xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c (revision 11696c5e)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <linux/module.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_ih.h"
29 #include "amdgpu_gfx.h"
30 #include "cikd.h"
31 #include "cik.h"
32 #include "cik_structs.h"
33 #include "atom.h"
34 #include "amdgpu_ucode.h"
35 #include "clearstate_ci.h"
36 
37 #include "dce/dce_8_0_d.h"
38 #include "dce/dce_8_0_sh_mask.h"
39 
40 #include "bif/bif_4_1_d.h"
41 #include "bif/bif_4_1_sh_mask.h"
42 
43 #include "gca/gfx_7_0_d.h"
44 #include "gca/gfx_7_2_enum.h"
45 #include "gca/gfx_7_2_sh_mask.h"
46 
47 #include "gmc/gmc_7_0_d.h"
48 #include "gmc/gmc_7_0_sh_mask.h"
49 
50 #include "oss/oss_2_0_d.h"
51 #include "oss/oss_2_0_sh_mask.h"
52 
53 #define NUM_SIMD_PER_CU 0x4 /* missing from the gfx_7 IP headers */
54 
55 #define GFX7_NUM_GFX_RINGS     1
56 #define GFX7_MEC_HPD_SIZE      2048
57 
58 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev);
59 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev);
60 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev);
61 
62 MODULE_FIRMWARE("amdgpu/bonaire_pfp.bin");
63 MODULE_FIRMWARE("amdgpu/bonaire_me.bin");
64 MODULE_FIRMWARE("amdgpu/bonaire_ce.bin");
65 MODULE_FIRMWARE("amdgpu/bonaire_rlc.bin");
66 MODULE_FIRMWARE("amdgpu/bonaire_mec.bin");
67 
68 MODULE_FIRMWARE("amdgpu/hawaii_pfp.bin");
69 MODULE_FIRMWARE("amdgpu/hawaii_me.bin");
70 MODULE_FIRMWARE("amdgpu/hawaii_ce.bin");
71 MODULE_FIRMWARE("amdgpu/hawaii_rlc.bin");
72 MODULE_FIRMWARE("amdgpu/hawaii_mec.bin");
73 
74 MODULE_FIRMWARE("amdgpu/kaveri_pfp.bin");
75 MODULE_FIRMWARE("amdgpu/kaveri_me.bin");
76 MODULE_FIRMWARE("amdgpu/kaveri_ce.bin");
77 MODULE_FIRMWARE("amdgpu/kaveri_rlc.bin");
78 MODULE_FIRMWARE("amdgpu/kaveri_mec.bin");
79 MODULE_FIRMWARE("amdgpu/kaveri_mec2.bin");
80 
81 MODULE_FIRMWARE("amdgpu/kabini_pfp.bin");
82 MODULE_FIRMWARE("amdgpu/kabini_me.bin");
83 MODULE_FIRMWARE("amdgpu/kabini_ce.bin");
84 MODULE_FIRMWARE("amdgpu/kabini_rlc.bin");
85 MODULE_FIRMWARE("amdgpu/kabini_mec.bin");
86 
87 MODULE_FIRMWARE("amdgpu/mullins_pfp.bin");
88 MODULE_FIRMWARE("amdgpu/mullins_me.bin");
89 MODULE_FIRMWARE("amdgpu/mullins_ce.bin");
90 MODULE_FIRMWARE("amdgpu/mullins_rlc.bin");
91 MODULE_FIRMWARE("amdgpu/mullins_mec.bin");
92 
93 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
94 {
95 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
96 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
97 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
98 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
99 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
100 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
101 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
102 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
103 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
104 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
105 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
106 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
107 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
108 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
109 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
110 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
111 };
112 
113 static const u32 spectre_rlc_save_restore_register_list[] =
114 {
115 	(0x0e00 << 16) | (0xc12c >> 2),
116 	0x00000000,
117 	(0x0e00 << 16) | (0xc140 >> 2),
118 	0x00000000,
119 	(0x0e00 << 16) | (0xc150 >> 2),
120 	0x00000000,
121 	(0x0e00 << 16) | (0xc15c >> 2),
122 	0x00000000,
123 	(0x0e00 << 16) | (0xc168 >> 2),
124 	0x00000000,
125 	(0x0e00 << 16) | (0xc170 >> 2),
126 	0x00000000,
127 	(0x0e00 << 16) | (0xc178 >> 2),
128 	0x00000000,
129 	(0x0e00 << 16) | (0xc204 >> 2),
130 	0x00000000,
131 	(0x0e00 << 16) | (0xc2b4 >> 2),
132 	0x00000000,
133 	(0x0e00 << 16) | (0xc2b8 >> 2),
134 	0x00000000,
135 	(0x0e00 << 16) | (0xc2bc >> 2),
136 	0x00000000,
137 	(0x0e00 << 16) | (0xc2c0 >> 2),
138 	0x00000000,
139 	(0x0e00 << 16) | (0x8228 >> 2),
140 	0x00000000,
141 	(0x0e00 << 16) | (0x829c >> 2),
142 	0x00000000,
143 	(0x0e00 << 16) | (0x869c >> 2),
144 	0x00000000,
145 	(0x0600 << 16) | (0x98f4 >> 2),
146 	0x00000000,
147 	(0x0e00 << 16) | (0x98f8 >> 2),
148 	0x00000000,
149 	(0x0e00 << 16) | (0x9900 >> 2),
150 	0x00000000,
151 	(0x0e00 << 16) | (0xc260 >> 2),
152 	0x00000000,
153 	(0x0e00 << 16) | (0x90e8 >> 2),
154 	0x00000000,
155 	(0x0e00 << 16) | (0x3c000 >> 2),
156 	0x00000000,
157 	(0x0e00 << 16) | (0x3c00c >> 2),
158 	0x00000000,
159 	(0x0e00 << 16) | (0x8c1c >> 2),
160 	0x00000000,
161 	(0x0e00 << 16) | (0x9700 >> 2),
162 	0x00000000,
163 	(0x0e00 << 16) | (0xcd20 >> 2),
164 	0x00000000,
165 	(0x4e00 << 16) | (0xcd20 >> 2),
166 	0x00000000,
167 	(0x5e00 << 16) | (0xcd20 >> 2),
168 	0x00000000,
169 	(0x6e00 << 16) | (0xcd20 >> 2),
170 	0x00000000,
171 	(0x7e00 << 16) | (0xcd20 >> 2),
172 	0x00000000,
173 	(0x8e00 << 16) | (0xcd20 >> 2),
174 	0x00000000,
175 	(0x9e00 << 16) | (0xcd20 >> 2),
176 	0x00000000,
177 	(0xae00 << 16) | (0xcd20 >> 2),
178 	0x00000000,
179 	(0xbe00 << 16) | (0xcd20 >> 2),
180 	0x00000000,
181 	(0x0e00 << 16) | (0x89bc >> 2),
182 	0x00000000,
183 	(0x0e00 << 16) | (0x8900 >> 2),
184 	0x00000000,
185 	0x3,
186 	(0x0e00 << 16) | (0xc130 >> 2),
187 	0x00000000,
188 	(0x0e00 << 16) | (0xc134 >> 2),
189 	0x00000000,
190 	(0x0e00 << 16) | (0xc1fc >> 2),
191 	0x00000000,
192 	(0x0e00 << 16) | (0xc208 >> 2),
193 	0x00000000,
194 	(0x0e00 << 16) | (0xc264 >> 2),
195 	0x00000000,
196 	(0x0e00 << 16) | (0xc268 >> 2),
197 	0x00000000,
198 	(0x0e00 << 16) | (0xc26c >> 2),
199 	0x00000000,
200 	(0x0e00 << 16) | (0xc270 >> 2),
201 	0x00000000,
202 	(0x0e00 << 16) | (0xc274 >> 2),
203 	0x00000000,
204 	(0x0e00 << 16) | (0xc278 >> 2),
205 	0x00000000,
206 	(0x0e00 << 16) | (0xc27c >> 2),
207 	0x00000000,
208 	(0x0e00 << 16) | (0xc280 >> 2),
209 	0x00000000,
210 	(0x0e00 << 16) | (0xc284 >> 2),
211 	0x00000000,
212 	(0x0e00 << 16) | (0xc288 >> 2),
213 	0x00000000,
214 	(0x0e00 << 16) | (0xc28c >> 2),
215 	0x00000000,
216 	(0x0e00 << 16) | (0xc290 >> 2),
217 	0x00000000,
218 	(0x0e00 << 16) | (0xc294 >> 2),
219 	0x00000000,
220 	(0x0e00 << 16) | (0xc298 >> 2),
221 	0x00000000,
222 	(0x0e00 << 16) | (0xc29c >> 2),
223 	0x00000000,
224 	(0x0e00 << 16) | (0xc2a0 >> 2),
225 	0x00000000,
226 	(0x0e00 << 16) | (0xc2a4 >> 2),
227 	0x00000000,
228 	(0x0e00 << 16) | (0xc2a8 >> 2),
229 	0x00000000,
230 	(0x0e00 << 16) | (0xc2ac  >> 2),
231 	0x00000000,
232 	(0x0e00 << 16) | (0xc2b0 >> 2),
233 	0x00000000,
234 	(0x0e00 << 16) | (0x301d0 >> 2),
235 	0x00000000,
236 	(0x0e00 << 16) | (0x30238 >> 2),
237 	0x00000000,
238 	(0x0e00 << 16) | (0x30250 >> 2),
239 	0x00000000,
240 	(0x0e00 << 16) | (0x30254 >> 2),
241 	0x00000000,
242 	(0x0e00 << 16) | (0x30258 >> 2),
243 	0x00000000,
244 	(0x0e00 << 16) | (0x3025c >> 2),
245 	0x00000000,
246 	(0x4e00 << 16) | (0xc900 >> 2),
247 	0x00000000,
248 	(0x5e00 << 16) | (0xc900 >> 2),
249 	0x00000000,
250 	(0x6e00 << 16) | (0xc900 >> 2),
251 	0x00000000,
252 	(0x7e00 << 16) | (0xc900 >> 2),
253 	0x00000000,
254 	(0x8e00 << 16) | (0xc900 >> 2),
255 	0x00000000,
256 	(0x9e00 << 16) | (0xc900 >> 2),
257 	0x00000000,
258 	(0xae00 << 16) | (0xc900 >> 2),
259 	0x00000000,
260 	(0xbe00 << 16) | (0xc900 >> 2),
261 	0x00000000,
262 	(0x4e00 << 16) | (0xc904 >> 2),
263 	0x00000000,
264 	(0x5e00 << 16) | (0xc904 >> 2),
265 	0x00000000,
266 	(0x6e00 << 16) | (0xc904 >> 2),
267 	0x00000000,
268 	(0x7e00 << 16) | (0xc904 >> 2),
269 	0x00000000,
270 	(0x8e00 << 16) | (0xc904 >> 2),
271 	0x00000000,
272 	(0x9e00 << 16) | (0xc904 >> 2),
273 	0x00000000,
274 	(0xae00 << 16) | (0xc904 >> 2),
275 	0x00000000,
276 	(0xbe00 << 16) | (0xc904 >> 2),
277 	0x00000000,
278 	(0x4e00 << 16) | (0xc908 >> 2),
279 	0x00000000,
280 	(0x5e00 << 16) | (0xc908 >> 2),
281 	0x00000000,
282 	(0x6e00 << 16) | (0xc908 >> 2),
283 	0x00000000,
284 	(0x7e00 << 16) | (0xc908 >> 2),
285 	0x00000000,
286 	(0x8e00 << 16) | (0xc908 >> 2),
287 	0x00000000,
288 	(0x9e00 << 16) | (0xc908 >> 2),
289 	0x00000000,
290 	(0xae00 << 16) | (0xc908 >> 2),
291 	0x00000000,
292 	(0xbe00 << 16) | (0xc908 >> 2),
293 	0x00000000,
294 	(0x4e00 << 16) | (0xc90c >> 2),
295 	0x00000000,
296 	(0x5e00 << 16) | (0xc90c >> 2),
297 	0x00000000,
298 	(0x6e00 << 16) | (0xc90c >> 2),
299 	0x00000000,
300 	(0x7e00 << 16) | (0xc90c >> 2),
301 	0x00000000,
302 	(0x8e00 << 16) | (0xc90c >> 2),
303 	0x00000000,
304 	(0x9e00 << 16) | (0xc90c >> 2),
305 	0x00000000,
306 	(0xae00 << 16) | (0xc90c >> 2),
307 	0x00000000,
308 	(0xbe00 << 16) | (0xc90c >> 2),
309 	0x00000000,
310 	(0x4e00 << 16) | (0xc910 >> 2),
311 	0x00000000,
312 	(0x5e00 << 16) | (0xc910 >> 2),
313 	0x00000000,
314 	(0x6e00 << 16) | (0xc910 >> 2),
315 	0x00000000,
316 	(0x7e00 << 16) | (0xc910 >> 2),
317 	0x00000000,
318 	(0x8e00 << 16) | (0xc910 >> 2),
319 	0x00000000,
320 	(0x9e00 << 16) | (0xc910 >> 2),
321 	0x00000000,
322 	(0xae00 << 16) | (0xc910 >> 2),
323 	0x00000000,
324 	(0xbe00 << 16) | (0xc910 >> 2),
325 	0x00000000,
326 	(0x0e00 << 16) | (0xc99c >> 2),
327 	0x00000000,
328 	(0x0e00 << 16) | (0x9834 >> 2),
329 	0x00000000,
330 	(0x0000 << 16) | (0x30f00 >> 2),
331 	0x00000000,
332 	(0x0001 << 16) | (0x30f00 >> 2),
333 	0x00000000,
334 	(0x0000 << 16) | (0x30f04 >> 2),
335 	0x00000000,
336 	(0x0001 << 16) | (0x30f04 >> 2),
337 	0x00000000,
338 	(0x0000 << 16) | (0x30f08 >> 2),
339 	0x00000000,
340 	(0x0001 << 16) | (0x30f08 >> 2),
341 	0x00000000,
342 	(0x0000 << 16) | (0x30f0c >> 2),
343 	0x00000000,
344 	(0x0001 << 16) | (0x30f0c >> 2),
345 	0x00000000,
346 	(0x0600 << 16) | (0x9b7c >> 2),
347 	0x00000000,
348 	(0x0e00 << 16) | (0x8a14 >> 2),
349 	0x00000000,
350 	(0x0e00 << 16) | (0x8a18 >> 2),
351 	0x00000000,
352 	(0x0600 << 16) | (0x30a00 >> 2),
353 	0x00000000,
354 	(0x0e00 << 16) | (0x8bf0 >> 2),
355 	0x00000000,
356 	(0x0e00 << 16) | (0x8bcc >> 2),
357 	0x00000000,
358 	(0x0e00 << 16) | (0x8b24 >> 2),
359 	0x00000000,
360 	(0x0e00 << 16) | (0x30a04 >> 2),
361 	0x00000000,
362 	(0x0600 << 16) | (0x30a10 >> 2),
363 	0x00000000,
364 	(0x0600 << 16) | (0x30a14 >> 2),
365 	0x00000000,
366 	(0x0600 << 16) | (0x30a18 >> 2),
367 	0x00000000,
368 	(0x0600 << 16) | (0x30a2c >> 2),
369 	0x00000000,
370 	(0x0e00 << 16) | (0xc700 >> 2),
371 	0x00000000,
372 	(0x0e00 << 16) | (0xc704 >> 2),
373 	0x00000000,
374 	(0x0e00 << 16) | (0xc708 >> 2),
375 	0x00000000,
376 	(0x0e00 << 16) | (0xc768 >> 2),
377 	0x00000000,
378 	(0x0400 << 16) | (0xc770 >> 2),
379 	0x00000000,
380 	(0x0400 << 16) | (0xc774 >> 2),
381 	0x00000000,
382 	(0x0400 << 16) | (0xc778 >> 2),
383 	0x00000000,
384 	(0x0400 << 16) | (0xc77c >> 2),
385 	0x00000000,
386 	(0x0400 << 16) | (0xc780 >> 2),
387 	0x00000000,
388 	(0x0400 << 16) | (0xc784 >> 2),
389 	0x00000000,
390 	(0x0400 << 16) | (0xc788 >> 2),
391 	0x00000000,
392 	(0x0400 << 16) | (0xc78c >> 2),
393 	0x00000000,
394 	(0x0400 << 16) | (0xc798 >> 2),
395 	0x00000000,
396 	(0x0400 << 16) | (0xc79c >> 2),
397 	0x00000000,
398 	(0x0400 << 16) | (0xc7a0 >> 2),
399 	0x00000000,
400 	(0x0400 << 16) | (0xc7a4 >> 2),
401 	0x00000000,
402 	(0x0400 << 16) | (0xc7a8 >> 2),
403 	0x00000000,
404 	(0x0400 << 16) | (0xc7ac >> 2),
405 	0x00000000,
406 	(0x0400 << 16) | (0xc7b0 >> 2),
407 	0x00000000,
408 	(0x0400 << 16) | (0xc7b4 >> 2),
409 	0x00000000,
410 	(0x0e00 << 16) | (0x9100 >> 2),
411 	0x00000000,
412 	(0x0e00 << 16) | (0x3c010 >> 2),
413 	0x00000000,
414 	(0x0e00 << 16) | (0x92a8 >> 2),
415 	0x00000000,
416 	(0x0e00 << 16) | (0x92ac >> 2),
417 	0x00000000,
418 	(0x0e00 << 16) | (0x92b4 >> 2),
419 	0x00000000,
420 	(0x0e00 << 16) | (0x92b8 >> 2),
421 	0x00000000,
422 	(0x0e00 << 16) | (0x92bc >> 2),
423 	0x00000000,
424 	(0x0e00 << 16) | (0x92c0 >> 2),
425 	0x00000000,
426 	(0x0e00 << 16) | (0x92c4 >> 2),
427 	0x00000000,
428 	(0x0e00 << 16) | (0x92c8 >> 2),
429 	0x00000000,
430 	(0x0e00 << 16) | (0x92cc >> 2),
431 	0x00000000,
432 	(0x0e00 << 16) | (0x92d0 >> 2),
433 	0x00000000,
434 	(0x0e00 << 16) | (0x8c00 >> 2),
435 	0x00000000,
436 	(0x0e00 << 16) | (0x8c04 >> 2),
437 	0x00000000,
438 	(0x0e00 << 16) | (0x8c20 >> 2),
439 	0x00000000,
440 	(0x0e00 << 16) | (0x8c38 >> 2),
441 	0x00000000,
442 	(0x0e00 << 16) | (0x8c3c >> 2),
443 	0x00000000,
444 	(0x0e00 << 16) | (0xae00 >> 2),
445 	0x00000000,
446 	(0x0e00 << 16) | (0x9604 >> 2),
447 	0x00000000,
448 	(0x0e00 << 16) | (0xac08 >> 2),
449 	0x00000000,
450 	(0x0e00 << 16) | (0xac0c >> 2),
451 	0x00000000,
452 	(0x0e00 << 16) | (0xac10 >> 2),
453 	0x00000000,
