xref: /openbmc/linux/drivers/gpu/drm/radeon/evergreen.c (revision 29d97219)
1 /*
2  * Copyright 2010 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  * Authors: Alex Deucher
23  */
24 
25 #include <linux/firmware.h>
26 #include <linux/pci.h>
27 #include <linux/slab.h>
28 
29 #include <drm/drm_vblank.h>
30 #include <drm/radeon_drm.h>
31 
32 #include "atom.h"
33 #include "avivod.h"
34 #include "cik.h"
35 #include "ni.h"
36 #include "rv770.h"
37 #include "evergreen.h"
38 #include "evergreen_blit_shaders.h"
39 #include "evergreen_reg.h"
40 #include "evergreend.h"
41 #include "radeon.h"
42 #include "radeon_asic.h"
43 #include "radeon_audio.h"
44 #include "radeon_ucode.h"
45 #include "si.h"
46 
47 #define DC_HPDx_CONTROL(x)        (DC_HPD1_CONTROL     + (x * 0xc))
48 #define DC_HPDx_INT_CONTROL(x)    (DC_HPD1_INT_CONTROL + (x * 0xc))
49 #define DC_HPDx_INT_STATUS_REG(x) (DC_HPD1_INT_STATUS  + (x * 0xc))
50 
51 /*
52  * Indirect registers accessor
53  */
54 u32 eg_cg_rreg(struct radeon_device *rdev, u32 reg)
55 {
56 	unsigned long flags;
57 	u32 r;
58 
59 	spin_lock_irqsave(&rdev->cg_idx_lock, flags);
60 	WREG32(EVERGREEN_CG_IND_ADDR, ((reg) & 0xffff));
61 	r = RREG32(EVERGREEN_CG_IND_DATA);
62 	spin_unlock_irqrestore(&rdev->cg_idx_lock, flags);
63 	return r;
64 }
65 
66 void eg_cg_wreg(struct radeon_device *rdev, u32 reg, u32 v)
67 {
68 	unsigned long flags;
69 
70 	spin_lock_irqsave(&rdev->cg_idx_lock, flags);
71 	WREG32(EVERGREEN_CG_IND_ADDR, ((reg) & 0xffff));
72 	WREG32(EVERGREEN_CG_IND_DATA, (v));
73 	spin_unlock_irqrestore(&rdev->cg_idx_lock, flags);
74 }
75 
76 u32 eg_pif_phy0_rreg(struct radeon_device *rdev, u32 reg)
77 {
78 	unsigned long flags;
79 	u32 r;
80 
81 	spin_lock_irqsave(&rdev->pif_idx_lock, flags);
82 	WREG32(EVERGREEN_PIF_PHY0_INDEX, ((reg) & 0xffff));
83 	r = RREG32(EVERGREEN_PIF_PHY0_DATA);
84 	spin_unlock_irqrestore(&rdev->pif_idx_lock, flags);
85 	return r;
86 }
87 
88 void eg_pif_phy0_wreg(struct radeon_device *rdev, u32 reg, u32 v)
89 {
90 	unsigned long flags;
91 
92 	spin_lock_irqsave(&rdev->pif_idx_lock, flags);
93 	WREG32(EVERGREEN_PIF_PHY0_INDEX, ((reg) & 0xffff));
94 	WREG32(EVERGREEN_PIF_PHY0_DATA, (v));
95 	spin_unlock_irqrestore(&rdev->pif_idx_lock, flags);
96 }
97 
98 u32 eg_pif_phy1_rreg(struct radeon_device *rdev, u32 reg)
99 {
100 	unsigned long flags;
101 	u32 r;
102 
103 	spin_lock_irqsave(&rdev->pif_idx_lock, flags);
104 	WREG32(EVERGREEN_PIF_PHY1_INDEX, ((reg) & 0xffff));
105 	r = RREG32(EVERGREEN_PIF_PHY1_DATA);
106 	spin_unlock_irqrestore(&rdev->pif_idx_lock, flags);
107 	return r;
108 }
109 
110 void eg_pif_phy1_wreg(struct radeon_device *rdev, u32 reg, u32 v)
111 {
112 	unsigned long flags;
113 
114 	spin_lock_irqsave(&rdev->pif_idx_lock, flags);
115 	WREG32(EVERGREEN_PIF_PHY1_INDEX, ((reg) & 0xffff));
116 	WREG32(EVERGREEN_PIF_PHY1_DATA, (v));
117 	spin_unlock_irqrestore(&rdev->pif_idx_lock, flags);
118 }
119 
120 static const u32 crtc_offsets[6] =
121 {
122 	EVERGREEN_CRTC0_REGISTER_OFFSET,
123 	EVERGREEN_CRTC1_REGISTER_OFFSET,
124 	EVERGREEN_CRTC2_REGISTER_OFFSET,
125 	EVERGREEN_CRTC3_REGISTER_OFFSET,
126 	EVERGREEN_CRTC4_REGISTER_OFFSET,
127 	EVERGREEN_CRTC5_REGISTER_OFFSET
128 };
129 
130 #include "clearstate_evergreen.h"
131 
132 static const u32 sumo_rlc_save_restore_register_list[] =
133 {
134 	0x98fc,
135 	0x9830,
136 	0x9834,
137 	0x9838,
138 	0x9870,
139 	0x9874,
140 	0x8a14,
141 	0x8b24,
142 	0x8bcc,
143 	0x8b10,
144 	0x8d00,
145 	0x8d04,
146 	0x8c00,
147 	0x8c04,
148 	0x8c08,
149 	0x8c0c,
150 	0x8d8c,
151 	0x8c20,
152 	0x8c24,
153 	0x8c28,
154 	0x8c18,
155 	0x8c1c,
156 	0x8cf0,
157 	0x8e2c,
158 	0x8e38,
159 	0x8c30,
160 	0x9508,
161 	0x9688,
162 	0x9608,
163 	0x960c,
164 	0x9610,
165 	0x9614,
166 	0x88c4,
167 	0x88d4,
168 	0xa008,
169 	0x900c,
170 	0x9100,
171 	0x913c,
172 	0x98f8,
173 	0x98f4,
174 	0x9b7c,
175 	0x3f8c,
176 	0x8950,
177 	0x8954,
178 	0x8a18,
179 	0x8b28,
180 	0x9144,
181 	0x9148,
182 	0x914c,
183 	0x3f90,
184 	0x3f94,
185 	0x915c,
186 	0x9160,
187 	0x9178,
188 	0x917c,
189 	0x9180,
190 	0x918c,
191 	0x9190,
192 	0x9194,
193 	0x9198,
194 	0x919c,
195 	0x91a8,
196 	0x91ac,
197 	0x91b0,
198 	0x91b4,
199 	0x91b8,
200 	0x91c4,
201 	0x91c8,
202 	0x91cc,
203 	0x91d0,
204 	0x91d4,
205 	0x91e0,
206 	0x91e4,
207 	0x91ec,
208 	0x91f0,
209 	0x91f4,
210 	0x9200,
211 	0x9204,
212 	0x929c,
213 	0x9150,
214 	0x802c,
215 };
216 
217 static void evergreen_gpu_init(struct radeon_device *rdev);
218 void evergreen_fini(struct radeon_device *rdev);
219 void evergreen_pcie_gen2_enable(struct radeon_device *rdev);
220 void evergreen_program_aspm(struct radeon_device *rdev);
221 
222 static const u32 evergreen_golden_registers[] =
223 {
224 	0x3f90, 0xffff0000, 0xff000000,
225 	0x9148, 0xffff0000, 0xff000000,
226 	0x3f94, 0xffff0000, 0xff000000,
227 	0x914c, 0xffff0000, 0xff000000,
228 	0x9b7c, 0xffffffff, 0x00000000,
229 	0x8a14, 0xffffffff, 0x00000007,
230 	0x8b10, 0xffffffff, 0x00000000,
231 	0x960c, 0xffffffff, 0x54763210,
232 	0x88c4, 0xffffffff, 0x000000c2,
233 	0x88d4, 0xffffffff, 0x00000010,
234 	0x8974, 0xffffffff, 0x00000000,
235 	0xc78, 0x00000080, 0x00000080,
236 	0x5eb4, 0xffffffff, 0x00000002,
237 	0x5e78, 0xffffffff, 0x001000f0,
238 	0x6104, 0x01000300, 0x00000000,
239 	0x5bc0, 0x00300000, 0x00000000,
240 	0x7030, 0xffffffff, 0x00000011,
241 	0x7c30, 0xffffffff, 0x00000011,
242 	0x10830, 0xffffffff, 0x00000011,
243 	0x11430, 0xffffffff, 0x00000011,
244 	0x12030, 0xffffffff, 0x00000011,
245 	0x12c30, 0xffffffff, 0x00000011,
246 	0xd02c, 0xffffffff, 0x08421000,
247 	0x240c, 0xffffffff, 0x00000380,
248 	0x8b24, 0xffffffff, 0x00ff0fff,
249 	0x28a4c, 0x06000000, 0x06000000,
250 	0x10c, 0x00000001, 0x00000001,
251 	0x8d00, 0xffffffff, 0x100e4848,
252 	0x8d04, 0xffffffff, 0x00164745,
253 	0x8c00, 0xffffffff, 0xe4000003,
254 	0x8c04, 0xffffffff, 0x40600060,
255 	0x8c08, 0xffffffff, 0x001c001c,
256 	0x8cf0, 0xffffffff, 0x08e00620,
257 	0x8c20, 0xffffffff, 0x00800080,
258 	0x8c24, 0xffffffff, 0x00800080,
259 	0x8c18, 0xffffffff, 0x20202078,
260 	0x8c1c, 0xffffffff, 0x00001010,
261 	0x28350, 0xffffffff, 0x00000000,
262 	0xa008, 0xffffffff, 0x00010000,
263 	0x5c4, 0xffffffff, 0x00000001,
264 	0x9508, 0xffffffff, 0x00000002,
265 	0x913c, 0x0000000f, 0x0000000a
266 };
267 
268 static const u32 evergreen_golden_registers2[] =
269 {
270 	0x2f4c, 0xffffffff, 0x00000000,
271 	0x54f4, 0xffffffff, 0x00000000,
272 	0x54f0, 0xffffffff, 0x00000000,
273 	0x5498, 0xffffffff, 0x00000000,
274 	0x549c, 0xffffffff, 0x00000000,
275 	0x5494, 0xffffffff, 0x00000000,
276 	0x53cc, 0xffffffff, 0x00000000,
277 	0x53c8, 0xffffffff, 0x00000000,
278 	0x53c4, 0xffffffff, 0x00000000,
279 	0x53c0, 0xffffffff, 0x00000000,
280 	0x53bc, 0xffffffff, 0x00000000,
281 	0x53b8, 0xffffffff, 0x00000000,
282 	0x53b4, 0xffffffff, 0x00000000,
283 	0x53b0, 0xffffffff, 0x00000000
284 };
285 
286 static const u32 cypress_mgcg_init[] =
287 {
288 	0x802c, 0xffffffff, 0xc0000000,
289 	0x5448, 0xffffffff, 0x00000100,
290 	0x55e4, 0xffffffff, 0x00000100,
291 	0x160c, 0xffffffff, 0x00000100,
292 	0x5644, 0xffffffff, 0x00000100,
293 	0xc164, 0xffffffff, 0x00000100,
294 	0x8a18, 0xffffffff, 0x00000100,
295 	0x897c, 0xffffffff, 0x06000100,
296 	0x8b28, 0xffffffff, 0x00000100,
297 	0x9144, 0xffffffff, 0x00000100,
298 	0x9a60, 0xffffffff, 0x00000100,
299 	0x9868, 0xffffffff, 0x00000100,
300 	0x8d58, 0xffffffff, 0x00000100,
301 	0x9510, 0xffffffff, 0x00000100,
302 	0x949c, 0xffffffff, 0x00000100,
303 	0x9654, 0xffffffff, 0x00000100,
304 	0x9030, 0xffffffff, 0x00000100,
305 	0x9034, 0xffffffff, 0x00000100,
306 	0x9038, 0xffffffff, 0x00000100,
307 	0x903c, 0xffffffff, 0x00000100,
308 	0x9040, 0xffffffff, 0x00000100,
309 	0xa200, 0xffffffff, 0x00000100,
310 	0xa204, 0xffffffff, 0x00000100,
311 	0xa208, 0xffffffff, 0x00000100,
312 	0xa20c, 0xffffffff, 0x00000100,
313 	0x971c, 0xffffffff, 0x00000100,
314 	0x977c, 0xffffffff, 0x00000100,
315 	0x3f80, 0xffffffff, 0x00000100,
316 	0xa210, 0xffffffff, 0x00000100,
317 	0xa214, 0xffffffff, 0x00000100,
318 	0x4d8, 0xffffffff, 0x00000100,
319 	0x9784, 0xffffffff, 0x00000100,
320 	0x9698, 0xffffffff, 0x00000100,
321 	0x4d4, 0xffffffff, 0x00000200,
322 	0x30cc, 0xffffffff, 0x00000100,
323 	0xd0c0, 0xffffffff, 0xff000100,
324 	0x802c, 0xffffffff, 0x40000000,
325 	0x915c, 0xffffffff, 0x00010000,
326 	0x9160, 0xffffffff, 0x00030002,
327 	0x9178, 0xffffffff, 0x00070000,
328 	0x917c, 0xffffffff, 0x00030002,
329 	0x9180, 0xffffffff, 0x00050004,
330 	0x918c, 0xffffffff, 0x00010006,
331 	0x9190, 0xffffffff, 0x00090008,
332 	0x9194, 0xffffffff, 0x00070000,
333 	0x9198, 0xffffffff, 0x00030002,
334 	0x919c, 0xffffffff, 0x00050004,
335 	0x91a8, 0xffffffff, 0x00010006,
336 	0x91ac, 0xffffffff, 0x00090008,
337 	0x91b0, 0xffffffff, 0x00070000,
338 	0x91b4, 0xffffffff, 0x00030002,
339 	0x91b8, 0xffffffff, 0x00050004,
340 	0x91c4, 0xffffffff, 0x00010006,
341 	0x91c8, 0xffffffff, 0x00090008,
342 	0x91cc, 0xffffffff, 0x00070000,
343 	0x91d0, 0xffffffff, 0x00030002,
344 	0x91d4, 0xffffffff, 0x00050004,
