << 16) | (0x30f0c >> 2),
343 	0x00000000,
344 	(0x0600 << 16) | (0x9b7c >> 2),
345 	0x00000000,
346 	(0x0e00 << 16) | (0x8a14 >> 2),
347 	0x00000000,
348 	(0x0e00 << 16) | (0x8a18 >> 2),
349 	0x00000000,
350 	(0x0600 << 16) | (0x30a00 >> 2),
351 	0x00000000,
352 	(0x0e00 << 16) | (0x8bf0 >> 2),
353 	0x00000000,
354 	(0x0e00 << 16) | (0x8bcc >> 2),
355 	0x00000000,
356 	(0x0e00 << 16) | (0x8b24 >> 2),
357 	0x00000000,
358 	(0x0e00 << 16) | (0x30a04 >> 2),
359 	0x00000000,
360 	(0x0600 << 16) | (0x30a10 >> 2),
361 	0x00000000,
362 	(0x0600 << 16) | (0x30a14 >> 2),
363 	0x00000000,
364 	(0x0600 << 16) | (0x30a18 >> 2),
365 	0x00000000,
366 	(0x0600 << 16) | (0x30a2c >> 2),
367 	0x00000000,
368 	(0x0e00 << 16) | (0xc700 >> 2),
369 	0x00000000,
370 	(0x0e00 << 16) | (0xc704 >> 2),
371 	0x00000000,
372 	(0x0e00 << 16) | (0xc708 >> 2),
373 	0x00000000,
374 	(0x0e00 << 16) | (0xc768 >> 2),
375 	0x00000000,
376 	(0x0400 << 16) | (0xc770 >> 2),
377 	0x00000000,
378 	(0x0400 << 16) | (0xc774 >> 2),
379 	0x00000000,
380 	(0x0400 << 16) | (0xc778 >> 2),
381 	0x00000000,
382 	(0x0400 << 16) | (0xc77c >> 2),
383 	0x00000000,
384 	(0x0400 << 16) | (0xc780 >> 2),
385 	0x00000000,
386 	(0x0400 << 16) | (0xc784 >> 2),
387 	0x00000000,
388 	(0x0400 << 16) | (0xc788 >> 2),
389 	0x00000000,
390 	(0x0400 << 16) | (0xc78c >> 2),
391 	0x00000000,
392 	(0x0400 << 16) | (0xc798 >> 2),
393 	0x00000000,
394 	(0x0400 << 16) | (0xc79c >> 2),
395 	0x00000000,
396 	(0x0400 << 16) | (0xc7a0 >> 2),
397 	0x00000000,
398 	(0x0400 << 16) | (0xc7a4 >> 2),
399 	0x00000000,
400 	(0x0400 << 16) | (0xc7a8 >> 2),
401 	0x00000000,
402 	(0x0400 << 16) | (0xc7ac >> 2),
403 	0x00000000,
404 	(0x0400 << 16) | (0xc7b0 >> 2),
405 	0x00000000,
406 	(0x0400 << 16) | (0xc7b4 >> 2),
407 	0x00000000,
408 	(0x0e00 << 16) | (0x9100 >> 2),
409 	0x00000000,
410 	(0x0e00 << 16) | (0x3c010 >> 2),
411 	0x00000000,
412 	(0x0e00 << 16) | (0x92a8 >> 2),
413 	0x00000000,
414 	(0x0e00 << 16) | (0x92ac >> 2),
415 	0x00000000,
416 	(0x0e00 << 16) | (0x92b4 >> 2),
417 	0x00000000,
418 	(0x0e00 << 16) | (0x92b8 >> 2),
419 	0x00000000,
420 	(0x0e00 << 16) | (0x92bc >> 2),
421 	0x00000000,
422 	(0x0e00 << 16) | (0x92c0 >> 2),
423 	0x00000000,
424 	(0x0e00 << 16) | (0x92c4 >> 2),
425 	0x00000000,
426 	(0x0e00 << 16) | (0x92c8 >> 2),
427 	0x00000000,
428 	(0x0e00 << 16) | (0x92cc >> 2),
429 	0x00000000,
430 	(0x0e00 << 16) | (0x92d0 >> 2),
431 	0x00000000,
432 	(0x0e00 << 16) | (0x8c00 >> 2),
433 	0x00000000,
434 	(0x0e00 << 16) | (0x8c04 >> 2),
435 	0x00000000,
436 	(0x0e00 << 16) | (0x8c20 >> 2),
437 	0x00000000,
438 	(0x0e00 << 16) | (0x8c38 >> 2),
439 	0x00000000,
440 	(0x0e00 << 16) | (0x8c3c >> 2),
441 	0x00000000,
442 	(0x0e00 << 16) | (0xae00 >> 2),
443 	0x00000000,
444 	(0x0e00 << 16) | (0x9604 >> 2),
445 	0x00000000,
446 	(0x0e00 << 16) | (0xac08 >> 2),
447 	0x00000000,
448 	(0x0e00 << 16) | (0xac0c >> 2),
449 	0x00000000,
450 	(0x0e00 << 16) | (0xac10 >> 2),
451 	0x00000000,
452 	(0x0e00 << 16) | (0xac14 >> 2),
453 	0x00000000,
454 	(0x0e00 << 16) | (0xac58 >> 2),
455 	0x00000000,
456 	(0x0e00 << 16) | (0xac68 >> 2),
457 	0x00000000,
458 	(0x0e00 << 16) | (0xac6c >> 2),
459 	0x00000000,
460 	(0x0e00 << 16) | (0xac70 >> 2),
461 	0x00000000,
462 	(0x0e00 << 16) | (0xac74 >> 2),
463 	0x00000000,
464 	(0x0e00 << 16) | (0xac78 >> 2),
465 	0x00000000,
466 	(0x0e00 << 16) | (0xac7c >> 2),
467 	0x00000000,
468 	(0x0e00 << 16) | (0xac80 >> 2),
469 	0x00000000,
470 	(0x0e00 << 16) | (0xac84 >> 2),
471 	0x00000000,
472 	(0x0e00 << 16) | (0xac88 >> 2),
473 	0x00000000,
474 	(0x0e00 << 16) | (0xac8c >> 2),
475 	0x00000000,
476 	(0x0e00 << 16) | (0x970c >> 2),
477 	0x00000000,
478 	(0x0e00 << 16) | (0x9714 >> 2),
479 	0x00000000,
480 	(0x0e00 << 16) | (0x9718 >> 2),
481 	0x00000000,
482 	(0x0e00 << 16) | (0x971c >> 2),
483 	0x00000000,
484 	(0x0e00 << 16) | (0x31068 >> 2),
485 	0x00000000,
486 	(0x4e00 << 16) | (0x31068 >> 2),
487 	0x00000000,
488 	(0x5e00 << 16) | (0x31068 >> 2),
489 	0x00000000,
490 	(0x6e00 << 16) | (0x31068 >> 2),
491 	0x00000000,
492 	(0x7e00 << 16) | (0x31068 >> 2),
493 	0x00000000,
494 	(0x8e00 << 16) | (0x31068 >> 2),
495 	0x00000000,
496 	(0x9e00 << 16) | (0x31068 >> 2),
497 	0x00000000,
498 	(0xae00 << 16) | (0x31068 >> 2),
499 	0x00000000,
500 	(0xbe00 << 16) | (0x31068 >> 2),
501 	0x00000000,
502 	(0x0e00 << 16) | (0xcd10 >> 2),
503 	0x00000000,
504 	(0x0e00 << 16) | (0xcd14 >> 2),
505 	0x00000000,
506 	(0x0e00 << 16) | (0x88b0 >> 2),
507 	0x00000000,
508 	(0x0e00 << 16) | (0x88b4 >> 2),
509 	0x00000000,
510 	(0x0e00 << 16) | (0x88b8 >> 2),
511 	0x00000000,
512 	(0x0e00 << 16) | (0x88bc >> 2),
513 	0x00000000,
514 	(0x0400 << 16) | (0x89c0 >> 2),
515 	0x00000000,
516 	(0x0e00 << 16) | (0x88c4 >> 2),
517 	0x00000000,
518 	(0x0e00 << 16) | (0x88c8 >> 2),
519 	0x00000000,
520 	(0x0e00 << 16) | (0x88d0 >> 2),
521 	0x00000000,
522 	(0x0e00 << 16) | (0x88d4 >> 2),
523 	0x00000000,
524 	(0x0e00 << 16) | (0x88d8 >> 2),
525 	0x00000000,
526 	(0x0e00 << 16) | (0x8980 >> 2),
527 	0x00000000,
528 	(0x0e00 << 16) | (0x30938 >> 2),
529 	0x00000000,
530 	(0x0e00 << 16) | (0x3093c >> 2),
531 	0x00000000,
532 	(0x0e00 << 16) | (0x30940 >> 2),
533 	0x00000000,
534 	(0x0e00 << 16) | (0x89a0 >> 2),
535 	0x00000000,
536 	(0x0e00 << 16) | (0x30900 >> 2),
537 	0x00000000,
538 	(0x0e00 << 16) | (0x30904 >> 2),
539 	0x00000000,
540 	(0x0e00 << 16) | (0x89b4 >> 2),
541 	0x00000000,
542 	(0x0e00 << 16) | (0x3c210 >> 2),
543 	0x00000000,
544 	(0x0e00 << 16) | (0x3c214 >> 2),
545 	0x00000000,
546 	(0x0e00 << 16) | (0x3c218 >> 2),
547 	0x00000000,
548 	(0x0e00 << 16) | (0x8904 >> 2),
549 	0x00000000,
550 	0x5,
551 	(0x0e00 << 16) | (0x8c28 >> 2),
552 	(0x0e00 << 16) | (0x8c2c >> 2),
553 	(0x0e00 << 16) | (0x8c30 >> 2),
554 	(0x0e00 << 16) | (0x8c34 >> 2),
555 	(0x0e00 << 16) | (0x9600 >> 2),
556 };
557 
558 static const u32 kalindi_rlc_save_restore_register_list[] = {
559 	(0x0e00 << 16) | (0xc12c >> 2),
560 	0x00000000,
561 	(0x0e00 << 16) | (0xc140 >> 2),
562 	0x00000000,
563 	(0x0e00 << 16) | (0xc150 >> 2),
564 	0x00000000,
565 	(0x0e00 << 16) | (0xc15c >> 2),
566 	0x00000000,
567 	(0x0e00 << 16) | (0xc168 >> 2),
568 	0x00000000,
569 	(0x0e00 << 16) | (0xc170 >> 2),
570 	0x00000000,
571 	(0x0e00 << 16) | (0xc204 >> 2),
572 	0x00000000,
573 	(0x0e00 << 16) | (0xc2b4 >> 2),
574 	0x00000000,
575 	(0x0e00 << 16) | (0xc2b8 >> 2),
576 	0x00000000,
577 	(0x0e00 << 16) | (0xc2bc >> 2),
578 	0x00000000,
579 	(0x0e00 << 16) | (0xc2c0 >> 2),
580 	0x00000000,
581 	(0x0e00 << 16) | (0x8228 >> 2),
582 	0x00000000,
583 	(0x0e00 << 16) | (0x829c >> 2),
584 	0x00000000,
585 	(0x0e00 << 16) | (0x869c >> 2),
586 	0x00000000,
587 	(0x0600 << 16) | (0x98f4 >> 2),
588 	0x00000000,
589 	(0x0e00 << 16) | (0x98f8 >> 2),
590 	0x00000000,
591 	(0x0e00 << 16) | (0x9900 >> 2),
592 	0x00000000,
593 	(0x0e00 << 16) | (0xc260 >> 2),
594 	0x00000000,
595 	(0x0e00 << 16) | (0x90e8 >> 2),
596 	0x00000000,
597 	(0x0e00 << 16) | (0x3c000 >> 2),
598 	0x00000000,
599 	(0x0e00 << 16) | (0x3c00c >> 2),
600 	0x00000000,
601 	(0x0e00 << 16) | (0x8c1c >> 2),
602 	0x00000000,
603 	(0x0e00 << 16) | (0x9700 >> 2),
604 	0x00000000,
605 	(0x0e00 << 16) | (0xcd20 >> 2),
606 	0x00000000,
607 	(0x4e00 << 16) | (0xcd20 >> 2),
608 	0x00000000,
609 	(0x5e00 << 16) | (0xcd20 >> 2),
610 	0x00000000,
611 	(0x6e00 << 16) | (0xcd20 >> 2),
612 	0x00000000,
613 	(0x7e00 << 16) | (0xcd20 >> 2),
614 	0x00000000,
615 	(0x0e00 << 16) | (0x89bc >> 2),
616 	0x00000000,
617 	(0x0e00 << 16) | (0x8900 >> 2),
618 	0x00000000,
619 	0x3,
620 	(0x0e00 << 16) | (0xc130 >> 2),
621 	0x00000000,
622 	(0x0e00 << 16) | (0xc134 >> 2),
623 	0x00000000,
624 	(0x0e00 << 16) | (0xc1fc >> 2),
625 	0x00000000,
626 	(0x0e00 << 16) | (0xc208 >> 2),
627 	0x00000000,
628 	(0x0e00 << 16) | (0xc264 >> 2),
629 	0x00000000,
630 	(0x0e00 << 16) | (0xc268 >> 2),
631 	0x00000000,
632 	(0x0e00 << 16) | (0xc26c >> 2),
633 	0x00000000,
634 	(0x0e00 << 16) | (0xc270 >> 2),
635 	0x00000000,
636 	(0x0e00 << 16) | (0xc274 >> 2),
637 	0x00000000,
638 	(0x0e00 << 16) | (0xc28c >> 2),
639 	0x00000000,
640 	(0x0e00 << 16) | (0xc290 >> 2),
641 	0x00000000,
642 	(0x0e00 << 16) | (0xc294 >> 2),
643 	0x00000000,
644 	(0x0e00 << 16) | (0xc298 >> 2),
645 	0x00000000,
646 	(0x0e00 << 16) | (0xc2a0 >> 2),
647 	0x00000000,
648 	(0x0e00 << 16) | (0xc2a4 >> 2),
649 	0x00000000,
650 	(0x0e00 << 16) | (0xc2a8 >> 2),
651 	0x00000000,
652 	(0x0e00 << 16) | (0xc2ac >> 2),
653 	0x00000000,
654 	(0x0e00 << 16) | (0x301d0 >> 2),
655 	0x00000000,
656 	(0x0e00 << 16) | (0x30238 >> 2),
657 	0x00000000,
658 	(0x0e00 << 16) | (0x30250 >> 2),
659 	0x00000000,
660 	(0x0e00 << 16) | (0x30254 >> 2),
661 	0x00000000,
662 	(0x0e00 << 16) | (0x30258 >> 2),
663 	0x00000000,
664 	(0x0e00 << 16) | (0x3025c >> 2),
665 	0x00000000,
666 	(0x4e00 << 16) | (0xc900 >> 2),
667 	0x00000000,
668 	(0x5e00 << 16) | (0xc900 >> 2),
669 	0x00000000,
670 	(0x6e00 << 16) | (0xc900 >> 2),
671 	0x00000000,
672 	(0x7e00 << 16) | (0xc900 >> 2),
673 	0x00000000,
674 	(0x4e00 << 16) | (0xc904 >> 2),
675 	0x00000000,
676 	(0x5e00 << 16) | (0xc904 >> 2),
677 	0x00000000,
678 	(0x6e00 << 16) | (0xc904 >> 2),
679 	0x00000000,
680 	(0x7e00 << 16) | (0xc904 >> 2),
681 	0x00000000,
682 	(0x4e00 << 16) | (0xc908 >> 2),
683 	0x00000000,
684 	(0x5e00 << 16) | (0xc908 >> 2),
685 	0x00000000,
686 	(0x6e00 << 16) | (0xc908 >> 2),
687 	0x00000000,
688 	(0x7e00 << 16) | (0xc908 >> 2),
689 	0x00000000,
690 	(0x4e00 << 16) | (0xc90c >> 2),
691 	0x00000000,
692 	(0x5e00 << 16) | (0xc90c >> 2),
693 	0x00000000,
694 	(0x6e00 << 16) | (0xc90c >> 2),
695 	0x00000000,
696 	(0x7e00 << 16) | (0xc90c >> 2),
697 	0x00000000,
698 	(0x4e00 << 16) | (0xc910 >> 2),
699 	0x00000000,
700 	(0x5e00 << 16) | (0xc910 >> 2),
701 	0x00000000,
702 	(0x6e00 << 16) | (0xc910 >> 2),
703 	0x00000000,
704 	(0x7e00 << 16) | (0xc910 >> 2),
705 	0x00000000,
706 	(0x0e00 << 16) | (0xc99c >> 2),
707 	0x00000000,
708 	(0x0e00 << 16) | (0x9834 >> 2),
709 	0x00000000,
710 	(0x0000 << 16) | (0x30f00 >> 2),
711 	0x00000000,
712 	(0x0000 << 16) | (0x30f04 >> 2),
713 	0x00000000,
714 	(0x0000 << 16) | (0x30f08 >> 2),
715 	0x00000000,
716 	(0x0000 << 16) | (0x30f0c >> 2),
717 	0x00000000,
718 	(0x0600 << 16) | (0x9b7c >> 2),
719 	0x00000000,
720 	(0x0e00 << 16) | (0x8a14 >> 2),
721 	0x00000000,
722 	(0x0e00 << 16) | (0x8a18 >> 2),
723 	0x00000000,
724 	(0x0600 << 16) | (0x30a00 >> 2),
725 	0x00000000,
726 	(0x0e00 << 16) | (0x8bf0 >> 2),
727 	0x00000000,
728 	(0x0e00 << 16) | (0x8bcc >> 2),
729 	0x00000000,
730 	(0x0e00 << 16) | (0x8b24 >> 2),
731 	0x00000000,
732 	(0x0e00 << 16) | (0x30a04 >> 2),
733 	0x00000000,
734 	(0x0600 << 16) | (0x30a10 >> 2),
735 	0x00000000,
736 	(0x0600 << 16) | (0x30a14 >> 2),
737 	0x00000000,
738 	(0x0600 << 16) | (0x30a18 >> 2),
739 	0x00000000,
740 	(0x0600 << 16) | (0x30a2c >> 2),
741 	0x00000000,
742 	(0x0e00 << 16) | (0xc700 >> 2),
743 	0x00000000,
744 	(0x0e00 << 16) | (0xc704 >> 2),
745 	0x00000000,
746 	(0x0e00 << 16) | (0xc708 >> 2),
747 	0x00000000,
748 	(0x0e00 << 16) | (0xc768 >> 2),
749 	0x00000000,
750 	(0x0400 << 16) | (0xc770 >> 2),
751 	0x00000000,
752 	(0x0400 << 16) | (0xc774 >> 2),
753 	0x00000000,
754 	(0x0400 << 16) | (0xc798 >> 2),
755 	0x00000000,
756 	(0x0400 << 16) | (0xc79c >> 2),
757 	0x00000000,
758 	(0x0e00 << 16) | (0x9100 >> 2),
759 	0x00000000,
760 	(0x0e00 << 16) | (0x3c010 >> 2),
761 	0x00000000,
762 	(0x0e00 << 16) | (0x8c00 >> 2),
763 	0x00000000,
764 	(0x0e00 << 16) | (0x8c04 >> 2),
765 	0x00000000,
766 	(0x0e00 << 16) | (0x8c20 >> 2),
767 	0x00000000,
768 	(0x0e00 << 16) | (0x8c38 >> 2),
769 	0x00000000,
770 	(0x0e00 << 16) | (0x8c3c >> 2),
771 	0x00000000,
772 	(0x0e00 << 16) | (0xae00 >> 2),
773 	0x00000000,
774 	(0x0e00 << 16) | (0x9604 >> 2),
775 	0x00000000,
776 	(0x0e00 << 16) | (0xac08 >> 2),
777 	0x00000000,
778 	(0x0e00 << 16) | (0xac0c >> 2),
779 	0x00000000,
780 	(0x0e00 << 16) | (0xac10 >> 2),
781 	0x00000000,
782 	(0x0e00 << 16) | (0xac14 >> 2),
783 	0x00000000,
784 	(0x0e00 << 16) | (0xac58 >> 2),
785 	0x00000000,
786 	(0x0e00 << 16) | (0xac68 >> 2),
787 	0x00000000,
788 	(0x0e00 << 16) | (0xac6c >> 2),
789 	0x00000000,
790 	(0x0e00 << 16) | (0xac70 >> 2),
791 	0x00000000,
792 	(0x0e00 << 16) | (0xac74 >> 2),
793 	0x00000000,
794 	(0x0e00 << 16) | (0xac78 >> 2),
795 	0x00000000,
796 	(0x0e00 << 16) | (0xac7c >> 2),
797 	0x00000000,
798 	(0x0e00 << 16) | (0xac80 >> 2),
799 	0x00000000,
800 	(0x0e00 << 16) | (0xac84 >> 2),
801 	0x00000000,
802 	(0x0e00 << 16) | (0xac88 >> 2),
803 	0x00000000,
804 	(0x0e00 << 16) | (0xac8c >> 2),
805 	0x00000000,
806 	(0x0e00 << 16) | (0x970c >> 2),
807 	0x00000000,
808 	(0x0e00 << 16) | (0x9714 >> 2),
809 	0x00000000,
810 	(0x0e00 << 16) | (0x9718 >> 2),
811 	0x00000000,
812 	(0x0e00 << 16) | (0x971c >> 2),
813 	0x00000000,
814 	(0x0e00 << 16) | (0x31068 >> 2),
815 	0x00000000,
816 	(0x4e00 << 16) | (0x31068 >> 2),
817 	0x00000000,
818 	(0x5e00 << 16) | (0x31068 >> 2),
819 	0x00000000,
820 	(0x6e00 << 16) | (0x31068 >> 2),
821 	0x00000000,
822 	(0x7e00 << 16) | (0x31068 >> 2),
823 	0x00000000,
824 	(0x0e00 << 16) | (0xcd10 >> 2),
825 	0x00000000,
826 	(0x0e00 << 16) | (0xcd14 >> 2),
827 	0x00000000,
828 	(0x0e00 << 16) | (0x88b0 >> 2),
829 	0x00000000,
830 	(0x0e00 << 16) | (0x88b4 >> 2),
831 	0x00000000,
832 	(0x0e00 << 16) | (0x88b8 >> 2),
833 	0x00000000,
834 	(0x0e00 << 16) | (0x88bc >> 2),
835 	0x00000000,
836 	(0x0400 << 16) | (0x89c0 >> 2),
837 	0x00000000,
838 	(0x0e00 << 16) | (0x88c4 >> 2),
839 	0x00000000,
840 	(0x0e00 << 16) | (0x88c8 >> 2),
841 	0x00000000,
842 	(0x0e00 << 16) | (0x88d0 >> 2),
843 	0x00000000,
844 	(0x0e00 << 16) | (0x88d4 >> 2),
845 	0x00000000,
846 	(0x0e00 << 16) | (0x88d8 >> 2),
847 	0x00000000,
848 	(0x0e00 << 16) | (0x8980 >> 2),
849 	0x00000000,
850 	(0x0e00 << 16) | (0x30938 >> 2),
851 	0x00000000,
852 	(0x0e00 << 16) | (0x3093c >> 2),
853 	0x00000000,
854 	(0x0e00 << 16) | (0x30940 >> 2),
855 	0x00000000,
856 	(0x0e00 << 16) | (0x89a0 >> 2),
857 	0x00000000,
858 	(0x0e00 << 16) | (0x30900 >> 2),
859 	0x00000000,
860 	(0x0e00 << 16) | (0x30904 >> 2),
861 	0x00000000,
862 	(0x0e00 << 16) | (0x89b4 >> 2),
863 	0x00000000,
864 	(0x0e00 << 16) | (0x3e1fc >> 2),
865 	0x00000000,
866 	(0x0e00 << 16) | (0x3c210 >> 2),
867 	0x00000000,
868 	(0x0e00 << 16) | (0x3c214 >> 2),
869 	0x00000000,
870 	(0x0e00 << 16) | (0x3c218 >> 2),
871 	0x00000000,
872 	(0x0e00 << 16) | (0x8904 >> 2),
873 	0x00000000,
874 	0x5,
875 	(0x0e00 << 16) | (0x8c28 >> 2),
876 	(0x0e00 << 16) | (0x8c2c >> 2),
877 	(0x0e00 << 16) | (0x8c30 >> 2),
878 	(0x0e00 << 16) | (0x8c34 >> 2),
879 	(0x0e00 << 16) | (0x9600 >> 2),
880 };
881 
882 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev);
883 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer);
884 static void gfx_v7_0_init_pg(struct amdgpu_device *adev);
885 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev);
886 
887 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev)
888 {
889 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
890 	amdgpu_ucode_release(&adev->gfx.me_fw);
891 	amdgpu_ucode_release(&adev->gfx.ce_fw);
892 	amdgpu_ucode_release(&adev->gfx.mec_fw);
893 	amdgpu_ucode_release(&adev->gfx.mec2_fw);
894 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
895 }
896 
897 /*
898  * Core functions
899  */
900 /**
901  * gfx_v7_0_init_microcode - load ucode images from disk
902  *
903  * @adev: amdgpu_device pointer
904  *
905  * Use the firmware interface to load the ucode images into
906  * the driver (not loaded into hw).
907  * Returns 0 on success, error on failure.
908  */
909 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev)
910 {
911 	const char *chip_name;
912 	char fw_name[30];
913 	int err;
914 
915 	DRM_DEBUG("\n");
916 
917 	switch (adev->asic_type) {
918 	case CHIP_BONAIRE:
919 		chip_name = "bonaire";
920 		break;
921 	case CHIP_HAWAII:
922 		chip_name = "hawaii";
923 		break;
924 	case CHIP_KAVERI:
925 		chip_name = "kaveri";
926 		break;
927 	case CHIP_KABINI:
928 		chip_name = "kabini";
929 		break;
930 	case CHIP_MULLINS:
931 		chip_name = "mullins";
932 		break;
933 	default:
934 		BUG();
935 	}
936 
937 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
938 	err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
939 	if (err)
940 		goto out;
941 
942 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
943 	err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
944 	if (err)
945 		goto out;
946 
947 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
948 	err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name);
949 	if (err)
950 		goto out;
951 
952 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
953 	err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
954 	if (err)
955 		goto out;
956 
957 	if (adev->asic_type == CHIP_KAVERI) {
958 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
959 		err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, fw_name);
960 		if (err)
961 			goto out;
962 	}
963 
964 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
965 	err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name);
966 	if (err)
967 		goto out;
968 out:
969 	if (err) {
970 		pr_err("gfx7: Failed to load firmware \"%s\"\n", fw_name);
971 		gfx_v7_0_free_microcode(adev);
972 	}
973 	return err;
974 }
975 
976 /**
977  * gfx_v7_0_tiling_mode_table_init - init the hw tiling table
978  *
979  * @adev: amdgpu_device pointer
980  *
981  * Starting with SI, the tiling setup is done globally in a
982  * set of 32 tiling modes.  Rather than selecting each set of
983  * parameters per surface as on older asics, we just select
984  * which index in the tiling table we want to use, and the
985  * surface uses those parameters (CIK).
986  */
987 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev)
988 {
989 	const u32 num_tile_mode_states =
990 			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
991 	const u32 num_secondary_tile_mode_states =
992 			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
993 	u32 reg_offset, split_equal_to_row_size;
994 	uint32_t *tile, *macrotile;
995 
996 	tile = adev->gfx.config.tile_mode_array;
997 	macrotile = adev->gfx.config.macrotile_mode_array;
998 
999 	switch (adev->gfx.config.mem_row_size_in_kb) {
1000 	case 1:
1001 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1002 		break;
1003 	case 2:
1004 	default:
1005 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1006 		break;
1007 	case 4:
1008 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1009 		break;
1010 	}
1011 
1012 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1013 		tile[reg_offset] = 0;
1014 	for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1015 		macrotile[reg_offset] = 0;
1016 
1017 	switch (adev->asic_type) {
1018 	case CHIP_BONAIRE:
1019 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1020 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1021 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1022 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1023 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1024 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1025 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1026 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1027 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1028 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1029 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1030 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1031 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1032 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1033 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1034 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1035 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1036 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1037 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1038 			   TILE_SPLIT(split_equal_to_row_size));
1039 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1040 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1041 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1042 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1043 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1044 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1045 			   TILE_SPLIT(split_equal_to_row_size));
1046 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1047 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1048 			   PIPE_CONFIG(ADDR_SURF_P4_16x16));
1049 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1050 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1051 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1052 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1053 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1054 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1055 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1056 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1057 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1058 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1059 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1060 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1061 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1062 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1063 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1064 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1065 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1066 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1067 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1068 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1069 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1070 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1071 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1072 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1073 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1074 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1075 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1076 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1077 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1078 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1079 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1080 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1081 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1082 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1083 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1084 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1085 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1086 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1087 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1088 		tile[21] =  (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1089 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1090 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1091 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1092 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1093 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1094 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1095 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1096 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1097 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1098 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1099 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1100 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1101 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1102 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1103 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1104 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1105 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1106 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1107 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1108 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1109 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1110 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1111 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1112 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1113 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1114 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1115 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1116 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1117 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1118 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1119 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1120 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1121 