454 	(0x0e00 << 16) | (0xac14 >> 2),
455 	0x00000000,
456 	(0x0e00 << 16) | (0xac58 >> 2),
457 	0x00000000,
458 	(0x0e00 << 16) | (0xac68 >> 2),
459 	0x00000000,
460 	(0x0e00 << 16) | (0xac6c >> 2),
461 	0x00000000,
462 	(0x0e00 << 16) | (0xac70 >> 2),
463 	0x00000000,
464 	(0x0e00 << 16) | (0xac74 >> 2),
465 	0x00000000,
466 	(0x0e00 << 16) | (0xac78 >> 2),
467 	0x00000000,
468 	(0x0e00 << 16) | (0xac7c >> 2),
469 	0x00000000,
470 	(0x0e00 << 16) | (0xac80 >> 2),
471 	0x00000000,
472 	(0x0e00 << 16) | (0xac84 >> 2),
473 	0x00000000,
474 	(0x0e00 << 16) | (0xac88 >> 2),
475 	0x00000000,
476 	(0x0e00 << 16) | (0xac8c >> 2),
477 	0x00000000,
478 	(0x0e00 << 16) | (0x970c >> 2),
479 	0x00000000,
480 	(0x0e00 << 16) | (0x9714 >> 2),
481 	0x00000000,
482 	(0x0e00 << 16) | (0x9718 >> 2),
483 	0x00000000,
484 	(0x0e00 << 16) | (0x971c >> 2),
485 	0x00000000,
486 	(0x0e00 << 16) | (0x31068 >> 2),
487 	0x00000000,
488 	(0x4e00 << 16) | (0x31068 >> 2),
489 	0x00000000,
490 	(0x5e00 << 16) | (0x31068 >> 2),
491 	0x00000000,
492 	(0x6e00 << 16) | (0x31068 >> 2),
493 	0x00000000,
494 	(0x7e00 << 16) | (0x31068 >> 2),
495 	0x00000000,
496 	(0x8e00 << 16) | (0x31068 >> 2),
497 	0x00000000,
498 	(0x9e00 << 16) | (0x31068 >> 2),
499 	0x00000000,
500 	(0xae00 << 16) | (0x31068 >> 2),
501 	0x00000000,
502 	(0xbe00 << 16) | (0x31068 >> 2),
503 	0x00000000,
504 	(0x0e00 << 16) | (0xcd10 >> 2),
505 	0x00000000,
506 	(0x0e00 << 16) | (0xcd14 >> 2),
507 	0x00000000,
508 	(0x0e00 << 16) | (0x88b0 >> 2),
509 	0x00000000,
510 	(0x0e00 << 16) | (0x88b4 >> 2),
511 	0x00000000,
512 	(0x0e00 << 16) | (0x88b8 >> 2),
513 	0x00000000,
514 	(0x0e00 << 16) | (0x88bc >> 2),
515 	0x00000000,
516 	(0x0400 << 16) | (0x89c0 >> 2),
517 	0x00000000,
518 	(0x0e00 << 16) | (0x88c4 >> 2),
519 	0x00000000,
520 	(0x0e00 << 16) | (0x88c8 >> 2),
521 	0x00000000,
522 	(0x0e00 << 16) | (0x88d0 >> 2),
523 	0x00000000,
524 	(0x0e00 << 16) | (0x88d4 >> 2),
525 	0x00000000,
526 	(0x0e00 << 16) | (0x88d8 >> 2),
527 	0x00000000,
528 	(0x0e00 << 16) | (0x8980 >> 2),
529 	0x00000000,
530 	(0x0e00 << 16) | (0x30938 >> 2),
531 	0x00000000,
532 	(0x0e00 << 16) | (0x3093c >> 2),
533 	0x00000000,
534 	(0x0e00 << 16) | (0x30940 >> 2),
535 	0x00000000,
536 	(0x0e00 << 16) | (0x89a0 >> 2),
537 	0x00000000,
538 	(0x0e00 << 16) | (0x30900 >> 2),
539 	0x00000000,
540 	(0x0e00 << 16) | (0x30904 >> 2),
541 	0x00000000,
542 	(0x0e00 << 16) | (0x89b4 >> 2),
543 	0x00000000,
544 	(0x0e00 << 16) | (0x3c210 >> 2),
545 	0x00000000,
546 	(0x0e00 << 16) | (0x3c214 >> 2),
547 	0x00000000,
548 	(0x0e00 << 16) | (0x3c218 >> 2),
549 	0x00000000,
550 	(0x0e00 << 16) | (0x8904 >> 2),
551 	0x00000000,
552 	0x5,
553 	(0x0e00 << 16) | (0x8c28 >> 2),
554 	(0x0e00 << 16) | (0x8c2c >> 2),
555 	(0x0e00 << 16) | (0x8c30 >> 2),
556 	(0x0e00 << 16) | (0x8c34 >> 2),
557 	(0x0e00 << 16) | (0x9600 >> 2),
558 };
559 
560 static const u32 kalindi_rlc_save_restore_register_list[] =
561 {
562 	(0x0e00 << 16) | (0xc12c >> 2),
563 	0x00000000,
564 	(0x0e00 << 16) | (0xc140 >> 2),
565 	0x00000000,
566 	(0x0e00 << 16) | (0xc150 >> 2),
567 	0x00000000,
568 	(0x0e00 << 16) | (0xc15c >> 2),
569 	0x00000000,
570 	(0x0e00 << 16) | (0xc168 >> 2),
571 	0x00000000,
572 	(0x0e00 << 16) | (0xc170 >> 2),
573 	0x00000000,
574 	(0x0e00 << 16) | (0xc204 >> 2),
575 	0x00000000,
576 	(0x0e00 << 16) | (0xc2b4 >> 2),
577 	0x00000000,
578 	(0x0e00 << 16) | (0xc2b8 >> 2),
579 	0x00000000,
580 	(0x0e00 << 16) | (0xc2bc >> 2),
581 	0x00000000,
582 	(0x0e00 << 16) | (0xc2c0 >> 2),
583 	0x00000000,
584 	(0x0e00 << 16) | (0x8228 >> 2),
585 	0x00000000,
586 	(0x0e00 << 16) | (0x829c >> 2),
587 	0x00000000,
588 	(0x0e00 << 16) | (0x869c >> 2),
589 	0x00000000,
590 	(0x0600 << 16) | (0x98f4 >> 2),
591 	0x00000000,
592 	(0x0e00 << 16) | (0x98f8 >> 2),
593 	0x00000000,
594 	(0x0e00 << 16) | (0x9900 >> 2),
595 	0x00000000,
596 	(0x0e00 << 16) | (0xc260 >> 2),
597 	0x00000000,
598 	(0x0e00 << 16) | (0x90e8 >> 2),
599 	0x00000000,
600 	(0x0e00 << 16) | (0x3c000 >> 2),
601 	0x00000000,
602 	(0x0e00 << 16) | (0x3c00c >> 2),
603 	0x00000000,
604 	(0x0e00 << 16) | (0x8c1c >> 2),
605 	0x00000000,
606 	(0x0e00 << 16) | (0x9700 >> 2),
607 	0x00000000,
608 	(0x0e00 << 16) | (0xcd20 >> 2),
609 	0x00000000,
610 	(0x4e00 << 16) | (0xcd20 >> 2),
611 	0x00000000,
612 	(0x5e00 << 16) | (0xcd20 >> 2),
613 	0x00000000,
614 	(0x6e00 << 16) | (0xcd20 >> 2),
615 	0x00000000,
616 	(0x7e00 << 16) | (0xcd20 >> 2),
617 	0x00000000,
618 	(0x0e00 << 16) | (0x89bc >> 2),
619 	0x00000000,
620 	(0x0e00 << 16) | (0x8900 >> 2),
621 	0x00000000,
622 	0x3,
623 	(0x0e00 << 16) | (0xc130 >> 2),
624 	0x00000000,
625 	(0x0e00 << 16) | (0xc134 >> 2),
626 	0x00000000,
627 	(0x0e00 << 16) | (0xc1fc >> 2),
628 	0x00000000,
629 	(0x0e00 << 16) | (0xc208 >> 2),
630 	0x00000000,
631 	(0x0e00 << 16) | (0xc264 >> 2),
632 	0x00000000,
633 	(0x0e00 << 16) | (0xc268 >> 2),
634 	0x00000000,
635 	(0x0e00 << 16) | (0xc26c >> 2),
636 	0x00000000,
637 	(0x0e00 << 16) | (0xc270 >> 2),
638 	0x00000000,
639 	(0x0e00 << 16) | (0xc274 >> 2),
640 	0x00000000,
641 	(0x0e00 << 16) | (0xc28c >> 2),
642 	0x00000000,
643 	(0x0e00 << 16) | (0xc290 >> 2),
644 	0x00000000,
645 	(0x0e00 << 16) | (0xc294 >> 2),
646 	0x00000000,
647 	(0x0e00 << 16) | (0xc298 >> 2),
648 	0x00000000,
649 	(0x0e00 << 16) | (0xc2a0 >> 2),
650 	0x00000000,
651 	(0x0e00 << 16) | (0xc2a4 >> 2),
652 	0x00000000,
653 	(0x0e00 << 16) | (0xc2a8 >> 2),
654 	0x00000000,
655 	(0x0e00 << 16) | (0xc2ac >> 2),
656 	0x00000000,
657 	(0x0e00 << 16) | (0x301d0 >> 2),
658 	0x00000000,
659 	(0x0e00 << 16) | (0x30238 >> 2),
660 	0x00000000,
661 	(0x0e00 << 16) | (0x30250 >> 2),
662 	0x00000000,
663 	(0x0e00 << 16) | (0x30254 >> 2),
664 	0x00000000,
665 	(0x0e00 << 16) | (0x30258 >> 2),
666 	0x00000000,
667 	(0x0e00 << 16) | (0x3025c >> 2),
668 	0x00000000,
669 	(0x4e00 << 16) | (0xc900 >> 2),
670 	0x00000000,
671 	(0x5e00 << 16) | (0xc900 >> 2),
672 	0x00000000,
673 	(0x6e00 << 16) | (0xc900 >> 2),
674 	0x00000000,
675 	(0x7e00 << 16) | (0xc900 >> 2),
676 	0x00000000,
677 	(0x4e00 << 16) | (0xc904 >> 2),
678 	0x00000000,
679 	(0x5e00 << 16) | (0xc904 >> 2),
680 	0x00000000,
681 	(0x6e00 << 16) | (0xc904 >> 2),
682 	0x00000000,
683 	(0x7e00 << 16) | (0xc904 >> 2),
684 	0x00000000,
685 	(0x4e00 << 16) | (0xc908 >> 2),
686 	0x00000000,
687 	(0x5e00 << 16) | (0xc908 >> 2),
688 	0x00000000,
689 	(0x6e00 << 16) | (0xc908 >> 2),
690 	0x00000000,
691 	(0x7e00 << 16) | (0xc908 >> 2),
692 	0x00000000,
693 	(0x4e00 << 16) | (0xc90c >> 2),
694 	0x00000000,
695 	(0x5e00 << 16) | (0xc90c >> 2),
696 	0x00000000,
697 	(0x6e00 << 16) | (0xc90c >> 2),
698 	0x00000000,
699 	(0x7e00 << 16) | (0xc90c >> 2),
700 	0x00000000,
701 	(0x4e00 << 16) | (0xc910 >> 2),
702 	0x00000000,
703 	(0x5e00 << 16) | (0xc910 >> 2),
704 	0x00000000,
705 	(0x6e00 << 16) | (0xc910 >> 2),
706 	0x00000000,
707 	(0x7e00 << 16) | (0xc910 >> 2),
708 	0x00000000,
709 	(0x0e00 << 16) | (0xc99c >> 2),
710 	0x00000000,
711 	(0x0e00 << 16) | (0x9834 >> 2),
712 	0x00000000,
713 	(0x0000 << 16) | (0x30f00 >> 2),
714 	0x00000000,
715 	(0x0000 << 16) | (0x30f04 >> 2),
716 	0x00000000,
717 	(0x0000 << 16) | (0x30f08 >> 2),
718 	0x00000000,
719 	(0x0000 << 16) | (0x30f0c >> 2),
720 	0x00000000,
721 	(0x0600 << 16) | (0x9b7c >> 2),
722 	0x00000000,
723 	(0x0e00 << 16) | (0x8a14 >> 2),
724 	0x00000000,
725 	(0x0e00 << 16) | (0x8a18 >> 2),
726 	0x00000000,
727 	(0x0600 << 16) | (0x30a00 >> 2),
728 	0x00000000,
729 	(0x0e00 << 16) | (0x8bf0 >> 2),
730 	0x00000000,
731 	(0x0e00 << 16) | (0x8bcc >> 2),
732 	0x00000000,
733 	(0x0e00 << 16) | (0x8b24 >> 2),
734 	0x00000000,
735 	(0x0e00 << 16) | (0x30a04 >> 2),
736 	0x00000000,
737 	(0x0600 << 16) | (0x30a10 >> 2),
738 	0x00000000,
739 	(0x0600 << 16) | (0x30a14 >> 2),
740 	0x00000000,
741 	(0x0600 << 16) | (0x30a18 >> 2),
742 	0x00000000,
743 	(0x0600 << 16) | (0x30a2c >> 2),
744 	0x00000000,
745 	(0x0e00 << 16) | (0xc700 >> 2),
746 	0x00000000,
747 	(0x0e00 << 16) | (0xc704 >> 2),
748 	0x00000000,
749 	(0x0e00 << 16) | (0xc708 >> 2),
750 	0x00000000,
751 	(0x0e00 << 16) | (0xc768 >> 2),
752 	0x00000000,
753 	(0x0400 << 16) | (0xc770 >> 2),
754 	0x00000000,
755 	(0x0400 << 16) | (0xc774 >> 2),
756 	0x00000000,
757 	(0x0400 << 16) | (0xc798 >> 2),
758 	0x00000000,
759 	(0x0400 << 16) | (0xc79c >> 2),
760 	0x00000000,
761 	(0x0e00 << 16) | (0x9100 >> 2),
762 	0x00000000,
763 	(0x0e00 << 16) | (0x3c010 >> 2),
764 	0x00000000,
765 	(0x0e00 << 16) | (0x8c00 >> 2),
766 	0x00000000,
767 	(0x0e00 << 16) | (0x8c04 >> 2),
768 	0x00000000,
769 	(0x0e00 << 16) | (0x8c20 >> 2),
770 	0x00000000,
771 	(0x0e00 << 16) | (0x8c38 >> 2),
772 	0x00000000,
773 	(0x0e00 << 16) | (0x8c3c >> 2),
774 	0x00000000,
775 	(0x0e00 << 16) | (0xae00 >> 2),
776 	0x00000000,
777 	(0x0e00 << 16) | (0x9604 >> 2),
778 	0x00000000,
779 	(0x0e00 << 16) | (0xac08 >> 2),
780 	0x00000000,
781 	(0x0e00 << 16) | (0xac0c >> 2),
782 	0x00000000,
783 	(0x0e00 << 16) | (0xac10 >> 2),
784 	0x00000000,
785 	(0x0e00 << 16) | (0xac14 >> 2),
786 	0x00000000,
787 	(0x0e00 << 16) | (0xac58 >> 2),
788 	0x00000000,
789 	(0x0e00 << 16) | (0xac68 >> 2),
790 	0x00000000,
791 	(0x0e00 << 16) | (0xac6c >> 2),
792 	0x00000000,
793 	(0x0e00 << 16) | (0xac70 >> 2),
794 	0x00000000,
795 	(0x0e00 << 16) | (0xac74 >> 2),
796 	0x00000000,
797 	(0x0e00 << 16) | (0xac78 >> 2),
798 	0x00000000,
799 	(0x0e00 << 16) | (0xac7c >> 2),
800 	0x00000000,
801 	(0x0e00 << 16) | (0xac80 >> 2),
802 	0x00000000,
803 	(0x0e00 << 16) | (0xac84 >> 2),
804 	0x00000000,
805 	(0x0e00 << 16) | (0xac88 >> 2),
806 	0x00000000,
807 	(0x0e00 << 16) | (0xac8c >> 2),
808 	0x00000000,
809 	(0x0e00 << 16) | (0x970c >> 2),
810 	0x00000000,
811 	(0x0e00 << 16) | (0x9714 >> 2),
812 	0x00000000,
813 	(0x0e00 << 16) | (0x9718 >> 2),
814 	0x00000000,
815 	(0x0e00 << 16) | (0x971c >> 2),
816 	0x00000000,
817 	(0x0e00 << 16) | (0x31068 >> 2),
818 	0x00000000,
819 	(0x4e00 << 16) | (0x31068 >> 2),
820 	0x00000000,
821 	(0x5e00 << 16) | (0x31068 >> 2),
822 	0x00000000,
823 	(0x6e00 << 16) | (0x31068 >> 2),
824 	0x00000000,
825 	(0x7e00 << 16) | (0x31068 >> 2),
826 	0x00000000,
827 	(0x0e00 << 16) | (0xcd10 >> 2),
828 	0x00000000,
829 	(0x0e00 << 16) | (0xcd14 >> 2),
830 	0x00000000,
831 	(0x0e00 << 16) | (0x88b0 >> 2),
832 	0x00000000,
833 	(0x0e00 << 16) | (0x88b4 >> 2),
834 	0x00000000,
835 	(0x0e00 << 16) | (0x88b8 >> 2),
836 	0x00000000,
837 	(0x0e00 << 16) | (0x88bc >> 2),
838 	0x00000000,
839 	(0x0400 << 16) | (0x89c0 >> 2),
840 	0x00000000,
841 	(0x0e00 << 16) | (0x88c4 >> 2),
842 	0x00000000,
843 	(0x0e00 << 16) | (0x88c8 >> 2),
844 	0x00000000,
845 	(0x0e00 << 16) | (0x88d0 >> 2),
846 	0x00000000,
847 	(0x0e00 << 16) | (0x88d4 >> 2),
848 	0x00000000,
849 	(0x0e00 << 16) | (0x88d8 >> 2),
850 	0x00000000,
851 	(0x0e00 << 16) | (0x8980 >> 2),
852 	0x00000000,
853 	(0x0e00 << 16) | (0x30938 >> 2),
854 	0x00000000,
855 	(0x0e00 << 16) | (0x3093c >> 2),
856 	0x00000000,
857 	(0x0e00 << 16) | (0x30940 >> 2),
858 	0x00000000,
859 	(0x0e00 << 16) | (0x89a0 >> 2),
860 	0x00000000,
861 	(0x0e00 << 16) | (0x30900 >> 2),
862 	0x00000000,
863 	(0x0e00 << 16) | (0x30904 >> 2),
864 	0x00000000,
865 	(0x0e00 << 16) | (0x89b4 >> 2),
866 	0x00000000,
867 	(0x0e00 << 16) | (0x3e1fc >> 2),
868 	0x00000000,
869 	(0x0e00 << 16) | (0x3c210 >> 2),
870 	0x00000000,
871 	(0x0e00 << 16) | (0x3c214 >> 2),
872 	0x00000000,
873 	(0x0e00 << 16) | (0x3c218 >> 2),
874 	0x00000000,
875 	(0x0e00 << 16) | (0x8904 >> 2),
876 	0x00000000,
877 	0x5,
878 	(0x0e00 << 16) | (0x8c28 >> 2),
879 	(0x0e00 << 16) | (0x8c2c >> 2),
880 	(0x0e00 << 16) | (0x8c30 >> 2),
881 	(0x0e00 << 16) | (0x8c34 >> 2),
882 	(0x0e00 << 16) | (0x9600 >> 2),
883 };
884 
885 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev);
886 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer);
887 static void gfx_v7_0_init_pg(struct amdgpu_device *adev);
888 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev);
889 
890 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev)
891 {
892 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
893 	amdgpu_ucode_release(&adev->gfx.me_fw);
894 	amdgpu_ucode_release(&adev->gfx.ce_fw);
895 	amdgpu_ucode_release(&adev->gfx.mec_fw);
896 	amdgpu_ucode_release(&adev->gfx.mec2_fw);
897 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
898 }
899 
900 /*
901  * Core functions
902  */
903 /**
904  * gfx_v7_0_init_microcode - load ucode images from disk
905  *
906  * @adev: amdgpu_device pointer
907  *
908  * Use the firmware interface to load the ucode images into
909  * the driver (not loaded into hw).