345 	0x91e0, 0xffffffff, 0x00010006,
346 	0x91e4, 0xffffffff, 0x00090008,
347 	0x91e8, 0xffffffff, 0x00000000,
348 	0x91ec, 0xffffffff, 0x00070000,
349 	0x91f0, 0xffffffff, 0x00030002,
350 	0x91f4, 0xffffffff, 0x00050004,
351 	0x9200, 0xffffffff, 0x00010006,
352 	0x9204, 0xffffffff, 0x00090008,
353 	0x9208, 0xffffffff, 0x00070000,
354 	0x920c, 0xffffffff, 0x00030002,
355 	0x9210, 0xffffffff, 0x00050004,
356 	0x921c, 0xffffffff, 0x00010006,
357 	0x9220, 0xffffffff, 0x00090008,
358 	0x9224, 0xffffffff, 0x00070000,
359 	0x9228, 0xffffffff, 0x00030002,
360 	0x922c, 0xffffffff, 0x00050004,
361 	0x9238, 0xffffffff, 0x00010006,
362 	0x923c, 0xffffffff, 0x00090008,
363 	0x9240, 0xffffffff, 0x00070000,
364 	0x9244, 0xffffffff, 0x00030002,
365 	0x9248, 0xffffffff, 0x00050004,
366 	0x9254, 0xffffffff, 0x00010006,
367 	0x9258, 0xffffffff, 0x00090008,
368 	0x925c, 0xffffffff, 0x00070000,
369 	0x9260, 0xffffffff, 0x00030002,
370 	0x9264, 0xffffffff, 0x00050004,
371 	0x9270, 0xffffffff, 0x00010006,
372 	0x9274, 0xffffffff, 0x00090008,
373 	0x9278, 0xffffffff, 0x00070000,
374 	0x927c, 0xffffffff, 0x00030002,
375 	0x9280, 0xffffffff, 0x00050004,
376 	0x928c, 0xffffffff, 0x00010006,
377 	0x9290, 0xffffffff, 0x00090008,
378 	0x9294, 0xffffffff, 0x00000000,
379 	0x929c, 0xffffffff, 0x00000001,
380 	0x802c, 0xffffffff, 0x40010000,
381 	0x915c, 0xffffffff, 0x00010000,
382 	0x9160, 0xffffffff, 0x00030002,
383 	0x9178, 0xffffffff, 0x00070000,
384 	0x917c, 0xffffffff, 0x00030002,
385 	0x9180, 0xffffffff, 0x00050004,
386 	0x918c, 0xffffffff, 0x00010006,
387 	0x9190, 0xffffffff, 0x00090008,
388 	0x9194, 0xffffffff, 0x00070000,
389 	0x9198, 0xffffffff, 0x00030002,
390 	0x919c, 0xffffffff, 0x00050004,
391 	0x91a8, 0xffffffff, 0x00010006,
392 	0x91ac, 0xffffffff, 0x00090008,
393 	0x91b0, 0xffffffff, 0x00070000,
394 	0x91b4, 0xffffffff, 0x00030002,
395 	0x91b8, 0xffffffff, 0x00050004,
396 	0x91c4, 0xffffffff, 0x00010006,
397 	0x91c8, 0xffffffff, 0x00090008,
398 	0x91cc, 0xffffffff, 0x00070000,
399 	0x91d0, 0xffffffff, 0x00030002,
400 	0x91d4, 0xffffffff, 0x00050004,
401 	0x91e0, 0xffffffff, 0x00010006,
402 	0x91e4, 0xffffffff, 0x00090008,
403 	0x91e8, 0xffffffff, 0x00000000,
404 	0x91ec, 0xffffffff, 0x00070000,
405 	0x91f0, 0xffffffff, 0x00030002,
406 	0x91f4, 0xffffffff, 0x00050004,
407 	0x9200, 0xffffffff, 0x00010006,
408 	0x9204, 0xffffffff, 0x00090008,
409 	0x9208, 0xffffffff, 0x00070000,
410 	0x920c, 0xffffffff, 0x00030002,
411 	0x9210, 0xffffffff, 0x00050004,
412 	0x921c, 0xffffffff, 0x00010006,
413 	0x9220, 0xffffffff, 0x00090008,
414 	0x9224, 0xffffffff, 0x00070000,
415 	0x9228, 0xffffffff, 0x00030002,
416 	0x922c, 0xffffffff, 0x00050004,
417 	0x9238, 0xffffffff, 0x00010006,
418 	0x923c, 0xffffffff, 0x00090008,
419 	0x9240, 0xffffffff, 0x00070000,
420 	0x9244, 0xffffffff, 0x00030002,
421 	0x9248, 0xffffffff, 0x00050004,
422 	0x9254, 0xffffffff, 0x00010006,
423 	0x9258, 0xffffffff, 0x00090008,
424 	0x925c, 0xffffffff, 0x00070000,
425 	0x9260, 0xffffffff, 0x00030002,
426 	0x9264, 0xffffffff, 0x00050004,
427 	0x9270, 0xffffffff, 0x00010006,
428 	0x9274, 0xffffffff, 0x00090008,
429 	0x9278, 0xffffffff, 0x00070000,
430 	0x927c, 0xffffffff, 0x00030002,
431 	0x9280, 0xffffffff, 0x00050004,
432 	0x928c, 0xffffffff, 0x00010006,
433 	0x9290, 0xffffffff, 0x00090008,
434 	0x9294, 0xffffffff, 0x00000000,
435 	0x929c, 0xffffffff, 0x00000001,
436 	0x802c, 0xffffffff, 0xc0000000
437 };
438 
439 static const u32 redwood_mgcg_init[] =
440 {
441 	0x802c, 0xffffffff, 0xc0000000,
442 	0x5448, 0xffffffff, 0x00000100,
443 	0x55e4, 0xffffffff, 0x00000100,
444 	0x160c, 0xffffffff, 0x00000100,
445 	0x5644, 0xffffffff, 0x00000100,
446 	0xc164, 0xffffffff, 0x00000100,
447 	0x8a18, 0xffffffff, 0x00000100,
448 	0x897c, 0xffffffff, 0x06000100,
449 	0x8b28, 0xffffffff, 0x00000100,
450 	0x9144, 0xffffffff, 0x00000100,
451 	0x9a60, 0xffffffff, 0x00000100,
452 	0x9868, 0xffffffff, 0x00000100,
453 	0x8d58, 0xffffffff, 0x00000100,
454 	0x9510, 0xffffffff, 0x00000100,
455 	0x949c, 0xffffffff, 0x00000100,
456 	0x9654, 0xffffffff, 0x00000100,
457 	0x9030, 0xffffffff, 0x00000100,
458 	0x9034, 0xffffffff, 0x00000100,
459 	0x9038, 0xffffffff, 0x00000100,
460 	0x903c, 0xffffffff, 0x00000100,
461 	0x9040, 0xffffffff, 0x00000100,
462 	0xa200, 0xffffffff, 0x00000100,
463 	0xa204, 0xffffffff, 0x00000100,
464 	0xa208, 0xffffffff, 0x00000100,
465 	0xa20c, 0xffffffff, 0x00000100,
466 	0x971c, 0xffffffff, 0x00000100,
467 	0x977c, 0xffffffff, 0x00000100,
468 	0x3f80, 0xffffffff, 0x00000100,
469 	0xa210, 0xffffffff, 0x00000100,
470 	0xa214, 0xffffffff, 0x00000100,
471 	0x4d8, 0xffffffff, 0x00000100,
472 	0x9784, 0xffffffff, 0x00000100,
473 	0x9698, 0xffffffff, 0x00000100,
474 	0x4d4, 0xffffffff, 0x00000200,
475 	0x30cc, 0xffffffff, 0x00000100,
476 	0xd0c0, 0xffffffff, 0xff000100,
477 	0x802c, 0xffffffff, 0x40000000,
478 	0x915c, 0xffffffff, 0x00010000,
479 	0x9160, 0xffffffff, 0x00030002,
480 	0x9178, 0xffffffff, 0x00070000,
481 	0x917c, 0xffffffff, 0x00030002,
482 	0x9180, 0xffffffff, 0x00050004,
483 	0x918c, 0xffffffff, 0x00010006,
484 	0x9190, 0xffffffff, 0x00090008,
485 	0x9194, 0xffffffff, 0x00070000,
486 	0x9198, 0xffffffff, 0x00030002,
487 	0x919c, 0xffffffff, 0x00050004,
488 	0x91a8, 0xffffffff, 0x00010006,
489 	0x91ac, 0xffffffff, 0x00090008,
490 	0x91b0, 0xffffffff, 0x00070000,
491 	0x91b4, 0xffffffff, 0x00030002,
492 	0x91b8, 0xffffffff, 0x00050004,
493 	0x91c4, 0xffffffff, 0x00010006,
494 	0x91c8, 0xffffffff, 0x00090008,
495 	0x91cc, 0xffffffff, 0x00070000,
496 	0x91d0, 0xffffffff, 0x00030002,
497 	0x91d4, 0xffffffff, 0x00050004,
498 	0x91e0, 0xffffffff, 0x00010006,
499 	0x91e4, 0xffffffff, 0x00090008,
500 	0x91e8, 0xffffffff, 0x00000000,
501 	0x91ec, 0xffffffff, 0x00070000,
502 	0x91f0, 0xffffffff, 0x00030002,
503 	0x91f4, 0xffffffff, 0x00050004,
504 	0x9200, 0xffffffff, 0x00010006,
505 	0x9204, 0xffffffff, 0x00090008,
506 	0x9294, 0xffffffff, 0x00000000,
507 	0x929c, 0xffffffff, 0x00000001,
508 	0x802c, 0xffffffff, 0xc0000000
509 };
510 
511 static const u32 cedar_golden_registers[] =
512 {
513 	0x3f90, 0xffff0000, 0xff000000,
514 	0x9148, 0xffff0000, 0xff000000,
515 	0x3f94, 0xffff0000, 0xff000000,
516 	0x914c, 0xffff0000, 0xff000000,
517 	0x9b7c, 0xffffffff, 0x00000000,
518 	0x8a14, 0xffffffff, 0x00000007,
519 	0x8b10, 0xffffffff, 0x00000000,
520 	0x960c, 0xffffffff, 0x54763210,
521 	0x88c4, 0xffffffff, 0x000000c2,
522 	0x88d4, 0xffffffff, 0x00000000,
523 	0x8974, 0xffffffff, 0x00000000,
524 	0xc78, 0x00000080, 0x00000080,
525 	0x5eb4, 0xffffffff, 0x00000002,
526 	0x5e78, 0xffffffff, 0x001000f0,
527 	0x6104, 0x01000300, 0x00000000,
528 	0x5bc0, 0x00300000, 0x00000000,
529 	0x7030, 0xffffffff, 0x00000011,
530 	0x7c30, 0xffffffff, 0x00000011,
531 	0x10830, 0xffffffff, 0x00000011,
532 	0x11430, 0xffffffff, 0x00000011,
533 	0xd02c, 0xffffffff, 0x08421000,
534 	0x240c, 0xffffffff, 0x00000380,
535 	0x8b24, 0xffffffff, 0x00ff0fff,
536 	0x28a4c, 0x06000000, 0x06000000,
537 	0x10c, 0x00000001, 0x00000001,
538 	0x8d00, 0xffffffff, 0x100e4848,
539 	0x8d04, 0xffffffff, 0x00164745,
540 	0x8c00, 0xffffffff, 0xe4000003,
541 	0x8c04, 0xffffffff, 0x40600060,
542 	0x8c08, 0xffffffff, 0x001c001c,
543 	0x8cf0, 0xffffffff, 0x08e00410,
544 	0x8c20, 0xffffffff, 0x00800080,
545 	0x8c24, 0xffffffff, 0x00800080,
546 	0x8c18, 0xffffffff, 0x20202078,
547 	0x8c1c, 0xffffffff, 0x00001010,
548 	0x28350, 0xffffffff, 0x00000000,
549 	0xa008, 0xffffffff, 0x00010000,
550 	0x5c4, 0xffffffff, 0x00000001,
551 	0x9508, 0xffffffff, 0x00000002
552 };
553 
554 static const u32 cedar_mgcg_init[] =
555 {
556 	0x802c, 0xffffffff, 0xc0000000,
557 	0x5448, 0xffffffff, 0x00000100,
558 	0x55e4, 0xffffffff, 0x00000100,
559 	0x160c, 0xffffffff, 0x00000100,
560 	0x5644, 0xffffffff, 0x00000100,
561 	0xc164, 0xffffffff, 0x00000100,
562 	0x8a18, 0xffffffff, 0x00000100,
563 	0x897c, 0xffffffff, 0x06000100,
564 	0x8b28, 0xffffffff, 0x00000100,
565 	0x9144, 0xffffffff, 0x00000100,
566 	0x9a60, 0xffffffff, 0x00000100,
567 	0x9868, 0xffffffff, 0x00000100,
568 	0x8d58, 0xffffffff, 0x00000100,
569 	0x9510, 0xffffffff, 0x00000100,
570 	0x949c, 0xffffffff, 0x00000100,
571 	0x9654, 0xffffffff, 0x00000100,
572 	0x9030, 0xffffffff, 0x00000100,
573 	0x9034, 0xffffffff, 0x00000100,
574 	0x9038, 0xffffffff, 0x00000100,
575 	0x903c, 0xffffffff, 0x00000100,
576 	0x9040, 0xffffffff, 0x00000100,
577 	0xa200, 0xffffffff, 0x00000100,
578 	0xa204, 0xffffffff, 0x00000100,
579 	0xa208, 0xffffffff, 0x00000100,
580 	0xa20c, 0xffffffff, 0x00000100,
581 	0x971c, 0xffffffff, 0x00000100,
582 	0x977c, 0xffffffff, 0x00000100,
583 	0x3f80, 0xffffffff, 0x00000100,
584 	0xa210, 0xffffffff, 0x00000100,
585 	0xa214, 0xffffffff, 0x00000100,
586 	0x4d8, 0xffffffff, 0x00000100,
587 	0x9784, 0xffffffff, 0x00000100,
588 	0x9698, 0xffffffff, 0x00000100,
589 	0x4d4, 0xffffffff, 0x00000200,
590 	0x30cc, 0xffffffff, 0x00000100,
591 	0xd0c0, 0xffffffff, 0xff000100,
592 	0x802c, 0xffffffff, 0x40000000,
593 	0x915c, 0xffffffff, 0x00010000,
594 	0x9178, 0xffffffff, 0x00050000,