1122 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1123 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1124 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1125 				NUM_BANKS(ADDR_SURF_16_BANK));
1126 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1127 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1128 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1129 				NUM_BANKS(ADDR_SURF_16_BANK));
1130 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1131 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1132 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1133 				NUM_BANKS(ADDR_SURF_16_BANK));
1134 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1135 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1136 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1137 				NUM_BANKS(ADDR_SURF_16_BANK));
1138 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1139 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1140 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1141 				NUM_BANKS(ADDR_SURF_16_BANK));
1142 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1143 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1144 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1145 				NUM_BANKS(ADDR_SURF_8_BANK));
1146 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1147 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1148 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1149 				NUM_BANKS(ADDR_SURF_4_BANK));
1150 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1151 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1152 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1153 				NUM_BANKS(ADDR_SURF_16_BANK));
1154 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1155 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1156 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1157 				NUM_BANKS(ADDR_SURF_16_BANK));
1158 		macrotile[10] = (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[11] = (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[12] = (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[13] = (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_8_BANK));
1174 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1175 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1176 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1177 				NUM_BANKS(ADDR_SURF_4_BANK));
1178 
1179 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1180 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1181 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1182 			if (reg_offset != 7)
1183 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1184 		break;
1185 	case CHIP_HAWAII:
1186 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1187 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1188 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1189 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1190 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1191 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1192 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1193 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1194 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1195 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1196 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1197 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1198 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1199 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1200 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1201 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1202 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1203 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1204 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1205 			   TILE_SPLIT(split_equal_to_row_size));
1206 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1207 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1208 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1209 			   TILE_SPLIT(split_equal_to_row_size));
1210 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1211 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1212 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1213 			   TILE_SPLIT(split_equal_to_row_size));
1214 		tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1215 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1216 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1217 			   TILE_SPLIT(split_equal_to_row_size));
1218 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1219 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
1220 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1221 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1222 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1223 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1224 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1225 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1226 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1227 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1228 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1229 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1230 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1231 		tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1232 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1233 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1234 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1235 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1236 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1237 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1238 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1239 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1240 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1241 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1242 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1243 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1244 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1245 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1246 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1247 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1248 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1249 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1250 		tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1251 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1252 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1253 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1254 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1255 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1256 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1257 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1258 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1259 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1260 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1261 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1262 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1263 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1264 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1265 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1266 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1267 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1268 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1269 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1270 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1271 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1272 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1273 		tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1274 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1275 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1276 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1277 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1278 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1279 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1280 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1281 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1282 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1283 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1284 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1285 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1286 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1287 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1288 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1289 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1290 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1291 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1292 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1293 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1294 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1295 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1296 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1297 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1298 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1299 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1300 		tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1301 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1302 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1303 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1304 
1305 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1306 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1307 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1308 				NUM_BANKS(ADDR_SURF_16_BANK));
1309 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1310 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1311 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1312 				NUM_BANKS(ADDR_SURF_16_BANK));
1313 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1314 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1315 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1316 				NUM_BANKS(ADDR_SURF_16_BANK));
1317 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1318 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1319 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1320 				NUM_BANKS(ADDR_SURF_16_BANK));
1321 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1322 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1323 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1324 				NUM_BANKS(ADDR_SURF_8_BANK));
1325 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1326 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1327 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1328 				NUM_BANKS(ADDR_SURF_4_BANK));
1329 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1330 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1331 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1332 				NUM_BANKS(ADDR_SURF_4_BANK));
1333 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1334 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1335 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1336 				NUM_BANKS(ADDR_SURF_16_BANK));
1337 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1338 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1339 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1340 				NUM_BANKS(ADDR_SURF_16_BANK));
1341 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1342 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1343 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1344 				NUM_BANKS(ADDR_SURF_16_BANK));
1345 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1346 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1347 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1348 				NUM_BANKS(ADDR_SURF_8_BANK));
1349 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1350 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1351 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1352 				NUM_BANKS(ADDR_SURF_16_BANK));
1353 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1354 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1355 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1356 				NUM_BANKS(ADDR_SURF_8_BANK));
1357 		macrotile[14] = (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_4_BANK));
1361 
1362 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1363 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1364 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1365 			if (reg_offset != 7)
1366 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1367 		break;
1368 	case CHIP_KABINI:
1369 	case CHIP_KAVERI:
1370 	case CHIP_MULLINS:
1371 	default:
1372 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1373 			   PIPE_CONFIG(ADDR_SURF_P2) |
1374 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1375 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1376 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1377 			   PIPE_CONFIG(ADDR_SURF_P2) |
1378 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1379 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1380 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1381 			   PIPE_CONFIG(ADDR_SURF_P2) |
1382 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1383 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1384 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1385 			   PIPE_CONFIG(ADDR_SURF_P2) |
1386 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1387 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1388 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1389 			   PIPE_CONFIG(ADDR_SURF_P2) |
1390 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1391 			   TILE_SPLIT(split_equal_to_row_size));
1392 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1393 			   PIPE_CONFIG(ADDR_SURF_P2) |
1394 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1395 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1396 			   PIPE_CONFIG(ADDR_SURF_P2) |
1397 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1398 			   TILE_SPLIT(split_equal_to_row_size));
1399 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1400 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1401 			   PIPE_CONFIG(ADDR_SURF_P2));
1402 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1403 			   PIPE_CONFIG(ADDR_SURF_P2) |
1404 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1405 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1406 			    PIPE_CONFIG(ADDR_SURF_P2) |
1407 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1408 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1409 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1410 			    PIPE_CONFIG(ADDR_SURF_P2) |
1411 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1412 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1413 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1414 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1415 			    PIPE_CONFIG(ADDR_SURF_P2) |
1416 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1417 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1418 			    PIPE_CONFIG(ADDR_SURF_P2) |
1419 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1420 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1421 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1422 			    PIPE_CONFIG(ADDR_SURF_P2) |
1423 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1424 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1425 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1426 			    PIPE_CONFIG(ADDR_SURF_P2) |
1427 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1428 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1429 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1430 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1431 			    PIPE_CONFIG(ADDR_SURF_P2) |
1432 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1433 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1434 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1435 			    PIPE_CONFIG(ADDR_SURF_P2) |
1436 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1437 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1438 			    PIPE_CONFIG(ADDR_SURF_P2) |
1439 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1440 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1441 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1442 			    PIPE_CONFIG(ADDR_SURF_P2) |
1443 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1444 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1445 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1446 			    PIPE_CONFIG(ADDR_SURF_P2) |
1447 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1448 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1449 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1450 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1451 			    PIPE_CONFIG(ADDR_SURF_P2) |
1452 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1453 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1454 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1455 			    PIPE_CONFIG(ADDR_SURF_P2) |
1456 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1457 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1458 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1459 			    PIPE_CONFIG(ADDR_SURF_P2) |
1460 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1461 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1462 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1463 			    PIPE_CONFIG(ADDR_SURF_P2) |
1464 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1465 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1466 			    PIPE_CONFIG(ADDR_SURF_P2) |
1467 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1468 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1469 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1470 			    PIPE_CONFIG(ADDR_SURF_P2) |
1471 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1472 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1473 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1474 
1475 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1476 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1477 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1478 				NUM_BANKS(ADDR_SURF_8_BANK));
1479 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1480 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1481 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1482 				NUM_BANKS(ADDR_SURF_8_BANK));
1483 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1484 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1485 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1486 				NUM_BANKS(ADDR_SURF_8_BANK));
1487 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1488 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1489 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1490 				NUM_BANKS(ADDR_SURF_8_BANK));
1491 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1492 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1493 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1494 				NUM_BANKS(ADDR_SURF_8_BANK));
1495 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1496 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1497 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1498 				NUM_BANKS(ADDR_SURF_8_BANK));
1499 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1500 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1501 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1502 				NUM_BANKS(ADDR_SURF_8_BANK));
1503 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1504 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1505 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1506 				NUM_BANKS(ADDR_SURF_16_BANK));
1507 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1508 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1509 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1510 				NUM_BANKS(ADDR_SURF_16_BANK));
1511 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1512 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1513 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1514 				NUM_BANKS(ADDR_SURF_16_BANK));
1515 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1516 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1517 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1518 				NUM_BANKS(ADDR_SURF_16_BANK));
1519 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1520 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1521 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1522 				NUM_BANKS(ADDR_SURF_16_BANK));
1523 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1524 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1525 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1526 				NUM_BANKS(ADDR_SURF_16_BANK));
1527 		macrotile[14] = (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 
1532 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1533 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1534 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1535 			if (reg_offset != 7)
1536 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1537 		break;
1538 	}
1539 }
1540 
1541 /**
1542  * gfx_v7_0_select_se_sh - select which SE, SH to address
1543  *
1544  * @adev: amdgpu_device pointer
1545  * @se_num: shader engine to address
1546  * @sh_num: sh block to address
1547  * @instance: Certain registers are instanced per SE or SH.