910  * Returns 0 on success, error on failure.
911  */
912 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev)
913 {
914 	const char *chip_name;
915 	char fw_name[30];
916 	int err;
917 
918 	DRM_DEBUG("\n");
919 
920 	switch (adev->asic_type) {
921 	case CHIP_BONAIRE:
922 		chip_name = "bonaire";
923 		break;
924 	case CHIP_HAWAII:
925 		chip_name = "hawaii";
926 		break;
927 	case CHIP_KAVERI:
928 		chip_name = "kaveri";
929 		break;
930 	case CHIP_KABINI:
931 		chip_name = "kabini";
932 		break;
933 	case CHIP_MULLINS:
934 		chip_name = "mullins";
935 		break;
936 	default: BUG();
937 	}
938 
939 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
940 	err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
941 	if (err)
942 		goto out;
943 
944 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
945 	err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
946 	if (err)
947 		goto out;
948 
949 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
950 	err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name);
951 	if (err)
952 		goto out;
953 
954 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
955 	err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
956 	if (err)
957 		goto out;
958 
959 	if (adev->asic_type == CHIP_KAVERI) {
960 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
961 		err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, fw_name);
962 		if (err)
963 			goto out;
964 	}
965 
966 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
967 	err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name);
968 	if (err)
969 		goto out;
970 out:
971 	if (err) {
972 		pr_err("gfx7: Failed to load firmware \"%s\"\n", fw_name);
973 		gfx_v7_0_free_microcode(adev);
974 	}
975 	return err;
976 }
977 
978 /**
979  * gfx_v7_0_tiling_mode_table_init - init the hw tiling table
980  *
981  * @adev: amdgpu_device pointer
982  *
983  * Starting with SI, the tiling setup is done globally in a
984  * set of 32 tiling modes.  Rather than selecting each set of
985  * parameters per surface as on older asics, we just select
986  * which index in the tiling table we want to use, and the
987  * surface uses those parameters (CIK).
988  */
989 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev)
990 {
991 	const u32 num_tile_mode_states =
992 			ARRAY_SIZE(adev->gfx.config.tile_mode_array);
993 	const u32 num_secondary_tile_mode_states =
994 			ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
995 	u32 reg_offset, split_equal_to_row_size;
996 	uint32_t *tile, *macrotile;
997 
998 	tile = adev->gfx.config.tile_mode_array;
999 	macrotile = adev->gfx.config.macrotile_mode_array;
1000 
1001 	switch (adev->gfx.config.mem_row_size_in_kb) {
1002 	case 1:
1003 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1004 		break;
1005 	case 2:
1006 	default:
1007 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1008 		break;
1009 	case 4:
1010 		split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1011 		break;
1012 	}
1013 
1014 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1015 		tile[reg_offset] = 0;
1016 	for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1017 		macrotile[reg_offset] = 0;
1018 
1019 	switch (adev->asic_type) {
1020 	case CHIP_BONAIRE:
1021 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1022 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1023 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1024 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1025 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1026 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1027 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1028 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1029 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1030 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1031 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1032 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1033 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1034 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1035 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1036 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1037 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1038 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1039 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1040 			   TILE_SPLIT(split_equal_to_row_size));
1041 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1042 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1043 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1044 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1045 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1046 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1047 			   TILE_SPLIT(split_equal_to_row_size));
1048 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1049 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1050 			   PIPE_CONFIG(ADDR_SURF_P4_16x16));
1051 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1052 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1053 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1054 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1055 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1056 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1057 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1058 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1059 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1060 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1061 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1062 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1063 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1064 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1065 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1066 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1067 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1068 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1069 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1070 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1071 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1072 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1073 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1074 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1075 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1076 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1077 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1078 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1079 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1080 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1081 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1082 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1083 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1084 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1085 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1086 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1087 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1088 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1089 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1090 		tile[21] =  (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1091 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1092 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1093 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1094 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1095 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1096 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1097 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1098 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1099 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1100 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1101 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1102 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1103 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1104 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1105 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1106 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1107 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1108 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1109 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1110 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1111 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1112 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1113 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1114 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1115 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1116 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1117 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1118 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1119 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1120 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1121 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1122 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1123 
1124 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1125 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1126 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1127 				NUM_BANKS(ADDR_SURF_16_BANK));
1128 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1129 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1130 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1131 				NUM_BANKS(ADDR_SURF_16_BANK));
1132 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1133 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1134 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1135 				NUM_BANKS(ADDR_SURF_16_BANK));
1136 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1137 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1138 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1139 				NUM_BANKS(ADDR_SURF_16_BANK));
1140 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1141 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1142 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1143 				NUM_BANKS(ADDR_SURF_16_BANK));
1144 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1145 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1146 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1147 				NUM_BANKS(ADDR_SURF_8_BANK));
1148 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1149 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1150 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1151 				NUM_BANKS(ADDR_SURF_4_BANK));
1152 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1153 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1154 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1155 				NUM_BANKS(ADDR_SURF_16_BANK));
1156 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1157 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1158 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1159 				NUM_BANKS(ADDR_SURF_16_BANK));
1160 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1161 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1162 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1163 				NUM_BANKS(ADDR_SURF_16_BANK));
1164 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1165 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1166 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1167 				NUM_BANKS(ADDR_SURF_16_BANK));
1168 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1169 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1170 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1171 				NUM_BANKS(ADDR_SURF_16_BANK));
1172 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1173 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1174 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1175 				NUM_BANKS(ADDR_SURF_8_BANK));
1176 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1177 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1178 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1179 				NUM_BANKS(ADDR_SURF_4_BANK));
1180 
1181 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1182 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1183 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1184 			if (reg_offset != 7)
1185 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1186 		break;
1187 	case CHIP_HAWAII:
1188 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1189 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1190 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1191 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1192 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1193 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1194 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1195 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1196 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1197 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1198 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1199 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1200 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1201 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1202 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1203 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1204 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1205 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1206 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1207 			   TILE_SPLIT(split_equal_to_row_size));
1208 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1209 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1210 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1211 			   TILE_SPLIT(split_equal_to_row_size));
1212 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1213 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1214 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1215 			   TILE_SPLIT(split_equal_to_row_size));
1216 		tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1217 			   PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1218 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1219 			   TILE_SPLIT(split_equal_to_row_size));
1220 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1221 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
1222 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1223 			   PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1224 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1225 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1226 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1227 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1228 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1229 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1230 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1231 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1232 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1233 		tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) |
1234 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1235 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1236 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1237 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1238 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1239 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1240 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1241 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1242 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1243 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1244 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1245 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1246 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1247 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1248 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1249 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1250 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1251 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1252 		tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1253 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1254 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1255 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1256 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1257 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1258 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1259 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1260 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1261 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1262 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1263 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1264 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1265 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1266 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1267 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1268 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1269 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1270 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1271 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1272 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1273 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1274 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1275 		tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1276 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1277 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1278 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1279 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1280 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1281 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1282 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1283 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1284 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1285 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1286 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1287 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1288 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1289 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1290 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1291 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1292 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1293 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1294 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1295 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1296 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1297 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1298 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1299 			    PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
1300 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1301 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1302 		tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1303 			    PIPE_CONFIG(ADDR_SURF_P4_16x16) |
1304 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1305 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1306 
1307 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1308 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1309 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1310 				NUM_BANKS(ADDR_SURF_16_BANK));
1311 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1312 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1313 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1314 				NUM_BANKS(ADDR_SURF_16_BANK));
1315 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1316 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1317 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1318 				NUM_BANKS(ADDR_SURF_16_BANK));
1319 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1320 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1321 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1322 				NUM_BANKS(ADDR_SURF_16_BANK));
1323 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1324 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1325 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1326 				NUM_BANKS(ADDR_SURF_8_BANK));
1327 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1328 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1329 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1330 				NUM_BANKS(ADDR_SURF_4_BANK));
1331 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1332 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1333 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1334 				NUM_BANKS(ADDR_SURF_4_BANK));
1335 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1336 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1337 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1338 				NUM_BANKS(ADDR_SURF_16_BANK));
1339 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1340 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1341 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1342 				NUM_BANKS(ADDR_SURF_16_BANK));
1343 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1344 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1345 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1346 				NUM_BANKS(ADDR_SURF_16_BANK));
1347 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1348 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1349 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1350 				NUM_BANKS(ADDR_SURF_8_BANK));
1351 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1352 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1353 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1354 				NUM_BANKS(ADDR_SURF_16_BANK));
1355 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1356 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1357 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1358 				NUM_BANKS(ADDR_SURF_8_BANK));
1359 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1360 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1361 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
1362 				NUM_BANKS(ADDR_SURF_4_BANK));
1363 
1364 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1365 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1366 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1367 			if (reg_offset != 7)
1368 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1369 		break;
1370 	case CHIP_KABINI:
1371 	case CHIP_KAVERI:
1372 	case CHIP_MULLINS:
1373 	default:
1374 		tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1375 			   PIPE_CONFIG(ADDR_SURF_P2) |
1376 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1377 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1378 		tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1379 			   PIPE_CONFIG(ADDR_SURF_P2) |
1380 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1381 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1382 		tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1383 			   PIPE_CONFIG(ADDR_SURF_P2) |
1384 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1385 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1386 		tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1387 			   PIPE_CONFIG(ADDR_SURF_P2) |
1388 			   TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1389 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1390 		tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1391 			   PIPE_CONFIG(ADDR_SURF_P2) |
1392 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1393 			   TILE_SPLIT(split_equal_to_row_size));
1394 		tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1395 			   PIPE_CONFIG(ADDR_SURF_P2) |
1396 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
1397 		tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1398 			   PIPE_CONFIG(ADDR_SURF_P2) |
1399 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) |
1400 			   TILE_SPLIT(split_equal_to_row_size));
1401 		tile[7] = (TILE_SPLIT(split_equal_to_row_size));
1402 		tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1403 			   PIPE_CONFIG(ADDR_SURF_P2));
1404 		tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1405 			   PIPE_CONFIG(ADDR_SURF_P2) |
1406 			   MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING));
1407 		tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1408 			    PIPE_CONFIG(ADDR_SURF_P2) |
1409 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1410 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1411 		tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1412 			    PIPE_CONFIG(ADDR_SURF_P2) |
1413 			    MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
1414 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1415 		tile[12] = (TILE_SPLIT(split_equal_to_row_size));
1416 		tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1417 			    PIPE_CONFIG(ADDR_SURF_P2) |
1418 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING));
1419 		tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1420 			    PIPE_CONFIG(ADDR_SURF_P2) |
1421 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1422 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1423 		tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
1424 			    PIPE_CONFIG(ADDR_SURF_P2) |
1425 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1426 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1427 		tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1428 			    PIPE_CONFIG(ADDR_SURF_P2) |
1429 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1430 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1431 		tile[17] = (TILE_SPLIT(split_equal_to_row_size));
1432 		tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1433 			    PIPE_CONFIG(ADDR_SURF_P2) |
1434 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1435 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1436 		tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
1437 			    PIPE_CONFIG(ADDR_SURF_P2) |
1438 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING));
1439 		tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1440 			    PIPE_CONFIG(ADDR_SURF_P2) |
1441 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1442 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1443 		tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
1444 			    PIPE_CONFIG(ADDR_SURF_P2) |
1445 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1446 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1447 		tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
1448 			    PIPE_CONFIG(ADDR_SURF_P2) |
1449 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1450 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1451 		tile[23] = (TILE_SPLIT(split_equal_to_row_size));
1452 		tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
1453 			    PIPE_CONFIG(ADDR_SURF_P2) |
1454 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
1455 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1456 		tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
1457 			    PIPE_CONFIG(ADDR_SURF_P2) |
1458 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1459 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1460 		tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
1461 			    PIPE_CONFIG(ADDR_SURF_P2) |
1462 			    MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
1463 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
1464 		tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1465 			    PIPE_CONFIG(ADDR_SURF_P2) |
1466 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING));
1467 		tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1468 			    PIPE_CONFIG(ADDR_SURF_P2) |
1469 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1470 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
1471 		tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
1472 			    PIPE_CONFIG(ADDR_SURF_P2) |
1473 			    MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
1474 			    SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
1475 		tile[30] = (TILE_SPLIT(split_equal_to_row_size));
1476 
1477 		macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1478 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1479 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1480 				NUM_BANKS(ADDR_SURF_8_BANK));
1481 		macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1482 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1483 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1484 				NUM_BANKS(ADDR_SURF_8_BANK));
1485 		macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1486 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1487 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1488 				NUM_BANKS(ADDR_SURF_8_BANK));
1489 		macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1490 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1491 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1492 				NUM_BANKS(ADDR_SURF_8_BANK));
1493 		macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1494 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1495 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1496 				NUM_BANKS(ADDR_SURF_8_BANK));
1497 		macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1498 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1499 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1500 				NUM_BANKS(ADDR_SURF_8_BANK));
1501 		macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1502 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1503 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1504 				NUM_BANKS(ADDR_SURF_8_BANK));
1505 		macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1506 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
1507 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1508 				NUM_BANKS(ADDR_SURF_16_BANK));
1509 		macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
1510 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1511 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1512 				NUM_BANKS(ADDR_SURF_16_BANK));
1513 		macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1514 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1515 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1516 				NUM_BANKS(ADDR_SURF_16_BANK));
1517 		macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1518 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1519 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1520 				NUM_BANKS(ADDR_SURF_16_BANK));
1521 		macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1522 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1523 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1524 				NUM_BANKS(ADDR_SURF_16_BANK));
1525 		macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1526 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1527 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
1528 				NUM_BANKS(ADDR_SURF_16_BANK));
1529 		macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1530 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1531 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
1532 				NUM_BANKS(ADDR_SURF_8_BANK));
1533 
1534 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
1535 			WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]);
1536 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
1537 			if (reg_offset != 7)
1538 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]);
1539 		break;
1540 	}
1541 }
1542 
1543 /**
1544  * gfx_v7_0_select_se_sh - select which SE, SH to address
1545  *
1546  * @adev: amdgpu_device pointer
1547  * @se_num: shader engine to address
1548  * @sh_num: sh block to address
1549  * @instance: Certain registers are instanced per SE or SH.
1550  *            0xffffffff means broadcast to all SEs or SHs (CIK).
1551  *
1552  * Select which SE, SH combinations to address.
1553  */
1554 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev,
1555 				  u32 se_num, u32 sh_num, u32 instance)
1556 {
1557 	u32 data;
1558 
1559 	if (instance == 0xffffffff)
1560 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
1561 	else
1562 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
1563 
1564 	if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1565 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1566 			GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK;
1567 	else if (se_num == 0xffffffff)
1568 		data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK |
1569 			(sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT);
1570 	else if (sh_num == 0xffffffff)
1571 		data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK |
1572 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1573 	else
1574 		data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) |
1575 			(se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT);
1576 	WREG32(mmGRBM_GFX_INDEX, data);
1577 }
1578 
1579 /**
1580  * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs
1581  *
1582  * @adev: amdgpu_device pointer
1583  *
1584  * Calculates the bitmask of enabled RBs (CIK).
1585  * Returns the enabled RB bitmask.