595 	0x917c, 0xffffffff, 0x00030002,
596 	0x918c, 0xffffffff, 0x00010004,
597 	0x9190, 0xffffffff, 0x00070006,
598 	0x9194, 0xffffffff, 0x00050000,
599 	0x9198, 0xffffffff, 0x00030002,
600 	0x91a8, 0xffffffff, 0x00010004,
601 	0x91ac, 0xffffffff, 0x00070006,
602 	0x91e8, 0xffffffff, 0x00000000,
603 	0x9294, 0xffffffff, 0x00000000,
604 	0x929c, 0xffffffff, 0x00000001,
605 	0x802c, 0xffffffff, 0xc0000000
606 };
607 
608 static const u32 juniper_mgcg_init[] =
609 {
610 	0x802c, 0xffffffff, 0xc0000000,
611 	0x5448, 0xffffffff, 0x00000100,
612 	0x55e4, 0xffffffff, 0x00000100,
613 	0x160c, 0xffffffff, 0x00000100,
614 	0x5644, 0xffffffff, 0x00000100,
615 	0xc164, 0xffffffff, 0x00000100,
616 	0x8a18, 0xffffffff, 0x00000100,
617 	0x897c, 0xffffffff, 0x06000100,
618 	0x8b28, 0xffffffff, 0x00000100,
619 	0x9144, 0xffffffff, 0x00000100,
620 	0x9a60, 0xffffffff, 0x00000100,
621 	0x9868, 0xffffffff, 0x00000100,
622 	0x8d58, 0xffffffff, 0x00000100,
623 	0x9510, 0xffffffff, 0x00000100,
624 	0x949c, 0xffffffff, 0x00000100,
625 	0x9654, 0xffffffff, 0x00000100,
626 	0x9030, 0xffffffff, 0x00000100,
627 	0x9034, 0xffffffff, 0x00000100,
628 	0x9038, 0xffffffff, 0x00000100,
629 	0x903c, 0xffffffff, 0x00000100,
630 	0x9040, 0xffffffff, 0x00000100,
631 	0xa200, 0xffffffff, 0x00000100,
632 	0xa204, 0xffffffff, 0x00000100,
633 	0xa208, 0xffffffff, 0x00000100,
634 	0xa20c, 0xffffffff, 0x00000100,
635 	0x971c, 0xffffffff, 0x00000100,
636 	0xd0c0, 0xffffffff, 0xff000100,
637 	0x802c, 0xffffffff, 0x40000000,
638 	0x915c, 0xffffffff, 0x00010000,
639 	0x9160, 0xffffffff, 0x00030002,
640 	0x9178, 0xffffffff, 0x00070000,
641 	0x917c, 0xffffffff, 0x00030002,
642 	0x9180, 0xffffffff, 0x00050004,
643 	0x918c, 0xffffffff, 0x00010006,
644 	0x9190, 0xffffffff, 0x00090008,
645 	0x9194, 0xffffffff, 0x00070000,
646 	0x9198, 0xffffffff, 0x00030002,
647 	0x919c, 0xffffffff, 0x00050004,
648 	0x91a8, 0xffffffff, 0x00010006,
649 	0x91ac, 0xffffffff, 0x00090008,
650 	0x91b0, 0xffffffff, 0x00070000,
651 	0x91b4, 0xffffffff, 0x00030002,
652 	0x91b8, 0xffffffff, 0x00050004,
653 	0x91c4, 0xffffffff, 0x00010006,
654 	0x91c8, 0xffffffff, 0x00090008,
655 	0x91cc, 0xffffffff, 0x00070000,
656 	0x91d0, 0xffffffff, 0x00030002,
657 	0x91d4, 0xffffffff, 0x00050004,
658 	0x91e0, 0xffffffff, 0x00010006,
659 	0x91e4, 0xffffffff, 0x00090008,
660 	0x91e8, 0xffffffff, 0x00000000,
661 	0x91ec, 0xffffffff, 0x00070000,
662 	0x91f0, 0xffffffff, 0x00030002,
663 	0x91f4, 0xffffffff, 0x00050004,
664 	0x9200, 0xffffffff, 0x00010006,
665 	0x9204, 0xffffffff, 0x00090008,
666 	0x9208, 0xffffffff, 0x00070000,
667 	0x920c, 0xffffffff, 0x00030002,
668 	0x9210, 0xffffffff, 0x00050004,
669 	0x921c, 0xffffffff, 0x00010006,
670 	0x9220, 0xffffffff, 0x00090008,
671 	0x9224, 0xffffffff, 0x00070000,
672 	0x9228, 0xffffffff, 0x00030002,
673 	0x922c, 0xffffffff, 0x00050004,
674 	0x9238, 0xffffffff, 0x00010006,
675 	0x923c, 0xffffffff, 0x00090008,
676 	0x9240, 0xffffffff, 0x00070000,
677 	0x9244, 0xffffffff, 0x00030002,
678 	0x9248, 0xffffffff, 0x00050004,
679 	0x9254, 0xffffffff, 0x00010006,
680 	0x9258, 0xffffffff, 0x00090008,
681 	0x925c, 0xffffffff, 0x00070000,
682 	0x9260, 0xffffffff, 0x00030002,
683 	0x9264, 0xffffffff, 0x00050004,
684 	0x9270, 0xffffffff, 0x00010006,
685 	0x9274, 0xffffffff, 0x00090008,
686 	0x9278, 0xffffffff, 0x00070000,
687 	0x927c, 0xffffffff, 0x00030002,
688 	0x9280, 0xffffffff, 0x00050004,
689 	0x928c, 0xffffffff, 0x00010006,
690 	0x9290, 0xffffffff, 0x00090008,
691 	0x9294, 0xffffffff, 0x00000000,
692 	0x929c, 0xffffffff, 0x00000001,
693 	0x802c, 0xffffffff, 0xc0000000,
694 	0x977c, 0xffffffff, 0x00000100,
695 	0x3f80, 0xffffffff, 0x00000100,
696 	0xa210, 0xffffffff, 0x00000100,
697 	0xa214, 0xffffffff, 0x00000100,
698 	0x4d8, 0xffffffff, 0x00000100,
699 	0x9784, 0xffffffff, 0x00000100,
700 	0x9698, 0xffffffff, 0x00000100,
701 	0x4d4, 0xffffffff, 0x00000200,
702 	0x30cc, 0xffffffff, 0x00000100,
703 	0x802c, 0xffffffff, 0xc0000000
704 };
705 
706 static const u32 supersumo_golden_registers[] =
707 {
708 	0x5eb4, 0xffffffff, 0x00000002,
709 	0x5c4, 0xffffffff, 0x00000001,
710 	0x7030, 0xffffffff, 0x00000011,
711 	0x7c30, 0xffffffff, 0x00000011,
712 	0x6104, 0x01000300, 0x00000000,
713 	0x5bc0, 0x00300000, 0x00000000,
714 	0x8c04, 0xffffffff, 0x40600060,
715 	0x8c08, 0xffffffff, 0x001c001c,
716 	0x8c20, 0xffffffff, 0x00800080,
717 	0x8c24, 0xffffffff, 0x00800080,
718 	0x8c18, 0xffffffff, 0x20202078,
719 	0x8c1c, 0xffffffff, 0x00001010,
720 	0x918c, 0xffffffff, 0x00010006,
721 	0x91a8, 0xffffffff, 0x00010006,
722 	0x91c4, 0xffffffff, 0x00010006,
723 	0x91e0, 0xffffffff, 0x00010006,
724 	0x9200, 0xffffffff, 0x00010006,
725 	0x9150, 0xffffffff, 0x6e944040,
726 	0x917c, 0xffffffff, 0x00030002,
727 	0x9180, 0xffffffff, 0x00050004,
728 	0x9198, 0xffffffff, 0x00030002,
729 	0x919c, 0xffffffff, 0x00050004,
730 	0x91b4, 0xffffffff, 0x00030002,
731 	0x91b8, 0xffffffff, 0x00050004,
732 	0x91d0, 0xffffffff, 0x00030002,
733 	0x91d4, 0xffffffff, 0x00050004,
734 	0x91f0, 0xffffffff, 0x00030002,
735 	0x91f4, 0xffffffff, 0x00050004,
736 	0x915c, 0xffffffff, 0x00010000,
737 	0x9160, 0xffffffff, 0x00030002,
738 	0x3f90, 0xffff0000, 0xff000000,
739 	0x9178, 0xffffffff, 0x00070000,
740 	0x9194, 0xffffffff, 0x00070000,
741 	0x91b0, 0xffffffff, 0x00070000,
742 	0x91cc, 0xffffffff, 0x00070000,
743 	0x91ec, 0xffffffff, 0x00070000,
744 	0x9148, 0xffff0000, 0xff000000,
745 	0x9190, 0xffffffff, 0x00090008,
746 	0x91ac, 0xffffffff, 0x00090008,
747 	0x91c8, 0xffffffff, 0x00090008,
748 	0x91e4, 0xffffffff, 0x00090008,
749 	0x9204, 0xffffffff, 0x00090008,
750 	0x3f94, 0xffff0000, 0xff000000,
751 	0x914c, 0xffff0000, 0xff000000,
752 	0x929c, 0xffffffff, 0x00000001,
753 	0x8a18, 0xffffffff, 0x00000100,
754 	0x8b28, 0xffffffff, 0x00000100,
755 	0x9144, 0xffffffff, 0x00000100,
756 	0x5644, 0xffffffff, 0x00000100,
757 	0x9b7c, 0xffffffff, 0x00000000,
758 	0x8030, 0xffffffff, 0x0000100a,
759 	0x8a14, 0xffffffff, 0x00000007,
760 	0x8b24, 0xffffffff, 0x00ff0fff,
761 	0x8b10, 0xffffffff, 0x00000000,
762 	0x28a4c, 0x06000000, 0x06000000,
763 	0x4d8, 0xffffffff, 0x00000100,
764 	0x913c, 0xffff000f, 0x0100000a,
765 	0x960c, 0xffffffff, 0x54763210,
766 	0x88c4, 0xffffffff, 0x000000c2,
767 	0x88d4, 0xffffffff, 0x00000010,
768 	0x8974, 0xffffffff, 0x00000000,
769 	0xc78, 0x00000080, 0x00000080,
770 	0x5e78, 0xffffffff, 0x001000f0,
771 	0xd02c, 0xffffffff, 0x08421000,
772 	0xa008, 0xffffffff, 0x00010000,
773 	0x8d00, 0xffffffff, 0x100e4848,
774 	0x8d04, 0xffffffff, 0x00164745,
775 	0x8c00, 0xffffffff, 0xe4000003,
776 	0x8cf0, 0x1fffffff, 0x08e00620,
777 	0x28350, 0xffffffff, 0x00000000,
778 	0x9508, 0xffffffff, 0x00000002
779 };
780 
781 static const u32 sumo_golden_registers[] =
782 {
783 	0x900c, 0x00ffffff, 0x0017071f,
784 	0x8c18, 0xffffffff, 0x10101060,
785 	0x8c1c, 0xffffffff, 0x00001010,
786 	0x8c30, 0x0000000f, 0x00000005,
787 	0x9688, 0x0000000f, 0x00000007
788 };
789 
790 static const u32 wrestler_golden_registers[] =
791 {
792 	0x5eb4, 0xffffffff, 0x00000002,
793 	0x5c4, 0xffffffff, 0x00000001,
794 	0x7030, 0xffffffff, 0x00000011,
795 	0x7c30, 0xffffffff, 0x00000011,
796 	0x6104, 0x01000300, 0x00000000,
797 	0x5bc0, 0x00300000, 0x00000000,
798 	0x918c, 0xffffffff, 0x00010006,
799 	0x91a8, 0xffffffff, 0x00010006,
800 	0x9150, 0xffffffff, 0x6e944040,
801 	0x917c, 0xffffffff, 0x00030002,
802 	0x9198, 0xffffffff, 0x00030002,
803 	0x915c, 0xffffffff, 0x00010000,
804 	0x3f90, 0xffff0000, 0xff000000,
805 	0x9178, 0xffffffff, 0x00070000,
806 	0x9194, 0xffffffff, 0x00070000,
807 	0x9148, 0xffff0000, 0xff000000,
808 	0x9190, 0xffffffff, 0x00090008,
809 	0x91ac, 0xffffffff, 0x00090008,
810 	0x3f94, 0xffff0000, 0xff000000,
811 	0x914c, 0xffff0000, 0xff000000,
812 	0x929c, 0xffffffff, 0x00000001,
813 	0x8a18, 0xffffffff, 0x00000100,
814 	0x8b28, 0xffffffff, 0x00000100,
815 	0x9144, 0xffffffff, 0x00000100,
816 	0x9b7c, 0xffffffff, 0x00000000,
817 	0x8030, 0xffffffff, 0x0000100a,
818 	0x8a14, 0xffffffff, 0x00000001,
819 	0x8b24, 0xffffffff, 0x00ff0fff,
820 	0x8b10, 0xffffffff, 0x00000000,
821 	0x28a4c, 0x06000000, 0x06000000,
822 	0x4d8, 0xffffffff, 0x00000100,
823 	0x913c, 0xffff000f, 0x0100000a,
824 	0x960c, 0xffffffff, 0x54763210,
825 	0x88c4, 0xffffffff, 0x000000c2,
826 	0x88d4, 0xffffffff, 0x00000010,
827 	0x8974, 0xffffffff, 0x00000000,
828 	0xc78, 0x00000080, 0x00000080,
829 	0x5e78, 0xffffffff, 0x001000f0,
830 	0xd02c, 0xffffffff, 0x08421000,
831 	0xa008, 0xffffffff, 0x00010000,
832 	0x8d00, 0xffffffff, 0x100e4848,
833 	0x8d04, 0xffffffff, 0x00164745,
834 	0x8c00, 0xffffffff, 0xe4000003,
835 	0x8cf0, 0x1fffffff, 0x08e00410,
836 	0x28350, 0xffffffff, 0x00000000,
837 	0x9508, 0xffffffff, 0x00000002,
838 	0x900c, 0xffffffff, 0x0017071f,
839 	0x8c18, 0xffffffff, 0x10101060,
840 	0x8c1c, 0xffffffff, 0x00001010
841 };
842 
843 static const u32 barts_golden_registers[] =
844 {
845 	0x5eb4, 0xffffffff, 0x00000002,
846 	0x5e78, 0x8f311ff1, 0x001000f0,
847 	0x3f90, 0xffff0000, 0xff000000,
848 	0x9148, 0xffff0000, 0xff000000,