1548  *            0xffffffff means broadcast to all SEs or SHs (CIK).
1549  * @xcc_id: xcc accelerated compute core id
1550  * Select which SE, SH combinations to address.
1551  */
1552 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev,
1553 				  u32 se_num, u32 sh_num, u32 instance,
1554 				  int xcc_id)
1555 {
1556 	u32 data;
1557 
1558 	if (instance == 0xffffffff)
1559 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1560 	else
1561 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1562 
1563 	if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1564 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1565 			GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
1566 	else if (se_num == 0xffffffff)
1567 		data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK |
1568 			(sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT);
1569 	else if (sh_num == 0xffffffff)
1570 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1571 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1572 	else
1573 		data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) |
1574 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1575 	WREG32(mmGRBM_GFX_INDEX, data);
1576 }
1577 
1578 /**
1579  * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs
1580  *
1581  * @adev: amdgpu_device pointer
1582  *
1583  * Calculates the bitmask of enabled RBs (CIK).
1584  * Returns the enabled RB bitmask.
1585  */
1586 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1587 {
1588 	u32 data, mask;
1589 
1590 	data = RREG32(mmCC_RB_BACKEND_DISABLE);
1591 	data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1592 
1593 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1594 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1595 
1596 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1597 					 adev->gfx.config.max_sh_per_se);
1598 
1599 	return (~data) & mask;
1600 }
1601 
1602 static void
1603 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
1604 {
1605 	switch (adev->asic_type) {
1606 	case CHIP_BONAIRE:
1607 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
1608 			  SE_XSEL(1) | SE_YSEL(1);
1609 		*rconf1 |= 0x0;
1610 		break;
1611 	case CHIP_HAWAII:
1612 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
1613 			  RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) |
1614 			  PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) |
1615 			  SE_YSEL(3);
1616 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
1617 			   SE_PAIR_YSEL(2);
1618 		break;
1619 	case CHIP_KAVERI:
1620 		*rconf |= RB_MAP_PKR0(2);
1621 		*rconf1 |= 0x0;
1622 		break;
1623 	case CHIP_KABINI:
1624 	case CHIP_MULLINS:
1625 		*rconf |= 0x0;
1626 		*rconf1 |= 0x0;
1627 		break;
1628 	default:
1629 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
1630 		break;
1631 	}
1632 }
1633 
1634 static void
1635 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev,
1636 					u32 raster_config, u32 raster_config_1,
1637 					unsigned rb_mask, unsigned num_rb)
1638 {
1639 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
1640 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
1641 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
1642 	unsigned rb_per_se = num_rb / num_se;
1643 	unsigned se_mask[4];
1644 	unsigned se;
1645 
1646 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
1647 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
1648 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
1649 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
1650 
1651 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
1652 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
1653 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
1654 
1655 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
1656 			     (!se_mask[2] && !se_mask[3]))) {
1657 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
1658 
1659 		if (!se_mask[0] && !se_mask[1]) {
1660 			raster_config_1 |=
1661 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
1662 		} else {
1663 			raster_config_1 |=
1664 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
1665 		}
1666 	}
1667 
1668 	for (se = 0; se < num_se; se++) {
1669 		unsigned raster_config_se = raster_config;
1670 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
1671 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
1672 		int idx = (se / 2) * 2;
1673 
1674 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
1675 			raster_config_se &= ~SE_MAP_MASK;
1676 
1677 			if (!se_mask[idx]) {
1678 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
1679 			} else {
1680 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
1681 			}
1682 		}
1683 
1684 		pkr0_mask &= rb_mask;
1685 		pkr1_mask &= rb_mask;
1686 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
1687 			raster_config_se &= ~PKR_MAP_MASK;
1688 
1689 			if (!pkr0_mask) {
1690 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
1691 			} else {
1692 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
1693 			}
1694 		}
1695 
1696 		if (rb_per_se >= 2) {
1697 			unsigned rb0_mask = 1 << (se * rb_per_se);
1698 			unsigned rb1_mask = rb0_mask << 1;
1699 
1700 			rb0_mask &= rb_mask;
1701 			rb1_mask &= rb_mask;
1702 			if (!rb0_mask || !rb1_mask) {
1703 				raster_config_se &= ~RB_MAP_PKR0_MASK;
1704 
1705 				if (!rb0_mask) {
1706 					raster_config_se |=
1707 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
1708 				} else {
1709 					raster_config_se |=
1710 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
1711 				}
1712 			}
1713 
1714 			if (rb_per_se > 2) {
1715 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
1716 				rb1_mask = rb0_mask << 1;
1717 				rb0_mask &= rb_mask;
1718 				rb1_mask &= rb_mask;
1719 				if (!rb0_mask || !rb1_mask) {
1720 					raster_config_se &= ~RB_MAP_PKR1_MASK;
1721 
1722 					if (!rb0_mask) {
1723 						raster_config_se |=
1724 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
1725 					} else {
1726 						raster_config_se |=
1727 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
1728 					}
1729 				}
1730 			}
1731 		}
1732 
1733 		/* GRBM_GFX_INDEX has a different offset on CI+ */
1734 		gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff, 0);
1735 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
1736 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1737 	}
1738 
1739 	/* GRBM_GFX_INDEX has a different offset on CI+ */
1740 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1741 }
1742 
1743 /**
1744  * gfx_v7_0_setup_rb - setup the RBs on the asic
1745  *
1746  * @adev: amdgpu_device pointer
1747  *
1748  * Configures per-SE/SH RB registers (CIK).
1749  */
1750 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1751 {
1752 	int i, j;
1753 	u32 data;
1754 	u32 raster_config = 0, raster_config_1 = 0;
1755 	u32 active_rbs = 0;
1756 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1757 					adev->gfx.config.max_sh_per_se;
1758 	unsigned num_rb_pipes;
1759 
1760 	mutex_lock(&adev->grbm_idx_mutex);
1761 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1762 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1763 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
1764 			data = gfx_v7_0_get_rb_active_bitmap(adev);
1765 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1766 					       rb_bitmap_width_per_sh);
1767 		}
1768 	}
1769 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1770 
1771 	adev->gfx.config.backend_enable_mask = active_rbs;
1772 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1773 
1774 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1775 			     adev->gfx.config.max_shader_engines, 16);
1776 
1777 	gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1778 
1779 	if (!adev->gfx.config.backend_enable_mask ||
1780 			adev->gfx.config.num_rbs >= num_rb_pipes) {
1781 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1782 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1783 	} else {
1784 		gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1785 							adev->gfx.config.backend_enable_mask,
1786 							num_rb_pipes);
1787 	}
1788 
1789 	/* cache the values for userspace */
1790 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1791 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1792 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
1793 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
1794 				RREG32(mmCC_RB_BACKEND_DISABLE);
1795 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
1796 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1797 			adev->gfx.config.rb_config[i][j].raster_config =
1798 				RREG32(mmPA_SC_RASTER_CONFIG);
1799 			adev->gfx.config.rb_config[i][j].raster_config_1 =
1800 				RREG32(mmPA_SC_RASTER_CONFIG_1);
1801 		}
1802 	}
1803 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1804 	mutex_unlock(&adev->grbm_idx_mutex);
1805 }
1806 
1807 #define DEFAULT_SH_MEM_BASES	(0x6000)
1808 /**
1809  * gfx_v7_0_init_compute_vmid - gart enable
1810  *
1811  * @adev: amdgpu_device pointer
1812  *
1813  * Initialize compute vmid sh_mem registers
1814  *
1815  */
1816 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1817 {
1818 	int i;
1819 	uint32_t sh_mem_config;
1820 	uint32_t sh_mem_bases;
1821 
1822 	/*
1823 	 * Configure apertures:
1824 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1825 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1826 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1827 	*/
1828 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1829 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1830 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1831 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1832 	mutex_lock(&adev->srbm_mutex);
1833 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1834 		cik_srbm_select(adev, 0, 0, 0, i);
1835 		/* CP and shaders */
1836 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1837 		WREG32(mmSH_MEM_APE1_BASE, 1);
1838 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1839 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1840 	}
1841 	cik_srbm_select(adev, 0, 0, 0, 0);
1842 	mutex_unlock(&adev->srbm_mutex);
1843 
1844 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1845 	   access. These should be enabled by FW for target VMIDs. */
1846 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1847 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
1848 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
1849 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
1850 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
1851 	}
1852 }
1853 
1854 static void gfx_v7_0_init_gds_vmid(struct amdgpu_device *adev)
1855 {
1856 	int vmid;
1857 
1858 	/*
1859 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1860 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1861 	 * the driver can enable them for graphics. VMID0 should maintain
1862 	 * access so that HWS firmware can save/restore entries.
1863 	 */
1864 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
1865 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
1866 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
1867 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
1868 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
1869 	}
1870 }
1871 
1872 static void gfx_v7_0_config_init(struct amdgpu_device *adev)
1873 {
1874 	adev->gfx.config.double_offchip_lds_buf = 1;
1875 }
1876 
1877 /**
1878  * gfx_v7_0_constants_init - setup the 3D engine
1879  *
1880  * @adev: amdgpu_device pointer
1881  *
1882  * init the gfx constants such as the 3D engine, tiling configuration
1883  * registers, maximum number of quad pipes, render backends...
1884  */
1885 static void gfx_v7_0_constants_init(struct amdgpu_device *adev)
1886 {
1887 	u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base;
1888 	u32 tmp;
1889 	int i;
1890 
1891 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1892 
1893 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1894 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1895 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1896 
1897 	gfx_v7_0_tiling_mode_table_init(adev);
1898 
1899 	gfx_v7_0_setup_rb(adev);
1900 	gfx_v7_0_get_cu_info(adev);
1901 	gfx_v7_0_config_init(adev);
1902 
1903 	/* set HW defaults for 3D engine */
1904 	WREG32(mmCP_MEQ_THRESHOLDS,
1905 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1906 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1907 
1908 	mutex_lock(&adev->grbm_idx_mutex);
1909 	/*
1910 	 * making sure that the following register writes will be broadcasted
1911 	 * to all the shaders
1912 	 */
1913 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1914 
1915 	/* XXX SH_MEM regs */
1916 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1917 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1918 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1919 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE,
1920 				   MTYPE_NC);
1921 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE,
1922 				   MTYPE_UC);
1923 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0);
1924 
1925 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
1926 				   SWIZZLE_ENABLE, 1);
1927 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1928 				   ELEMENT_SIZE, 1);
1929 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1930 				   INDEX_STRIDE, 3);
1931 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
1932 
1933 	mutex_lock(&adev->srbm_mutex);
1934 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
1935 		if (i == 0)
1936 			sh_mem_base = 0;
1937 		else
1938 			sh_mem_base = adev->gmc.shared_aperture_start >> 48;
1939 		cik_srbm_select(adev, 0, 0, 0, i);
1940 		/* CP and shaders */
1941 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1942 		WREG32(mmSH_MEM_APE1_BASE, 1);
1943 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1944 		WREG32(mmSH_MEM_BASES, sh_mem_base);
1945 	}
1946 	cik_srbm_select(adev, 0, 0, 0, 0);
1947 	mutex_unlock(&adev->srbm_mutex);
1948 
1949 	gfx_v7_0_init_compute_vmid(adev);
1950 	gfx_v7_0_init_gds_vmid(adev);
1951 
1952 	WREG32(mmSX_DEBUG_1, 0x20);
1953 
1954 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1955 
1956 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1957 	tmp |= 0x03000000;
1958 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1959 
1960 	WREG32(mmSQ_CONFIG, 1);
1961 
1962 	WREG32(mmDB_DEBUG, 0);
1963 
1964 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
1965 	tmp |= 0x00000400;
1966 	WREG32(mmDB_DEBUG2, tmp);
1967 
1968 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
1969 	tmp |= 0x00020200;
1970 	WREG32(mmDB_DEBUG3, tmp);
1971 
1972 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
1973 	tmp |= 0x00018208;
1974 	WREG32(mmCB_HW_CONTROL, tmp);
1975 
1976 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
1977 
1978 	WREG32(mmPA_SC_FIFO_SIZE,
1979 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
1980 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
1981 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
1982 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
1983 
1984 	WREG32(mmVGT_NUM_INSTANCES, 1);
1985 
1986 	WREG32(mmCP_PERFMON_CNTL, 0);
1987 
1988 	WREG32(mmSQ_CONFIG, 0);
1989 
1990 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
1991 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
1992 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
1993 
1994 	WREG32(mmVGT_CACHE_INVALIDATION,
1995 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
1996 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
1997 
1998 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
1999 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
2000 
2001 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
2002 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
2003 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
2004 
2005 	tmp = RREG32(mmSPI_ARB_PRIORITY);
2006 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
2007 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
2008 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
2009 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
2010 	WREG32(mmSPI_ARB_PRIORITY, tmp);
2011 
2012 	mutex_unlock(&adev->grbm_idx_mutex);
2013 
2014 	udelay(50);
2015 }
2016 
2017 /**
2018  * gfx_v7_0_ring_test_ring - basic gfx ring test
2019  *
2020  * @ring: amdgpu_ring structure holding ring information
2021  *
2022  * Allocate a scratch register and write to it using the gfx ring (CIK).
2023  * Provides a basic gfx ring test to verify that the ring is working.
2024  * Used by gfx_v7_0_cp_gfx_resume();
2025  * Returns 0 on success, error on failure.
2026  */
2027 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2028 {
2029 	struct amdgpu_device *adev = ring->adev;
2030 	uint32_t tmp = 0;
2031 	unsigned i;
2032 	int r;
2033 
2034 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2035 	r = amdgpu_ring_alloc(ring, 3);
2036 	if (r)
2037 		return r;
2038 
2039 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2040 	amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START);
2041 	amdgpu_ring_write(ring, 0xDEADBEEF);
2042 	amdgpu_ring_commit(ring);
2043 
2044 	for (i = 0; i < adev->usec_timeout; i++) {
2045 		tmp = RREG32(mmSCRATCH_REG0);
2046 		if (tmp == 0xDEADBEEF)
2047 			break;
2048 		udelay(1);
2049 	}
2050 	if (i >= adev->usec_timeout)
2051 		r = -ETIMEDOUT;
2052 	return r;
2053 }
2054 
2055 /**
2056  * gfx_v7_0_ring_emit_hdp_flush - emit an hdp flush on the cp
2057  *
2058  * @ring: amdgpu_ring structure holding ring information
2059  *
2060  * Emits an hdp flush on the cp.