1586  */
1587 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1588 {
1589 	u32 data, mask;
1590 
1591 	data = RREG32(mmCC_RB_BACKEND_DISABLE);
1592 	data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1593 
1594 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1595 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1596 
1597 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1598 					 adev->gfx.config.max_sh_per_se);
1599 
1600 	return (~data) & mask;
1601 }
1602 
1603 static void
1604 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
1605 {
1606 	switch (adev->asic_type) {
1607 	case CHIP_BONAIRE:
1608 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
1609 			  SE_XSEL(1) | SE_YSEL(1);
1610 		*rconf1 |= 0x0;
1611 		break;
1612 	case CHIP_HAWAII:
1613 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
1614 			  RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) |
1615 			  PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) |
1616 			  SE_YSEL(3);
1617 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
1618 			   SE_PAIR_YSEL(2);
1619 		break;
1620 	case CHIP_KAVERI:
1621 		*rconf |= RB_MAP_PKR0(2);
1622 		*rconf1 |= 0x0;
1623 		break;
1624 	case CHIP_KABINI:
1625 	case CHIP_MULLINS:
1626 		*rconf |= 0x0;
1627 		*rconf1 |= 0x0;
1628 		break;
1629 	default:
1630 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
1631 		break;
1632 	}
1633 }
1634 
1635 static void
1636 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev,
1637 					u32 raster_config, u32 raster_config_1,
1638 					unsigned rb_mask, unsigned num_rb)
1639 {
1640 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
1641 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
1642 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
1643 	unsigned rb_per_se = num_rb / num_se;
1644 	unsigned se_mask[4];
1645 	unsigned se;
1646 
1647 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
1648 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
1649 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
1650 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
1651 
1652 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
1653 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
1654 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
1655 
1656 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
1657 			     (!se_mask[2] && !se_mask[3]))) {
1658 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
1659 
1660 		if (!se_mask[0] && !se_mask[1]) {
1661 			raster_config_1 |=
1662 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
1663 		} else {
1664 			raster_config_1 |=
1665 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
1666 		}
1667 	}
1668 
1669 	for (se = 0; se < num_se; se++) {
1670 		unsigned raster_config_se = raster_config;
1671 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
1672 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
1673 		int idx = (se / 2) * 2;
1674 
1675 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
1676 			raster_config_se &= ~SE_MAP_MASK;
1677 
1678 			if (!se_mask[idx]) {
1679 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
1680 			} else {
1681 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
1682 			}
1683 		}
1684 
1685 		pkr0_mask &= rb_mask;
1686 		pkr1_mask &= rb_mask;
1687 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
1688 			raster_config_se &= ~PKR_MAP_MASK;
1689 
1690 			if (!pkr0_mask) {
1691 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
1692 			} else {
1693 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
1694 			}
1695 		}
1696 
1697 		if (rb_per_se >= 2) {
1698 			unsigned rb0_mask = 1 << (se * rb_per_se);
1699 			unsigned rb1_mask = rb0_mask << 1;
1700 
1701 			rb0_mask &= rb_mask;
1702 			rb1_mask &= rb_mask;
1703 			if (!rb0_mask || !rb1_mask) {
1704 				raster_config_se &= ~RB_MAP_PKR0_MASK;
1705 
1706 				if (!rb0_mask) {
1707 					raster_config_se |=
1708 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
1709 				} else {
1710 					raster_config_se |=
1711 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
1712 				}
1713 			}
1714 
1715 			if (rb_per_se > 2) {
1716 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
1717 				rb1_mask = rb0_mask << 1;
1718 				rb0_mask &= rb_mask;
1719 				rb1_mask &= rb_mask;
1720 				if (!rb0_mask || !rb1_mask) {
1721 					raster_config_se &= ~RB_MAP_PKR1_MASK;
1722 
1723 					if (!rb0_mask) {
1724 						raster_config_se |=
1725 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
1726 					} else {
1727 						raster_config_se |=
1728 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
1729 					}
1730 				}
1731 			}
1732 		}
1733 
1734 		/* GRBM_GFX_INDEX has a different offset on CI+ */
1735 		gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
1736 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
1737 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1738 	}
1739 
1740 	/* GRBM_GFX_INDEX has a different offset on CI+ */
1741 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1742 }
1743 
1744 /**
1745  * gfx_v7_0_setup_rb - setup the RBs on the asic
1746  *
1747  * @adev: amdgpu_device pointer
1748  *
1749  * Configures per-SE/SH RB registers (CIK).
1750  */
1751 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev)
1752 {
1753 	int i, j;
1754 	u32 data;
1755 	u32 raster_config = 0, raster_config_1 = 0;
1756 	u32 active_rbs = 0;
1757 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1758 					adev->gfx.config.max_sh_per_se;
1759 	unsigned num_rb_pipes;
1760 
1761 	mutex_lock(&adev->grbm_idx_mutex);
1762 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1763 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1764 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1765 			data = gfx_v7_0_get_rb_active_bitmap(adev);
1766 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1767 					       rb_bitmap_width_per_sh);
1768 		}
1769 	}
1770 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1771 
1772 	adev->gfx.config.backend_enable_mask = active_rbs;
1773 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1774 
1775 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
1776 			     adev->gfx.config.max_shader_engines, 16);
1777 
1778 	gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1);
1779 
1780 	if (!adev->gfx.config.backend_enable_mask ||
1781 			adev->gfx.config.num_rbs >= num_rb_pipes) {
1782 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
1783 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
1784 	} else {
1785 		gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
1786 							adev->gfx.config.backend_enable_mask,
1787 							num_rb_pipes);
1788 	}
1789 
1790 	/* cache the values for userspace */
1791 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1792 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1793 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
1794 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
1795 				RREG32(mmCC_RB_BACKEND_DISABLE);
1796 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
1797 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
1798 			adev->gfx.config.rb_config[i][j].raster_config =
1799 				RREG32(mmPA_SC_RASTER_CONFIG);
1800 			adev->gfx.config.rb_config[i][j].raster_config_1 =
1801 				RREG32(mmPA_SC_RASTER_CONFIG_1);
1802 		}
1803 	}
1804 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1805 	mutex_unlock(&adev->grbm_idx_mutex);
1806 }
1807 
1808 #define DEFAULT_SH_MEM_BASES	(0x6000)
1809 /**
1810  * gfx_v7_0_init_compute_vmid - gart enable
1811  *
1812  * @adev: amdgpu_device pointer
1813  *
1814  * Initialize compute vmid sh_mem registers
1815  *
1816  */
1817 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev)
1818 {
1819 	int i;
1820 	uint32_t sh_mem_config;
1821 	uint32_t sh_mem_bases;
1822 
1823 	/*
1824 	 * Configure apertures:
1825 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1826 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1827 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1828 	*/
1829 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1830 	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
1831 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
1832 	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
1833 	mutex_lock(&adev->srbm_mutex);
1834 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1835 		cik_srbm_select(adev, 0, 0, 0, i);
1836 		/* CP and shaders */
1837 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
1838 		WREG32(mmSH_MEM_APE1_BASE, 1);
1839 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1840 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
1841 	}
1842 	cik_srbm_select(adev, 0, 0, 0, 0);
1843 	mutex_unlock(&adev->srbm_mutex);
1844 
1845 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1846 	   access. These should be enabled by FW for target VMIDs. */
1847 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1848 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
1849 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
1850 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
1851 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
1852 	}
1853 }
1854 
1855 static void gfx_v7_0_init_gds_vmid(struct amdgpu_device *adev)
1856 {
1857 	int vmid;
1858 
1859 	/*
1860 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1861 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1862 	 * the driver can enable them for graphics. VMID0 should maintain
1863 	 * access so that HWS firmware can save/restore entries.
1864 	 */
1865 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
1866 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
1867 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
1868 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
1869 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
1870 	}
1871 }
1872 
1873 static void gfx_v7_0_config_init(struct amdgpu_device *adev)
1874 {
1875 	adev->gfx.config.double_offchip_lds_buf = 1;
1876 }
1877 
1878 /**
1879  * gfx_v7_0_constants_init - setup the 3D engine
1880  *
1881  * @adev: amdgpu_device pointer
1882  *
1883  * init the gfx constants such as the 3D engine, tiling configuration
1884  * registers, maximum number of quad pipes, render backends...
1885  */
1886 static void gfx_v7_0_constants_init(struct amdgpu_device *adev)
1887 {
1888 	u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base;
1889 	u32 tmp;
1890 	int i;
1891 
1892 	WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT));
1893 
1894 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1895 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
1896 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
1897 
1898 	gfx_v7_0_tiling_mode_table_init(adev);
1899 
1900 	gfx_v7_0_setup_rb(adev);
1901 	gfx_v7_0_get_cu_info(adev);
1902 	gfx_v7_0_config_init(adev);
1903 
1904 	/* set HW defaults for 3D engine */
1905 	WREG32(mmCP_MEQ_THRESHOLDS,
1906 	       (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) |
1907 	       (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT));
1908 
1909 	mutex_lock(&adev->grbm_idx_mutex);
1910 	/*
1911 	 * making sure that the following register writes will be broadcasted
1912 	 * to all the shaders
1913 	 */
1914 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1915 
1916 	/* XXX SH_MEM regs */
1917 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1918 	sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
1919 				   SH_MEM_ALIGNMENT_MODE_UNALIGNED);
1920 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE,
1921 				   MTYPE_NC);
1922 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE,
1923 				   MTYPE_UC);
1924 	sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0);
1925 
1926 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
1927 				   SWIZZLE_ENABLE, 1);
1928 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1929 				   ELEMENT_SIZE, 1);
1930 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
1931 				   INDEX_STRIDE, 3);
1932 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
1933 
1934 	mutex_lock(&adev->srbm_mutex);
1935 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
1936 		if (i == 0)
1937 			sh_mem_base = 0;
1938 		else
1939 			sh_mem_base = adev->gmc.shared_aperture_start >> 48;
1940 		cik_srbm_select(adev, 0, 0, 0, i);
1941 		/* CP and shaders */
1942 		WREG32(mmSH_MEM_CONFIG, sh_mem_cfg);
1943 		WREG32(mmSH_MEM_APE1_BASE, 1);
1944 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
1945 		WREG32(mmSH_MEM_BASES, sh_mem_base);
1946 	}
1947 	cik_srbm_select(adev, 0, 0, 0, 0);
1948 	mutex_unlock(&adev->srbm_mutex);
1949 
1950 	gfx_v7_0_init_compute_vmid(adev);
1951 	gfx_v7_0_init_gds_vmid(adev);
1952 
1953 	WREG32(mmSX_DEBUG_1, 0x20);
1954 
1955 	WREG32(mmTA_CNTL_AUX, 0x00010000);
1956 
1957 	tmp = RREG32(mmSPI_CONFIG_CNTL);
1958 	tmp |= 0x03000000;
1959 	WREG32(mmSPI_CONFIG_CNTL, tmp);
1960 
1961 	WREG32(mmSQ_CONFIG, 1);
1962 
1963 	WREG32(mmDB_DEBUG, 0);
1964 
1965 	tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff;
1966 	tmp |= 0x00000400;
1967 	WREG32(mmDB_DEBUG2, tmp);
1968 
1969 	tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c;
1970 	tmp |= 0x00020200;
1971 	WREG32(mmDB_DEBUG3, tmp);
1972 
1973 	tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000;
1974 	tmp |= 0x00018208;
1975 	WREG32(mmCB_HW_CONTROL, tmp);
1976 
1977 	WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT));
1978 
1979 	WREG32(mmPA_SC_FIFO_SIZE,
1980 		((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
1981 		(adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
1982 		(adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
1983 		(adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT)));
1984 
1985 	WREG32(mmVGT_NUM_INSTANCES, 1);
1986 
1987 	WREG32(mmCP_PERFMON_CNTL, 0);
1988 
1989 	WREG32(mmSQ_CONFIG, 0);
1990 
1991 	WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS,
1992 		((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) |
1993 		(255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT)));
1994 
1995 	WREG32(mmVGT_CACHE_INVALIDATION,
1996 		(VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) |
1997 		(ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT));
1998 
1999 	WREG32(mmVGT_GS_VERTEX_REUSE, 16);
2000 	WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0);
2001 
2002 	WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK |
2003 			(3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT));
2004 	WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK);
2005 
2006 	tmp = RREG32(mmSPI_ARB_PRIORITY);
2007 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
2008 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
2009 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
2010 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
2011 	WREG32(mmSPI_ARB_PRIORITY, tmp);
2012 
2013 	mutex_unlock(&adev->grbm_idx_mutex);
2014 
2015 	udelay(50);
2016 }
2017 
2018 /**
2019  * gfx_v7_0_ring_test_ring - basic gfx ring test
2020  *
2021  * @ring: amdgpu_ring structure holding ring information
2022  *
2023  * Allocate a scratch register and write to it using the gfx ring (CIK).
2024  * Provides a basic gfx ring test to verify that the ring is working.
2025  * Used by gfx_v7_0_cp_gfx_resume();
2026  * Returns 0 on success, error on failure.
2027  */
2028 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
2029 {
2030 	struct amdgpu_device *adev = ring->adev;
2031 	uint32_t tmp = 0;
2032 	unsigned i;
2033 	int r;
2034 
2035 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2036 	r = amdgpu_ring_alloc(ring, 3);
2037 	if (r)
2038 		return r;
2039 
2040 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
2041 	amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START);
2042 	amdgpu_ring_write(ring, 0xDEADBEEF);
2043 	amdgpu_ring_commit(ring);
2044 
2045 	for (i = 0; i < adev->usec_timeout; i++) {
2046 		tmp = RREG32(mmSCRATCH_REG0);
2047 		if (tmp == 0xDEADBEEF)
2048 			break;
2049 		udelay(1);
2050 	}
2051 	if (i >= adev->usec_timeout)
2052 		r = -ETIMEDOUT;
2053 	return r;
2054 }
2055 
2056 /**
2057  * gfx_v7_0_ring_emit_hdp_flush - emit an hdp flush on the cp
2058  *
2059  * @ring: amdgpu_ring structure holding ring information
2060  *
2061  * Emits an hdp flush on the cp.
2062  */
2063 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2064 {
2065 	u32 ref_and_mask;
2066 	int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
2067 
2068 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2069 		switch (ring->me) {
2070 		case 1:
2071 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
2072 			break;
2073 		case 2:
2074 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
2075 			break;
2076 		default:
2077 			return;
2078 		}
2079 	} else {
2080 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
2081 	}
2082 
2083 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
2084 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
2085 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
2086 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
2087 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
2088 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
2089 	amdgpu_ring_write(ring, ref_and_mask);
2090 	amdgpu_ring_write(ring, ref_and_mask);
2091 	amdgpu_ring_write(ring, 0x20); /* poll interval */
2092 }
2093 
2094 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
2095 {
2096 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2097 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
2098 		EVENT_INDEX(4));
2099 
2100 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2101 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
2102 		EVENT_INDEX(0));
2103 }
2104 
2105 /**
2106  * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring
2107  *
2108  * @ring: amdgpu_ring structure holding ring information
2109  * @addr: address
2110  * @seq: sequence number
2111  * @flags: fence related flags
2112  *
2113  * Emits a fence sequence number on the gfx ring and flushes
2114  * GPU caches.
2115  */
2116 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
2117 					 u64 seq, unsigned flags)
2118 {
2119 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2120 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2121 	/* Workaround for cache flush problems. First send a dummy EOP
2122 	 * event down the pipe with seq one below.
2123 	 */
2124 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2125 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2126 				 EOP_TC_ACTION_EN |
2127 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2128 				 EVENT_INDEX(5)));
2129 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2130 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2131 				DATA_SEL(1) | INT_SEL(0));
2132 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
2133 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
2134 
2135 	/* Then send the real EOP event down the pipe. */
2136 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2137 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2138 				 EOP_TC_ACTION_EN |
2139 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2140 				 EVENT_INDEX(5)));
2141 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2142 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
2143 				DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2144 	amdgpu_ring_write(ring, lower_32_bits(seq));
2145 	amdgpu_ring_write(ring, upper_32_bits(seq));
2146 }
2147 
2148 /**
2149  * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring
2150  *
2151  * @ring: amdgpu_ring structure holding ring information
2152  * @addr: address
2153  * @seq: sequence number
2154  * @flags: fence related flags
2155  *
2156  * Emits a fence sequence number on the compute ring and flushes
2157  * GPU caches.
2158  */
2159 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
2160 					     u64 addr, u64 seq,
2161 					     unsigned flags)
2162 {
2163 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2164 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2165 
2166 	/* RELEASE_MEM - flush caches, send int */
2167 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
2168 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
2169 				 EOP_TC_ACTION_EN |
2170 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2171 				 EVENT_INDEX(5)));
2172 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2173 	amdgpu_ring_write(ring, addr & 0xfffffffc);
2174 	amdgpu_ring_write(ring, upper_32_bits(addr));
2175 	amdgpu_ring_write(ring, lower_32_bits(seq));
2176 	amdgpu_ring_write(ring, upper_32_bits(seq));
2177 }
2178 
2179 /*
2180  * IB stuff
2181  */
2182 /**
2183  * gfx_v7_0_ring_emit_ib_gfx - emit an IB (Indirect Buffer) on the ring
2184  *
2185  * @ring: amdgpu_ring structure holding ring information
2186  * @job: job to retrieve vmid from
2187  * @ib: amdgpu indirect buffer object
2188  * @flags: options (AMDGPU_HAVE_CTX_SWITCH)
2189  *
2190  * Emits an DE (drawing engine) or CE (constant engine) IB
2191  * on the gfx ring.  IBs are usually generated by userspace
2192  * acceleration drivers and submitted to the kernel for
2193  * scheduling on the ring.  This function schedules the IB
2194  * on the gfx ring for execution by the GPU.
2195  */
2196 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
2197 					struct amdgpu_job *job,
2198 					struct amdgpu_ib *ib,
2199 					uint32_t flags)
2200 {
2201 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2202 	u32 header, control = 0;
2203 
2204 	/* insert SWITCH_BUFFER packet before first IB in the ring frame */
2205 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2206 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
2207 		amdgpu_ring_write(ring, 0);
2208 	}
2209 
2210 	if (ib->flags & AMDGPU_IB_FLAG_CE)
2211 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
2212 	else
2213 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
2214 
2215 	control |= ib->length_dw | (vmid << 24);
2216 
2217 	amdgpu_ring_write(ring, header);
2218 	amdgpu_ring_write(ring,
2219 #ifdef __BIG_ENDIAN
2220 			  (2 << 0) |
2221 #endif
2222 			  (ib->gpu_addr & 0xFFFFFFFC));
2223 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2224 	amdgpu_ring_write(ring, control);
2225 }
2226 
2227 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
2228 					  struct amdgpu_job *job,
2229 					  struct amdgpu_ib *ib,
2230 					  uint32_t flags)
2231 {
2232 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2233 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
2234 
2235 	/* Currently, there is a high possibility to get wave ID mismatch
2236 	 * between ME and GDS, leading to a hw deadlock, because ME generates
2237 	 * different wave IDs than the GDS expects. This situation happens
2238 	 * randomly when at least 5 compute pipes use GDS ordered append.