849 	0x3f94, 0xffff0000, 0xff000000,
850 	0x914c, 0xffff0000, 0xff000000,
851 	0xc78, 0x00000080, 0x00000080,
852 	0xbd4, 0x70073777, 0x00010001,
853 	0xd02c, 0xbfffff1f, 0x08421000,
854 	0xd0b8, 0x03773777, 0x02011003,
855 	0x5bc0, 0x00200000, 0x50100000,
856 	0x98f8, 0x33773777, 0x02011003,
857 	0x98fc, 0xffffffff, 0x76543210,
858 	0x7030, 0x31000311, 0x00000011,
859 	0x2f48, 0x00000007, 0x02011003,
860 	0x6b28, 0x00000010, 0x00000012,
861 	0x7728, 0x00000010, 0x00000012,
862 	0x10328, 0x00000010, 0x00000012,
863 	0x10f28, 0x00000010, 0x00000012,
864 	0x11b28, 0x00000010, 0x00000012,
865 	0x12728, 0x00000010, 0x00000012,
866 	0x240c, 0x000007ff, 0x00000380,
867 	0x8a14, 0xf000001f, 0x00000007,
868 	0x8b24, 0x3fff3fff, 0x00ff0fff,
869 	0x8b10, 0x0000ff0f, 0x00000000,
870 	0x28a4c, 0x07ffffff, 0x06000000,
871 	0x10c, 0x00000001, 0x00010003,
872 	0xa02c, 0xffffffff, 0x0000009b,
873 	0x913c, 0x0000000f, 0x0100000a,
874 	0x8d00, 0xffff7f7f, 0x100e4848,
875 	0x8d04, 0x00ffffff, 0x00164745,
876 	0x8c00, 0xfffc0003, 0xe4000003,
877 	0x8c04, 0xf8ff00ff, 0x40600060,
878 	0x8c08, 0x00ff00ff, 0x001c001c,
879 	0x8cf0, 0x1fff1fff, 0x08e00620,
880 	0x8c20, 0x0fff0fff, 0x00800080,
881 	0x8c24, 0x0fff0fff, 0x00800080,
882 	0x8c18, 0xffffffff, 0x20202078,
883 	0x8c1c, 0x0000ffff, 0x00001010,
884 	0x28350, 0x00000f01, 0x00000000,
885 	0x9508, 0x3700001f, 0x00000002,
886 	0x960c, 0xffffffff, 0x54763210,
887 	0x88c4, 0x001f3ae3, 0x000000c2,
888 	0x88d4, 0x0000001f, 0x00000010,
889 	0x8974, 0xffffffff, 0x00000000
890 };
891 
892 static const u32 turks_golden_registers[] =
893 {
894 	0x5eb4, 0xffffffff, 0x00000002,
895 	0x5e78, 0x8f311ff1, 0x001000f0,
896 	0x8c8, 0x00003000, 0x00001070,
897 	0x8cc, 0x000fffff, 0x00040035,
898 	0x3f90, 0xffff0000, 0xfff00000,
899 	0x9148, 0xffff0000, 0xfff00000,
900 	0x3f94, 0xffff0000, 0xfff00000,
901 	0x914c, 0xffff0000, 0xfff00000,
902 	0xc78, 0x00000080, 0x00000080,
903 	0xbd4, 0x00073007, 0x00010002,
904 	0xd02c, 0xbfffff1f, 0x08421000,
905 	0xd0b8, 0x03773777, 0x02010002,
906 	0x5bc0, 0x00200000, 0x50100000,
907 	0x98f8, 0x33773777, 0x00010002,
908 	0x98fc, 0xffffffff, 0x33221100,
909 	0x7030, 0x31000311, 0x00000011,
910 	0x2f48, 0x33773777, 0x00010002,
911 	0x6b28, 0x00000010, 0x00000012,
912 	0x7728, 0x00000010, 0x00000012,
913 	0x10328, 0x00000010, 0x00000012,
914 	0x10f28, 0x00000010, 0x00000012,
915 	0x11b28, 0x00000010, 0x00000012,
916 	0x12728, 0x00000010, 0x00000012,
917 	0x240c, 0x000007ff, 0x00000380,
918 	0x8a14, 0xf000001f, 0x00000007,
919 	0x8b24, 0x3fff3fff, 0x00ff0fff,
920 	0x8b10, 0x0000ff0f, 0x00000000,
921 	0x28a4c, 0x07ffffff, 0x06000000,
922 	0x10c, 0x00000001, 0x00010003,
923 	0xa02c, 0xffffffff, 0x0000009b,
924 	0x913c, 0x0000000f, 0x0100000a,
925 	0x8d00, 0xffff7f7f, 0x100e4848,
926 	0x8d04, 0x00ffffff, 0x00164745,
927 	0x8c00, 0xfffc0003, 0xe4000003,
928 	0x8c04, 0xf8ff00ff, 0x40600060,
929 	0x8c08, 0x00ff00ff, 0x001c001c,
930 	0x8cf0, 0x1fff1fff, 0x08e00410,
931 	0x8c20, 0x0fff0fff, 0x00800080,
932 	0x8c24, 0x0fff0fff, 0x00800080,
933 	0x8c18, 0xffffffff, 0x20202078,
934 	0x8c1c, 0x0000ffff, 0x00001010,
935 	0x28350, 0x00000f01, 0x00000000,
936 	0x9508, 0x3700001f, 0x00000002,
937 	0x960c, 0xffffffff, 0x54763210,
938 	0x88c4, 0x001f3ae3, 0x000000c2,
939 	0x88d4, 0x0000001f, 0x00000010,
940 	0x8974, 0xffffffff, 0x00000000
941 };
942 
943 static const u32 caicos_golden_registers[] =
944 {
945 	0x5eb4, 0xffffffff, 0x00000002,
946 	0x5e78, 0x8f311ff1, 0x001000f0,
947 	0x8c8, 0x00003420, 0x00001450,
948 	0x8cc, 0x000fffff, 0x00040035,
949 	0x3f90, 0xffff0000, 0xfffc0000,
950 	0x9148, 0xffff0000, 0xfffc0000,
951 	0x3f94, 0xffff0000, 0xfffc0000,
952 	0x914c, 0xffff0000, 0xfffc0000,
953 	0xc78, 0x00000080, 0x00000080,
954 	0xbd4, 0x00073007, 0x00010001,
955 	0xd02c, 0xbfffff1f, 0x08421000,
956 	0xd0b8, 0x03773777, 0x02010001,
957 	0x5bc0, 0x00200000, 0x50100000,
958 	0x98f8, 0x33773777, 0x02010001,
959 	0x98fc, 0xffffffff, 0x33221100,
960 	0x7030, 0x31000311, 0x00000011,
961 	0x2f48, 0x33773777, 0x02010001,
962 	0x6b28, 0x00000010, 0x00000012,
963 	0x7728, 0x00000010, 0x00000012,
964 	0x10328, 0x00000010, 0x00000012,
965 	0x10f28, 0x00000010, 0x00000012,
966 	0x11b28, 0x00000010, 0x00000012,
967 	0x12728, 0x00000010, 0x00000012,
968 	0x240c, 0x000007ff, 0x00000380,
969 	0x8a14, 0xf000001f, 0x00000001,
970 	0x8b24, 0x3fff3fff, 0x00ff0fff,
971 	0x8b10, 0x0000ff0f, 0x00000000,
972 	0x28a4c, 0x07ffffff, 0x06000000,
973 	0x10c, 0x00000001, 0x00010003,
974 	0xa02c, 0xffffffff, 0x0000009b,
975 	0x913c, 0x0000000f, 0x0100000a,
976 	0x8d00, 0xffff7f7f, 0x100e4848,
977 	0x8d04, 0x00ffffff, 0x00164745,
978 	0x8c00, 0xfffc0003, 0xe4000003,
979 	0x8c04, 0xf8ff00ff, 0x40600060,
980 	0x8c08, 0x00ff00ff, 0x001c001c,
981 	0x8cf0, 0x1fff1fff, 0x08e00410,
982 	0x8c20, 0x0fff0fff, 0x00800080,
983 	0x8c24, 0x0fff0fff, 0x00800080,
984 	0x8c18, 0xffffffff, 0x20202078,
985 	0x8c1c, 0x0000ffff, 0x00001010,
986 	0x28350, 0x00000f01, 0x00000000,
987 	0x9508, 0x3700001f, 0x00000002,
988 	0x960c, 0xffffffff, 0x54763210,
989 	0x88c4, 0x001f3ae3, 0x000000c2,
990 	0x88d4, 0x0000001f, 0x00000010,
991 	0x8974, 0xffffffff, 0x00000000
992 };
993 
994 static void evergreen_init_golden_registers(struct radeon_device *rdev)
995 {
996 	switch (rdev->family) {
997 	case CHIP_CYPRESS:
998 	case CHIP_HEMLOCK:
999 		radeon_program_register_sequence(rdev,
1000 						 evergreen_golden_registers,
1001 						 (const u32)ARRAY_SIZE(evergreen_golden_registers));
1002 		radeon_program_register_sequence(rdev,
1003 						 evergreen_golden_registers2,
1004 						 (const u32)ARRAY_SIZE(evergreen_golden_registers2));
1005 		radeon_program_register_sequence(rdev,
1006 						 cypress_mgcg_init,
1007 						 (const u32)ARRAY_SIZE(cypress_mgcg_init));
1008 		break;
1009 	case CHIP_JUNIPER:
1010 		radeon_program_register_sequence(rdev,
1011 						 evergreen_golden_registers,
1012 						 (const u32)ARRAY_SIZE(evergreen_golden_registers));
1013 		radeon_program_register_sequence(rdev,
1014 						 evergreen_golden_registers2,
1015 						 (const u32)ARRAY_SIZE(evergreen_golden_registers2));
1016 		radeon_program_register_sequence(rdev,
1017 						 juniper_mgcg_init,
1018 						 (const u32)ARRAY_SIZE(juniper_mgcg_init));
1019 		break;
1020 	case CHIP_REDWOOD:
1021 		radeon_program_register_sequence(rdev,
1022 						 evergreen_golden_registers,
1023 						 (const u32)ARRAY_SIZE(evergreen_golden_registers));
1024 		radeon_program_register_sequence(rdev,
1025 						 evergreen_golden_registers2,
1026 						 (const u32)ARRAY_SIZE(evergreen_golden_registers2));
1027 		radeon_program_register_sequence(rdev,
1028 						 redwood_mgcg_init,
1029 						 (const u32)ARRAY_SIZE(redwood_mgcg_init));
1030 		break;
1031 	case CHIP_CEDAR:
1032 		radeon_program_register_sequence(rdev,
1033 						 cedar_golden_registers,
1034 						 (const u32)ARRAY_SIZE(cedar_golden_registers));
1035 		radeon_program_register_sequence(rdev,
1036 						 evergreen_golden_registers2,
1037 						 (const u32)ARRAY_SIZE(evergreen_golden_registers2));
1038 		radeon_program_register_sequence(rdev,
1039 						 cedar_mgcg_init,
1040 						 (const u32)ARRAY_SIZE(cedar_mgcg_init));
1041 		break;
1042 	case CHIP_PALM:
1043 		radeon_program_register_sequence(rdev,
1044 						 wrestler_golden_registers,
1045 						 (const u32)ARRAY_SIZE(wrestler_golden_registers));
1046 		break;
1047 	case CHIP_SUMO:
1048 		radeon_program_register_sequence(rdev,
1049 						 supersumo_golden_registers,
1050 						 (const u32)ARRAY_SIZE(supersumo_golden_registers));
1051 		break;
1052 	case CHIP_SUMO2:
1053 		radeon_program_register_sequence(rdev,
1054 						 supersumo_golden_registers,
1055 						 (const u32)ARRAY_SIZE(supersumo_golden_registers));
1056 		radeon_program_register_sequence(rdev,
1057 						 sumo_golden_registers,
1058 						 (const u32)ARRAY_SIZE(sumo_golden_registers));
1059 		break;
1060 	case CHIP_BARTS:
1061 		radeon_program_register_sequence(rdev,
1062 						 barts_golden_registers,
1063 						 (const u32)ARRAY_SIZE(barts_golden_registers));
1064 		break;
1065 	case CHIP_TURKS:
1066 		radeon_program_register_sequence(rdev,
1067 						 turks_golden_registers,
1068 						 (const u32)ARRAY_SIZE(turks_golden_registers));
1069 		break;
1070 	case CHIP_CAICOS:
1071 		radeon_program_register_sequence(rdev,
1072 						 caicos_golden_registers,
1073 						 (const u32)ARRAY_SIZE(caicos_golden_registers));
1074 		break;
1075 	default:
1076 		break;
1077 	}
1078 }
1079 
1080 /**
1081  * evergreen_get_allowed_info_register - fetch the register for the info ioctl
1082  *
1083  * @rdev: radeon_device pointer
1084  * @reg: register offset in bytes
1085  * @val: register value
1086  *
1087  * Returns 0 for success or -EINVAL for an invalid register
1088  *
1089  */
1090 int evergreen_get_allowed_info_register(struct radeon_device *rdev,
1091 					u32 reg, u32 *val)
1092 {
1093 	switch (reg) {
1094 	case GRBM_STATUS:
1095 	case GRBM_STATUS_SE0:
1096 	case GRBM_STATUS_SE1:
1097 	case SRBM_STATUS:
1098 	case SRBM_STATUS2:
1099 	case DMA_STATUS_REG:
1100 	case UVD_STATUS:
1101 		*val = RREG32(reg);
1102 		return 0;
1103 	default:
1104 		return -EINVAL;
1105 	}
1106 }
1107 
1108 void evergreen_tiling_fields(unsigned tiling_flags, unsigned *bankw,
1109 			     unsigned *bankh, unsigned *mtaspect,
1110 			     unsigned *tile_split)
1111 {
1112 	*bankw = (tiling_flags >> RADEON_TILING_EG_BANKW_SHIFT) & RADEON_TILING_EG_BANKW_MASK;
1113 	*bankh = (tiling_flags >> RADEON_TILING_EG_BANKH_SHIFT) & RADEON_TILING_EG_BANKH_MASK;
1114 	*mtaspect = (tiling_flags >> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK;
1115 	*tile_split = (tiling_flags >> RADEON_TILING_EG_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_TILE_SPLIT_MASK;
1116 	switch (*bankw) {
1117 	default:
1118 	case 1: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_1; break;
1119 	case 2: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_2; break;
1120 	case 4: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_4; break;
1121 	case 8: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_8; break;
1122 	}
1123 	switch (*bankh) {
1124 	default:
1125 	case 1: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_1; break;
1126 	case 2: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_2; break;
1127 	case 4: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_4; break;
1128 	case 8: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_8; break;
1129 	}
1130 	switch (*mtaspect) {
1131 	default:
1132 	case 1: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_1; break;
1133 	case 2: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_2; break;
1134 	case 4: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_4; break;
1135 	case 8: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_8; break;
1136 	}
1137 }
1138 
1139 static int sumo_set_uvd_clock(struct radeon_device *rdev, u32 clock,
1140 			      u32 cntl_reg, u32 status_reg)
1141 {
1142 	int r, i;
1143 	struct atom_clock_dividers dividers;
1144 
1145 	r = radeon_atom_get_clock_dividers(rdev, COMPUTE_ENGINE_PLL_PARAM,
1146 					   clock, false, &dividers);
1147 	if (r)
1148 		return r;
1149 
1150 	WREG32_P(cntl_reg, dividers.post_div, ~(DCLK_DIR_CNTL_EN|DCLK_DIVIDER_MASK));
1151 
1152 	for (i = 0; i < 100; i++) {
1153 		if (RREG32(status_reg) & DCLK_STATUS)
1154 			break;
1155 		mdelay(10);
1156 	}
1157 	if (i == 100)
1158 		return -ETIMEDOUT;
1159 
1160 	return 0;
1161 }
1162 
1163 int sumo_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk)
1164 {
1165 	int r = 0;
1166 	u32 cg_scratch = RREG32(CG_SCRATCH1);
1167 
1168 	r = sumo_set_uvd_clock(rdev, vclk, CG_VCLK_CNTL, CG_VCLK_STATUS);
1169 	if (r)
1170 		goto done;
1171 	cg_scratch &= 0xffff0000;
1172 	cg_scratch |= vclk / 100; /* Mhz */
1173 
1174 	r = sumo_set_uvd_clock(rdev, dclk, CG_DCLK_CNTL, CG_DCLK_STATUS);
1175 	if (r)
1176 		goto done;
1177 	cg_scratch &= 0x0000ffff;
1178 	cg_scratch |= (dclk / 100) << 16; /* Mhz */
1179 
1180 done:
1181 	WREG32(CG_SCRATCH1, cg_scratch);
1182 
1183 	return r;
1184 }
1185 
1186 int evergreen_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk)
1187 {
1188 	/* start off with something large */
1189 	unsigned fb_div = 0, vclk_div = 0, dclk_div = 0;
1190 	int r;
1191 
1192 	/* bypass vclk and dclk with bclk */
1193 	WREG32_P(CG_UPLL_FUNC_CNTL_2,
1194 		VCLK_SRC_SEL(1) | DCLK_SRC_SEL(1),
1195 		~(VCLK_SRC_SEL_MASK | DCLK_SRC_SEL_MASK));
1196 
1197 	/* put PLL in bypass mode */
1198 	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_BYPASS_EN_MASK, ~UPLL_BYPASS_EN_MASK);
1199 
1200 	if (!vclk || !dclk) {
1201 		/* keep the Bypass mode, put PLL to sleep */
1202 		WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_SLEEP_MASK, ~UPLL_SLEEP_MASK);
1203 		return 0;
1204 	}
1205 
1206 	r = radeon_uvd_calc_upll_dividers(rdev, vclk, dclk, 125000, 250000,
1207 					  16384, 0x03FFFFFF, 0, 128, 5,
1208 					  &fb_div, &vclk_div, &dclk_div);
1209 	if (r)
1210 		return r;
1211 
1212 	/* set VCO_MODE to 1 */
1213 	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_VCO_MODE_MASK, ~UPLL_VCO_MODE_MASK);
1214 
1215 	/* toggle UPLL_SLEEP to 1 then back to 0 */
1216 	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_SLEEP_MASK, ~UPLL_SLEEP_MASK);
1217 	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_SLEEP_MASK);
1218 
1219 	/* deassert UPLL_RESET */
1220 	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_RESET_MASK);
1221 
1222 	mdelay(1);
1223 
1224 	r = radeon_uvd_send_upll_ctlreq(rdev, CG_UPLL_FUNC_CNTL);
1225 	if (r)
1226 		return r;
1227 
1228 	/* assert UPLL_RESET again */
1229 	WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_RESET_MASK, ~UPLL_RESET_MASK);
1230 
1231 	/* disable spread spectrum. */
1232 	WREG32_P(CG_UPLL_SPREAD_SPECTRUM, 0, ~SSEN_MASK);
1233 
1234 	/* set feedback divider */
1235 	WREG32_P(CG_UPLL_FUNC_CNTL_3, UPLL_FB_DIV(fb_div), ~UPLL_FB_DIV_MASK);
1236 
1237 	/* set ref divider to 0 */
1238 	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_REF_DIV_MASK);
1239 
1240 	if (fb_div < 307200)
1241 		WREG32_P(CG_UPLL_FUNC_CNTL_4, 0, ~UPLL_SPARE_ISPARE9);
1242 	else
1243 		WREG32_P(CG_UPLL_FUNC_CNTL_4, UPLL_SPARE_ISPARE9, ~UPLL_SPARE_ISPARE9);
1244 
1245 	/* set PDIV_A and PDIV_B */
1246 	WREG32_P(CG_UPLL_FUNC_CNTL_2,
1247 		UPLL_PDIV_A(vclk_div) | UPLL_PDIV_B(dclk_div),
1248 		~(UPLL_PDIV_A_MASK | UPLL_PDIV_B_MASK));
1249 
1250 	/* give the PLL some time to settle */
1251 	mdelay(15);
1252 
1253 	/* deassert PLL_RESET */
1254 	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_RESET_MASK);
1255 
1256 	mdelay(15);
1257 
1258 	/* switch from bypass mode to normal mode */
1259 	WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_BYPASS_EN_MASK);
1260 
1261 	r = radeon_uvd_send_upll_ctlreq(rdev, CG_UPLL_FUNC_CNTL);
1262 	if (r)
1263 		return r;
1264 
1265 	/* switch VCLK and DCLK selection */
1266 	WREG32_P(CG_UPLL_FUNC_CNTL_2,
1267 		VCLK_SRC_SEL(2) | DCLK_SRC_SEL(2),
1268 		~(VCLK_SRC_SEL_MASK | DCLK_SRC_SEL_MASK));
1269 
1270 	mdelay(100);
1271 
1272 	return 0;
1273 }
1274 
1275 void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev)
1276 {
1277 	int readrq;
1278 	u16 v;
1279 
1280 	readrq = pcie_get_readrq(rdev->pdev);
1281 	v = ffs(readrq) - 8;
1282 	/* if bios or OS sets MAX_READ_REQUEST_SIZE to an invalid value, fix it
1283 	 * to avoid hangs or perfomance issues
1284 	 */
1285 	if ((v == 0) || (v == 6) || (v == 7))
1286 		pcie_set_readrq(rdev->pdev, 512);
1287 }
1288 
1289 void dce4_program_fmt(struct drm_encoder *encoder)
1290 {
1291 	struct drm_device *dev = encoder->dev;
1292 	struct radeon_device *rdev = dev->dev_private;
1293 	struct radeon_encoder *radeon_encoder = to_radeon_encoder(encoder);
1294 	struct radeon_crtc *radeon_crtc = to_radeon_crtc(encoder->crtc);
1295 	struct drm_connector *connector = radeon_get_connector_for_encoder(encoder);
1296 	int bpc = 0;
1297 	u32 tmp = 0;
1298 	enum radeon_connector_dither dither = RADEON_FMT_DITHER_DISABLE;
1299 
1300 	if (connector) {
1301 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
1302 		bpc = radeon_get_monitor_bpc(connector);
1303 		dither = radeon_connector->dither;
1304 	}
1305 
1306 	/* LVDS/eDP FMT is set up by atom */
1307 	if (radeon_encoder->devices & ATOM_DEVICE_LCD_SUPPORT)
1308 		return;
1309 
1310 	/* not needed for analog */
1311 	if ((radeon_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC1) ||
1312 	    (radeon_encoder->encoder_id == ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC2))
1313 		return;
1314 
1315 	if (bpc == 0)
1316 		return;
1317 
1318 	switch (bpc) {
1319 	case 6:
1320 		if (dither == RADEON_FMT_DITHER_ENABLE)
1321 			/* XXX sort out optimal dither settings */
1322 			tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
1323 				FMT_SPATIAL_DITHER_EN);
1324 		else
1325 			tmp |= FMT_TRUNCATE_EN;
1326 		break;
1327 	case 8:
1328 		if (dither == RADEON_FMT_DITHER_ENABLE)
1329 			/* XXX sort out optimal dither settings */
1330 			tmp |= (FMT_FRAME_RANDOM_ENABLE | FMT_HIGHPASS_RANDOM_ENABLE |
1331 				FMT_RGB_RANDOM_ENABLE |
1332 				FMT_SPATIAL_DITHER_EN | FMT_SPATIAL_DITHER_DEPTH);
1333 		else
1334 			tmp |= (FMT_TRUNCATE_EN | FMT_TRUNCATE_DEPTH);
1335 		break;
1336 	case 10:
1337 	default:
1338 		/* not needed */
1339 		break;
1340 	}
1341 
1342 	WREG32(FMT_BIT_DEPTH_CONTROL + radeon_crtc->crtc_offset, tmp);
1343 }
1344 
1345 static bool dce4_is_in_vblank(struct radeon_device *rdev, int crtc)
1346 {
1347 	if (RREG32(EVERGREEN_CRTC_STATUS + crtc_offsets[crtc]) & EVERGREEN_CRTC_V_BLANK)
1348 		return true;
1349 	else
1350 		return false;
1351 }
1352 
1353 static bool dce4_is_counter_moving(struct radeon_device *rdev, int crtc)
1354 {
1355 	u32 pos1, pos2;
1356 
1357 	pos1 = RREG32(EVERGREEN_CRTC_STATUS_POSITION + crtc_offsets[crtc]);
1358 	pos2 = RREG32(EVERGREEN_CRTC_STATUS_POSITION + crtc_offsets[crtc]);
1359 
1360 	if (pos1 != pos2)
1361 		return true;
1362 	else
1363 		return false;
1364 }
1365 
1366 /**
1367  * dce4_wait_for_vblank - vblank wait asic callback.