2061  */
2062 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2063 {
2064 	u32 ref_and_mask;
2065 	int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2066 
2067 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2068 		switch (ring->me) {
2069 		case 1:
2070 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2071 			break;
2072 		case 2:
2073 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2074 			break;
2075 		default:
2076 			return;
2077 		}
2078 	} else {
2079 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2080 	}
2081 
2082 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2083 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2084 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2085 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2086 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2087 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2088 	amdgpu_ring_write(ring, ref_and_mask);
2089 	amdgpu_ring_write(ring, ref_and_mask);
2090 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2091 }
2092 
2093 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2094 {
2095 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2096 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2097 		EVENT_INDEX(4));
2098 
2099 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2100 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2101 		EVENT_INDEX(0));
2102 }
2103 
2104 /**
2105  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2106  *
2107  * @ring: amdgpu_ring structure holding ring information
2108  * @addr: address
2109  * @seq: sequence number
2110  * @flags: fence related flags
2111  *
2112  * Emits a fence sequence number on the gfx ring and flushes
2113  * GPU caches.
2114  */
2115 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2116 					 u64 seq, unsigned flags)
2117 {
2118 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2119 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2120 	/* Workaround for cache flush problems. First send a dummy EOP
2121 	 * event down the pipe with seq one below.
2122 	 */
2123 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2124 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2125 				 EOP_TC_ACTION_EN |
2126 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2127 				 EVENT_INDEX(5)));
2128 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2129 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2130 				DATA_SEL(1) | INT_SEL(0));
2131 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2132 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2133 
2134 	/* Then send the real EOP event down the pipe. */
2135 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2136 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2137 				 EOP_TC_ACTION_EN |
2138 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2139 				 EVENT_INDEX(5)));
2140 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2141 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2142 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2143 	amdgpu_ring_write(ring, lower_32_bits(seq));
2144 	amdgpu_ring_write(ring, upper_32_bits(seq));
2145 }
2146 
2147 /**
2148  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2149  *
2150  * @ring: amdgpu_ring structure holding ring information
2151  * @addr: address
2152  * @seq: sequence number
2153  * @flags: fence related flags
2154  *
2155  * Emits a fence sequence number on the compute ring and flushes
2156  * GPU caches.
2157  */
2158 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2159 					     u64 addr, u64 seq,
2160 					     unsigned flags)
2161 {
2162 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2163 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2164 
2165 	/* RELEASE_MEM - flush caches, send int */
2166 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2167 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2168 				 EOP_TC_ACTION_EN |
2169 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2170 				 EVENT_INDEX(5)));
2171 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2172 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2173 	amdgpu_ring_write(ring, upper_32_bits(addr));
2174 	amdgpu_ring_write(ring, lower_32_bits(seq));
2175 	amdgpu_ring_write(ring, upper_32_bits(seq));
2176 }
2177 
2178 /*
2179  * IB stuff
2180  */
2181 /**
2182  * gfx_v7_0_ring_emit_ib_gfx - emit an IB (Indirect Buffer) on the ring
2183  *
2184  * @ring: amdgpu_ring structure holding ring information
2185  * @job: job to retrieve vmid from
2186  * @ib: amdgpu indirect buffer object
2187  * @flags: options (AMDGPU_HAVE_CTX_SWITCH)
2188  *
2189  * Emits an DE (drawing engine) or CE (constant engine) IB
2190  * on the gfx ring.  IBs are usually generated by userspace
2191  * acceleration drivers and submitted to the kernel for
2192  * scheduling on the ring.  This function schedules the IB
2193  * on the gfx ring for execution by the GPU.
2194  */
2195 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2196 					struct amdgpu_job *job,
2197 					struct amdgpu_ib *ib,
2198 					uint32_t flags)
2199 {
2200 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2201 	u32 header, control = 0;
2202 
2203 	/* insert SWITCH_BUFFER packet before first IB in the ring frame */
2204 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2205 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2206 		amdgpu_ring_write(ring, 0);
2207 	}
2208 
2209 	if (ib->flags & AMDGPU_IB_FLAG_CE)
2210 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2211 	else
2212 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2213 
2214 	control |= ib->length_dw | (vmid << 24);
2215 
2216 	amdgpu_ring_write(ring, header);
2217 	amdgpu_ring_write(ring,
2218 #ifdef __BIG_ENDIAN
2219 			  (2 << 0) |
2220 #endif
2221 			  (ib->gpu_addr & 0xFFFFFFFC));
2222 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2223 	amdgpu_ring_write(ring, control);
2224 }
2225 
2226 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
2227 					  struct amdgpu_job *job,
2228 					  struct amdgpu_ib *ib,
2229 					  uint32_t flags)
2230 {
2231 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2232 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
2233 
2234 	/* Currently, there is a high possibility to get wave ID mismatch
2235 	 * between ME and GDS, leading to a hw deadlock, because ME generates
2236 	 * different wave IDs than the GDS expects. This situation happens
2237 	 * randomly when at least 5 compute pipes use GDS ordered append.
2238 	 * The wave IDs generated by ME are also wrong after suspend/resume.
2239 	 * Those are probably bugs somewhere else in the kernel driver.
2240 	 *
2241 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
2242 	 * GDS to 0 for this ring (me/pipe).
2243 	 */
2244 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
2245 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2246 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
2247 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
2248 	}
2249 
2250 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2251 	amdgpu_ring_write(ring,
2252 #ifdef __BIG_ENDIAN
2253 					  (2 << 0) |
2254 #endif
2255 					  (ib->gpu_addr & 0xFFFFFFFC));
2256 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2257 	amdgpu_ring_write(ring, control);
2258 }
2259 
2260 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2261 {
2262 	uint32_t dw2 = 0;
2263 
2264 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2265 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2266 		gfx_v7_0_ring_emit_vgt_flush(ring);
2267 		/* set load_global_config & load_global_uconfig */
2268 		dw2 |= 0x8001;
2269 		/* set load_cs_sh_regs */
2270 		dw2 |= 0x01000000;
2271 		/* set load_per_context_state & load_gfx_sh_regs */
2272 		dw2 |= 0x10002;
2273 	}
2274 
2275 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2276 	amdgpu_ring_write(ring, dw2);
2277 	amdgpu_ring_write(ring, 0);
2278 }
2279 
2280 /**
2281  * gfx_v7_0_ring_test_ib - basic ring IB test
2282  *
2283  * @ring: amdgpu_ring structure holding ring information
2284  * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT
2285  *
2286  * Allocate an IB and execute it on the gfx ring (CIK).
2287  * Provides a basic gfx ring test to verify that IBs are working.
2288  * Returns 0 on success, error on failure.
2289  */
2290 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2291 {
2292 	struct amdgpu_device *adev = ring->adev;
2293 	struct amdgpu_ib ib;
2294 	struct dma_fence *f = NULL;
2295 	uint32_t tmp = 0;
2296 	long r;
2297 
2298 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2299 	memset(&ib, 0, sizeof(ib));
2300 	r = amdgpu_ib_get(adev, NULL, 256, AMDGPU_IB_POOL_DIRECT, &ib);
2301 	if (r)
2302 		return r;
2303 
2304 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2305 	ib.ptr[1] = mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START;
2306 	ib.ptr[2] = 0xDEADBEEF;
2307 	ib.length_dw = 3;
2308 
2309 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2310 	if (r)
2311 		goto error;
2312 
2313 	r = dma_fence_wait_timeout(f, false, timeout);
2314 	if (r == 0) {
2315 		r = -ETIMEDOUT;
2316 		goto error;
2317 	} else if (r < 0) {
2318 		goto error;
2319 	}
2320 	tmp = RREG32(mmSCRATCH_REG0);
2321 	if (tmp == 0xDEADBEEF)
2322 		r = 0;
2323 	else
2324 		r = -EINVAL;
2325 
2326 error:
2327 	amdgpu_ib_free(adev, &ib, NULL);
2328 	dma_fence_put(f);
2329 	return r;
2330 }
2331 
2332 /*
2333  * CP.
2334  * On CIK, gfx and compute now have independent command processors.
2335  *
2336  * GFX
2337  * Gfx consists of a single ring and can process both gfx jobs and
2338  * compute jobs.  The gfx CP consists of three microengines (ME):
2339  * PFP - Pre-Fetch Parser
2340  * ME - Micro Engine
2341  * CE - Constant Engine
2342  * The PFP and ME make up what is considered the Drawing Engine (DE).
2343  * The CE is an asynchronous engine used for updating buffer desciptors
2344  * used by the DE so that they can be loaded into cache in parallel
2345  * while the DE is processing state update packets.
2346  *
2347  * Compute
2348  * The compute CP consists of two microengines (ME):
2349  * MEC1 - Compute MicroEngine 1
2350  * MEC2 - Compute MicroEngine 2
2351  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2352  * The queues are exposed to userspace and are programmed directly
2353  * by the compute runtime.
2354  */
2355 /**
2356  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2357  *
2358  * @adev: amdgpu_device pointer
2359  * @enable: enable or disable the MEs
2360  *
2361  * Halts or unhalts the gfx MEs.
2362  */
2363 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2364 {
2365 	if (enable)
2366 		WREG32(mmCP_ME_CNTL, 0);
2367 	else
2368 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK |
2369 				      CP_ME_CNTL__PFP_HALT_MASK |
2370 				      CP_ME_CNTL__CE_HALT_MASK));
2371 	udelay(50);
2372 }
2373 
2374 /**
2375  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2376  *
2377  * @adev: amdgpu_device pointer
2378  *
2379  * Loads the gfx PFP, ME, and CE ucode.
2380  * Returns 0 for success, -EINVAL if the ucode is not available.
2381  */
2382 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2383 {
2384 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2385 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2386 	const struct gfx_firmware_header_v1_0 *me_hdr;
2387 	const __le32 *fw_data;
2388 	unsigned i, fw_size;
2389 
2390 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2391 		return -EINVAL;
2392 
2393 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2394 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2395 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2396 
2397 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2398 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2399 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2400 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2401 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2402 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2403 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2404 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2405 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2406 
2407 	gfx_v7_0_cp_gfx_enable(adev, false);
2408 
2409 	/* PFP */
2410 	fw_data = (const __le32 *)
2411 		(adev->gfx.pfp_fw->data +
2412 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2413 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2414 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2415 	for (i = 0; i < fw_size; i++)
2416 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2417 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2418 
2419 	/* CE */
2420 	fw_data = (const __le32 *)
2421 		(adev->gfx.ce_fw->data +
2422 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2423 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2424 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2425 	for (i = 0; i < fw_size; i++)
2426 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2427 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2428 
2429 	/* ME */
2430 	fw_data = (const __le32 *)
2431 		(adev->gfx.me_fw->data +
2432 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2433 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2434 	WREG32(mmCP_ME_RAM_WADDR, 0);
2435 	for (i = 0; i < fw_size; i++)
2436 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2437 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2438 
2439 	return 0;
2440 }
2441 
2442 /**
2443  * gfx_v7_0_cp_gfx_start - start the gfx ring
2444  *
2445  * @adev: amdgpu_device pointer
2446  *
2447  * Enables the ring and loads the clear state context and other
2448  * packets required to init the ring.
2449  * Returns 0 for success, error for failure.
2450  */
2451 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2452 {
2453 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2454 	const struct cs_section_def *sect = NULL;
2455 	const struct cs_extent_def *ext = NULL;
2456 	int r, i;
2457 
2458 	/* init the CP */
2459 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2460 	WREG32(mmCP_ENDIAN_SWAP, 0);
2461 	WREG32(mmCP_DEVICE_ID, 1);
2462 
2463 	gfx_v7_0_cp_gfx_enable(adev, true);
2464 
2465 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2466 	if (r) {
2467 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2468 		return r;
2469 	}
2470 
2471 	/* init the CE partitions.  CE only used for gfx on CIK */
2472 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2473 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2474 	amdgpu_ring_write(ring, 0x8000);
2475 	amdgpu_ring_write(ring, 0x8000);
2476 
2477 	/* clear state buffer */
2478 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2479 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2480 
2481 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2482 	amdgpu_ring_write(ring, 0x80000000);
2483 	amdgpu_ring_write(ring, 0x80000000);
2484 
2485 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2486 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2487 			if (sect->id == SECT_CONTEXT) {
2488 				amdgpu_ring_write(ring,
2489 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2490 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2491 				for (i = 0; i < ext->reg_count; i++)
2492 					amdgpu_ring_write(ring, ext->extent[i]);
2493 			}
2494 		}
2495 	}
2496 
2497 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2498 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2499 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
2500 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
2501 
2502 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2503 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2504 
2505 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2506 	amdgpu_ring_write(ring, 0);
2507 
2508 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2509 	amdgpu_ring_write(ring, 0x00000316);
2510 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2511 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2512 
2513 	amdgpu_ring_commit(ring);
2514 
2515 	return 0;
2516 }
2517 
2518 /**
2519  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2520  *
2521  * @adev: amdgpu_device pointer
2522  *
2523  * Program the location and size of the gfx ring buffer
2524  * and test it to make sure it's working.
2525  * Returns 0 for success, error for failure.
2526  */
2527 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2528 {
2529 	struct amdgpu_ring *ring;
2530 	u32 tmp;
2531 	u32 rb_bufsz;
2532 	u64 rb_addr, rptr_addr;
2533 	int r;
2534 
2535 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2536 	if (adev->asic_type != CHIP_HAWAII)
2537 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2538 
2539 	/* Set the write pointer delay */
2540 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2541 
2542 	/* set the RB to use vmid 0 */
2543 	WREG32(mmCP_RB_VMID, 0);
2544 
2545 	WREG32(mmSCRATCH_ADDR, 0);
2546 
2547 	/* ring 0 - compute and gfx */
2548 	/* Set ring buffer size */
2549 	ring = &adev->gfx.gfx_ring[0];
2550 	rb_bufsz = order_base_2(ring->ring_size / 8);
2551 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2552 #ifdef __BIG_ENDIAN
2553 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2554 #endif
2555 	WREG32(mmCP_RB0_CNTL, tmp);
2556 
2557 	/* Initialize the ring buffer's read and write pointers */
2558 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2559 	ring->wptr = 0;
2560 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2561 
2562 	/* set the wb address wether it's enabled or not */
2563 	rptr_addr = ring->rptr_gpu_addr;
2564 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2565 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2566 
2567 	/* scratch register shadowing is no longer supported */
2568 	WREG32(mmSCRATCH_UMSK, 0);
2569 
2570 	mdelay(1);
2571 	WREG32(mmCP_RB0_CNTL, tmp);
2572 
2573 	rb_addr = ring->gpu_addr >> 8;
2574 	WREG32(mmCP_RB0_BASE, rb_addr);
2575 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2576 
2577 	/* start the ring */
2578 	gfx_v7_0_cp_gfx_start(adev);
2579 	r = amdgpu_ring_test_helper(ring);
2580 	if (r)
2581 		return r;
2582 
2583 	return 0;
2584 }
2585 
2586 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2587 {
2588 	return *ring->rptr_cpu_addr;
2589 }
2590 
2591 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2592 {
2593 	struct amdgpu_device *adev = ring->adev;
2594 
2595 	return RREG32(mmCP_RB0_WPTR);
2596 }
2597 
2598 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2599 {
2600 	struct amdgpu_device *adev = ring->adev;
2601 
2602 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2603 	(void)RREG32(mmCP_RB0_WPTR);
2604 }
2605 
2606 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2607 {
2608 	/* XXX check if swapping is necessary on BE */
2609 	return *ring->wptr_cpu_addr;
2610 }
2611 
2612 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2613 {
2614 	struct amdgpu_device *adev = ring->adev;
2615 
2616 	/* XXX check if swapping is necessary on BE */
2617 	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
2618 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2619 }
2620 
2621 /**
2622  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2623  *
2624  * @adev: amdgpu_device pointer
2625  * @enable: enable or disable the MEs
2626  *
2627  * Halts or unhalts the compute MEs.
2628  */
2629 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2630 {
2631 	if (enable)
2632 		WREG32(mmCP_MEC_CNTL, 0);
2633 	else
2634 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK |
2635 				       CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2636 	udelay(50);
2637 }
2638 
2639 /**
2640  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2641  *
2642  * @adev: amdgpu_device pointer
2643  *
2644  * Loads the compute MEC1&2 ucode.
2645  * Returns 0 for success, -EINVAL if the ucode is not available.