2239 	 * The wave IDs generated by ME are also wrong after suspend/resume.
2240 	 * Those are probably bugs somewhere else in the kernel driver.
2241 	 *
2242 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
2243 	 * GDS to 0 for this ring (me/pipe).
2244 	 */
2245 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
2246 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2247 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
2248 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
2249 	}
2250 
2251 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2252 	amdgpu_ring_write(ring,
2253 #ifdef __BIG_ENDIAN
2254 					  (2 << 0) |
2255 #endif
2256 					  (ib->gpu_addr & 0xFFFFFFFC));
2257 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
2258 	amdgpu_ring_write(ring, control);
2259 }
2260 
2261 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
2262 {
2263 	uint32_t dw2 = 0;
2264 
2265 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
2266 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
2267 		gfx_v7_0_ring_emit_vgt_flush(ring);
2268 		/* set load_global_config & load_global_uconfig */
2269 		dw2 |= 0x8001;
2270 		/* set load_cs_sh_regs */
2271 		dw2 |= 0x01000000;
2272 		/* set load_per_context_state & load_gfx_sh_regs */
2273 		dw2 |= 0x10002;
2274 	}
2275 
2276 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2277 	amdgpu_ring_write(ring, dw2);
2278 	amdgpu_ring_write(ring, 0);
2279 }
2280 
2281 /**
2282  * gfx_v7_0_ring_test_ib - basic ring IB test
2283  *
2284  * @ring: amdgpu_ring structure holding ring information
2285  * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT
2286  *
2287  * Allocate an IB and execute it on the gfx ring (CIK).
2288  * Provides a basic gfx ring test to verify that IBs are working.
2289  * Returns 0 on success, error on failure.
2290  */
2291 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
2292 {
2293 	struct amdgpu_device *adev = ring->adev;
2294 	struct amdgpu_ib ib;
2295 	struct dma_fence *f = NULL;
2296 	uint32_t tmp = 0;
2297 	long r;
2298 
2299 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
2300 	memset(&ib, 0, sizeof(ib));
2301 	r = amdgpu_ib_get(adev, NULL, 256, AMDGPU_IB_POOL_DIRECT, &ib);
2302 	if (r)
2303 		return r;
2304 
2305 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
2306 	ib.ptr[1] = mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START;
2307 	ib.ptr[2] = 0xDEADBEEF;
2308 	ib.length_dw = 3;
2309 
2310 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
2311 	if (r)
2312 		goto error;
2313 
2314 	r = dma_fence_wait_timeout(f, false, timeout);
2315 	if (r == 0) {
2316 		r = -ETIMEDOUT;
2317 		goto error;
2318 	} else if (r < 0) {
2319 		goto error;
2320 	}
2321 	tmp = RREG32(mmSCRATCH_REG0);
2322 	if (tmp == 0xDEADBEEF)
2323 		r = 0;
2324 	else
2325 		r = -EINVAL;
2326 
2327 error:
2328 	amdgpu_ib_free(adev, &ib, NULL);
2329 	dma_fence_put(f);
2330 	return r;
2331 }
2332 
2333 /*
2334  * CP.
2335  * On CIK, gfx and compute now have independent command processors.
2336  *
2337  * GFX
2338  * Gfx consists of a single ring and can process both gfx jobs and
2339  * compute jobs.  The gfx CP consists of three microengines (ME):
2340  * PFP - Pre-Fetch Parser
2341  * ME - Micro Engine
2342  * CE - Constant Engine
2343  * The PFP and ME make up what is considered the Drawing Engine (DE).
2344  * The CE is an asynchronous engine used for updating buffer desciptors
2345  * used by the DE so that they can be loaded into cache in parallel
2346  * while the DE is processing state update packets.
2347  *
2348  * Compute
2349  * The compute CP consists of two microengines (ME):
2350  * MEC1 - Compute MicroEngine 1
2351  * MEC2 - Compute MicroEngine 2
2352  * Each MEC supports 4 compute pipes and each pipe supports 8 queues.
2353  * The queues are exposed to userspace and are programmed directly
2354  * by the compute runtime.
2355  */
2356 /**
2357  * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs
2358  *
2359  * @adev: amdgpu_device pointer
2360  * @enable: enable or disable the MEs
2361  *
2362  * Halts or unhalts the gfx MEs.
2363  */
2364 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2365 {
2366 	if (enable)
2367 		WREG32(mmCP_ME_CNTL, 0);
2368 	else
2369 		WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK |
2370 				      CP_ME_CNTL__PFP_HALT_MASK |
2371 				      CP_ME_CNTL__CE_HALT_MASK));
2372 	udelay(50);
2373 }
2374 
2375 /**
2376  * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode
2377  *
2378  * @adev: amdgpu_device pointer
2379  *
2380  * Loads the gfx PFP, ME, and CE ucode.
2381  * Returns 0 for success, -EINVAL if the ucode is not available.
2382  */
2383 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2384 {
2385 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2386 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2387 	const struct gfx_firmware_header_v1_0 *me_hdr;
2388 	const __le32 *fw_data;
2389 	unsigned i, fw_size;
2390 
2391 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2392 		return -EINVAL;
2393 
2394 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
2395 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
2396 	me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
2397 
2398 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2399 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2400 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2401 	adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version);
2402 	adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version);
2403 	adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version);
2404 	adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version);
2405 	adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version);
2406 	adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version);
2407 
2408 	gfx_v7_0_cp_gfx_enable(adev, false);
2409 
2410 	/* PFP */
2411 	fw_data = (const __le32 *)
2412 		(adev->gfx.pfp_fw->data +
2413 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2414 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2415 	WREG32(mmCP_PFP_UCODE_ADDR, 0);
2416 	for (i = 0; i < fw_size; i++)
2417 		WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2418 	WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2419 
2420 	/* CE */
2421 	fw_data = (const __le32 *)
2422 		(adev->gfx.ce_fw->data +
2423 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2424 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2425 	WREG32(mmCP_CE_UCODE_ADDR, 0);
2426 	for (i = 0; i < fw_size; i++)
2427 		WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2428 	WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2429 
2430 	/* ME */
2431 	fw_data = (const __le32 *)
2432 		(adev->gfx.me_fw->data +
2433 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2434 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2435 	WREG32(mmCP_ME_RAM_WADDR, 0);
2436 	for (i = 0; i < fw_size; i++)
2437 		WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2438 	WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2439 
2440 	return 0;
2441 }
2442 
2443 /**
2444  * gfx_v7_0_cp_gfx_start - start the gfx ring
2445  *
2446  * @adev: amdgpu_device pointer
2447  *
2448  * Enables the ring and loads the clear state context and other
2449  * packets required to init the ring.
2450  * Returns 0 for success, error for failure.
2451  */
2452 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev)
2453 {
2454 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2455 	const struct cs_section_def *sect = NULL;
2456 	const struct cs_extent_def *ext = NULL;
2457 	int r, i;
2458 
2459 	/* init the CP */
2460 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2461 	WREG32(mmCP_ENDIAN_SWAP, 0);
2462 	WREG32(mmCP_DEVICE_ID, 1);
2463 
2464 	gfx_v7_0_cp_gfx_enable(adev, true);
2465 
2466 	r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8);
2467 	if (r) {
2468 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2469 		return r;
2470 	}
2471 
2472 	/* init the CE partitions.  CE only used for gfx on CIK */
2473 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2474 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2475 	amdgpu_ring_write(ring, 0x8000);
2476 	amdgpu_ring_write(ring, 0x8000);
2477 
2478 	/* clear state buffer */
2479 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2480 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2481 
2482 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2483 	amdgpu_ring_write(ring, 0x80000000);
2484 	amdgpu_ring_write(ring, 0x80000000);
2485 
2486 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
2487 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2488 			if (sect->id == SECT_CONTEXT) {
2489 				amdgpu_ring_write(ring,
2490 						  PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
2491 				amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2492 				for (i = 0; i < ext->reg_count; i++)
2493 					amdgpu_ring_write(ring, ext->extent[i]);
2494 			}
2495 		}
2496 	}
2497 
2498 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2499 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
2500 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
2501 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
2502 
2503 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2504 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2505 
2506 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2507 	amdgpu_ring_write(ring, 0);
2508 
2509 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2510 	amdgpu_ring_write(ring, 0x00000316);
2511 	amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2512 	amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2513 
2514 	amdgpu_ring_commit(ring);
2515 
2516 	return 0;
2517 }
2518 
2519 /**
2520  * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers
2521  *
2522  * @adev: amdgpu_device pointer
2523  *
2524  * Program the location and size of the gfx ring buffer
2525  * and test it to make sure it's working.
2526  * Returns 0 for success, error for failure.
2527  */
2528 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev)
2529 {
2530 	struct amdgpu_ring *ring;
2531 	u32 tmp;
2532 	u32 rb_bufsz;
2533 	u64 rb_addr, rptr_addr;
2534 	int r;
2535 
2536 	WREG32(mmCP_SEM_WAIT_TIMER, 0x0);
2537 	if (adev->asic_type != CHIP_HAWAII)
2538 		WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2539 
2540 	/* Set the write pointer delay */
2541 	WREG32(mmCP_RB_WPTR_DELAY, 0);
2542 
2543 	/* set the RB to use vmid 0 */
2544 	WREG32(mmCP_RB_VMID, 0);
2545 
2546 	WREG32(mmSCRATCH_ADDR, 0);
2547 
2548 	/* ring 0 - compute and gfx */
2549 	/* Set ring buffer size */
2550 	ring = &adev->gfx.gfx_ring[0];
2551 	rb_bufsz = order_base_2(ring->ring_size / 8);
2552 	tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2553 #ifdef __BIG_ENDIAN
2554 	tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT;
2555 #endif
2556 	WREG32(mmCP_RB0_CNTL, tmp);
2557 
2558 	/* Initialize the ring buffer's read and write pointers */
2559 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
2560 	ring->wptr = 0;
2561 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2562 
2563 	/* set the wb address wether it's enabled or not */
2564 	rptr_addr = ring->rptr_gpu_addr;
2565 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2566 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
2567 
2568 	/* scratch register shadowing is no longer supported */
2569 	WREG32(mmSCRATCH_UMSK, 0);
2570 
2571 	mdelay(1);
2572 	WREG32(mmCP_RB0_CNTL, tmp);
2573 
2574 	rb_addr = ring->gpu_addr >> 8;
2575 	WREG32(mmCP_RB0_BASE, rb_addr);
2576 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2577 
2578 	/* start the ring */
2579 	gfx_v7_0_cp_gfx_start(adev);
2580 	r = amdgpu_ring_test_helper(ring);
2581 	if (r)
2582 		return r;
2583 
2584 	return 0;
2585 }
2586 
2587 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
2588 {
2589 	return *ring->rptr_cpu_addr;
2590 }
2591 
2592 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
2593 {
2594 	struct amdgpu_device *adev = ring->adev;
2595 
2596 	return RREG32(mmCP_RB0_WPTR);
2597 }
2598 
2599 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
2600 {
2601 	struct amdgpu_device *adev = ring->adev;
2602 
2603 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2604 	(void)RREG32(mmCP_RB0_WPTR);
2605 }
2606 
2607 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
2608 {
2609 	/* XXX check if swapping is necessary on BE */
2610 	return *ring->wptr_cpu_addr;
2611 }
2612 
2613 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
2614 {
2615 	struct amdgpu_device *adev = ring->adev;
2616 
2617 	/* XXX check if swapping is necessary on BE */
2618 	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
2619 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2620 }
2621 
2622 /**
2623  * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs
2624  *
2625  * @adev: amdgpu_device pointer
2626  * @enable: enable or disable the MEs
2627  *
2628  * Halts or unhalts the compute MEs.
2629  */
2630 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2631 {
2632 	if (enable)
2633 		WREG32(mmCP_MEC_CNTL, 0);
2634 	else
2635 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK |
2636 				       CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2637 	udelay(50);
2638 }
2639 
2640 /**
2641  * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode
2642  *
2643  * @adev: amdgpu_device pointer
2644  *
2645  * Loads the compute MEC1&2 ucode.
2646  * Returns 0 for success, -EINVAL if the ucode is not available.
2647  */
2648 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2649 {
2650 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2651 	const __le32 *fw_data;
2652 	unsigned i, fw_size;
2653 
2654 	if (!adev->gfx.mec_fw)
2655 		return -EINVAL;
2656 
2657 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2658 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2659 	adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version);
2660 	adev->gfx.mec_feature_version = le32_to_cpu(
2661 					mec_hdr->ucode_feature_version);
2662 
2663 	gfx_v7_0_cp_compute_enable(adev, false);
2664 
2665 	/* MEC1 */
2666 	fw_data = (const __le32 *)
2667 		(adev->gfx.mec_fw->data +
2668 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2669 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
2670 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2671 	for (i = 0; i < fw_size; i++)
2672 		WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++));
2673 	WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0);
2674 
2675 	if (adev->asic_type == CHIP_KAVERI) {
2676 		const struct gfx_firmware_header_v1_0 *mec2_hdr;
2677 
2678 		if (!adev->gfx.mec2_fw)
2679 			return -EINVAL;
2680 
2681 		mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data;
2682 		amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header);
2683 		adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version);
2684 		adev->gfx.mec2_feature_version = le32_to_cpu(
2685 				mec2_hdr->ucode_feature_version);
2686 
2687 		/* MEC2 */
2688 		fw_data = (const __le32 *)
2689 			(adev->gfx.mec2_fw->data +
2690 			 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes));
2691 		fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4;
2692 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2693 		for (i = 0; i < fw_size; i++)
2694 			WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++));
2695 		WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0);
2696 	}
2697 
2698 	return 0;
2699 }
2700 
2701 /**
2702  * gfx_v7_0_cp_compute_fini - stop the compute queues
2703  *
2704  * @adev: amdgpu_device pointer
2705  *
2706  * Stop the compute queues and tear down the driver queue
2707  * info.