1368  *
1369  * @rdev: radeon_device pointer
1370  * @crtc: crtc to wait for vblank on
1371  *
1372  * Wait for vblank on the requested crtc (evergreen+).
1373  */
1374 void dce4_wait_for_vblank(struct radeon_device *rdev, int crtc)
1375 {
1376 	unsigned i = 0;
1377 
1378 	if (crtc >= rdev->num_crtc)
1379 		return;
1380 
1381 	if (!(RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[crtc]) & EVERGREEN_CRTC_MASTER_EN))
1382 		return;
1383 
1384 	/* depending on when we hit vblank, we may be close to active; if so,
1385 	 * wait for another frame.
1386 	 */
1387 	while (dce4_is_in_vblank(rdev, crtc)) {
1388 		if (i++ % 100 == 0) {
1389 			if (!dce4_is_counter_moving(rdev, crtc))
1390 				break;
1391 		}
1392 	}
1393 
1394 	while (!dce4_is_in_vblank(rdev, crtc)) {
1395 		if (i++ % 100 == 0) {
1396 			if (!dce4_is_counter_moving(rdev, crtc))
1397 				break;
1398 		}
1399 	}
1400 }
1401 
1402 /**
1403  * evergreen_page_flip - pageflip callback.
1404  *
1405  * @rdev: radeon_device pointer
1406  * @crtc_id: crtc to cleanup pageflip on
1407  * @crtc_base: new address of the crtc (GPU MC address)
1408  * @async: asynchronous flip
1409  *
1410  * Triggers the actual pageflip by updating the primary
1411  * surface base address (evergreen+).
1412  */
1413 void evergreen_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base,
1414 			 bool async)
1415 {
1416 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
1417 
1418 	/* update the scanout addresses */
1419 	WREG32(EVERGREEN_GRPH_FLIP_CONTROL + radeon_crtc->crtc_offset,
1420 	       async ? EVERGREEN_GRPH_SURFACE_UPDATE_H_RETRACE_EN : 0);
1421 	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + radeon_crtc->crtc_offset,
1422 	       upper_32_bits(crtc_base));
1423 	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
1424 	       (u32)crtc_base);
1425 	/* post the write */
1426 	RREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset);
1427 }
1428 
1429 /**
1430  * evergreen_page_flip_pending - check if page flip is still pending
1431  *
1432  * @rdev: radeon_device pointer
1433  * @crtc_id: crtc to check
1434  *
1435  * Returns the current update pending status.
1436  */
1437 bool evergreen_page_flip_pending(struct radeon_device *rdev, int crtc_id)
1438 {
1439 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
1440 
1441 	/* Return current update_pending status: */
1442 	return !!(RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset) &
1443 		EVERGREEN_GRPH_SURFACE_UPDATE_PENDING);
1444 }
1445 
1446 /* get temperature in millidegrees */
1447 int evergreen_get_temp(struct radeon_device *rdev)
1448 {
1449 	u32 temp, toffset;
1450 	int actual_temp = 0;
1451 
1452 	if (rdev->family == CHIP_JUNIPER) {
1453 		toffset = (RREG32(CG_THERMAL_CTRL) & TOFFSET_MASK) >>
1454 			TOFFSET_SHIFT;
1455 		temp = (RREG32(CG_TS0_STATUS) & TS0_ADC_DOUT_MASK) >>
1456 			TS0_ADC_DOUT_SHIFT;
1457 
1458 		if (toffset & 0x100)
1459 			actual_temp = temp / 2 - (0x200 - toffset);
1460 		else
1461 			actual_temp = temp / 2 + toffset;
1462 
1463 		actual_temp = actual_temp * 1000;
1464 
1465 	} else {
1466 		temp = (RREG32(CG_MULT_THERMAL_STATUS) & ASIC_T_MASK) >>
1467 			ASIC_T_SHIFT;
1468 
1469 		if (temp & 0x400)
1470 			actual_temp = -256;
1471 		else if (temp & 0x200)
1472 			actual_temp = 255;
1473 		else if (temp & 0x100) {
1474 			actual_temp = temp & 0x1ff;
1475 			actual_temp |= ~0x1ff;
1476 		} else
1477 			actual_temp = temp & 0xff;
1478 
1479 		actual_temp = (actual_temp * 1000) / 2;
1480 	}
1481 
1482 	return actual_temp;
1483 }
1484 
1485 int sumo_get_temp(struct radeon_device *rdev)
1486 {
1487 	u32 temp = RREG32(CG_THERMAL_STATUS) & 0xff;
1488 	int actual_temp = temp - 49;
1489 
1490 	return actual_temp * 1000;
1491 }
1492 
1493 /**
1494  * sumo_pm_init_profile - Initialize power profiles callback.
1495  *
1496  * @rdev: radeon_device pointer
1497  *
1498  * Initialize the power states used in profile mode
1499  * (sumo, trinity, SI).
1500  * Used for profile mode only.
1501  */
1502 void sumo_pm_init_profile(struct radeon_device *rdev)
1503 {
1504 	int idx;
1505 
1506 	/* default */
1507 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
1508 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
1509 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
1510 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
1511 
1512 	/* low,mid sh/mh */
1513 	if (rdev->flags & RADEON_IS_MOBILITY)
1514 		idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_BATTERY, 0);
1515 	else
1516 		idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
1517 
1518 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = idx;
1519 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = idx;
1520 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
1521 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
1522 
1523 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = idx;
1524 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = idx;
1525 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
1526 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
1527 
1528 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = idx;
1529 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = idx;
1530 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
1531 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
1532 
1533 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = idx;
1534 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = idx;
1535 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
1536 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
1537 
1538 	/* high sh/mh */
1539 	idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
1540 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = idx;
1541 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = idx;
1542 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
1543 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx =
1544 		rdev->pm.power_state[idx].num_clock_modes - 1;
1545 
1546 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = idx;
1547 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = idx;
1548 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
1549 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx =
1550 		rdev->pm.power_state[idx].num_clock_modes - 1;
1551 }
1552 
1553 /**
1554  * btc_pm_init_profile - Initialize power profiles callback.
1555  *
1556  * @rdev: radeon_device pointer
1557  *
1558  * Initialize the power states used in profile mode
1559  * (BTC, cayman).
1560  * Used for profile mode only.
1561  */
1562 void btc_pm_init_profile(struct radeon_device *rdev)
1563 {
1564 	int idx;
1565 
1566 	/* default */
1567 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
1568 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
1569 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
1570 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 2;
1571 	/* starting with BTC, there is one state that is used for both
1572 	 * MH and SH.  Difference is that we always use the high clock index for
1573 	 * mclk.
1574 	 */
1575 	if (rdev->flags & RADEON_IS_MOBILITY)
1576 		idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_BATTERY, 0);
1577 	else
1578 		idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
1579 	/* low sh */
1580 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = idx;
1581 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = idx;
1582 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
1583 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
1584 	/* mid sh */
1585 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = idx;
1586 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = idx;
1587 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
1588 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 1;
1589 	/* high sh */
1590 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = idx;
1591 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = idx;
1592 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
1593 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 2;
1594 	/* low mh */
1595 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = idx;
1596 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = idx;
1597 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
1598 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
1599 	/* mid mh */
1600 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = idx;
1601 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = idx;
1602 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
1603 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 1;
1604 	/* high mh */
1605 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = idx;
1606 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = idx;
1607 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
1608 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 2;
1609 }
1610 
1611 /**
1612  * evergreen_pm_misc - set additional pm hw parameters callback.
1613  *
1614  * @rdev: radeon_device pointer
1615  *
1616  * Set non-clock parameters associated with a power state
1617  * (voltage, etc.) (evergreen+).
1618  */
1619 void evergreen_pm_misc(struct radeon_device *rdev)
1620 {
1621 	int req_ps_idx = rdev->pm.requested_power_state_index;
1622 	int req_cm_idx = rdev->pm.requested_clock_mode_index;
1623 	struct radeon_power_state *ps = &rdev->pm.power_state[req_ps_idx];
1624 	struct radeon_voltage *voltage = &ps->clock_info[req_cm_idx].voltage;
1625 
1626 	if (voltage->type == VOLTAGE_SW) {
1627 		/* 0xff0x are flags rather then an actual voltage */
1628 		if ((voltage->voltage & 0xff00) == 0xff00)
1629 			return;
1630 		if (voltage->voltage && (voltage->voltage != rdev->pm.current_vddc)) {
1631 			radeon_atom_set_voltage(rdev, voltage->voltage, SET_VOLTAGE_TYPE_ASIC_VDDC);
1632 			rdev->pm.current_vddc = voltage->voltage;
1633 			DRM_DEBUG("Setting: vddc: %d\n", voltage->voltage);
1634 		}
1635 
1636 		/* starting with BTC, there is one state that is used for both
1637 		 * MH and SH.  Difference is that we always use the high clock index for
1638 		 * mclk and vddci.
1639 		 */
1640 		if ((rdev->pm.pm_method == PM_METHOD_PROFILE) &&
1641 		    (rdev->family >= CHIP_BARTS) &&
1642 		    rdev->pm.active_crtc_count &&
1643 		    ((rdev->pm.profile_index == PM_PROFILE_MID_MH_IDX) ||
1644 		     (rdev->pm.profile_index == PM_PROFILE_LOW_MH_IDX)))
1645 			voltage = &rdev->pm.power_state[req_ps_idx].
1646 				clock_info[rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx].voltage;
1647 
1648 		/* 0xff0x are flags rather then an actual voltage */
1649 		if ((voltage->vddci & 0xff00) == 0xff00)
1650 			return;
1651 		if (voltage->vddci && (voltage->vddci != rdev->pm.current_vddci)) {
1652 			radeon_atom_set_voltage(rdev, voltage->vddci, SET_VOLTAGE_TYPE_ASIC_VDDCI);
1653 			rdev->pm.current_vddci = voltage->vddci;
1654 			DRM_DEBUG("Setting: vddci: %d\n", voltage->vddci);
1655 		}
1656 	}
1657 }
1658 
1659 /**
1660  * evergreen_pm_prepare - pre-power state change callback.
1661  *
1662  * @rdev: radeon_device pointer
1663  *
1664  * Prepare for a power state change (evergreen+).
1665  */
1666 void evergreen_pm_prepare(struct radeon_device *rdev)
1667 {
1668 	struct drm_device *ddev = rdev->ddev;
1669 	struct drm_crtc *crtc;
1670 	struct radeon_crtc *radeon_crtc;
1671 	u32 tmp;
1672 
1673 	/* disable any active CRTCs */
1674 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
1675 		radeon_crtc = to_radeon_crtc(crtc);
1676 		if (radeon_crtc->enabled) {
1677 			tmp = RREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset);
1678 			tmp |= EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
1679 			WREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset, tmp);
1680 		}
1681 	}
1682 }
1683 
1684 /**
1685  * evergreen_pm_finish - post-power state change callback.
1686  *
1687  * @rdev: radeon_device pointer
1688  *
1689  * Clean up after a power state change (evergreen+).
1690  */
1691 void evergreen_pm_finish(struct radeon_device *rdev)
1692 {
1693 	struct drm_device *ddev = rdev->ddev;
1694 	struct drm_crtc *crtc;
1695 	struct radeon_crtc *radeon_crtc;
1696 	u32 tmp;
1697 
1698 	/* enable any active CRTCs */
1699 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
1700 		radeon_crtc = to_radeon_crtc(crtc);
1701 		if (radeon_crtc->enabled) {
1702 			tmp = RREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset);
1703 			tmp &= ~EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
1704 			WREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset, tmp);
1705 		}
1706 	}
1707 }
1708 
1709 /**
1710  * evergreen_hpd_sense - hpd sense callback.
1711  *
1712  * @rdev: radeon_device pointer
1713  * @hpd: hpd (hotplug detect) pin
1714  *
1715  * Checks if a digital monitor is connected (evergreen+).
1716  * Returns true if connected, false if not connected.
1717  */
1718 bool evergreen_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
1719 {
1720 	if (hpd == RADEON_HPD_NONE)
1721 		return false;
1722 
1723 	return !!(RREG32(DC_HPDx_INT_STATUS_REG(hpd)) & DC_HPDx_SENSE);
1724 }
1725 
1726 /**
1727  * evergreen_hpd_set_polarity - hpd set polarity callback.
1728  *
1729  * @rdev: radeon_device pointer
1730  * @hpd: hpd (hotplug detect) pin
1731  *
1732  * Set the polarity of the hpd pin (evergreen+).
1733  */
1734 void evergreen_hpd_set_polarity(struct radeon_device *rdev,
1735 				enum radeon_hpd_id hpd)
1736 {
1737 	bool connected = evergreen_hpd_sense(rdev, hpd);
1738 
1739 	if (hpd == RADEON_HPD_NONE)
1740 		return;
1741 
1742 	if (connected)
1743 		WREG32_AND(DC_HPDx_INT_CONTROL(hpd), ~DC_HPDx_INT_POLARITY);
1744 	else
1745 		WREG32_OR(DC_HPDx_INT_CONTROL(hpd), DC_HPDx_INT_POLARITY);
1746 }
1747 
1748 /**
1749  * evergreen_hpd_init - hpd setup callback.