2646  */
2647 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2648 {
2649 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2650 	const __le32 *fw_data;
2651 	unsigned i, fw_size;
2652 
2653 	if (!adev->gfx.mec_fw)
2654 		return -EINVAL;
2655 
2656 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2657 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2658 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2659 	adev->gfx.mec_feature_version = le32_to_cpu(
2660 					mec_hdr->ucode_feature_version);
2661 
2662 	gfx_v7_0_cp_compute_enable(adev, false);
2663 
2664 	/* MEC1 */
2665 	fw_data = (const __le32 *)
2666 		(adev->gfx.mec_fw->data +
2667 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2668 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2669 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2670 	for (i = 0; i < fw_size; i++)
2671 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2672 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2673 
2674 	if (adev->asic_type == CHIP_KAVERI) {
2675 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2676 
2677 		if (!adev->gfx.mec2_fw)
2678 			return -EINVAL;
2679 
2680 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2681 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2682 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2683 		adev->gfx.mec2_feature_version = le32_to_cpu(
2684 				mec2_hdr->ucode_feature_version);
2685 
2686 		/* MEC2 */
2687 		fw_data = (const __le32 *)
2688 			(adev->gfx.mec2_fw->data +
2689 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2690 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2691 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2692 		for (i = 0; i < fw_size; i++)
2693 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2694 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2695 	}
2696 
2697 	return 0;
2698 }
2699 
2700 /**
2701  * gfx_v7_0_cp_compute_fini - stop the compute queues
2702  *
2703  * @adev: amdgpu_device pointer
2704  *
2705  * Stop the compute queues and tear down the driver queue
2706  * info.
2707  */
2708 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2709 {
2710 	int i;
2711 
2712 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2713 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2714 
2715 		amdgpu_bo_free_kernel(&ring->mqd_obj, NULL, NULL);
2716 	}
2717 }
2718 
2719 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2720 {
2721 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
2722 }
2723 
2724 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2725 {
2726 	int r;
2727 	u32 *hpd;
2728 	size_t mec_hpd_size;
2729 
2730 	bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
2731 
2732 	/* take ownership of the relevant compute queues */
2733 	amdgpu_gfx_compute_queue_acquire(adev);
2734 
2735 	/* allocate space for ALL pipes (even the ones we don't own) */
2736 	mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec
2737 		* GFX7_MEC_HPD_SIZE * 2;
2738 
2739 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
2740 				      AMDGPU_GEM_DOMAIN_VRAM |
2741 				      AMDGPU_GEM_DOMAIN_GTT,
2742 				      &adev->gfx.mec.hpd_eop_obj,
2743 				      &adev->gfx.mec.hpd_eop_gpu_addr,
2744 				      (void **)&hpd);
2745 	if (r) {
2746 		dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r);
2747 		gfx_v7_0_mec_fini(adev);
2748 		return r;
2749 	}
2750 
2751 	/* clear memory.  Not sure if this is required or not */
2752 	memset(hpd, 0, mec_hpd_size);
2753 
2754 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2755 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2756 
2757 	return 0;
2758 }
2759 
2760 struct hqd_registers {
2761 	u32 cp_mqd_base_addr;
2762 	u32 cp_mqd_base_addr_hi;
2763 	u32 cp_hqd_active;
2764 	u32 cp_hqd_vmid;
2765 	u32 cp_hqd_persistent_state;
2766 	u32 cp_hqd_pipe_priority;
2767 	u32 cp_hqd_queue_priority;
2768 	u32 cp_hqd_quantum;
2769 	u32 cp_hqd_pq_base;
2770 	u32 cp_hqd_pq_base_hi;
2771 	u32 cp_hqd_pq_rptr;
2772 	u32 cp_hqd_pq_rptr_report_addr;
2773 	u32 cp_hqd_pq_rptr_report_addr_hi;
2774 	u32 cp_hqd_pq_wptr_poll_addr;
2775 	u32 cp_hqd_pq_wptr_poll_addr_hi;
2776 	u32 cp_hqd_pq_doorbell_control;
2777 	u32 cp_hqd_pq_wptr;
2778 	u32 cp_hqd_pq_control;
2779 	u32 cp_hqd_ib_base_addr;
2780 	u32 cp_hqd_ib_base_addr_hi;
2781 	u32 cp_hqd_ib_rptr;
2782 	u32 cp_hqd_ib_control;
2783 	u32 cp_hqd_iq_timer;
2784 	u32 cp_hqd_iq_rptr;
2785 	u32 cp_hqd_dequeue_request;
2786 	u32 cp_hqd_dma_offload;
2787 	u32 cp_hqd_sema_cmd;
2788 	u32 cp_hqd_msg_type;
2789 	u32 cp_hqd_atomic0_preop_lo;
2790 	u32 cp_hqd_atomic0_preop_hi;
2791 	u32 cp_hqd_atomic1_preop_lo;
2792 	u32 cp_hqd_atomic1_preop_hi;
2793 	u32 cp_hqd_hq_scheduler0;
2794 	u32 cp_hqd_hq_scheduler1;
2795 	u32 cp_mqd_control;
2796 };
2797 
2798 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev,
2799 				       int mec, int pipe)
2800 {
2801 	u64 eop_gpu_addr;
2802 	u32 tmp;
2803 	size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe)
2804 			    * GFX7_MEC_HPD_SIZE * 2;
2805 
2806 	mutex_lock(&adev->srbm_mutex);
2807 	eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset;
2808 
2809 	cik_srbm_select(adev, mec + 1, pipe, 0, 0);
2810 
2811 	/* write the EOP addr */
2812 	WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2813 	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2814 
2815 	/* set the VMID assigned */
2816 	WREG32(mmCP_HPD_EOP_VMID, 0);
2817 
2818 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2819 	tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2820 	tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2821 	tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8);
2822 	WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2823 
2824 	cik_srbm_select(adev, 0, 0, 0, 0);
2825 	mutex_unlock(&adev->srbm_mutex);
2826 }
2827 
2828 static int gfx_v7_0_mqd_deactivate(struct amdgpu_device *adev)
2829 {
2830 	int i;
2831 
2832 	/* disable the queue if it's active */
2833 	if (RREG32(mmCP_HQD_ACTIVE) & 1) {
2834 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
2835 		for (i = 0; i < adev->usec_timeout; i++) {
2836 			if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
2837 				break;
2838 			udelay(1);
2839 		}
2840 
2841 		if (i == adev->usec_timeout)
2842 			return -ETIMEDOUT;
2843 
2844 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
2845 		WREG32(mmCP_HQD_PQ_RPTR, 0);
2846 		WREG32(mmCP_HQD_PQ_WPTR, 0);
2847 	}
2848 
2849 	return 0;
2850 }
2851 
2852 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev,
2853 			     struct cik_mqd *mqd,
2854 			     uint64_t mqd_gpu_addr,
2855 			     struct amdgpu_ring *ring)
2856 {
2857 	u64 hqd_gpu_addr;
2858 	u64 wb_gpu_addr;
2859 
2860 	/* init the mqd struct */
2861 	memset(mqd, 0, sizeof(struct cik_mqd));
2862 
2863 	mqd->header = 0xC0310800;
2864 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2865 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2866 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2867 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2868 
2869 	/* enable doorbell? */
2870 	mqd->cp_hqd_pq_doorbell_control =
2871 		RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2872 	if (ring->use_doorbell)
2873 		mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2874 	else
2875 		mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2876 
2877 	/* set the pointer to the MQD */
2878 	mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc;
2879 	mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
2880 
2881 	/* set MQD vmid to 0 */
2882 	mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
2883 	mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
2884 
2885 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2886 	hqd_gpu_addr = ring->gpu_addr >> 8;
2887 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2888 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2889 
2890 	/* set up the HQD, this is similar to CP_RB0_CNTL */
2891 	mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
2892 	mqd->cp_hqd_pq_control &=
2893 		~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
2894 				CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
2895 
2896 	mqd->cp_hqd_pq_control |=
2897 		order_base_2(ring->ring_size / 8);
2898 	mqd->cp_hqd_pq_control |=
2899 		(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
2900 #ifdef __BIG_ENDIAN
2901 	mqd->cp_hqd_pq_control |=
2902 		2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
2903 #endif
2904 	mqd->cp_hqd_pq_control &=
2905 		~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
2906 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
2907 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
2908 	mqd->cp_hqd_pq_control |=
2909 		CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
2910 		CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
2911 
2912 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2913 	wb_gpu_addr = ring->wptr_gpu_addr;
2914 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2915 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2916 
2917 	/* set the wb address wether it's enabled or not */
2918 	wb_gpu_addr = ring->rptr_gpu_addr;
2919 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2920 	mqd->cp_hqd_pq_rptr_report_addr_hi =
2921 		upper_32_bits(wb_gpu_addr) & 0xffff;
2922 
2923 	/* enable the doorbell if requested */
2924 	if (ring->use_doorbell) {
2925 		mqd->cp_hqd_pq_doorbell_control =
2926 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2927 		mqd->cp_hqd_pq_doorbell_control &=
2928 			~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
2929 		mqd->cp_hqd_pq_doorbell_control |=
2930 			(ring->doorbell_index <<
2931 			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
2932 		mqd->cp_hqd_pq_doorbell_control |=
2933 			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2934 		mqd->cp_hqd_pq_doorbell_control &=
2935 			~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
2936 					CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
2937 
2938 	} else {
2939 		mqd->cp_hqd_pq_doorbell_control = 0;
2940 	}
2941 
2942 	/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
2943 	ring->wptr = 0;
2944 	mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr);
2945 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2946 
2947 	/* set the vmid for the queue */
2948 	mqd->cp_hqd_vmid = 0;
2949 
2950 	/* defaults */
2951 	mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL);
2952 	mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR);
2953 	mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI);
2954 	mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR);
2955 	mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE);
2956 	mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD);
2957 	mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE);
2958 	mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO);
2959 	mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI);
2960 	mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO);
2961 	mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI);
2962 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2963 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
2964 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
2965 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
2966 	mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR);
2967 
2968 	/* activate the queue */
2969 	mqd->cp_hqd_active = 1;
2970 }
2971 
2972 static int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd)
2973 {
2974 	uint32_t tmp;
2975 	uint32_t mqd_reg;
2976 	uint32_t *mqd_data;
2977 
2978 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */
2979 	mqd_data = &mqd->cp_mqd_base_addr_lo;
2980 
2981 	/* disable wptr polling */
2982 	tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
2983 	tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0);
2984 	WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
2985 
2986 	/* program all HQD registers */
2987 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++)
2988 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
2989 
2990 	/* activate the HQD */
2991 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
2992 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
2993 
2994 	return 0;
2995 }
2996 
2997 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id)
2998 {
2999 	int r;
3000 	u64 mqd_gpu_addr;
3001 	struct cik_mqd *mqd;
3002 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
3003 
3004 	r = amdgpu_bo_create_reserved(adev, sizeof(struct cik_mqd), PAGE_SIZE,
3005 				      AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj,
3006 				      &mqd_gpu_addr, (void **)&mqd);
3007 	if (r) {
3008 		dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
3009 		return r;
3010 	}
3011 
3012 	mutex_lock(&adev->srbm_mutex);
3013 	cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3014 
3015 	gfx_v7_0_mqd_init(adev, mqd, mqd_gpu_addr, ring);
3016 	gfx_v7_0_mqd_deactivate(adev);
3017 	gfx_v7_0_mqd_commit(adev, mqd);
3018 
3019 	cik_srbm_select(adev, 0, 0, 0, 0);
3020 	mutex_unlock(&adev->srbm_mutex);
3021 
3022 	amdgpu_bo_kunmap(ring->mqd_obj);
3023 	amdgpu_bo_unreserve(ring->mqd_obj);
3024 	return 0;
3025 }
3026 
3027 /**
3028  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
3029  *
3030  * @adev: amdgpu_device pointer
3031  *
3032  * Program the compute queues and test them to make sure they
3033  * are working.
3034  * Returns 0 for success, error for failure.
3035  */
3036 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
3037 {
3038 	int r, i, j;
3039 	u32 tmp;
3040 	struct amdgpu_ring *ring;
3041 
3042 	/* fix up chicken bits */
3043 	tmp = RREG32(mmCP_CPF_DEBUG);
3044 	tmp |= (1 << 23);
3045 	WREG32(mmCP_CPF_DEBUG, tmp);
3046 
3047 	/* init all pipes (even the ones we don't own) */
3048 	for (i = 0; i < adev->gfx.mec.num_mec; i++)
3049 		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++)
3050 			gfx_v7_0_compute_pipe_init(adev, i, j);
3051 
3052 	/* init the queues */
3053 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3054 		r = gfx_v7_0_compute_queue_init(adev, i);
3055 		if (r) {
3056 			gfx_v7_0_cp_compute_fini(adev);
3057 			return r;
3058 		}
3059 	}
3060 
3061 	gfx_v7_0_cp_compute_enable(adev, true);
3062 
3063 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3064 		ring = &adev->gfx.compute_ring[i];
3065 		amdgpu_ring_test_helper(ring);
3066 	}
3067 
3068 	return 0;
3069 }
3070 
3071 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3072 {
3073 	gfx_v7_0_cp_gfx_enable(adev, enable);
3074 	gfx_v7_0_cp_compute_enable(adev, enable);
3075 }
3076 
3077 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3078 {
3079 	int r;
3080 
3081 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3082 	if (r)
3083 		return r;
3084 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3085 	if (r)
3086 		return r;
3087 
3088 	return 0;
3089 }
3090 
3091 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3092 					       bool enable)
3093 {
3094 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3095 
3096 	if (enable)
3097 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3098 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3099 	else
3100 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3101 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3102 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3103 }
3104 
3105 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3106 {
3107 	int r;
3108 
3109 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3110 
3111 	r = gfx_v7_0_cp_load_microcode(adev);
3112 	if (r)
3113 		return r;
3114 
3115 	r = gfx_v7_0_cp_gfx_resume(adev);
3116 	if (r)
3117 		return r;
3118 	r = gfx_v7_0_cp_compute_resume(adev);
3119 	if (r)
3120 		return r;
3121 
3122 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3123 
3124 	return 0;
3125 }
3126 
3127 /**
3128  * gfx_v7_0_ring_emit_pipeline_sync - cik vm flush using the CP
3129  *
3130  * @ring: the ring to emit the commands to
3131  *
3132  * Sync the command pipeline with the PFP. E.g. wait for everything
3133  * to be completed.
3134  */
3135 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3136 {
3137 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3138 	uint32_t seq = ring->fence_drv.sync_seq;
3139 	uint64_t addr = ring->fence_drv.gpu_addr;
3140 
3141 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3142 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3143 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3144 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3145 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3146 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3147 	amdgpu_ring_write(ring, seq);
3148 	amdgpu_ring_write(ring, 0xffffffff);
3149 	amdgpu_ring_write(ring, 4); /* poll interval */
3150 
3151 	if (usepfp) {
3152 		/* sync CE with ME to prevent CE fetch CEIB before context switch done */
3153 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3154 		amdgpu_ring_write(ring, 0);
3155 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3156 		amdgpu_ring_write(ring, 0);
3157 	}
3158 }
3159 
3160 /*
3161  * vm
3162  * VMID 0 is the physical GPU addresses as used by the kernel.
3163  * VMIDs 1-15 are used for userspace clients and are handled
3164  * by the amdgpu vm/hsa code.
3165  */
3166 /**
3167  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3168  *
3169  * @ring: amdgpu_ring pointer
3170  * @vmid: vmid number to use
3171  * @pd_addr: address
3172  *
3173  * Update the page table base and flush the VM TLB
3174  * using the CP (CIK).
3175  */
3176 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3177 					unsigned vmid, uint64_t pd_addr)
3178 {
3179 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3180 
3181 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3182 
3183 	/* wait for the invalidate to complete */
3184 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3185 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3186 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3187 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3188 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3189 	amdgpu_ring_write(ring, 0);
3190 	amdgpu_ring_write(ring, 0); /* ref */
3191 	amdgpu_ring_write(ring, 0); /* mask */
3192 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3193 
3194 	/* compute doesn't have PFP */
3195 	if (usepfp) {
3196 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3197 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3198 		amdgpu_ring_write(ring, 0x0);
3199 
3200 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3201 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3202 		amdgpu_ring_write(ring, 0);
3203 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3204 		amdgpu_ring_write(ring, 0);
3205 	}
3206 }
3207 
3208 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
3209 				    uint32_t reg, uint32_t val)
3210 {
3211 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3212 
3213 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3214 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3215 				 WRITE_DATA_DST_SEL(0)));
3216 	amdgpu_ring_write(ring, reg);
3217 	amdgpu_ring_write(ring, 0);
3218 	amdgpu_ring_write(ring, val);
3219 }
3220 
3221 /*
3222  * RLC
3223  * The RLC is a multi-purpose microengine that handles a
3224  * variety of functions.