2708  */
2709 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev)
2710 {
2711 	int i;
2712 
2713 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2714 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
2715 
2716 		amdgpu_bo_free_kernel(&ring->mqd_obj, NULL, NULL);
2717 	}
2718 }
2719 
2720 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev)
2721 {
2722 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
2723 }
2724 
2725 static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
2726 {
2727 	int r;
2728 	u32 *hpd;
2729 	size_t mec_hpd_size;
2730 
2731 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
2732 
2733 	/* take ownership of the relevant compute queues */
2734 	amdgpu_gfx_compute_queue_acquire(adev);
2735 
2736 	/* allocate space for ALL pipes (even the ones we don't own) */
2737 	mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec
2738 		* GFX7_MEC_HPD_SIZE * 2;
2739 
2740 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
2741 				      AMDGPU_GEM_DOMAIN_VRAM |
2742 				      AMDGPU_GEM_DOMAIN_GTT,
2743 				      &adev->gfx.mec.hpd_eop_obj,
2744 				      &adev->gfx.mec.hpd_eop_gpu_addr,
2745 				      (void **)&hpd);
2746 	if (r) {
2747 		dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r);
2748 		gfx_v7_0_mec_fini(adev);
2749 		return r;
2750 	}
2751 
2752 	/* clear memory.  Not sure if this is required or not */
2753 	memset(hpd, 0, mec_hpd_size);
2754 
2755 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
2756 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
2757 
2758 	return 0;
2759 }
2760 
2761 struct hqd_registers
2762 {
2763 	u32 cp_mqd_base_addr;
2764 	u32 cp_mqd_base_addr_hi;
2765 	u32 cp_hqd_active;
2766 	u32 cp_hqd_vmid;
2767 	u32 cp_hqd_persistent_state;
2768 	u32 cp_hqd_pipe_priority;
2769 	u32 cp_hqd_queue_priority;
2770 	u32 cp_hqd_quantum;
2771 	u32 cp_hqd_pq_base;
2772 	u32 cp_hqd_pq_base_hi;
2773 	u32 cp_hqd_pq_rptr;
2774 	u32 cp_hqd_pq_rptr_report_addr;
2775 	u32 cp_hqd_pq_rptr_report_addr_hi;
2776 	u32 cp_hqd_pq_wptr_poll_addr;
2777 	u32 cp_hqd_pq_wptr_poll_addr_hi;
2778 	u32 cp_hqd_pq_doorbell_control;
2779 	u32 cp_hqd_pq_wptr;
2780 	u32 cp_hqd_pq_control;
2781 	u32 cp_hqd_ib_base_addr;
2782 	u32 cp_hqd_ib_base_addr_hi;
2783 	u32 cp_hqd_ib_rptr;
2784 	u32 cp_hqd_ib_control;
2785 	u32 cp_hqd_iq_timer;
2786 	u32 cp_hqd_iq_rptr;
2787 	u32 cp_hqd_dequeue_request;
2788 	u32 cp_hqd_dma_offload;
2789 	u32 cp_hqd_sema_cmd;
2790 	u32 cp_hqd_msg_type;
2791 	u32 cp_hqd_atomic0_preop_lo;
2792 	u32 cp_hqd_atomic0_preop_hi;
2793 	u32 cp_hqd_atomic1_preop_lo;
2794 	u32 cp_hqd_atomic1_preop_hi;
2795 	u32 cp_hqd_hq_scheduler0;
2796 	u32 cp_hqd_hq_scheduler1;
2797 	u32 cp_mqd_control;
2798 };
2799 
2800 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev,
2801 				       int mec, int pipe)
2802 {
2803 	u64 eop_gpu_addr;
2804 	u32 tmp;
2805 	size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe)
2806 			    * GFX7_MEC_HPD_SIZE * 2;
2807 
2808 	mutex_lock(&adev->srbm_mutex);
2809 	eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset;
2810 
2811 	cik_srbm_select(adev, mec + 1, pipe, 0, 0);
2812 
2813 	/* write the EOP addr */
2814 	WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
2815 	WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
2816 
2817 	/* set the VMID assigned */
2818 	WREG32(mmCP_HPD_EOP_VMID, 0);
2819 
2820 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
2821 	tmp = RREG32(mmCP_HPD_EOP_CONTROL);
2822 	tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK;
2823 	tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8);
2824 	WREG32(mmCP_HPD_EOP_CONTROL, tmp);
2825 
2826 	cik_srbm_select(adev, 0, 0, 0, 0);
2827 	mutex_unlock(&adev->srbm_mutex);
2828 }
2829 
2830 static int gfx_v7_0_mqd_deactivate(struct amdgpu_device *adev)
2831 {
2832 	int i;
2833 
2834 	/* disable the queue if it's active */
2835 	if (RREG32(mmCP_HQD_ACTIVE) & 1) {
2836 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1);
2837 		for (i = 0; i < adev->usec_timeout; i++) {
2838 			if (!(RREG32(mmCP_HQD_ACTIVE) & 1))
2839 				break;
2840 			udelay(1);
2841 		}
2842 
2843 		if (i == adev->usec_timeout)
2844 			return -ETIMEDOUT;
2845 
2846 		WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
2847 		WREG32(mmCP_HQD_PQ_RPTR, 0);
2848 		WREG32(mmCP_HQD_PQ_WPTR, 0);
2849 	}
2850 
2851 	return 0;
2852 }
2853 
2854 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev,
2855 			     struct cik_mqd *mqd,
2856 			     uint64_t mqd_gpu_addr,
2857 			     struct amdgpu_ring *ring)
2858 {
2859 	u64 hqd_gpu_addr;
2860 	u64 wb_gpu_addr;
2861 
2862 	/* init the mqd struct */
2863 	memset(mqd, 0, sizeof(struct cik_mqd));
2864 
2865 	mqd->header = 0xC0310800;
2866 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
2867 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
2868 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
2869 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
2870 
2871 	/* enable doorbell? */
2872 	mqd->cp_hqd_pq_doorbell_control =
2873 		RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2874 	if (ring->use_doorbell)
2875 		mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2876 	else
2877 		mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2878 
2879 	/* set the pointer to the MQD */
2880 	mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc;
2881 	mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr);
2882 
2883 	/* set MQD vmid to 0 */
2884 	mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL);
2885 	mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK;
2886 
2887 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
2888 	hqd_gpu_addr = ring->gpu_addr >> 8;
2889 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
2890 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
2891 
2892 	/* set up the HQD, this is similar to CP_RB0_CNTL */
2893 	mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL);
2894 	mqd->cp_hqd_pq_control &=
2895 		~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK |
2896 				CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK);
2897 
2898 	mqd->cp_hqd_pq_control |=
2899 		order_base_2(ring->ring_size / 8);
2900 	mqd->cp_hqd_pq_control |=
2901 		(order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8);
2902 #ifdef __BIG_ENDIAN
2903 	mqd->cp_hqd_pq_control |=
2904 		2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT;
2905 #endif
2906 	mqd->cp_hqd_pq_control &=
2907 		~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK |
2908 				CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK |
2909 				CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK);
2910 	mqd->cp_hqd_pq_control |=
2911 		CP_HQD_PQ_CONTROL__PRIV_STATE_MASK |
2912 		CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */
2913 
2914 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
2915 	wb_gpu_addr = ring->wptr_gpu_addr;
2916 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
2917 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
2918 
2919 	/* set the wb address wether it's enabled or not */
2920 	wb_gpu_addr = ring->rptr_gpu_addr;
2921 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
2922 	mqd->cp_hqd_pq_rptr_report_addr_hi =
2923 		upper_32_bits(wb_gpu_addr) & 0xffff;
2924 
2925 	/* enable the doorbell if requested */
2926 	if (ring->use_doorbell) {
2927 		mqd->cp_hqd_pq_doorbell_control =
2928 			RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
2929 		mqd->cp_hqd_pq_doorbell_control &=
2930 			~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK;
2931 		mqd->cp_hqd_pq_doorbell_control |=
2932 			(ring->doorbell_index <<
2933 			 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT);
2934 		mqd->cp_hqd_pq_doorbell_control |=
2935 			CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK;
2936 		mqd->cp_hqd_pq_doorbell_control &=
2937 			~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK |
2938 					CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK);
2939 
2940 	} else {
2941 		mqd->cp_hqd_pq_doorbell_control = 0;
2942 	}
2943 
2944 	/* read and write pointers, similar to CP_RB0_WPTR/_RPTR */
2945 	ring->wptr = 0;
2946 	mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr);
2947 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2948 
2949 	/* set the vmid for the queue */
2950 	mqd->cp_hqd_vmid = 0;
2951 
2952 	/* defaults */
2953 	mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL);
2954 	mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR);
2955 	mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI);
2956 	mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR);
2957 	mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE);
2958 	mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD);
2959 	mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE);
2960 	mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO);
2961 	mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI);
2962 	mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO);
2963 	mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI);
2964 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
2965 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
2966 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
2967 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
2968 	mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR);
2969 
2970 	/* activate the queue */
2971 	mqd->cp_hqd_active = 1;
2972 }
2973 
2974 static int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd)
2975 {
2976 	uint32_t tmp;
2977 	uint32_t mqd_reg;
2978 	uint32_t *mqd_data;
2979 
2980 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */
2981 	mqd_data = &mqd->cp_mqd_base_addr_lo;
2982 
2983 	/* disable wptr polling */
2984 	tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL);
2985 	tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0);
2986 	WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp);
2987 
2988 	/* program all HQD registers */
2989 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++)
2990 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
2991 
2992 	/* activate the HQD */
2993 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
2994 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
2995 
2996 	return 0;
2997 }
2998 
2999 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id)
3000 {
3001 	int r;
3002 	u64 mqd_gpu_addr;
3003 	struct cik_mqd *mqd;
3004 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
3005 
3006 	r = amdgpu_bo_create_reserved(adev, sizeof(struct cik_mqd), PAGE_SIZE,
3007 				      AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj,
3008 				      &mqd_gpu_addr, (void **)&mqd);
3009 	if (r) {
3010 		dev_warn(adev->dev, "(%d) create MQD bo failed\n", r);
3011 		return r;
3012 	}
3013 
3014 	mutex_lock(&adev->srbm_mutex);
3015 	cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3016 
3017 	gfx_v7_0_mqd_init(adev, mqd, mqd_gpu_addr, ring);
3018 	gfx_v7_0_mqd_deactivate(adev);
3019 	gfx_v7_0_mqd_commit(adev, mqd);
3020 
3021 	cik_srbm_select(adev, 0, 0, 0, 0);
3022 	mutex_unlock(&adev->srbm_mutex);
3023 
3024 	amdgpu_bo_kunmap(ring->mqd_obj);
3025 	amdgpu_bo_unreserve(ring->mqd_obj);
3026 	return 0;
3027 }
3028 
3029 /**
3030  * gfx_v7_0_cp_compute_resume - setup the compute queue registers
3031  *
3032  * @adev: amdgpu_device pointer
3033  *
3034  * Program the compute queues and test them to make sure they
3035  * are working.
3036  * Returns 0 for success, error for failure.
3037  */
3038 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev)
3039 {
3040 	int r, i, j;
3041 	u32 tmp;
3042 	struct amdgpu_ring *ring;
3043 
3044 	/* fix up chicken bits */
3045 	tmp = RREG32(mmCP_CPF_DEBUG);
3046 	tmp |= (1 << 23);
3047 	WREG32(mmCP_CPF_DEBUG, tmp);
3048 
3049 	/* init all pipes (even the ones we don't own) */
3050 	for (i = 0; i < adev->gfx.mec.num_mec; i++)
3051 		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++)
3052 			gfx_v7_0_compute_pipe_init(adev, i, j);
3053 
3054 	/* init the queues */
3055 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3056 		r = gfx_v7_0_compute_queue_init(adev, i);
3057 		if (r) {
3058 			gfx_v7_0_cp_compute_fini(adev);
3059 			return r;
3060 		}
3061 	}
3062 
3063 	gfx_v7_0_cp_compute_enable(adev, true);
3064 
3065 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3066 		ring = &adev->gfx.compute_ring[i];
3067 		amdgpu_ring_test_helper(ring);
3068 	}
3069 
3070 	return 0;
3071 }
3072 
3073 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable)
3074 {
3075 	gfx_v7_0_cp_gfx_enable(adev, enable);
3076 	gfx_v7_0_cp_compute_enable(adev, enable);
3077 }
3078 
3079 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev)
3080 {
3081 	int r;
3082 
3083 	r = gfx_v7_0_cp_gfx_load_microcode(adev);
3084 	if (r)
3085 		return r;
3086 	r = gfx_v7_0_cp_compute_load_microcode(adev);
3087 	if (r)
3088 		return r;
3089 
3090 	return 0;
3091 }
3092 
3093 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3094 					       bool enable)
3095 {
3096 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3097 
3098 	if (enable)
3099 		tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3100 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3101 	else
3102 		tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK |
3103 				CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK);
3104 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3105 }
3106 
3107 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev)
3108 {
3109 	int r;
3110 
3111 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3112 
3113 	r = gfx_v7_0_cp_load_microcode(adev);
3114 	if (r)
3115 		return r;
3116 
3117 	r = gfx_v7_0_cp_gfx_resume(adev);
3118 	if (r)
3119 		return r;
3120 	r = gfx_v7_0_cp_compute_resume(adev);
3121 	if (r)
3122 		return r;
3123 
3124 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3125 
3126 	return 0;
3127 }
3128 
3129 /**
3130  * gfx_v7_0_ring_emit_pipeline_sync - cik vm flush using the CP
3131  *
3132  * @ring: the ring to emit the commands to
3133  *
3134  * Sync the command pipeline with the PFP. E.g. wait for everything
3135  * to be completed.
3136  */
3137 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
3138 {
3139 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3140 	uint32_t seq = ring->fence_drv.sync_seq;
3141 	uint64_t addr = ring->fence_drv.gpu_addr;
3142 
3143 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3144 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
3145 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
3146 				 WAIT_REG_MEM_ENGINE(usepfp)));   /* pfp or me */
3147 	amdgpu_ring_write(ring, addr & 0xfffffffc);
3148 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
3149 	amdgpu_ring_write(ring, seq);
3150 	amdgpu_ring_write(ring, 0xffffffff);
3151 	amdgpu_ring_write(ring, 4); /* poll interval */
3152 
3153 	if (usepfp) {
3154 		/* sync CE with ME to prevent CE fetch CEIB before context switch done */
3155 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3156 		amdgpu_ring_write(ring, 0);
3157 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3158 		amdgpu_ring_write(ring, 0);
3159 	}
3160 }
3161 
3162 /*
3163  * vm
3164  * VMID 0 is the physical GPU addresses as used by the kernel.
3165  * VMIDs 1-15 are used for userspace clients and are handled
3166  * by the amdgpu vm/hsa code.
3167  */
3168 /**
3169  * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP
3170  *
3171  * @ring: amdgpu_ring pointer
3172  * @vmid: vmid number to use
3173  * @pd_addr: address
3174  *
3175  * Update the page table base and flush the VM TLB
3176  * using the CP (CIK).
3177  */
3178 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
3179 					unsigned vmid, uint64_t pd_addr)
3180 {
3181 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3182 
3183 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
3184 
3185 	/* wait for the invalidate to complete */
3186 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
3187 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
3188 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
3189 				 WAIT_REG_MEM_ENGINE(0))); /* me */
3190 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
3191 	amdgpu_ring_write(ring, 0);
3192 	amdgpu_ring_write(ring, 0); /* ref */
3193 	amdgpu_ring_write(ring, 0); /* mask */
3194 	amdgpu_ring_write(ring, 0x20); /* poll interval */
3195 
3196 	/* compute doesn't have PFP */
3197 	if (usepfp) {
3198 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
3199 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3200 		amdgpu_ring_write(ring, 0x0);
3201 
3202 		/* synce CE with ME to prevent CE fetch CEIB before context switch done */
3203 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3204 		amdgpu_ring_write(ring, 0);
3205 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
3206 		amdgpu_ring_write(ring, 0);
3207 	}
3208 }
3209 
3210 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
3211 				    uint32_t reg, uint32_t val)
3212 {
3213 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
3214 
3215 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3216 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) |
3217 				 WRITE_DATA_DST_SEL(0)));
3218 	amdgpu_ring_write(ring, reg);
3219 	amdgpu_ring_write(ring, 0);
3220 	amdgpu_ring_write(ring, val);
3221 }
3222 
3223 /*
3224  * RLC
3225  * The RLC is a multi-purpose microengine that handles a
3226  * variety of functions.
3227  */
3228 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
3229 {
3230 	const u32 *src_ptr;
3231 	u32 dws;
3232 	const struct cs_section_def *cs_data;
3233 	int r;
3234 
3235 	/* allocate rlc buffers */
3236 	if (adev->flags & AMD_IS_APU) {
3237 		if (adev->asic_type == CHIP_KAVERI) {
3238 			adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list;
3239 			adev->gfx.rlc.reg_list_size =
3240 				(u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list);
3241 		} else {
3242 			adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list;
3243 			adev->gfx.rlc.reg_list_size =
3244 				(u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list);
3245 		}
3246 	}
3247 	adev->gfx.rlc.cs_data = ci_cs_data;
3248 	adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */
3249 	adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */
3250 
3251 	src_ptr = adev->gfx.rlc.reg_list;
3252 	dws = adev->gfx.rlc.reg_list_size;
3253 	dws += (5 * 16) + 48 + 48 + 64;
3254 
3255 	cs_data = adev->gfx.rlc.cs_data;
3256 
3257 	if (src_ptr) {
3258 		/* init save restore block */
3259 		r = amdgpu_gfx_rlc_init_sr(adev, dws);
3260 		if (r)
3261 			return r;
3262 	}
3263 
3264 	if (cs_data) {
3265 		/* init clear state block */
3266 		r = amdgpu_gfx_rlc_init_csb(adev);
3267 		if (r)
3268 			return r;
3269 	}
3270 
3271 	if (adev->gfx.rlc.cp_table_size) {
3272 		r = amdgpu_gfx_rlc_init_cpt(adev);
3273 		if (r)
3274 			return r;
3275 	}
3276 
3277 	/* init spm vmid with 0xf */
3278 	if (adev->gfx.rlc.funcs->update_spm_vmid)
3279 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
3280 
3281 	return 0;
3282 }
3283 
3284 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
3285 {
3286 	u32 tmp;
3287 
3288 	tmp = RREG32(mmRLC_LB_CNTL);
3289 	if (enable)
3290 		tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3291 	else
3292 		tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK;
3293 	WREG32(mmRLC_LB_CNTL, tmp);
3294 }
3295 
3296 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3297 {
3298 	u32 i, j, k;
3299 	u32 mask;
3300 
3301 	mutex_lock(&adev->grbm_idx_mutex);
3302 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3303 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3304 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
3305 			for (k = 0; k < adev->usec_timeout; k++) {
3306 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3307 					break;
3308 				udelay(1);
3309 			}
3310 		}
3311 	}
3312 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3313 	mutex_unlock(&adev->grbm_idx_mutex);
3314 
3315 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3316 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3317 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3318 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3319 	for (k = 0; k < adev->usec_timeout; k++) {
3320 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3321 			break;
3322 		udelay(1);
3323 	}
3324 }
3325 
3326 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc)
3327 {
3328 	u32 tmp;
3329 
3330 	tmp = RREG32(mmRLC_CNTL);
3331 	if (tmp != rlc)
3332 		WREG32(mmRLC_CNTL, rlc);
3333 }
3334 
3335 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev)
3336 {
3337 	u32 data, orig;
3338 
3339 	orig = data = RREG32(mmRLC_CNTL);
3340 
3341 	if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) {
3342 		u32 i;
3343 
3344 		data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK;
3345 		WREG32(mmRLC_CNTL, data);
3346 
3347 		for (i = 0; i < adev->usec_timeout; i++) {
3348 			if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0)
3349 				break;
3350 			udelay(1);
3351 		}
3352 
3353 		gfx_v7_0_wait_for_rlc_serdes(adev);
3354 	}
3355 
3356 	return orig;
3357 }
3358 
3359 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev)
3360 {
3361 	return true;
3362 }
3363 
3364 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev)
3365 {
3366 	u32 tmp, i, mask;
3367 
3368 	tmp = 0x1 | (1 << 1);
3369 	WREG32(mmRLC_GPR_REG2, tmp);
3370 
3371 	mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK |
3372 		RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK;
3373 	for (i = 0; i < adev->usec_timeout; i++) {
3374 		if ((RREG32(mmRLC_GPM_STAT) & mask) == mask)
3375 			break;
3376 		udelay(1);
3377 	}
3378 
3379 	for (i = 0; i < adev->usec_timeout; i++) {
3380 		if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0)
3381 			break;
3382 		udelay(1);
3383 	}
3384 }
3385 
3386 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev)
3387 {
3388 	u32 tmp;
3389 
3390 	tmp = 0x1 | (0 << 1);
3391 	WREG32(mmRLC_GPR_REG2, tmp);
3392 }
3393 
3394 /**
3395  * gfx_v7_0_rlc_stop - stop the RLC ME
3396  *
3397  * @adev: amdgpu_device pointer
3398  *
3399  * Halt the RLC ME (MicroEngine) (CIK).
3400  */
3401 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev)
3402 {
3403 	WREG32(mmRLC_CNTL, 0);
3404 
3405 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3406 
3407 	gfx_v7_0_wait_for_rlc_serdes(adev);
3408 }
3409 
3410 /**
3411  * gfx_v7_0_rlc_start - start the RLC ME
3412  *
3413  * @adev: amdgpu_device pointer
3414  *
3415  * Unhalt the RLC ME (MicroEngine) (CIK).
3416  */
3417 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev)
3418 {
3419 	WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK);
3420 
3421 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3422 
3423 	udelay(50);
3424 }
3425 
3426 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev)
3427 {
3428 	u32 tmp = RREG32(mmGRBM_SOFT_RESET);
3429 
3430 	tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3431 	WREG32(mmGRBM_SOFT_RESET, tmp);
3432 	udelay(50);
3433 	tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK;
3434 	WREG32(mmGRBM_SOFT_RESET, tmp);
3435 	udelay(50);
3436 }
3437 
3438 /**
3439  * gfx_v7_0_rlc_resume - setup the RLC hw
3440  *
3441  * @adev: amdgpu_device pointer
3442  *
3443  * Initialize the RLC registers, load the ucode,
3444  * and start the RLC (CIK).
3445  * Returns 0 for success, -EINVAL if the ucode is not available.