1750  *
1751  * @rdev: radeon_device pointer
1752  *
1753  * Setup the hpd pins used by the card (evergreen+).
1754  * Enable the pin, set the polarity, and enable the hpd interrupts.
1755  */
1756 void evergreen_hpd_init(struct radeon_device *rdev)
1757 {
1758 	struct drm_device *dev = rdev->ddev;
1759 	struct drm_connector *connector;
1760 	unsigned enabled = 0;
1761 	u32 tmp = DC_HPDx_CONNECTION_TIMER(0x9c4) |
1762 		DC_HPDx_RX_INT_TIMER(0xfa) | DC_HPDx_EN;
1763 
1764 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
1765 		enum radeon_hpd_id hpd =
1766 			to_radeon_connector(connector)->hpd.hpd;
1767 
1768 		if (connector->connector_type == DRM_MODE_CONNECTOR_eDP ||
1769 		    connector->connector_type == DRM_MODE_CONNECTOR_LVDS) {
1770 			/* don't try to enable hpd on eDP or LVDS avoid breaking the
1771 			 * aux dp channel on imac and help (but not completely fix)
1772 			 * https://bugzilla.redhat.com/show_bug.cgi?id=726143
1773 			 * also avoid interrupt storms during dpms.
1774 			 */
1775 			continue;
1776 		}
1777 
1778 		if (hpd == RADEON_HPD_NONE)
1779 			continue;
1780 
1781 		WREG32(DC_HPDx_CONTROL(hpd), tmp);
1782 		enabled |= 1 << hpd;
1783 
1784 		radeon_hpd_set_polarity(rdev, hpd);
1785 	}
1786 	radeon_irq_kms_enable_hpd(rdev, enabled);
1787 }
1788 
1789 /**
1790  * evergreen_hpd_fini - hpd tear down callback.
1791  *
1792  * @rdev: radeon_device pointer
1793  *
1794  * Tear down the hpd pins used by the card (evergreen+).
1795  * Disable the hpd interrupts.
1796  */
1797 void evergreen_hpd_fini(struct radeon_device *rdev)
1798 {
1799 	struct drm_device *dev = rdev->ddev;
1800 	struct drm_connector *connector;
1801 	unsigned disabled = 0;
1802 
1803 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
1804 		enum radeon_hpd_id hpd =
1805 			to_radeon_connector(connector)->hpd.hpd;
1806 
1807 		if (hpd == RADEON_HPD_NONE)
1808 			continue;
1809 
1810 		WREG32(DC_HPDx_CONTROL(hpd), 0);
1811 		disabled |= 1 << hpd;
1812 	}
1813 	radeon_irq_kms_disable_hpd(rdev, disabled);
1814 }
1815 
1816 /* watermark setup */
1817 
1818 static u32 evergreen_line_buffer_adjust(struct radeon_device *rdev,
1819 					struct radeon_crtc *radeon_crtc,
1820 					struct drm_display_mode *mode,
1821 					struct drm_display_mode *other_mode)
1822 {
1823 	u32 tmp, buffer_alloc, i;
1824 	u32 pipe_offset = radeon_crtc->crtc_id * 0x20;
1825 	/*
1826 	 * Line Buffer Setup
1827 	 * There are 3 line buffers, each one shared by 2 display controllers.
1828 	 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
1829 	 * the display controllers.  The paritioning is done via one of four
1830 	 * preset allocations specified in bits 2:0:
1831 	 * first display controller
1832 	 *  0 - first half of lb (3840 * 2)
1833 	 *  1 - first 3/4 of lb (5760 * 2)
1834 	 *  2 - whole lb (7680 * 2), other crtc must be disabled
1835 	 *  3 - first 1/4 of lb (1920 * 2)
1836 	 * second display controller
1837 	 *  4 - second half of lb (3840 * 2)
1838 	 *  5 - second 3/4 of lb (5760 * 2)
1839 	 *  6 - whole lb (7680 * 2), other crtc must be disabled
1840 	 *  7 - last 1/4 of lb (1920 * 2)
1841 	 */
1842 	/* this can get tricky if we have two large displays on a paired group
1843 	 * of crtcs.  Ideally for multiple large displays we'd assign them to
1844 	 * non-linked crtcs for maximum line buffer allocation.
1845 	 */
1846 	if (radeon_crtc->base.enabled && mode) {
1847 		if (other_mode) {
1848 			tmp = 0; /* 1/2 */
1849 			buffer_alloc = 1;
1850 		} else {
1851 			tmp = 2; /* whole */
1852 			buffer_alloc = 2;
1853 		}
1854 	} else {
1855 		tmp = 0;
1856 		buffer_alloc = 0;
1857 	}
1858 
1859 	/* second controller of the pair uses second half of the lb */
1860 	if (radeon_crtc->crtc_id % 2)
1861 		tmp += 4;
1862 	WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset, tmp);
1863 
1864 	if (ASIC_IS_DCE41(rdev) || ASIC_IS_DCE5(rdev)) {
1865 		WREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset,
1866 		       DMIF_BUFFERS_ALLOCATED(buffer_alloc));
1867 		for (i = 0; i < rdev->usec_timeout; i++) {
1868 			if (RREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset) &
1869 			    DMIF_BUFFERS_ALLOCATED_COMPLETED)
1870 				break;
1871 			udelay(1);
1872 		}
1873 	}
1874 
1875 	if (radeon_crtc->base.enabled && mode) {
1876 		switch (tmp) {
1877 		case 0:
1878 		case 4:
1879 		default:
1880 			if (ASIC_IS_DCE5(rdev))
1881 				return 4096 * 2;
1882 			else
1883 				return 3840 * 2;
1884 		case 1:
1885 		case 5:
1886 			if (ASIC_IS_DCE5(rdev))
1887 				return 6144 * 2;
1888 			else
1889 				return 5760 * 2;
1890 		case 2:
1891 		case 6:
1892 			if (ASIC_IS_DCE5(rdev))
1893 				return 8192 * 2;
1894 			else
1895 				return 7680 * 2;
1896 		case 3:
1897 		case 7:
1898 			if (ASIC_IS_DCE5(rdev))
1899 				return 2048 * 2;
1900 			else
1901 				return 1920 * 2;
1902 		}
1903 	}
1904 
1905 	/* controller not enabled, so no lb used */
1906 	return 0;
1907 }
1908 
1909 u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev)
1910 {
1911 	u32 tmp = RREG32(MC_SHARED_CHMAP);
1912 
1913 	switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
1914 	case 0:
1915 	default:
1916 		return 1;
1917 	case 1:
1918 		return 2;
1919 	case 2:
1920 		return 4;
1921 	case 3:
1922 		return 8;
1923 	}
1924 }
1925 
1926 struct evergreen_wm_params {
1927 	u32 dram_channels; /* number of dram channels */
1928 	u32 yclk;          /* bandwidth per dram data pin in kHz */
1929 	u32 sclk;          /* engine clock in kHz */
1930 	u32 disp_clk;      /* display clock in kHz */
1931 	u32 src_width;     /* viewport width */
1932 	u32 active_time;   /* active display time in ns */
1933 	u32 blank_time;    /* blank time in ns */
1934 	bool interlaced;    /* mode is interlaced */
1935 	fixed20_12 vsc;    /* vertical scale ratio */
1936 	u32 num_heads;     /* number of active crtcs */
1937 	u32 bytes_per_pixel; /* bytes per pixel display + overlay */
1938 	u32 lb_size;       /* line buffer allocated to pipe */
1939 	u32 vtaps;         /* vertical scaler taps */
1940 };
1941 
1942 static u32 evergreen_dram_bandwidth(struct evergreen_wm_params *wm)
1943 {
1944 	/* Calculate DRAM Bandwidth and the part allocated to display. */
1945 	fixed20_12 dram_efficiency; /* 0.7 */
1946 	fixed20_12 yclk, dram_channels, bandwidth;
1947 	fixed20_12 a;
1948 
1949 	a.full = dfixed_const(1000);
1950 	yclk.full = dfixed_const(wm->yclk);
1951 	yclk.full = dfixed_div(yclk, a);
1952 	dram_channels.full = dfixed_const(wm->dram_channels * 4);
1953 	a.full = dfixed_const(10);
1954 	dram_efficiency.full = dfixed_const(7);
1955 	dram_efficiency.full = dfixed_div(dram_efficiency, a);
1956 	bandwidth.full = dfixed_mul(dram_channels, yclk);
1957 	bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
1958 
1959 	return dfixed_trunc(bandwidth);
1960 }
1961 
1962 static u32 evergreen_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
1963 {
1964 	/* Calculate DRAM Bandwidth and the part allocated to display. */
1965 	fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
1966 	fixed20_12 yclk, dram_channels, bandwidth;
1967 	fixed20_12 a;
1968 
1969 	a.full = dfixed_const(1000);
1970 	yclk.full = dfixed_const(wm->yclk);
1971 	yclk.full = dfixed_div(yclk, a);
1972 	dram_channels.full = dfixed_const(wm->dram_channels * 4);
1973 	a.full = dfixed_const(10);
1974 	disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
1975 	disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
1976 	bandwidth.full = dfixed_mul(dram_channels, yclk);
1977 	bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
1978 
1979 	return dfixed_trunc(bandwidth);
1980 }
1981 
1982 static u32 evergreen_data_return_bandwidth(struct evergreen_wm_params *wm)
1983 {
1984 	/* Calculate the display Data return Bandwidth */
1985 	fixed20_12 return_efficiency; /* 0.8 */
1986 	fixed20_12 sclk, bandwidth;
1987 	fixed20_12 a;
1988 
1989 	a.full = dfixed_const(1000);
1990 	sclk.full = dfixed_const(wm->sclk);
1991 	sclk.full = dfixed_div(sclk, a);
1992 	a.full = dfixed_const(10);
1993 	return_efficiency.full = dfixed_const(8);
1994 	return_efficiency.full = dfixed_div(return_efficiency, a);
1995 	a.full = dfixed_const(32);
1996 	bandwidth.full = dfixed_mul(a, sclk);
1997 	bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
1998 
1999 	return dfixed_trunc(bandwidth);
2000 }
2001 
2002 static u32 evergreen_dmif_request_bandwidth(struct evergreen_wm_params *wm)
2003 {
2004 	/* Calculate the DMIF Request Bandwidth */
2005 	fixed20_12 disp_clk_request_efficiency; /* 0.8 */
2006 	fixed20_12 disp_clk, bandwidth;
2007 	fixed20_12 a;
2008 
2009 	a.full = dfixed_const(1000);
2010 	disp_clk.full = dfixed_const(wm->disp_clk);
2011 	disp_clk.full = dfixed_div(disp_clk, a);
2012 	a.full = dfixed_const(10);
2013 	disp_clk_request_efficiency.full = dfixed_const(8);
2014 	disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
2015 	a.full = dfixed_const(32);
2016 	bandwidth.full = dfixed_mul(a, disp_clk);
2017 	bandwidth.full = dfixed_mul(bandwidth, disp_clk_request_efficiency);
2018 
2019 	return dfixed_trunc(bandwidth);
2020 }
2021 
2022 static u32 evergreen_available_bandwidth(struct evergreen_wm_params *wm)
2023 {
2024 	/* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
2025 	u32 dram_bandwidth = evergreen_dram_bandwidth(wm);
2026 	u32 data_return_bandwidth = evergreen_data_return_bandwidth(wm);
2027 	u32 dmif_req_bandwidth = evergreen_dmif_request_bandwidth(wm);
2028 
2029 	return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
2030 }
2031 
2032 static u32 evergreen_average_bandwidth(struct evergreen_wm_params *wm)
2033 {
2034 	/* Calculate the display mode Average Bandwidth
2035 	 * DisplayMode should contain the source and destination dimensions,
2036 	 * timing, etc.