3225  */
3226 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3227 {
3228 	const u32 *src_ptr;
3229 	u32 dws;
3230 	const struct cs_section_def *cs_data;
3231 	int r;
3232 
3233 	/* allocate rlc buffers */
3234 	if (adev->flags & AMD_IS_APU) {
3235 		if (adev->asic_type == CHIP_KAVERI) {
3236 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3237 			adev->gfx.rlc.reg_list_size =
3238 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3239 		} else {
3240 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3241 			adev->gfx.rlc.reg_list_size =
3242 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3243 		}
3244 	}
3245 	adev->gfx.rlc.cs_data = ci_cs_data;
3246 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3247 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3248 
3249 	src_ptr = adev->gfx.rlc.reg_list;
3250 	dws = adev->gfx.rlc.reg_list_size;
3251 	dws += (5 * 16) + 48 + 48 + 64;
3252 
3253 	cs_data = adev->gfx.rlc.cs_data;
3254 
3255 	if (src_ptr) {
3256 		/* init save restore block */
3257 		r = amdgpu_gfx_rlc_init_sr(adev, dws);
3258 		if (r)
3259 			return r;
3260 	}
3261 
3262 	if (cs_data) {
3263 		/* init clear state block */
3264 		r = amdgpu_gfx_rlc_init_csb(adev);
3265 		if (r)
3266 			return r;
3267 	}
3268 
3269 	if (adev->gfx.rlc.cp_table_size) {
3270 		r = amdgpu_gfx_rlc_init_cpt(adev);
3271 		if (r)
3272 			return r;
3273 	}
3274 
3275 	/* init spm vmid with 0xf */
3276 	if (adev->gfx.rlc.funcs->update_spm_vmid)
3277 		adev->gfx.rlc.funcs->update_spm_vmid(adev, NULL, 0xf);
3278 
3279 	return 0;
3280 }
3281 
3282 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3283 {
3284 	u32 tmp;
3285 
3286 	tmp = RREG32(mmRLC_LB_CNTL);
3287 	if (enable)
3288 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3289 	else
3290 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3291 	WREG32(mmRLC_LB_CNTL, tmp);
3292 }
3293 
3294 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3295 {
3296 	u32 i, j, k;
3297 	u32 mask;
3298 
3299 	mutex_lock(&adev->grbm_idx_mutex);
3300 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3301 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3302 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3303 			for (k = 0; k < adev->usec_timeout; k++) {
3304 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3305 					break;
3306 				udelay(1);
3307 			}
3308 		}
3309 	}
3310 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3311 	mutex_unlock(&adev->grbm_idx_mutex);
3312 
3313 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3314 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3315 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3316 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3317 	for (k = 0; k < adev->usec_timeout; k++) {
3318 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3319 			break;
3320 		udelay(1);
3321 	}
3322 }
3323 
3324 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3325 {
3326 	u32 tmp;
3327 
3328 	tmp = RREG32(mmRLC_CNTL);
3329 	if (tmp != rlc)
3330 		WREG32(mmRLC_CNTL, rlc);
3331 }
3332 
3333 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3334 {
3335 	u32 data, orig;
3336 
3337 	orig = data = RREG32(mmRLC_CNTL);
3338 
3339 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3340 		u32 i;
3341 
3342 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3343 		WREG32(mmRLC_CNTL, data);
3344 
3345 		for (i = 0; i < adev->usec_timeout; i++) {
3346 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3347 				break;
3348 			udelay(1);
3349 		}
3350 
3351 		gfx_v7_0_wait_for_rlc_serdes(adev);
3352 	}
3353 
3354 	return orig;
3355 }
3356 
3357 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev)
3358 {
3359 	return true;
3360 }
3361 
3362 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
3363 {
3364 	u32 tmp, i, mask;
3365 
3366 	tmp = 0x1 | (1 << 1);
3367 	WREG32(mmRLC_GPR_REG2, tmp);
3368 
3369 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3370 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3371 	for (i = 0; i < adev->usec_timeout; i++) {
3372 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3373 			break;
3374 		udelay(1);
3375 	}
3376 
3377 	for (i = 0; i < adev->usec_timeout; i++) {
3378 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3379 			break;
3380 		udelay(1);
3381 	}
3382 }
3383 
3384 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
3385 {
3386 	u32 tmp;
3387 
3388 	tmp = 0x1 | (0 << 1);
3389 	WREG32(mmRLC_GPR_REG2, tmp);
3390 }
3391 
3392 /**
3393  * gfx_v7_0_rlc_stop - stop the RLC ME
3394  *
3395  * @adev: amdgpu_device pointer
3396  *
3397  * Halt the RLC ME (MicroEngine) (CIK).
3398  */
3399 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3400 {
3401 	WREG32(mmRLC_CNTL, 0);
3402 
3403 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3404 
3405 	gfx_v7_0_wait_for_rlc_serdes(adev);
3406 }
3407 
3408 /**
3409  * gfx_v7_0_rlc_start - start the RLC ME
3410  *
3411  * @adev: amdgpu_device pointer
3412  *
3413  * Unhalt the RLC ME (MicroEngine) (CIK).
3414  */
3415 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3416 {
3417 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3418 
3419 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3420 
3421 	udelay(50);
3422 }
3423 
3424 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3425 {
3426 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3427 
3428 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3429 	WREG32(mmGRBM_SOFT_RESET, tmp);
3430 	udelay(50);
3431 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3432 	WREG32(mmGRBM_SOFT_RESET, tmp);
3433 	udelay(50);
3434 }
3435 
3436 /**
3437  * gfx_v7_0_rlc_resume - setup the RLC hw
3438  *
3439  * @adev: amdgpu_device pointer
3440  *
3441  * Initialize the RLC registers, load the ucode,
3442  * and start the RLC (CIK).
3443  * Returns 0 for success, -EINVAL if the ucode is not available.
3444  */
3445 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3446 {
3447 	const struct rlc_firmware_header_v1_0 *hdr;
3448 	const __le32 *fw_data;
3449 	unsigned i, fw_size;
3450 	u32 tmp;
3451 
3452 	if (!adev->gfx.rlc_fw)
3453 		return -EINVAL;
3454 
3455 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3456 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3457 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3458 	adev->gfx.rlc_feature_version = le32_to_cpu(
3459 					hdr->ucode_feature_version);
3460 
3461 	adev->gfx.rlc.funcs->stop(adev);
3462 
3463 	/* disable CG */
3464 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3465 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3466 
3467 	adev->gfx.rlc.funcs->reset(adev);
3468 
3469 	gfx_v7_0_init_pg(adev);
3470 
3471 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3472 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3473 
3474 	mutex_lock(&adev->grbm_idx_mutex);
3475 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3476 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3477 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3478 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3479 	mutex_unlock(&adev->grbm_idx_mutex);
3480 
3481 	WREG32(mmRLC_MC_CNTL, 0);
3482 	WREG32(mmRLC_UCODE_CNTL, 0);
3483 
3484 	fw_data = (const __le32 *)
3485 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3486 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3487 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3488 	for (i = 0; i < fw_size; i++)
3489 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3490 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3491 
3492 	/* XXX - find out what chips support lbpw */
3493 	gfx_v7_0_enable_lbpw(adev, false);
3494 
3495 	if (adev->asic_type == CHIP_BONAIRE)
3496 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3497 
3498 	adev->gfx.rlc.funcs->start(adev);
3499 
3500 	return 0;
3501 }
3502 
3503 static void gfx_v7_0_update_spm_vmid(struct amdgpu_device *adev, struct amdgpu_ring *ring, unsigned vmid)
3504 {
3505 	u32 data;
3506 
3507 	amdgpu_gfx_off_ctrl(adev, false);
3508 
3509 	data = RREG32(mmRLC_SPM_VMID);
3510 
3511 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
3512 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
3513 
3514 	WREG32(mmRLC_SPM_VMID, data);
3515 
3516 	amdgpu_gfx_off_ctrl(adev, true);
3517 }
3518 
3519 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3520 {
3521 	u32 data, orig, tmp, tmp2;
3522 
3523 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3524 
3525 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3526 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3527 
3528 		tmp = gfx_v7_0_halt_rlc(adev);
3529 
3530 		mutex_lock(&adev->grbm_idx_mutex);
3531 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3532 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3533 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3534 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3535 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3536 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3537 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3538 		mutex_unlock(&adev->grbm_idx_mutex);
3539 
3540 		gfx_v7_0_update_rlc(adev, tmp);
3541 
3542 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3543 		if (orig != data)
3544 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3545 
3546 	} else {
3547 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3548 
3549 		RREG32(mmCB_CGTT_SCLK_CTRL);
3550 		RREG32(mmCB_CGTT_SCLK_CTRL);
3551 		RREG32(mmCB_CGTT_SCLK_CTRL);
3552 		RREG32(mmCB_CGTT_SCLK_CTRL);
3553 
3554 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3555 		if (orig != data)
3556 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3557 
3558 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3559 	}
3560 }
3561 
3562 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3563 {
3564 	u32 data, orig, tmp = 0;
3565 
3566 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3567 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3568 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3569 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3570 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3571 				if (orig != data)
3572 					WREG32(mmCP_MEM_SLP_CNTL, data);
3573 			}
3574 		}
3575 
3576 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3577 		data |= 0x00000001;
3578 		data &= 0xfffffffd;
3579 		if (orig != data)
3580 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
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, 0);
3586 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3587 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3588 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3589 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3590 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3591 		mutex_unlock(&adev->grbm_idx_mutex);
3592 
3593 		gfx_v7_0_update_rlc(adev, tmp);
3594 
3595 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3596 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3597 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3598 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3599 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3600 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3601 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3602 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3603 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3604 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3605 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3606 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3607 			if (orig != data)
3608 				WREG32(mmCGTS_SM_CTRL_REG, data);
3609 		}
3610 	} else {
3611 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3612 		data |= 0x00000003;
3613 		if (orig != data)
3614 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3615 
3616 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3617 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3618 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3619 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3620 		}
3621 
3622 		data = RREG32(mmCP_MEM_SLP_CNTL);
3623 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3624 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3625 			WREG32(mmCP_MEM_SLP_CNTL, data);
3626 		}
3627 
3628 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3629 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3630 		if (orig != data)
3631 			WREG32(mmCGTS_SM_CTRL_REG, data);
3632 
3633 		tmp = gfx_v7_0_halt_rlc(adev);
3634 
3635 		mutex_lock(&adev->grbm_idx_mutex);
3636 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3637 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3638 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3639 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3640 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3641 		mutex_unlock(&adev->grbm_idx_mutex);
3642 
3643 		gfx_v7_0_update_rlc(adev, tmp);
3644 	}
3645 }
3646 
3647 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3648 			       bool enable)
3649 {
3650 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3651 	/* order matters! */
3652 	if (enable) {
3653 		gfx_v7_0_enable_mgcg(adev, true);
3654 		gfx_v7_0_enable_cgcg(adev, true);
3655 	} else {
3656 		gfx_v7_0_enable_cgcg(adev, false);
3657 		gfx_v7_0_enable_mgcg(adev, false);
3658 	}
3659 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3660 }
3661 
3662 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3663 						bool enable)
3664 {
3665 	u32 data, orig;
3666 
3667 	orig = data = RREG32(mmRLC_PG_CNTL);
3668 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3669 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3670 	else
3671 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3672 	if (orig != data)
3673 		WREG32(mmRLC_PG_CNTL, data);
3674 }
3675 
3676 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3677 						bool enable)
3678 {
3679 	u32 data, orig;
3680 
3681 	orig = data = RREG32(mmRLC_PG_CNTL);
3682 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3683 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3684 	else
3685 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3686 	if (orig != data)
3687 		WREG32(mmRLC_PG_CNTL, data);
3688 }
3689 
3690 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3691 {
3692 	u32 data, orig;
3693 
3694 	orig = data = RREG32(mmRLC_PG_CNTL);
3695 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3696 		data &= ~0x8000;
3697 	else
3698 		data |= 0x8000;
3699 	if (orig != data)
3700 		WREG32(mmRLC_PG_CNTL, data);
3701 }
3702 
3703 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3704 {
3705 	u32 data, orig;
3706 
3707 	orig = data = RREG32(mmRLC_PG_CNTL);
3708 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3709 		data &= ~0x2000;
3710 	else
3711 		data |= 0x2000;
3712 	if (orig != data)
3713 		WREG32(mmRLC_PG_CNTL, data);
3714 }
3715 
3716 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev)
3717 {
3718 	if (adev->asic_type == CHIP_KAVERI)
3719 		return 5;
3720 	else
3721 		return 4;
3722 }
3723 
3724 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
3725 				     bool enable)
3726 {
3727 	u32 data, orig;
3728 
3729 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
3730 		orig = data = RREG32(mmRLC_PG_CNTL);
3731 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3732 		if (orig != data)
3733 			WREG32(mmRLC_PG_CNTL, data);
3734 
3735 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3736 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3737 		if (orig != data)
3738 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3739 	} else {
3740 		orig = data = RREG32(mmRLC_PG_CNTL);
3741 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3742 		if (orig != data)
3743 			WREG32(mmRLC_PG_CNTL, data);
3744 
3745 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3746 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3747 		if (orig != data)
3748 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3749 
3750 		data = RREG32(mmDB_RENDER_CONTROL);
3751 	}
3752 }
3753 
3754 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
3755 						 u32 bitmap)
3756 {
3757 	u32 data;
3758 
3759 	if (!bitmap)
3760 		return;
3761 
3762 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3763 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3764 
3765 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
3766 }
3767 
3768 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
3769 {
3770 	u32 data, mask;
3771 
3772 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
3773 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
3774 
3775 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3776 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3777 
3778 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
3779 
3780 	return (~data) & mask;
3781 }
3782 
3783 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
3784 {
3785 	u32 tmp;
3786 
3787 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
3788 
3789 	tmp = RREG32(mmRLC_MAX_PG_CU);
3790 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
3791 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
3792 	WREG32(mmRLC_MAX_PG_CU, tmp);
3793 }
3794 
3795 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
3796 					    bool enable)
3797 {
3798 	u32 data, orig;
3799 
3800 	orig = data = RREG32(mmRLC_PG_CNTL);
3801 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
3802 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3803 	else
3804 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3805 	if (orig != data)
3806 		WREG32(mmRLC_PG_CNTL, data);
3807 }
3808 
3809 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
3810 					     bool enable)
3811 {
3812 	u32 data, orig;
3813 
3814 	orig = data = RREG32(mmRLC_PG_CNTL);
3815 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
3816 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3817 	else
3818 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3819 	if (orig != data)
3820 		WREG32(mmRLC_PG_CNTL, data);
3821 }
3822 
3823 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
3824 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
3825 
3826 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
3827 {
3828 	u32 data, orig;
3829 	u32 i;
3830 
3831 	if (adev->gfx.rlc.cs_data) {
3832 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3833 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3834 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3835 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
3836 	} else {
3837 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3838 		for (i = 0; i < 3; i++)
3839 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
3840 	}
3841 	if (adev->gfx.rlc.reg_list) {
3842 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
3843 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3844 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
3845 	}
3846 
3847 	orig = data = RREG32(mmRLC_PG_CNTL);
3848 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
3849 	if (orig != data)
3850 		WREG32(mmRLC_PG_CNTL, data);
3851 
3852 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
3853 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
3854 
3855 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
3856 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
3857 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3858 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
3859 
3860 	data = 0x10101010;
3861 	WREG32(mmRLC_PG_DELAY, data);
3862 
3863 	data = RREG32(mmRLC_PG_DELAY_2);
3864 	data &= ~0xff;
3865 	data |= 0x3;
3866 	WREG32(mmRLC_PG_DELAY_2, data);
3867 
3868 	data = RREG32(mmRLC_AUTO_PG_CTRL);
3869 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
3870 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
3871 	WREG32(mmRLC_AUTO_PG_CTRL, data);
3872 
3873 }
3874 
3875 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
3876 {
3877 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
3878 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
3879 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
3880 }
3881 
3882 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
3883 {
3884 	u32 count = 0;
3885 	const struct cs_section_def *sect = NULL;
3886 	const struct cs_extent_def *ext = NULL;
3887 
3888 	if (adev->gfx.rlc.cs_data == NULL)
3889 		return 0;
3890 
3891 	/* begin clear state */
3892 	count += 2;
3893 	/* context control state */
3894 	count += 3;
3895 
3896 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3897 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3898 			if (sect->id == SECT_CONTEXT)
3899 				count += 2 + ext->reg_count;
3900 			else
3901 				return 0;
3902 		}
3903 	}
3904 	/* pa_sc_raster_config/pa_sc_raster_config1 */
3905 	count += 4;
3906 	/* end clear state */
3907 	count += 2;
3908 	/* clear state */
3909 	count += 2;
3910 
3911 	return count;
3912 }
3913 
3914 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
3915 				    volatile u32 *buffer)
3916 {
3917 	u32 count = 0, i;
3918 	const struct cs_section_def *sect = NULL;
3919 	const struct cs_extent_def *ext = NULL;
3920 
3921 	if (adev->gfx.rlc.cs_data == NULL)
3922 		return;
3923 	if (buffer == NULL)
3924 		return;
3925 
3926 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3927 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3928 
3929 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3930 	buffer[count++] = cpu_to_le32(0x80000000);
3931 	buffer[count++] = cpu_to_le32(0x80000000);
3932 
3933 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3934 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3935 			if (sect->id == SECT_CONTEXT) {
3936 				buffer[count++] =
3937 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
3938 				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3939 				for (i = 0; i < ext->reg_count; i++)
3940 					buffer[count++] = cpu_to_le32(ext->extent[i]);
3941 			} else {
3942 				return;
3943 			}
3944 		}
3945 	}
3946 
3947 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
3948 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
3949 	switch (adev->asic_type) {
3950 	case CHIP_BONAIRE:
3951 		buffer[count++] = cpu_to_le32(0x16000012);
3952 		buffer[count++] = cpu_to_le32(0x00000000);
3953 		break;
3954 	case CHIP_KAVERI:
3955 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3956 		buffer[count++] = cpu_to_le32(0x00000000);
3957 		break;
3958 	case CHIP_KABINI:
3959 	case CHIP_MULLINS:
3960 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3961 		buffer[count++] = cpu_to_le32(0x00000000);
3962 		break;
3963 	case CHIP_HAWAII:
3964 		buffer[count++] = cpu_to_le32(0x3a00161a);
3965 		buffer[count++] = cpu_to_le32(0x0000002e);
3966 		break;
3967 	default:
3968 		buffer[count++] = cpu_to_le32(0x00000000);
3969 		buffer[count++] = cpu_to_le32(0x00000000);
3970 		break;
3971 	}
3972 
3973 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3974 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
3975 
3976 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
3977 	buffer[count++] = cpu_to_le32(0);
3978 }
3979 
3980 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
3981 {
3982 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
3983 			      AMD_PG_SUPPORT_GFX_SMG |
3984 			      AMD_PG_SUPPORT_GFX_DMG |
3985 			      AMD_PG_SUPPORT_CP |
3986 			      AMD_PG_SUPPORT_GDS |
3987 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
3988 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
3989 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
3990 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
3991 			gfx_v7_0_init_gfx_cgpg(adev);
3992 			gfx_v7_0_enable_cp_pg(adev, true);
3993 			gfx_v7_0_enable_gds_pg(adev, true);
3994 		}
3995 		gfx_v7_0_init_ao_cu_mask(adev);
3996 		gfx_v7_0_update_gfx_pg(adev, true);
3997 	}
3998 }
3999 
4000 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4001 {
4002 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4003 			      AMD_PG_SUPPORT_GFX_SMG |
4004 			      AMD_PG_SUPPORT_GFX_DMG |
4005 			      AMD_PG_SUPPORT_CP |
4006 			      AMD_PG_SUPPORT_GDS |
4007 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4008 		gfx_v7_0_update_gfx_pg(adev, false);
4009 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4010 			gfx_v7_0_enable_cp_pg(adev, false);
4011 			gfx_v7_0_enable_gds_pg(adev, false);
4012 		}
4013 	}
4014 }
4015 
4016 /**
4017  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4018  *
4019  * @adev: amdgpu_device pointer
4020  *
4021  * Fetches a GPU clock counter snapshot (SI).