3446  */
3447 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev)
3448 {
3449 	const struct rlc_firmware_header_v1_0 *hdr;
3450 	const __le32 *fw_data;
3451 	unsigned i, fw_size;
3452 	u32 tmp;
3453 
3454 	if (!adev->gfx.rlc_fw)
3455 		return -EINVAL;
3456 
3457 	hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data;
3458 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3459 	adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version);
3460 	adev->gfx.rlc_feature_version = le32_to_cpu(
3461 					hdr->ucode_feature_version);
3462 
3463 	adev->gfx.rlc.funcs->stop(adev);
3464 
3465 	/* disable CG */
3466 	tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc;
3467 	WREG32(mmRLC_CGCG_CGLS_CTRL, tmp);
3468 
3469 	adev->gfx.rlc.funcs->reset(adev);
3470 
3471 	gfx_v7_0_init_pg(adev);
3472 
3473 	WREG32(mmRLC_LB_CNTR_INIT, 0);
3474 	WREG32(mmRLC_LB_CNTR_MAX, 0x00008000);
3475 
3476 	mutex_lock(&adev->grbm_idx_mutex);
3477 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3478 	WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff);
3479 	WREG32(mmRLC_LB_PARAMS, 0x00600408);
3480 	WREG32(mmRLC_LB_CNTL, 0x80000004);
3481 	mutex_unlock(&adev->grbm_idx_mutex);
3482 
3483 	WREG32(mmRLC_MC_CNTL, 0);
3484 	WREG32(mmRLC_UCODE_CNTL, 0);
3485 
3486 	fw_data = (const __le32 *)
3487 		(adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3488 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3489 	WREG32(mmRLC_GPM_UCODE_ADDR, 0);
3490 	for (i = 0; i < fw_size; i++)
3491 		WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3492 	WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3493 
3494 	/* XXX - find out what chips support lbpw */
3495 	gfx_v7_0_enable_lbpw(adev, false);
3496 
3497 	if (adev->asic_type == CHIP_BONAIRE)
3498 		WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0);
3499 
3500 	adev->gfx.rlc.funcs->start(adev);
3501 
3502 	return 0;
3503 }
3504 
3505 static void gfx_v7_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
3506 {
3507 	u32 data;
3508 
3509 	amdgpu_gfx_off_ctrl(adev, false);
3510 
3511 	data = RREG32(mmRLC_SPM_VMID);
3512 
3513 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
3514 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
3515 
3516 	WREG32(mmRLC_SPM_VMID, data);
3517 
3518 	amdgpu_gfx_off_ctrl(adev, true);
3519 }
3520 
3521 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable)
3522 {
3523 	u32 data, orig, tmp, tmp2;
3524 
3525 	orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
3526 
3527 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
3528 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3529 
3530 		tmp = gfx_v7_0_halt_rlc(adev);
3531 
3532 		mutex_lock(&adev->grbm_idx_mutex);
3533 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3534 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3535 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3536 		tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3537 			RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK |
3538 			RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK;
3539 		WREG32(mmRLC_SERDES_WR_CTRL, tmp2);
3540 		mutex_unlock(&adev->grbm_idx_mutex);
3541 
3542 		gfx_v7_0_update_rlc(adev, tmp);
3543 
3544 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
3545 		if (orig != data)
3546 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3547 
3548 	} else {
3549 		gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3550 
3551 		RREG32(mmCB_CGTT_SCLK_CTRL);
3552 		RREG32(mmCB_CGTT_SCLK_CTRL);
3553 		RREG32(mmCB_CGTT_SCLK_CTRL);
3554 		RREG32(mmCB_CGTT_SCLK_CTRL);
3555 
3556 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
3557 		if (orig != data)
3558 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
3559 
3560 		gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3561 	}
3562 }
3563 
3564 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable)
3565 {
3566 	u32 data, orig, tmp = 0;
3567 
3568 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
3569 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
3570 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
3571 				orig = data = RREG32(mmCP_MEM_SLP_CNTL);
3572 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3573 				if (orig != data)
3574 					WREG32(mmCP_MEM_SLP_CNTL, data);
3575 			}
3576 		}
3577 
3578 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3579 		data |= 0x00000001;
3580 		data &= 0xfffffffd;
3581 		if (orig != data)
3582 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3583 
3584 		tmp = gfx_v7_0_halt_rlc(adev);
3585 
3586 		mutex_lock(&adev->grbm_idx_mutex);
3587 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3588 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3589 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3590 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK |
3591 			RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK;
3592 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3593 		mutex_unlock(&adev->grbm_idx_mutex);
3594 
3595 		gfx_v7_0_update_rlc(adev, tmp);
3596 
3597 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
3598 			orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3599 			data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK;
3600 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
3601 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
3602 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
3603 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
3604 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
3605 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3606 			data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK;
3607 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
3608 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
3609 			if (orig != data)
3610 				WREG32(mmCGTS_SM_CTRL_REG, data);
3611 		}
3612 	} else {
3613 		orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
3614 		data |= 0x00000003;
3615 		if (orig != data)
3616 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
3617 
3618 		data = RREG32(mmRLC_MEM_SLP_CNTL);
3619 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
3620 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
3621 			WREG32(mmRLC_MEM_SLP_CNTL, data);
3622 		}
3623 
3624 		data = RREG32(mmCP_MEM_SLP_CNTL);
3625 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
3626 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
3627 			WREG32(mmCP_MEM_SLP_CNTL, data);
3628 		}
3629 
3630 		orig = data = RREG32(mmCGTS_SM_CTRL_REG);
3631 		data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
3632 		if (orig != data)
3633 			WREG32(mmCGTS_SM_CTRL_REG, data);
3634 
3635 		tmp = gfx_v7_0_halt_rlc(adev);
3636 
3637 		mutex_lock(&adev->grbm_idx_mutex);
3638 		gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3639 		WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
3640 		WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
3641 		data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK;
3642 		WREG32(mmRLC_SERDES_WR_CTRL, data);
3643 		mutex_unlock(&adev->grbm_idx_mutex);
3644 
3645 		gfx_v7_0_update_rlc(adev, tmp);
3646 	}
3647 }
3648 
3649 static void gfx_v7_0_update_cg(struct amdgpu_device *adev,
3650 			       bool enable)
3651 {
3652 	gfx_v7_0_enable_gui_idle_interrupt(adev, false);
3653 	/* order matters! */
3654 	if (enable) {
3655 		gfx_v7_0_enable_mgcg(adev, true);
3656 		gfx_v7_0_enable_cgcg(adev, true);
3657 	} else {
3658 		gfx_v7_0_enable_cgcg(adev, false);
3659 		gfx_v7_0_enable_mgcg(adev, false);
3660 	}
3661 	gfx_v7_0_enable_gui_idle_interrupt(adev, true);
3662 }
3663 
3664 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev,
3665 						bool enable)
3666 {
3667 	u32 data, orig;
3668 
3669 	orig = data = RREG32(mmRLC_PG_CNTL);
3670 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3671 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3672 	else
3673 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK;
3674 	if (orig != data)
3675 		WREG32(mmRLC_PG_CNTL, data);
3676 }
3677 
3678 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev,
3679 						bool enable)
3680 {
3681 	u32 data, orig;
3682 
3683 	orig = data = RREG32(mmRLC_PG_CNTL);
3684 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS))
3685 		data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3686 	else
3687 		data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK;
3688 	if (orig != data)
3689 		WREG32(mmRLC_PG_CNTL, data);
3690 }
3691 
3692 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable)
3693 {
3694 	u32 data, orig;
3695 
3696 	orig = data = RREG32(mmRLC_PG_CNTL);
3697 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP))
3698 		data &= ~0x8000;
3699 	else
3700 		data |= 0x8000;
3701 	if (orig != data)
3702 		WREG32(mmRLC_PG_CNTL, data);
3703 }
3704 
3705 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable)
3706 {
3707 	u32 data, orig;
3708 
3709 	orig = data = RREG32(mmRLC_PG_CNTL);
3710 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS))
3711 		data &= ~0x2000;
3712 	else
3713 		data |= 0x2000;
3714 	if (orig != data)
3715 		WREG32(mmRLC_PG_CNTL, data);
3716 }
3717 
3718 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev)
3719 {
3720 	if (adev->asic_type == CHIP_KAVERI)
3721 		return 5;
3722 	else
3723 		return 4;
3724 }
3725 
3726 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev,
3727 				     bool enable)
3728 {
3729 	u32 data, orig;
3730 
3731 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
3732 		orig = data = RREG32(mmRLC_PG_CNTL);
3733 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3734 		if (orig != data)
3735 			WREG32(mmRLC_PG_CNTL, data);
3736 
3737 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3738 		data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3739 		if (orig != data)
3740 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3741 	} else {
3742 		orig = data = RREG32(mmRLC_PG_CNTL);
3743 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
3744 		if (orig != data)
3745 			WREG32(mmRLC_PG_CNTL, data);
3746 
3747 		orig = data = RREG32(mmRLC_AUTO_PG_CTRL);
3748 		data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK;
3749 		if (orig != data)
3750 			WREG32(mmRLC_AUTO_PG_CTRL, data);
3751 
3752 		data = RREG32(mmDB_RENDER_CONTROL);
3753 	}
3754 }
3755 
3756 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
3757 						 u32 bitmap)
3758 {
3759 	u32 data;
3760 
3761 	if (!bitmap)
3762 		return;
3763 
3764 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3765 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3766 
3767 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
3768 }
3769 
3770 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev)
3771 {
3772 	u32 data, mask;
3773 
3774 	data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG);
3775 	data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
3776 
3777 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
3778 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
3779 
3780 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
3781 
3782 	return (~data) & mask;
3783 }
3784 
3785 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev)
3786 {
3787 	u32 tmp;
3788 
3789 	WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
3790 
3791 	tmp = RREG32(mmRLC_MAX_PG_CU);
3792 	tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK;
3793 	tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT);
3794 	WREG32(mmRLC_MAX_PG_CU, tmp);
3795 }
3796 
3797 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev,
3798 					    bool enable)
3799 {
3800 	u32 data, orig;
3801 
3802 	orig = data = RREG32(mmRLC_PG_CNTL);
3803 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG))
3804 		data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3805 	else
3806 		data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK;
3807 	if (orig != data)
3808 		WREG32(mmRLC_PG_CNTL, data);
3809 }
3810 
3811 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev,
3812 					     bool enable)
3813 {
3814 	u32 data, orig;
3815 
3816 	orig = data = RREG32(mmRLC_PG_CNTL);
3817 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG))
3818 		data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3819 	else
3820 		data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK;
3821 	if (orig != data)
3822 		WREG32(mmRLC_PG_CNTL, data);
3823 }
3824 
3825 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90
3826 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET    0x3D
3827 
3828 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev)
3829 {
3830 	u32 data, orig;
3831 	u32 i;
3832 
3833 	if (adev->gfx.rlc.cs_data) {
3834 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3835 		WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3836 		WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr));
3837 		WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size);
3838 	} else {
3839 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET);
3840 		for (i = 0; i < 3; i++)
3841 			WREG32(mmRLC_GPM_SCRATCH_DATA, 0);
3842 	}
3843 	if (adev->gfx.rlc.reg_list) {
3844 		WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET);
3845 		for (i = 0; i < adev->gfx.rlc.reg_list_size; i++)
3846 			WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]);
3847 	}
3848 
3849 	orig = data = RREG32(mmRLC_PG_CNTL);
3850 	data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK;
3851 	if (orig != data)
3852 		WREG32(mmRLC_PG_CNTL, data);
3853 
3854 	WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8);
3855 	WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
3856 
3857 	data = RREG32(mmCP_RB_WPTR_POLL_CNTL);
3858 	data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
3859 	data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
3860 	WREG32(mmCP_RB_WPTR_POLL_CNTL, data);
3861 
3862 	data = 0x10101010;
3863 	WREG32(mmRLC_PG_DELAY, data);
3864 
3865 	data = RREG32(mmRLC_PG_DELAY_2);
3866 	data &= ~0xff;
3867 	data |= 0x3;
3868 	WREG32(mmRLC_PG_DELAY_2, data);
3869 
3870 	data = RREG32(mmRLC_AUTO_PG_CTRL);
3871 	data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
3872 	data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
3873 	WREG32(mmRLC_AUTO_PG_CTRL, data);
3874 
3875 }
3876 
3877 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable)
3878 {
3879 	gfx_v7_0_enable_gfx_cgpg(adev, enable);
3880 	gfx_v7_0_enable_gfx_static_mgpg(adev, enable);
3881 	gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable);
3882 }
3883 
3884 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev)
3885 {
3886 	u32 count = 0;
3887 	const struct cs_section_def *sect = NULL;
3888 	const struct cs_extent_def *ext = NULL;
3889 
3890 	if (adev->gfx.rlc.cs_data == NULL)
3891 		return 0;
3892 
3893 	/* begin clear state */
3894 	count += 2;
3895 	/* context control state */
3896 	count += 3;
3897 
3898 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3899 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3900 			if (sect->id == SECT_CONTEXT)
3901 				count += 2 + ext->reg_count;
3902 			else
3903 				return 0;
3904 		}
3905 	}
3906 	/* pa_sc_raster_config/pa_sc_raster_config1 */
3907 	count += 4;
3908 	/* end clear state */
3909 	count += 2;
3910 	/* clear state */
3911 	count += 2;
3912 
3913 	return count;
3914 }
3915 
3916 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev,
3917 				    volatile u32 *buffer)
3918 {
3919 	u32 count = 0, i;
3920 	const struct cs_section_def *sect = NULL;
3921 	const struct cs_extent_def *ext = NULL;
3922 
3923 	if (adev->gfx.rlc.cs_data == NULL)
3924 		return;
3925 	if (buffer == NULL)
3926 		return;
3927 
3928 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3929 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3930 
3931 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3932 	buffer[count++] = cpu_to_le32(0x80000000);
3933 	buffer[count++] = cpu_to_le32(0x80000000);
3934 
3935 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
3936 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3937 			if (sect->id == SECT_CONTEXT) {
3938 				buffer[count++] =
3939 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
3940 				buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3941 				for (i = 0; i < ext->reg_count; i++)
3942 					buffer[count++] = cpu_to_le32(ext->extent[i]);
3943 			} else {
3944 				return;
3945 			}
3946 		}
3947 	}
3948 
3949 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
3950 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
3951 	switch (adev->asic_type) {
3952 	case CHIP_BONAIRE:
3953 		buffer[count++] = cpu_to_le32(0x16000012);
3954 		buffer[count++] = cpu_to_le32(0x00000000);
3955 		break;
3956 	case CHIP_KAVERI:
3957 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3958 		buffer[count++] = cpu_to_le32(0x00000000);
3959 		break;
3960 	case CHIP_KABINI:
3961 	case CHIP_MULLINS:
3962 		buffer[count++] = cpu_to_le32(0x00000000); /* XXX */
3963 		buffer[count++] = cpu_to_le32(0x00000000);
3964 		break;
3965 	case CHIP_HAWAII:
3966 		buffer[count++] = cpu_to_le32(0x3a00161a);
3967 		buffer[count++] = cpu_to_le32(0x0000002e);
3968 		break;
3969 	default:
3970 		buffer[count++] = cpu_to_le32(0x00000000);
3971 		buffer[count++] = cpu_to_le32(0x00000000);
3972 		break;
3973 	}
3974 
3975 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3976 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
3977 
3978 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
3979 	buffer[count++] = cpu_to_le32(0);
3980 }
3981 
3982 static void gfx_v7_0_init_pg(struct amdgpu_device *adev)
3983 {
3984 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
3985 			      AMD_PG_SUPPORT_GFX_SMG |
3986 			      AMD_PG_SUPPORT_GFX_DMG |
3987 			      AMD_PG_SUPPORT_CP |
3988 			      AMD_PG_SUPPORT_GDS |
3989 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
3990 		gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true);
3991 		gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true);
3992 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
3993 			gfx_v7_0_init_gfx_cgpg(adev);
3994 			gfx_v7_0_enable_cp_pg(adev, true);
3995 			gfx_v7_0_enable_gds_pg(adev, true);
3996 		}
3997 		gfx_v7_0_init_ao_cu_mask(adev);
3998 		gfx_v7_0_update_gfx_pg(adev, true);
3999 	}
4000 }
4001 
4002 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev)
4003 {
4004 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
4005 			      AMD_PG_SUPPORT_GFX_SMG |
4006 			      AMD_PG_SUPPORT_GFX_DMG |
4007 			      AMD_PG_SUPPORT_CP |
4008 			      AMD_PG_SUPPORT_GDS |
4009 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
4010 		gfx_v7_0_update_gfx_pg(adev, false);
4011 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) {
4012 			gfx_v7_0_enable_cp_pg(adev, false);
4013 			gfx_v7_0_enable_gds_pg(adev, false);
4014 		}
4015 	}
4016 }
4017 
4018 /**
4019  * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot
4020  *
4021  * @adev: amdgpu_device pointer
4022  *
4023  * Fetches a GPU clock counter snapshot (SI).
4024  * Returns the 64 bit clock counter snapshot.
4025  */
4026 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4027 {
4028 	uint64_t clock;
4029 
4030 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4031 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4032 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
4033 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4034 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4035 	return clock;
4036 }
4037 
4038 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4039 					  uint32_t vmid,
4040 					  uint32_t gds_base, uint32_t gds_size,
4041 					  uint32_t gws_base, uint32_t gws_size,
4042 					  uint32_t oa_base, uint32_t oa_size)
4043 {
4044 	/* GDS Base */
4045 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4046 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4047 				WRITE_DATA_DST_SEL(0)));
4048 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
4049 	amdgpu_ring_write(ring, 0);
4050 	amdgpu_ring_write(ring, gds_base);
4051 
4052 	/* GDS Size */
4053 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4054 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4055 				WRITE_DATA_DST_SEL(0)));
4056 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
4057 	amdgpu_ring_write(ring, 0);
4058 	amdgpu_ring_write(ring, gds_size);
4059 
4060 	/* GWS */
4061 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4062 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4063 				WRITE_DATA_DST_SEL(0)));
4064 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
4065 	amdgpu_ring_write(ring, 0);
4066 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4067 
4068 	/* OA */
4069 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4070 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4071 				WRITE_DATA_DST_SEL(0)));
4072 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
4073 	amdgpu_ring_write(ring, 0);
4074 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
4075 }
4076 
4077 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
4078 {
4079 	struct amdgpu_device *adev = ring->adev;
4080 	uint32_t value = 0;
4081 
4082 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
4083 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
4084 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
4085 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
4086 	WREG32(mmSQ_CMD, value);
4087 }
4088 
4089 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
4090 {
4091 	WREG32(mmSQ_IND_INDEX,
4092 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4093 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4094 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
4095 		(SQ_IND_INDEX__FORCE_READ_MASK));
4096 	return RREG32(mmSQ_IND_DATA);
4097 }
4098 
4099 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
4100 			   uint32_t wave, uint32_t thread,
4101 			   uint32_t regno, uint32_t num, uint32_t *out)
4102 {
4103 	WREG32(mmSQ_IND_INDEX,
4104 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
4105 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
4106 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
4107 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
4108 		(SQ_IND_INDEX__FORCE_READ_MASK) |
4109 		(SQ_IND_INDEX__AUTO_INCR_MASK));
4110 	while (num--)
4111 		*(out++) = RREG32(mmSQ_IND_DATA);
4112 }
4113 
4114 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
4115 {
4116 	/* type 0 wave data */
4117 	dst[(*no_fields)++] = 0;
4118 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
4119 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
4120 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
4121 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
4122 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
4123 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
4124 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
4125 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
4126 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
4127 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
4128 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
4129 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
4130 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
4131 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
4132 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
4133 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
4134 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
4135 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
4136 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
4137 }
4138 
4139 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
4140 				     uint32_t wave, uint32_t start,
4141 				     uint32_t size, uint32_t *dst)
4142 {
4143 	wave_read_regs(
4144 		adev, simd, wave, 0,
4145 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
4146 }
4147 
4148 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
4149 				  u32 me, u32 pipe, u32 q, u32 vm)
4150 {
4151 	cik_srbm_select(adev, me, pipe, q, vm);
4152 }
4153 
4154 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
4155 	.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
4156 	.select_se_sh = &gfx_v7_0_select_se_sh,
4157 	.read_wave_data = &gfx_v7_0_read_wave_data,
4158 	.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
4159 	.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q
4160 };
4161 
4162 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {
4163 	.is_rlc_enabled = gfx_v7_0_is_rlc_enabled,
4164 	.set_safe_mode = gfx_v7_0_set_safe_mode,
4165 	.unset_safe_mode = gfx_v7_0_unset_safe_mode,
4166 	.init = gfx_v7_0_rlc_init,
4167 	.get_csb_size = gfx_v7_0_get_csb_size,
4168 	.get_csb_buffer = gfx_v7_0_get_csb_buffer,
4169 	.get_cp_table_num = gfx_v7_0_cp_pg_table_num,
4170 	.resume = gfx_v7_0_rlc_resume,
4171 	.stop = gfx_v7_0_rlc_stop,
4172 	.reset = gfx_v7_0_rlc_reset,
4173 	.start = gfx_v7_0_rlc_start,
4174 	.update_spm_vmid = gfx_v7_0_update_spm_vmid
4175 };
4176 
4177 static int gfx_v7_0_early_init(void *handle)
4178 {
4179 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4180 
4181 	adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS;
4182 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4183 					  AMDGPU_MAX_COMPUTE_RINGS);
4184 	adev->gfx.funcs = &gfx_v7_0_gfx_funcs;
4185 	adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs;
4186 	gfx_v7_0_set_ring_funcs(adev);
4187 	gfx_v7_0_set_irq_funcs(adev);
4188 	gfx_v7_0_set_gds_init(adev);
4189 
4190 	return 0;
4191 }
4192 
4193 static int gfx_v7_0_late_init(void *handle)
4194 {
4195 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4196 	int r;
4197 
4198 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4199 	if (r)
4200 		return r;
4201 
4202 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4203 	if (r)
4204 		return r;
4205 
4206 	return 0;
4207 }
4208 
4209 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev)
4210 {
4211 	u32 gb_addr_config;
4212 	u32 mc_arb_ramcfg;
4213 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
4214 	u32 tmp;
4215 
4216 	switch (adev->asic_type) {
4217 	case CHIP_BONAIRE:
4218 		adev->gfx.config.max_shader_engines = 2;
4219 		adev->gfx.config.max_tile_pipes = 4;
4220 		adev->gfx.config.max_cu_per_sh = 7;
4221 		adev->gfx.config.max_sh_per_se = 1;
4222 		adev->gfx.config.max_backends_per_se = 2;
4223 		adev->gfx.config.max_texture_channel_caches = 4;
4224 		adev->gfx.config.max_gprs = 256;
4225 		adev->gfx.config.max_gs_threads = 32;
4226 		adev->gfx.config.max_hw_contexts = 8;
4227 
4228 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4229 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4230 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4231 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4232 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4233 		break;
4234 	case CHIP_HAWAII:
4235 		adev->gfx.config.max_shader_engines = 4;
4236 		adev->gfx.config.max_tile_pipes = 16;
4237 		adev->gfx.config.max_cu_per_sh = 11;
4238 		adev->gfx.config.max_sh_per_se = 1;
4239 		adev->gfx.config.max_backends_per_se = 4;
4240 		adev->gfx.config.max_texture_channel_caches = 16;
4241 		adev->gfx.config.max_gprs = 256;
4242 		adev->gfx.config.max_gs_threads = 32;
4243 		adev->gfx.config.max_hw_contexts = 8;
4244 
4245 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4246 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4247 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4248 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4249 		gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN;
4250 		break;
4251 	case CHIP_KAVERI:
4252 		adev->gfx.config.max_shader_engines = 1;
4253 		adev->gfx.config.max_tile_pipes = 4;
4254 		adev->gfx.config.max_cu_per_sh = 8;
4255 		adev->gfx.config.max_backends_per_se = 2;
4256 		adev->gfx.config.max_sh_per_se = 1;
4257 		adev->gfx.config.max_texture_channel_caches = 4;
4258 		adev->gfx.config.max_gprs = 256;
4259 		adev->gfx.config.max_gs_threads = 16;
4260 		adev->gfx.config.max_hw_contexts = 8;
4261 
4262 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4263 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4264 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4265 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4266 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4267 		break;
4268 	case CHIP_KABINI:
4269 	case CHIP_MULLINS:
4270 	default:
4271 		adev->gfx.config.max_shader_engines = 1;
4272 		adev->gfx.config.max_tile_pipes = 2;
4273 		adev->gfx.config.max_cu_per_sh = 2;
4274 		adev->gfx.config.max_sh_per_se = 1;
4275 		adev->gfx.config.max_backends_per_se = 1;
4276 		adev->gfx.config.max_texture_channel_caches = 2;
4277 		adev->gfx.config.max_gprs = 256;
4278 		adev->gfx.config.max_gs_threads = 16;
4279 		adev->gfx.config.max_hw_contexts = 8;
4280 
4281 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
4282 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
4283 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
4284 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
4285 		gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN;
4286 		break;
4287 	}
4288 
4289 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
4290 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
4291 
4292 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
4293 				MC_ARB_RAMCFG, NOOFBANK);
4294 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
4295 				MC_ARB_RAMCFG, NOOFRANKS);
4296 
4297 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
4298 	adev->gfx.config.mem_max_burst_length_bytes = 256;
4299 	if (adev->flags & AMD_IS_APU) {
4300 		/* Get memory bank mapping mode. */
4301 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
4302 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4303 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4304 
4305 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
4306 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
4307 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
4308 
4309 		/* Validate settings in case only one DIMM installed. */
4310 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
4311 			dimm00_addr_map = 0;
4312 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
4313 			dimm01_addr_map = 0;
4314 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
4315 			dimm10_addr_map = 0;
4316 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
4317 			dimm11_addr_map = 0;
4318 
4319 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
4320 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
4321 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
4322 			adev->gfx.config.mem_row_size_in_kb = 2;
4323 		else
4324 			adev->gfx.config.mem_row_size_in_kb = 1;
4325 	} else {
4326 		tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT;
4327 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
4328 		if (adev->gfx.config.mem_row_size_in_kb > 4)
4329 			adev->gfx.config.mem_row_size_in_kb = 4;
4330 	}
4331 	/* XXX use MC settings? */
4332 	adev->gfx.config.shader_engine_tile_size = 32;
4333 	adev->gfx.config.num_gpus = 1;
4334 	adev->gfx.config.multi_gpu_tile_size = 64;
4335 
4336 	/* fix up row size */
4337 	gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK;
4338 	switch (adev->gfx.config.mem_row_size_in_kb) {
4339 	case 1:
4340 	default:
4341 		gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4342 		break;
4343 	case 2:
4344 		gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4345 		break;
4346 	case 4:
4347 		gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT);
4348 		break;
4349 	}
4350 	adev->gfx.config.gb_addr_config = gb_addr_config;
4351 }
4352 
4353 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
4354 					int mec, int pipe, int queue)
4355 {
4356 	int r;
4357 	unsigned irq_type;
4358 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
4359 
4360 	/* mec0 is me1 */
4361 	ring->me = mec + 1;
4362 	ring->pipe = pipe;
4363 	ring->queue = queue;
4364 
4365 	ring->ring_obj = NULL;
4366 	ring->use_doorbell = true;
4367 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
4368 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
4369 
4370 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
4371 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
4372 		+ ring->pipe;
4373 
4374 	/* type-2 packets are deprecated on MEC, use type-3 instead */
4375 	r = amdgpu_ring_init(adev, ring, 1024,
4376 			     &adev->gfx.eop_irq, irq_type,
4377 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
4378 	if (r)
4379 		return r;
4380 
4381 
4382 	return 0;
4383 }
4384 
4385 static int gfx_v7_0_sw_init(void *handle)
4386 {
4387 	struct amdgpu_ring *ring;
4388 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4389 	int i, j, k, r, ring_id;
4390 
4391 	switch (adev->asic_type) {
4392 	case CHIP_KAVERI:
4393 		adev->gfx.mec.num_mec = 2;
4394 		break;
4395 	case CHIP_BONAIRE:
4396 	case CHIP_HAWAII:
4397 	case CHIP_KABINI:
4398 	case CHIP_MULLINS:
4399 	default:
4400 		adev->gfx.mec.num_mec = 1;
4401 		break;
4402 	}
4403 	adev->gfx.mec.num_pipe_per_mec = 4;
4404 	adev->gfx.mec.num_queue_per_pipe = 8;
4405 
4406 	/* EOP Event */
4407 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq);
4408 	if (r)
4409 		return r;
4410 
4411 	/* Privileged reg */
4412 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184,
4413 			      &adev->gfx.priv_reg_irq);
4414 	if (r)
4415 		return r;
4416 
4417 	/* Privileged inst */
4418 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185,
4419 			      &adev->gfx.priv_inst_irq);
4420 	if (r)
4421 		return r;
4422 
4423 	r = gfx_v7_0_init_microcode(adev);
4424 	if (r) {
4425 		DRM_ERROR("Failed to load gfx firmware!\n");
4426 		return r;
4427 	}
4428 
4429 	r = adev->gfx.rlc.funcs->init(adev);
4430 	if (r) {
4431 		DRM_ERROR("Failed to init rlc BOs!\n");
4432 		return r;
4433 	}
4434 
4435 	/* allocate mec buffers */
4436 	r = gfx_v7_0_mec_init(adev);
4437 	if (r) {
4438 		DRM_ERROR("Failed to init MEC BOs!\n");
4439 		return r;
4440 	}
4441 
4442 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4443 		ring = &adev->gfx.gfx_ring[i];
4444 		ring->ring_obj = NULL;
4445 		sprintf(ring->name, "gfx");
4446 		r = amdgpu_ring_init(adev, ring, 1024,
4447 				     &adev->gfx.eop_irq,
4448 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
4449 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
4450 		if (r)
4451 			return r;
4452 	}
4453 
4454 	/* set up the compute queues - allocate horizontally across pipes */
4455 	ring_id = 0;
4456 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4457 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4458 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4459 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, 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 };
4981 
4982 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
4983 	.type = AMDGPU_RING_TYPE_GFX,
4984 	.align_mask = 0xff,
4985 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
4986 	.support_64bit_ptrs = false,
4987 	.get_rptr = gfx_v7_0_ring_get_rptr,
4988 	.get_wptr = gfx_v7_0_ring_get_wptr_gfx,
4989 	.set_wptr = gfx_v7_0_ring_set_wptr_gfx,
4990 	.emit_frame_size =
4991 		20 + /* gfx_v7_0_ring_emit_gds_switch */
4992 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
4993 		5 + /* hdp invalidate */
4994 		12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
4995 		7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
4996 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
4997 		3 + 4 + /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
4998 		5, /* SURFACE_SYNC */
4999 	.emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
5000 	.emit_ib = gfx_v7_0_ring_emit_ib_gfx,
5001 	.emit_fence = gfx_v7_0_ring_emit_fence_gfx,
5002 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5003 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5004 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5005 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5006 	.test_ring = gfx_v7_0_ring_test_ring,
5007 	.test_ib = gfx_v7_0_ring_test_ib,
5008 	.insert_nop = amdgpu_ring_insert_nop,
5009 	.pad_ib = amdgpu_ring_generic_pad_ib,
5010 	.emit_cntxcntl = gfx_v7_ring_emit_cntxcntl,
5011 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5012 	.soft_recovery = gfx_v7_0_ring_soft_recovery,
5013 	.emit_mem_sync = gfx_v7_0_emit_mem_sync,
5014 };
5015 
5016 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
5017 	.type = AMDGPU_RING_TYPE_COMPUTE,
5018 	.align_mask = 0xff,
5019 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5020 	.support_64bit_ptrs = false,
5021 	.get_rptr = gfx_v7_0_ring_get_rptr,
5022 	.get_wptr = gfx_v7_0_ring_get_wptr_compute,
5023 	.set_wptr = gfx_v7_0_ring_set_wptr_compute,
5024 	.emit_frame_size =
5025 		20 + /* gfx_v7_0_ring_emit_gds_switch */
5026 		7 + /* gfx_v7_0_ring_emit_hdp_flush */
5027 		5 + /* hdp invalidate */
5028 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
5029 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
5030 		7 + 7 + 7 + /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
5031 		7, /* gfx_v7_0_emit_mem_sync_compute */
5032 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
5033 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
5034 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
5035 	.emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync,
5036 	.emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
5037 	.emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
5038 	.emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
5039 	.test_ring = gfx_v7_0_ring_test_ring,
5040 	.test_ib = gfx_v7_0_ring_test_ib,
5041 	.insert_nop = amdgpu_ring_insert_nop,
5042 	.pad_ib = amdgpu_ring_generic_pad_ib,
5043 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
5044 	.emit_mem_sync = gfx_v7_0_emit_mem_sync_compute,
5045 };
5046 
5047 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)
5048 {
5049 	int i;
5050 
5051 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5052 		adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx;
5053 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5054 		adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute;
5055 }
5056 
5057 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = {
5058 	.set = gfx_v7_0_set_eop_interrupt_state,
5059 	.process = gfx_v7_0_eop_irq,
5060 };
5061 
5062 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = {
5063 	.set = gfx_v7_0_set_priv_reg_fault_state,
5064 	.process = gfx_v7_0_priv_reg_irq,
5065 };
5066 
5067 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = {
5068 	.set = gfx_v7_0_set_priv_inst_fault_state,
5069 	.process = gfx_v7_0_priv_inst_irq,
5070 };
5071 
5072 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev)
5073 {
5074 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5075 	adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs;
5076 
5077 	adev->gfx.priv_reg_irq.num_types = 1;
5078 	adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs;
5079 
5080 	adev->gfx.priv_inst_irq.num_types = 1;
5081 	adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs;
5082 }
5083 
5084 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev)
5085 {
5086 	/* init asci gds info */
5087 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
5088 	adev->gds.gws_size = 64;
5089 	adev->gds.oa_size = 16;
5090 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
5091 }
5092 
5093 
5094 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
5095 {
5096 	int i, j, k, counter, active_cu_number = 0;
5097 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5098 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
5099 	unsigned disable_masks[4 * 2];
5100 	u32 ao_cu_num;
5101 
5102 	if (adev->flags & AMD_IS_APU)
5103 		ao_cu_num = 2;
5104 	else
5105 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
5106 
5107 	memset(cu_info, 0, sizeof(*cu_info));
5108 
5109 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5110 
5111 	mutex_lock(&adev->grbm_idx_mutex);
5112 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5113 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5114 			mask = 1;
5115 			ao_bitmap = 0;
5116 			counter = 0;
5117 			gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff);
5118 			if (i < 4 && j < 2)
5119 				gfx_v7_0_set_user_cu_inactive_bitmap(
5120 					adev, disable_masks[i * 2 + j]);
5121 			bitmap = gfx_v7_0_get_cu_active_bitmap(adev);
5122 			cu_info->bitmap[i][j] = bitmap;
5123 
5124 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
5125 				if (bitmap & mask) {
5126 					if (counter < ao_cu_num)
5127 						ao_bitmap |= mask;
5128 					counter ++;
5129 				}
5130 				mask <<= 1;
5131 			}
5132 			active_cu_number += counter;
5133 			if (i < 2 && j < 2)
5134 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5135 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5136 		}
5137 	}
5138 	gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5139 	mutex_unlock(&adev->grbm_idx_mutex);
5140 
5141 	cu_info->number = active_cu_number;
5142 	cu_info->ao_cu_mask = ao_cu_mask;
5143 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5144 	cu_info->max_waves_per_simd = 10;
5145 	cu_info->max_scratch_slots_per_cu = 32;
5146 	cu_info->wave_front_size = 64;
5147 	cu_info->lds_size = 64;
5148 }
5149 
5150 const struct amdgpu_ip_block_version gfx_v7_1_ip_block =
5151 {
5152 	.type = AMD_IP_BLOCK_TYPE_GFX,
5153 	.major = 7,
5154 	.minor = 1,
5155 	.rev = 0,
5156 	.funcs = &gfx_v7_0_ip_funcs,
5157 };
5158 
5159 const struct amdgpu_ip_block_version gfx_v7_2_ip_block =
5160 {
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 {
5170 	.type = AMD_IP_BLOCK_TYPE_GFX,
5171 	.major = 7,
5172 	.minor = 3,
5173 	.rev = 0,
5174 	.funcs = &gfx_v7_0_ip_funcs,
5175 };
5176