4022  * Returns the 64 bit clock counter snapshot.
4023  */
4024 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4025 {
4026 	uint64_t clock;
4027 
4028 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4029 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4030 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4031 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4032 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4033 	return clock;
4034 }
4035 
4036 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4037 					  uint32_t vmid,
4038 					  uint32_t gds_base, uint32_t gds_size,
4039 					  uint32_t gws_base, uint32_t gws_size,
4040 					  uint32_t oa_base, uint32_t oa_size)
4041 {
4042 	/* GDS Base */
4043 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4044 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4045 				WRITE_DATA_DST_SEL(0)));
4046 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4047 	amdgpu_ring_write(ring, 0);
4048 	amdgpu_ring_write(ring, gds_base);
4049 
4050 	/* GDS Size */
4051 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4052 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4053 				WRITE_DATA_DST_SEL(0)));
4054 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
4055 	amdgpu_ring_write(ring, 0);
4056 	amdgpu_ring_write(ring, gds_size);
4057 
4058 	/* GWS */
4059 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4060 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4061 				WRITE_DATA_DST_SEL(0)));
4062 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
4063 	amdgpu_ring_write(ring, 0);
4064 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4065 
4066 	/* OA */
4067 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4068 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4069 				WRITE_DATA_DST_SEL(0)));
4070 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
4071 	amdgpu_ring_write(ring, 0);
4072 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4073 }
4074 
4075 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
4076 {
4077 	struct amdgpu_device *adev = ring->adev;
4078 	uint32_t value = 0;
4079 
4080 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
4081 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
4082 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
4083 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
4084 	WREG32(mmSQ_CMD, value);
4085 }
4086 
4087 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4088 {
4089 	WREG32(mmSQ_IND_INDEX,
4090 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4091 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4092 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4093 		(SQ_IND_INDEX__FORCE_READ_MASK));
4094 	return RREG32(mmSQ_IND_DATA);
4095 }
4096 
4097 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4098 			   uint32_t wave, uint32_t thread,
4099 			   uint32_t regno, uint32_t num, uint32_t *out)
4100 {
4101 	WREG32(mmSQ_IND_INDEX,
4102 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4103 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4104 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4105 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4106 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4107 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4108 	while (num--)
4109 		*(out++) = RREG32(mmSQ_IND_DATA);
4110 }
4111 
4112 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4113 {
4114 	/* type 0 wave data */
4115 	dst[(*no_fields)++] = 0;
4116 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4117 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4118 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4119 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4120 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4121 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4122 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4123 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4124 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4125 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4126 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4127 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4128 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4129 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4130 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4131 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4132 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4133 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4134 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
4135 }
4136 
4137 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
4138 				     uint32_t wave, uint32_t start,
4139 				     uint32_t size, uint32_t *dst)
4140 {
4141 	wave_read_regs(
4142 		adev, simd, wave, 0,
4143 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4144 }
4145 
4146 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
4147 				  u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
4148 {
4149 	cik_srbm_select(adev, me, pipe, q, vm);
4150 }
4151 
4152 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4153 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4154 	.select_se_sh = &gfx_v7_0_select_se_sh,
4155 	.read_wave_data = &gfx_v7_0_read_wave_data,
4156 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4157 	.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q
4158 };
4159 
4160 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4161 	.is_rlc_enabled = gfx_v7_0_is_rlc_enabled,
4162 	.set_safe_mode = gfx_v7_0_set_safe_mode,
4163 	.unset_safe_mode = gfx_v7_0_unset_safe_mode,
4164 	.init = gfx_v7_0_rlc_init,
4165 	.get_csb_size = gfx_v7_0_get_csb_size,
4166 	.get_csb_buffer = gfx_v7_0_get_csb_buffer,
4167 	.get_cp_table_num = gfx_v7_0_cp_pg_table_num,
4168 	.resume = gfx_v7_0_rlc_resume,
4169 	.stop = gfx_v7_0_rlc_stop,
4170 	.reset = gfx_v7_0_rlc_reset,
4171 	.start = gfx_v7_0_rlc_start,
4172 	.update_spm_vmid = gfx_v7_0_update_spm_vmid
4173 };
4174 
4175 static int gfx_v7_0_early_init(void *handle)
4176 {
4177 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4178 
4179 	adev->gfx.xcc_mask = 1;
4180 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4181 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4182 					  AMDGPU_MAX_COMPUTE_RINGS);
4183 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4184 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4185 	gfx_v7_0_set_ring_funcs(adev);
4186 	gfx_v7_0_set_irq_funcs(adev);
4187 	gfx_v7_0_set_gds_init(adev);
4188 
4189 	return 0;
4190 }
4191 
4192 static int gfx_v7_0_late_init(void *handle)
4193 {
4194 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4195 	int r;
4196 
4197 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4198 	if (r)
4199 		return r;
4200 
4201 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4202 	if (r)
4203 		return r;
4204 
4205 	return 0;
4206 }
4207 
4208 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4209 {
4210 	u32 gb_addr_config;
4211 	u32 mc_arb_ramcfg;
4212 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4213 	u32 tmp;
4214 
4215 	switch (adev->asic_type) {
4216 	case CHIP_BONAIRE:
4217 		adev->gfx.config.max_shader_engines = 2;
4218 		adev->gfx.config.max_tile_pipes = 4;
4219 		adev->gfx.config.max_cu_per_sh = 7;
4220 		adev->gfx.config.max_sh_per_se = 1;
4221 		adev->gfx.config.max_backends_per_se = 2;
4222 		adev->gfx.config.max_texture_channel_caches = 4;
4223 		adev->gfx.config.max_gprs = 256;
4224 		adev->gfx.config.max_gs_threads = 32;
4225 		adev->gfx.config.max_hw_contexts = 8;
4226 
4227 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4228 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4229 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4230 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4231 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4232 		break;
4233 	case CHIP_HAWAII:
4234 		adev->gfx.config.max_shader_engines = 4;
4235 		adev->gfx.config.max_tile_pipes = 16;
4236 		adev->gfx.config.max_cu_per_sh = 11;
4237 		adev->gfx.config.max_sh_per_se = 1;
4238 		adev->gfx.config.max_backends_per_se = 4;
4239 		adev->gfx.config.max_texture_channel_caches = 16;
4240 		adev->gfx.config.max_gprs = 256;
4241 		adev->gfx.config.max_gs_threads = 32;
4242 		adev->gfx.config.max_hw_contexts = 8;
4243 
4244 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4245 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4246 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4247 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4248 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4249 		break;
4250 	case CHIP_KAVERI:
4251 		adev->gfx.config.max_shader_engines = 1;
4252 		adev->gfx.config.max_tile_pipes = 4;
4253 		adev->gfx.config.max_cu_per_sh = 8;
4254 		adev->gfx.config.max_backends_per_se = 2;
4255 		adev->gfx.config.max_sh_per_se = 1;
4256 		adev->gfx.config.max_texture_channel_caches = 4;
4257 		adev->gfx.config.max_gprs = 256;
4258 		adev->gfx.config.max_gs_threads = 16;
4259 		adev->gfx.config.max_hw_contexts = 8;
4260 
4261 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4262 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4263 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4264 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4265 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4266 		break;
4267 	case CHIP_KABINI:
4268 	case CHIP_MULLINS:
4269 	default:
4270 		adev->gfx.config.max_shader_engines = 1;
4271 		adev->gfx.config.max_tile_pipes = 2;
4272 		adev->gfx.config.max_cu_per_sh = 2;
4273 		adev->gfx.config.max_sh_per_se = 1;
4274 		adev->gfx.config.max_backends_per_se = 1;
4275 		adev->gfx.config.max_texture_channel_caches = 2;
4276 		adev->gfx.config.max_gprs = 256;
4277 		adev->gfx.config.max_gs_threads = 16;
4278 		adev->gfx.config.max_hw_contexts = 8;
4279 
4280 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4281 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4282 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4283 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4284 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4285 		break;
4286 	}
4287 
4288 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4289 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4290 
4291 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
4292 				MC_ARB_RAMCFG, NOOFBANK);
4293 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
4294 				MC_ARB_RAMCFG, NOOFRANKS);
4295 
4296 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4297 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4298 	if (adev->flags & AMD_IS_APU) {
4299 		/* Get memory bank mapping mode. */
4300 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4301 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4302 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4303 
4304 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4305 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4306 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4307 
4308 		/* Validate settings in case only one DIMM installed. */
4309 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4310 			dimm00_addr_map = 0;
4311 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4312 			dimm01_addr_map = 0;
4313 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4314 			dimm10_addr_map = 0;
4315 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4316 			dimm11_addr_map = 0;
4317 
4318 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4319 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4320 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4321 			adev->gfx.config.mem_row_size_in_kb = 2;
4322 		else
4323 			adev->gfx.config.mem_row_size_in_kb = 1;
4324 	} else {
4325 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4326 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4327 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4328 			adev->gfx.config.mem_row_size_in_kb = 4;
4329 	}
4330 	/* XXX use MC settings? */
4331 	adev->gfx.config.shader_engine_tile_size = 32;
4332 	adev->gfx.config.num_gpus = 1;
4333 	adev->gfx.config.multi_gpu_tile_size = 64;
4334 
4335 	/* fix up row size */
4336 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4337 	switch (adev->gfx.config.mem_row_size_in_kb) {
4338 	case 1:
4339 	default:
4340 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4341 		break;
4342 	case 2:
4343 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4344 		break;
4345 	case 4:
4346 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4347 		break;
4348 	}
4349 	adev->gfx.config.gb_addr_config = gb_addr_config;
4350 }
4351 
4352 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
4353 					int mec, int pipe, int queue)
4354 {
4355 	int r;
4356 	unsigned irq_type;
4357 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
4358 
4359 	/* mec0 is me1 */
4360 	ring->me = mec + 1;
4361 	ring->pipe = pipe;
4362 	ring->queue = queue;
4363 
4364 	ring->ring_obj = NULL;
4365 	ring->use_doorbell = true;
4366 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
4367 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4368 
4369 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
4370 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
4371 		+ ring->pipe;
4372 
4373 	/* type-2 packets are deprecated on MEC, use type-3 instead */
4374 	r = amdgpu_ring_init(adev, ring, 1024,
4375 			     &adev->gfx.eop_irq, irq_type,
4376 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
4377 	if (r)
4378 		return r;
4379 
4380 
4381 	return 0;
4382 }
4383 
4384 static int gfx_v7_0_sw_init(void *handle)
4385 {
4386 	struct amdgpu_ring *ring;
4387 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4388 	int i, j, k, r, ring_id;
4389 
4390 	switch (adev->asic_type) {
4391 	case CHIP_KAVERI:
4392 		adev->gfx.mec.num_mec = 2;
4393 		break;
4394 	case CHIP_BONAIRE:
4395 	case CHIP_HAWAII:
4396 	case CHIP_KABINI:
4397 	case CHIP_MULLINS:
4398 	default:
4399 		adev->gfx.mec.num_mec = 1;
4400 		break;
4401 	}
4402 	adev->gfx.mec.num_pipe_per_mec = 4;
4403 	adev->gfx.mec.num_queue_per_pipe = 8;
4404 
4405 	/* EOP Event */
4406 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq);
4407 	if (r)
4408 		return r;
4409 
4410 	/* Privileged reg */
4411 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184,
4412 			      &adev->gfx.priv_reg_irq);
4413 	if (r)
4414 		return r;
4415 
4416 	/* Privileged inst */
4417 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185,
4418 			      &adev->gfx.priv_inst_irq);
4419 	if (r)
4420 		return r;
4421 
4422 	r = gfx_v7_0_init_microcode(adev);
4423 	if (r) {
4424 		DRM_ERROR("Failed to load gfx firmware!\n");
4425 		return r;
4426 	}
4427 
4428 	r = adev->gfx.rlc.funcs->init(adev);
4429 	if (r) {
4430 		DRM_ERROR("Failed to init rlc BOs!\n");
4431 		return r;
4432 	}
4433 
4434 	/* allocate mec buffers */
4435 	r = gfx_v7_0_mec_init(adev);
4436 	if (r) {
4437 		DRM_ERROR("Failed to init MEC BOs!\n");
4438 		return r;
4439 	}
4440 
4441 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4442 		ring = &adev->gfx.gfx_ring[i];
4443 		ring->ring_obj = NULL;
4444 		sprintf(ring->name, "gfx");
4445 		r = amdgpu_ring_init(adev, ring, 1024,
4446 				     &adev->gfx.eop_irq,
4447 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
4448 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
4449 		if (r)
4450 			return r;
4451 	}
4452 
4453 	/* set up the compute queues - allocate horizontally across pipes */
4454 	ring_id = 0;
4455 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4456 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4457 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4458 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
4459 								     k, j))
4460 					continue;
4461 
4462 				r = gfx_v7_0_compute_ring_init(adev,
4463 								ring_id,
4464 								i, k, j);
4465 				if (r)
4466 					return r;
4467 
4468 				ring_id++;
4469 			}
4470 		}
4471 	}
4472 
4473 	adev->gfx.ce_ram_size = 0x8000;
4474 
4475 	gfx_v7_0_gpu_early_init(adev);
4476 
4477 	return r;
4478 }
4479 
4480 static int gfx_v7_0_sw_fini(void *handle)
4481 {
4482 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4483 	int i;
4484 
4485 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4486 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
4487 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
4488 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
4489 
4490 	gfx_v7_0_cp_compute_fini(adev);
4491 	amdgpu_gfx_rlc_fini(adev);
4492 	gfx_v7_0_mec_fini(adev);
4493 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
4494 				&adev->gfx.rlc.clear_state_gpu_addr,
4495 				(void **)&adev->gfx.rlc.cs_ptr);
4496 	if (adev->gfx.rlc.cp_table_size) {
4497 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
4498 				&adev->gfx.rlc.cp_table_gpu_addr,
4499 				(void **)&adev->gfx.rlc.cp_table_ptr);
4500 	}
4501 	gfx_v7_0_free_microcode(adev);
4502 
4503 	return 0;
4504 }
4505 
4506 static int gfx_v7_0_hw_init(void *handle)
4507 {
4508 	int r;
4509 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4510 
4511 	gfx_v7_0_constants_init(adev);
4512 
4513 	/* init CSB */
4514 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
4515 	/* init rlc */
4516 	r = adev->gfx.rlc.funcs->resume(adev);
4517 	if (r)
4518 		return r;
4519 
4520 	r = gfx_v7_0_cp_resume(adev);
4521 	if (r)
4522 		return r;
4523 
4524 	return r;
4525 }
4526 
4527 static int gfx_v7_0_hw_fini(void *handle)
4528 {
4529 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4530 
4531 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4532 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4533 	gfx_v7_0_cp_enable(adev, false);
4534 	adev->gfx.rlc.funcs->stop(adev);
4535 	gfx_v7_0_fini_pg(adev);
4536 
4537 	return 0;
4538 }
4539 
4540 static int gfx_v7_0_suspend(void *handle)
4541 {
4542 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4543 
4544 	return gfx_v7_0_hw_fini(adev);
4545 }
4546 
4547 static int gfx_v7_0_resume(void *handle)
4548 {
4549 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4550 
4551 	return gfx_v7_0_hw_init(adev);
4552 }
4553 
4554 static bool gfx_v7_0_is_idle(void *handle)
4555 {
4556 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4557 
4558 	if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK)
4559 		return false;
4560 	else
4561 		return true;
4562 }
4563 
4564 static int gfx_v7_0_wait_for_idle(void *handle)
4565 {
4566 	unsigned i;
4567 	u32 tmp;
4568 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4569 
4570 	for (i = 0; i < adev->usec_timeout; i++) {
4571 		/* read MC_STATUS */
4572 		tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK;
4573 
4574 		if (!tmp)
4575 			return 0;
4576 		udelay(1);
4577 	}
4578 	return -ETIMEDOUT;
4579 }
4580 
4581 static int gfx_v7_0_soft_reset(void *handle)
4582 {
4583 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4584 	u32 tmp;
4585 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4586 
4587 	/* GRBM_STATUS */
4588 	tmp = RREG32(mmGRBM_STATUS);
4589 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4590 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4591 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4592 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4593 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4594 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK))
4595 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK |
4596 			GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK;
4597 
4598 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4599 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK;
4600 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4601 	}
4602 
4603 	/* GRBM_STATUS2 */
4604 	tmp = RREG32(mmGRBM_STATUS2);
4605 	if (tmp & GRBM_STATUS2__RLC_BUSY_MASK)
4606 		grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
4607 
4608 	/* SRBM_STATUS */
4609 	tmp = RREG32(mmSRBM_STATUS);
4610 	if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK)
4611 		srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK;
4612 
4613 	if (grbm_soft_reset || srbm_soft_reset) {
4614 		/* disable CG/PG */
4615 		gfx_v7_0_fini_pg(adev);
4616 		gfx_v7_0_update_cg(adev, false);
4617 
4618 		/* stop the rlc */
4619 		adev->gfx.rlc.funcs->stop(adev);
4620 
4621 		/* Disable GFX parsing/prefetching */
4622 		WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK);
4623 
4624 		/* Disable MEC parsing/prefetching */
4625 		WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK);
4626 
4627 		if (grbm_soft_reset) {
4628 			tmp = RREG32(mmGRBM_SOFT_RESET);
4629 			tmp |= grbm_soft_reset;
4630 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4631 			WREG32(mmGRBM_SOFT_RESET, tmp);
4632 			tmp = RREG32(mmGRBM_SOFT_RESET);
4633 
4634 			udelay(50);
4635 
4636 			tmp &= ~grbm_soft_reset;
4637 			WREG32(mmGRBM_SOFT_RESET, tmp);
4638 			tmp = RREG32(mmGRBM_SOFT_RESET);
4639 		}
4640 
4641 		if (srbm_soft_reset) {
4642 			tmp = RREG32(mmSRBM_SOFT_RESET);
4643 			tmp |= srbm_soft_reset;
4644 			dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
4645 			WREG32(mmSRBM_SOFT_RESET, tmp);
4646 			tmp = RREG32(mmSRBM_SOFT_RESET);
4647 
4648 			udelay(50);
4649 
4650 			tmp &= ~srbm_soft_reset;
4651 			WREG32(mmSRBM_SOFT_RESET, tmp);
4652 			tmp = RREG32(mmSRBM_SOFT_RESET);
4653 		}
4654 		/* Wait a little for things to settle down */
4655 		udelay(50);
4656 	}
4657 	return 0;
4658 }
4659 
4660 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4661 						 enum amdgpu_interrupt_state state)
4662 {
4663 	u32 cp_int_cntl;
4664 
4665 	switch (state) {
4666 	case AMDGPU_IRQ_STATE_DISABLE:
4667 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4668 		cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4669 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4670 		break;
4671 	case AMDGPU_IRQ_STATE_ENABLE:
4672 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4673 		cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4674 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4675 		break;
4676 	default:
4677 		break;
4678 	}
4679 }
4680 
4681 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4682 						     int me, int pipe,
4683 						     enum amdgpu_interrupt_state state)
4684 {
4685 	u32 mec_int_cntl, mec_int_cntl_reg;
4686 
4687 	/*
4688 	 * amdgpu controls only the first MEC. That's why this function only
4689 	 * handles the setting of interrupts for this specific MEC. All other
4690 	 * pipes' interrupts are set by amdkfd.
4691 	 */
4692 
4693 	if (me == 1) {
4694 		switch (pipe) {
4695 		case 0:
4696 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
4697 			break;
4698 		case 1:
4699 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
4700 			break;
4701 		case 2:
4702 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
4703 			break;
4704 		case 3:
4705 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
4706 			break;
4707 		default:
4708 			DRM_DEBUG("invalid pipe %d\n", pipe);
4709 			return;
4710 		}
4711 	} else {
4712 		DRM_DEBUG("invalid me %d\n", me);
4713 		return;
4714 	}
4715 
4716 	switch (state) {
4717 	case AMDGPU_IRQ_STATE_DISABLE:
4718 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4719 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4720 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4721 		break;
4722 	case AMDGPU_IRQ_STATE_ENABLE:
4723 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4724 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
4725 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4726 		break;
4727 	default:
4728 		break;
4729 	}
4730 }
4731 
4732 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4733 					     struct amdgpu_irq_src *src,
4734 					     unsigned type,
4735 					     enum amdgpu_interrupt_state state)
4736 {
4737 	u32 cp_int_cntl;
4738 
4739 	switch (state) {
4740 	case AMDGPU_IRQ_STATE_DISABLE:
4741 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4742 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4743 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4744 		break;
4745 	case AMDGPU_IRQ_STATE_ENABLE:
4746 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4747 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK;
4748 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4749 		break;
4750 	default:
4751 		break;
4752 	}
4753 
4754 	return 0;
4755 }
4756 
4757 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4758 					      struct amdgpu_irq_src *src,
4759 					      unsigned type,
4760 					      enum amdgpu_interrupt_state state)
4761 {
4762 	u32 cp_int_cntl;
4763 
4764 	switch (state) {
4765 	case AMDGPU_IRQ_STATE_DISABLE:
4766 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4767 		cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4768 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4769 		break;
4770 	case AMDGPU_IRQ_STATE_ENABLE:
4771 		cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0);
4772 		cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK;
4773 		WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl);
4774 		break;
4775 	default:
4776 		break;
4777 	}
4778 
4779 	return 0;
4780 }
4781 
4782 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4783 					    struct amdgpu_irq_src *src,
4784 					    unsigned type,
4785 					    enum amdgpu_interrupt_state state)
4786 {
4787 	switch (type) {
4788 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
4789 		gfx_v7_0_set_gfx_eop_interrupt_state(adev, state);
4790 		break;
4791 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4792 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4793 		break;
4794 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4795 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4796 		break;
4797 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4798 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4799 		break;
4800 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4801 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4802 		break;
4803 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4804 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4805 		break;
4806 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4807 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4808 		break;
4809 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4810 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4811 		break;
4812 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4813 		gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4814 		break;
4815 	default:
4816 		break;
4817 	}
4818 	return 0;
4819 }
4820 
4821 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev,
4822 			    struct amdgpu_irq_src *source,
4823 			    struct amdgpu_iv_entry *entry)
4824 {
4825 	u8 me_id, pipe_id;
4826 	struct amdgpu_ring *ring;
4827 	int i;
4828 
4829 	DRM_DEBUG("IH: CP EOP\n");
4830 	me_id = (entry->ring_id & 0x0c) >> 2;
4831 	pipe_id = (entry->ring_id & 0x03) >> 0;
4832 	switch (me_id) {
4833 	case 0:
4834 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4835 		break;
4836 	case 1:
4837 	case 2:
4838 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4839 			ring = &adev->gfx.compute_ring[i];
4840 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4841 				amdgpu_fence_process(ring);
4842 		}
4843 		break;
4844 	}
4845 	return 0;
4846 }
4847 
4848 static void gfx_v7_0_fault(struct amdgpu_device *adev,
4849 			   struct amdgpu_iv_entry *entry)
4850 {
4851 	struct amdgpu_ring *ring;
4852 	u8 me_id, pipe_id;
4853 	int i;
4854 
4855 	me_id = (entry->ring_id & 0x0c) >> 2;
4856 	pipe_id = (entry->ring_id & 0x03) >> 0;
4857 	switch (me_id) {
4858 	case 0:
4859 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
4860 		break;
4861 	case 1:
4862 	case 2:
4863 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4864 			ring = &adev->gfx.compute_ring[i];
4865 			if ((ring->me == me_id) && (ring->pipe == pipe_id))
4866 				drm_sched_fault(&ring->sched);
4867 		}
4868 		break;
4869 	}
4870 }
4871 
4872 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev,
4873 				 struct amdgpu_irq_src *source,
4874 				 struct amdgpu_iv_entry *entry)
4875 {
4876 	DRM_ERROR("Illegal register access in command stream\n");
4877 	gfx_v7_0_fault(adev, entry);
4878 	return 0;
4879 }
4880 
4881 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev,
4882 				  struct amdgpu_irq_src *source,
4883 				  struct amdgpu_iv_entry *entry)
4884 {
4885 	DRM_ERROR("Illegal instruction in command stream\n");
4886 	// XXX soft reset the gfx block only
4887 	gfx_v7_0_fault(adev, entry);
4888 	return 0;
4889 }
4890 
4891 static int gfx_v7_0_set_clockgating_state(void *handle,
4892 					  enum amd_clockgating_state state)
4893 {
4894 	bool gate = false;
4895 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4896 
4897 	if (state == AMD_CG_STATE_GATE)
4898 		gate = true;
4899 
4900 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
4901 	/* order matters! */
4902 	if (gate) {
4903 		gfx_v7_0_enable_mgcg(adev, true);
4904 		gfx_v7_0_enable_cgcg(adev, true);
4905 	} else {
4906 		gfx_v7_0_enable_cgcg(adev, false);
4907 		gfx_v7_0_enable_mgcg(adev, false);
4908 	}
4909 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
4910 
4911 	return 0;
4912 }
4913 
4914 static int gfx_v7_0_set_powergating_state(void *handle,
4915 					  enum amd_powergating_state state)
4916 {
4917 	bool gate = false;
4918 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4919 
4920 	if (state == AMD_PG_STATE_GATE)
4921 		gate = true;
4922 
4923 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4924 			      AMD_PG_SUPPORT_GFX_SMG |
4925 			      AMD_PG_SUPPORT_GFX_DMG |
4926 			      AMD_PG_SUPPORT_CP |
4927 			      AMD_PG_SUPPORT_GDS |
4928 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4929 		gfx_v7_0_update_gfx_pg(adev, gate);
4930 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4931 			gfx_v7_0_enable_cp_pg(adev, gate);
4932 			gfx_v7_0_enable_gds_pg(adev, gate);
4933 		}
4934 	}
4935 
4936 	return 0;
4937 }
4938 
4939 static void gfx_v7_0_emit_mem_sync(struct amdgpu_ring *ring)
4940 {
4941 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
4942 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
4943 			  PACKET3_TC_ACTION_ENA |
4944 			  PACKET3_SH_KCACHE_ACTION_ENA |
4945 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
4946 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
4947 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
4948 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
4949 }
4950 
4951 static void gfx_v7_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
4952 {
4953 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
4954 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
4955 			  PACKET3_TC_ACTION_ENA |
4956 			  PACKET3_SH_KCACHE_ACTION_ENA |
4957 			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
4958 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
4959 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
4960 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
4961 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
4962 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
4963 }
4964 
4965 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
4966 	.name = "gfx_v7_0",
4967 	.early_init = gfx_v7_0_early_init,
4968 	.late_init = gfx_v7_0_late_init,
4969 	.sw_init = gfx_v7_0_sw_init,
4970 	.sw_fini = gfx_v7_0_sw_fini,
4971 	.hw_init = gfx_v7_0_hw_init,
4972 	.hw_fini = gfx_v7_0_hw_fini,
4973 	.suspend = gfx_v7_0_suspend,
4974 	.resume = gfx_v7_0_resume,
4975 	.is_idle = gfx_v7_0_is_idle,
4976 	.wait_for_idle = gfx_v7_0_wait_for_idle,
4977 	.soft_reset = gfx_v7_0_soft_reset,
4978 	.set_clockgating_state = gfx_v7_0_set_clockgating_state,
4979 	.set_powergating_state = gfx_v7_0_set_powergating_state,
4980 	.dump_ip_state = NULL,
4981 	.print_ip_state = NULL,
4982 };
4983 
4984 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
4985 	.type = AMDGPU_RING_TYPE_GFX,
4986 	.align_mask = 0xff,
4987 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
4988 	.support_64bit_ptrs = false,
4989 	.get_rptr = gfx_v7_0_ring_get_rptr,
4990 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
4991 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
4992 	.emit_frame_size =
4993 		20 + /* gfx_v7_0_ring_emit_gds_switch */
4994 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
4995 		5 + /* hdp invalidate */
4996 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
4997 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
4998 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
4999 		3 + 4 + /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
5000 		5, /* SURFACE_SYNC */
5001 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
5002 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
5003 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
5004 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5005 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5006 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5007 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5008 	.test_ring = gfx_v7_0_ring_test_ring,
5009 	.test_ib = gfx_v7_0_ring_test_ib,
5010 	.insert_nop = amdgpu_ring_insert_nop,
5011 	.pad_ib = amdgpu_ring_generic_pad_ib,
5012 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
5013 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5014 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
5015 	.emit_mem_sync = gfx_v7_0_emit_mem_sync,
5016 };
5017 
5018 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5019 	.type = AMDGPU_RING_TYPE_COMPUTE,
5020 	.align_mask = 0xff,
5021 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5022 	.support_64bit_ptrs = false,
5023 	.get_rptr = gfx_v7_0_ring_get_rptr,
5024 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5025 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5026 	.emit_frame_size =
5027 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5028 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5029 		5 + /* hdp invalidate */
5030 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5031 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
5032 		7 + 7 + 7 + /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5033 		7, /* gfx_v7_0_emit_mem_sync_compute */
5034 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
5035 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5036 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5037 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5038 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5039 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5040 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5041 	.test_ring = gfx_v7_0_ring_test_ring,
5042 	.test_ib = gfx_v7_0_ring_test_ib,
5043 	.insert_nop = amdgpu_ring_insert_nop,
5044 	.pad_ib = amdgpu_ring_generic_pad_ib,
5045 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5046 	.emit_mem_sync = gfx_v7_0_emit_mem_sync_compute,
5047 };
5048 
5049 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5050 {
5051 	int i;
5052 
5053 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5054 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5055 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5056 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5057 }
5058 
5059 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5060 	.set = gfx_v7_0_set_eop_interrupt_state,
5061 	.process = gfx_v7_0_eop_irq,
5062 };
5063 
5064 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5065 	.set = gfx_v7_0_set_priv_reg_fault_state,
5066 	.process = gfx_v7_0_priv_reg_irq,
5067 };
5068 
5069 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5070 	.set = gfx_v7_0_set_priv_inst_fault_state,
5071 	.process = gfx_v7_0_priv_inst_irq,
5072 };
5073 
5074 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5075 {
5076 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5077 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5078 
5079 	adev->gfx.priv_reg_irq.num_types = 1;
5080 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5081 
5082 	adev->gfx.priv_inst_irq.num_types = 1;
5083 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5084 }
5085 
5086 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5087 {
5088 	/* init asci gds info */
5089 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
5090 	adev->gds.gws_size = 64;
5091 	adev->gds.oa_size = 16;
5092 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
5093 }
5094 
5095 
5096 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5097 {
5098 	int i, j, k, counter, active_cu_number = 0;
5099 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5100 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5101 	unsigned disable_masks[4 * 2];
5102 	u32 ao_cu_num;
5103 
5104 	if (adev->flags & AMD_IS_APU)
5105 		ao_cu_num = 2;
5106 	else
5107 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
5108 
5109 	memset(cu_info, 0, sizeof(*cu_info));
5110 
5111 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5112 
5113 	mutex_lock(&adev->grbm_idx_mutex);
5114 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5115 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5116 			mask = 1;
5117 			ao_bitmap = 0;
5118 			counter = 0;
5119 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0);
5120 			if (i < 4 && j < 2)
5121 				gfx_v7_0_set_user_cu_inactive_bitmap(
5122 					adev, disable_masks[i * 2 + j]);
5123 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5124 			cu_info->bitmap[0][i][j] = bitmap;
5125 
5126 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) {
5127 				if (bitmap & mask) {
5128 					if (counter < ao_cu_num)
5129 						ao_bitmap |= mask;
5130 					counter++;
5131 				}
5132 				mask <<= 1;
5133 			}
5134 			active_cu_number += counter;
5135 			if (i < 2 && j < 2)
5136 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5137 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5138 		}
5139 	}
5140 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
5141 	mutex_unlock(&adev->grbm_idx_mutex);
5142 
5143 	cu_info->number = active_cu_number;
5144 	cu_info->ao_cu_mask = ao_cu_mask;
5145 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5146 	cu_info->max_waves_per_simd = 10;
5147 	cu_info->max_scratch_slots_per_cu = 32;
5148 	cu_info->wave_front_size = 64;
5149 	cu_info->lds_size = 64;
5150 }
5151 
5152 const struct amdgpu_ip_block_version gfx_v7_1_ip_block = {
5153 	.type = AMD_IP_BLOCK_TYPE_GFX,
5154 	.major = 7,
5155 	.minor = 1,
5156 	.rev = 0,
5157 	.funcs = &gfx_v7_0_ip_funcs,
5158 };
5159 
5160 const struct amdgpu_ip_block_version gfx_v7_2_ip_block = {
5161 	.type = AMD_IP_BLOCK_TYPE_GFX,
5162 	.major = 7,
5163 	.minor = 2,
5164 	.rev = 0,
5165 	.funcs = &gfx_v7_0_ip_funcs,
5166 };
5167 
5168 const struct amdgpu_ip_block_version gfx_v7_3_ip_block = {
5169 	.type = AMD_IP_BLOCK_TYPE_GFX,
5170 	.major = 7,
5171 	.minor = 3,
5172 	.rev = 0,
5173 	.funcs = &gfx_v7_0_ip_funcs,
5174 };
5175