2037 	 */
2038 	fixed20_12 bpp;
2039 	fixed20_12 line_time;
2040 	fixed20_12 src_width;
2041 	fixed20_12 bandwidth;
2042 	fixed20_12 a;
2043 
2044 	a.full = dfixed_const(1000);
2045 	line_time.full = dfixed_const(wm->active_time + wm->blank_time);
2046 	line_time.full = dfixed_div(line_time, a);
2047 	bpp.full = dfixed_const(wm->bytes_per_pixel);
2048 	src_width.full = dfixed_const(wm->src_width);
2049 	bandwidth.full = dfixed_mul(src_width, bpp);
2050 	bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
2051 	bandwidth.full = dfixed_div(bandwidth, line_time);
2052 
2053 	return dfixed_trunc(bandwidth);
2054 }
2055 
2056 static u32 evergreen_latency_watermark(struct evergreen_wm_params *wm)
2057 {
2058 	/* First calcualte the latency in ns */
2059 	u32 mc_latency = 2000; /* 2000 ns. */
2060 	u32 available_bandwidth = evergreen_available_bandwidth(wm);
2061 	u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
2062 	u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
2063 	u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
2064 	u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
2065 		(wm->num_heads * cursor_line_pair_return_time);
2066 	u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
2067 	u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
2068 	fixed20_12 a, b, c;
2069 
2070 	if (wm->num_heads == 0)
2071 		return 0;
2072 
2073 	a.full = dfixed_const(2);
2074 	b.full = dfixed_const(1);
2075 	if ((wm->vsc.full > a.full) ||
2076 	    ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
2077 	    (wm->vtaps >= 5) ||
2078 	    ((wm->vsc.full >= a.full) && wm->interlaced))
2079 		max_src_lines_per_dst_line = 4;
2080 	else
2081 		max_src_lines_per_dst_line = 2;
2082 
2083 	a.full = dfixed_const(available_bandwidth);
2084 	b.full = dfixed_const(wm->num_heads);
2085 	a.full = dfixed_div(a, b);
2086 
2087 	lb_fill_bw = min(dfixed_trunc(a), wm->disp_clk * wm->bytes_per_pixel / 1000);
2088 
2089 	a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
2090 	b.full = dfixed_const(1000);
2091 	c.full = dfixed_const(lb_fill_bw);
2092 	b.full = dfixed_div(c, b);
2093 	a.full = dfixed_div(a, b);
2094 	line_fill_time = dfixed_trunc(a);
2095 
2096 	if (line_fill_time < wm->active_time)
2097 		return latency;
2098 	else
2099 		return latency + (line_fill_time - wm->active_time);
2100 
2101 }
2102 
2103 static bool evergreen_average_bandwidth_vs_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
2104 {
2105 	if (evergreen_average_bandwidth(wm) <=
2106 	    (evergreen_dram_bandwidth_for_display(wm) / wm->num_heads))
2107 		return true;
2108 	else
2109 		return false;
2110 };
2111 
2112 static bool evergreen_average_bandwidth_vs_available_bandwidth(struct evergreen_wm_params *wm)
2113 {
2114 	if (evergreen_average_bandwidth(wm) <=
2115 	    (evergreen_available_bandwidth(wm) / wm->num_heads))
2116 		return true;
2117 	else
2118 		return false;
2119 };
2120 
2121 static bool evergreen_check_latency_hiding(struct evergreen_wm_params *wm)
2122 {
2123 	u32 lb_partitions = wm->lb_size / wm->src_width;
2124 	u32 line_time = wm->active_time + wm->blank_time;
2125 	u32 latency_tolerant_lines;
2126 	u32 latency_hiding;
2127 	fixed20_12 a;
2128 
2129 	a.full = dfixed_const(1);
2130 	if (wm->vsc.full > a.full)
2131 		latency_tolerant_lines = 1;
2132 	else {
2133 		if (lb_partitions <= (wm->vtaps + 1))
2134 			latency_tolerant_lines = 1;
2135 		else
2136 			latency_tolerant_lines = 2;
2137 	}
2138 
2139 	latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
2140 
2141 	if (evergreen_latency_watermark(wm) <= latency_hiding)
2142 		return true;
2143 	else
2144 		return false;
2145 }
2146 
2147 static void evergreen_program_watermarks(struct radeon_device *rdev,
2148 					 struct radeon_crtc *radeon_crtc,
2149 					 u32 lb_size, u32 num_heads)
2150 {
2151 	struct drm_display_mode *mode = &radeon_crtc->base.mode;
2152 	struct evergreen_wm_params wm_low, wm_high;
2153 	u32 dram_channels;
2154 	u32 active_time;
2155 	u32 line_time = 0;
2156 	u32 latency_watermark_a = 0, latency_watermark_b = 0;
2157 	u32 priority_a_mark = 0, priority_b_mark = 0;
2158 	u32 priority_a_cnt = PRIORITY_OFF;
2159 	u32 priority_b_cnt = PRIORITY_OFF;
2160 	u32 pipe_offset = radeon_crtc->crtc_id * 16;
2161 	u32 tmp, arb_control3;
2162 	fixed20_12 a, b, c;
2163 
2164 	if (radeon_crtc->base.enabled && num_heads && mode) {
2165 		active_time = (u32) div_u64((u64)mode->crtc_hdisplay * 1000000,
2166 					    (u32)mode->clock);
2167 		line_time = (u32) div_u64((u64)mode->crtc_htotal * 1000000,
2168 					  (u32)mode->clock);
2169 		line_time = min(line_time, (u32)65535);
2170 		priority_a_cnt = 0;
2171 		priority_b_cnt = 0;
2172 		dram_channels = evergreen_get_number_of_dram_channels(rdev);
2173 
2174 		/* watermark for high clocks */
2175 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
2176 			wm_high.yclk =
2177 				radeon_dpm_get_mclk(rdev, false) * 10;
2178 			wm_high.sclk =
2179 				radeon_dpm_get_sclk(rdev, false) * 10;
2180 		} else {
2181 			wm_high.yclk = rdev->pm.current_mclk * 10;
2182 			wm_high.sclk = rdev->pm.current_sclk * 10;
2183 		}
2184 
2185 		wm_high.disp_clk = mode->clock;
2186 		wm_high.src_width = mode->crtc_hdisplay;
2187 		wm_high.active_time = active_time;
2188 		wm_high.blank_time = line_time - wm_high.active_time;
2189 		wm_high.interlaced = false;
2190 		if (mode->flags & DRM_MODE_FLAG_INTERLACE)
2191 			wm_high.interlaced = true;
2192 		wm_high.vsc = radeon_crtc->vsc;
2193 		wm_high.vtaps = 1;
2194 		if (radeon_crtc->rmx_type != RMX_OFF)
2195 			wm_high.vtaps = 2;
2196 		wm_high.bytes_per_pixel = 4; /* XXX: get this from fb config */
2197 		wm_high.lb_size = lb_size;
2198 		wm_high.dram_channels = dram_channels;
2199 		wm_high.num_heads = num_heads;
2200 
2201 		/* watermark for low clocks */
2202 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
2203 			wm_low.yclk =
2204 				radeon_dpm_get_mclk(rdev, true) * 10;
2205 			wm_low.sclk =
2206 				radeon_dpm_get_sclk(rdev, true) * 10;
2207 		} else {
2208 			wm_low.yclk = rdev->pm.current_mclk * 10;
2209 			wm_low.sclk = rdev->pm.current_sclk * 10;
2210 		}
2211 
2212 		wm_low.disp_clk = mode->clock;
2213 		wm_low.src_width = mode->crtc_hdisplay;
2214 		wm_low.active_time = active_time;
2215 		wm_low.blank_time = line_time - wm_low.active_time;
2216 		wm_low.interlaced = false;
2217 		if (mode->flags & DRM_MODE_FLAG_INTERLACE)
2218 			wm_low.interlaced = true;
2219 		wm_low.vsc = radeon_crtc->vsc;
2220 		wm_low.vtaps = 1;
2221 		if (radeon_crtc->rmx_type != RMX_OFF)
2222 			wm_low.vtaps = 2;
2223 		wm_low.bytes_per_pixel = 4; /* XXX: get this from fb config */
2224 		wm_low.lb_size = lb_size;
2225 		wm_low.dram_channels = dram_channels;
2226 		wm_low.num_heads = num_heads;
2227 
2228 		/* set for high clocks */
2229 		latency_watermark_a = min(evergreen_latency_watermark(&wm_high), (u32)65535);
2230 		/* set for low clocks */
2231 		latency_watermark_b = min(evergreen_latency_watermark(&wm_low), (u32)65535);
2232 
2233 		/* possibly force display priority to high */
2234 		/* should really do this at mode validation time... */
2235 		if (!evergreen_average_bandwidth_vs_dram_bandwidth_for_display(&wm_high) ||
2236 		    !evergreen_average_bandwidth_vs_available_bandwidth(&wm_high) ||
2237 		    !evergreen_check_latency_hiding(&wm_high) ||
2238 		    (rdev->disp_priority == 2)) {
2239 			DRM_DEBUG_KMS("force priority a to high\n");
2240 			priority_a_cnt |= PRIORITY_ALWAYS_ON;
2241 		}
2242 		if (!evergreen_average_bandwidth_vs_dram_bandwidth_for_display(&wm_low) ||
2243 		    !evergreen_average_bandwidth_vs_available_bandwidth(&wm_low) ||
2244 		    !evergreen_check_latency_hiding(&wm_low) ||
2245 		    (rdev->disp_priority == 2)) {
2246 			DRM_DEBUG_KMS("force priority b to high\n");
2247 			priority_b_cnt |= PRIORITY_ALWAYS_ON;
2248 		}
2249 
2250 		a.full = dfixed_const(1000);
2251 		b.full = dfixed_const(mode->clock);
2252 		b.full = dfixed_div(b, a);
2253 		c.full = dfixed_const(latency_watermark_a);
2254 		c.full = dfixed_mul(c, b);
2255 		c.full = dfixed_mul(c, radeon_crtc->hsc);
2256 		c.full = dfixed_div(c, a);
2257 		a.full = dfixed_const(16);
2258 		c.full = dfixed_div(c, a);
2259 		priority_a_mark = dfixed_trunc(c);
2260 		priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
2261 
2262 		a.full = dfixed_const(1000);
2263 		b.full = dfixed_const(mode->clock);
2264 		b.full = dfixed_div(b, a);
2265 		c.full = dfixed_const(latency_watermark_b);
2266 		c.full = dfixed_mul(c, b);
2267 		c.full = dfixed_mul(c, radeon_crtc->hsc);
2268 		c.full = dfixed_div(c, a);
2269 		a.full = dfixed_const(16);
2270 		c.full = dfixed_div(c, a);
2271 		priority_b_mark = dfixed_trunc(c);
2272 		priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
2273 
2274 		/* Save number of lines the linebuffer leads before the scanout */
2275 		radeon_crtc->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay);
2276 	}
2277 
2278 	/* select wm A */
2279 	arb_control3 = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
2280 	tmp = arb_control3;
2281 	tmp &= ~LATENCY_WATERMARK_MASK(3);
2282 	tmp |= LATENCY_WATERMARK_MASK(1);
2283 	WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
2284 	WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
2285 	       (LATENCY_LOW_WATERMARK(latency_watermark_a) |
2286 		LATENCY_HIGH_WATERMARK(line_time)));
2287 	/* select wm B */
2288 	tmp = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
2289 	tmp &= ~LATENCY_WATERMARK_MASK(3);
2290 	tmp |= LATENCY_WATERMARK_MASK(2);
2291 	WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
2292 	WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
2293 	       (LATENCY_LOW_WATERMARK(latency_watermark_b) |
2294 		LATENCY_HIGH_WATERMARK(line_time)));
2295 	/* restore original selection */
2296 	WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, arb_control3);
2297 
2298 	/* write the priority marks */
2299 	WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
2300 	WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
2301 
2302 	/* save values for DPM */
2303 	radeon_crtc->line_time = line_time;
2304 	radeon_crtc->wm_high = latency_watermark_a;
2305 	radeon_crtc->wm_low = latency_watermark_b;
2306 }
2307 
2308 /**
2309  * evergreen_bandwidth_update - update display watermarks callback.
2310  *
2311  * @rdev: radeon_device pointer
2312  *
2313  * Update the display watermarks based on the requested mode(s)
2314  * (evergreen+).
2315  */
2316 void evergreen_bandwidth_update(struct radeon_device *rdev)
2317 {
2318 	struct drm_display_mode *mode0 = NULL;
2319 	struct drm_display_mode *mode1 = NULL;
2320 	u32 num_heads = 0, lb_size;
2321 	int i;
2322 
2323 	if (!rdev->mode_info.mode_config_initialized)
2324 		return;
2325 
2326 	radeon_update_display_priority(rdev);
2327 
2328 	for (i = 0; i < rdev->num_crtc; i++) {
2329 		if (rdev->mode_info.crtcs[i]->base.enabled)
2330 			num_heads++;
2331 	}
2332 	for (i = 0; i < rdev->num_crtc; i += 2) {
2333 		mode0 = &rdev->mode_info.crtcs[i]->base.mode;
2334 		mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
2335 		lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
2336 		evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
2337 		lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
2338 		evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
2339 	}
2340 }
2341 
2342 /**
2343  * evergreen_mc_wait_for_idle - wait for MC idle callback.
2344  *
2345  * @rdev: radeon_device pointer
2346  *
2347  * Wait for the MC (memory controller) to be idle.
2348  * (evergreen+).
2349  * Returns 0 if the MC is idle, -1 if not.
2350  */
2351 int evergreen_mc_wait_for_idle(struct radeon_device *rdev)
2352 {
2353 	unsigned i;
2354 	u32 tmp;
2355 
2356 	for (i = 0; i < rdev->usec_timeout; i++) {
2357 		/* read MC_STATUS */
2358 		tmp = RREG32(SRBM_STATUS) & 0x1F00;
2359 		if (!tmp)
2360 			return 0;
2361 		udelay(1);
2362 	}
2363 	return -1;
2364 }
2365 
2366 /*
2367  * GART
2368  */
2369 void evergreen_pcie_gart_tlb_flush(struct radeon_device *rdev)
2370 {
2371 	unsigned i;
2372 	u32 tmp;
2373 
2374 	WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2375 
2376 	WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
2377 	for (i = 0; i < rdev->usec_timeout; i++) {
2378 		/* read MC_STATUS */
2379 		tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
2380 		tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
2381 		if (tmp == 2) {
2382 			pr_warn("[drm] r600 flush TLB failed\n");
2383 			return;
2384 		}
2385 		if (tmp) {
2386 			return;
2387 		}
2388 		udelay(1);
2389 	}
2390 }
2391 
2392 static int evergreen_pcie_gart_enable(struct radeon_device *rdev)
2393 {
2394 	u32 tmp;
2395 	int r;
2396 
2397 	if (rdev->gart.robj == NULL) {
2398 		dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
2399 		return -EINVAL;
2400 	}
2401 	r = radeon_gart_table_vram_pin(rdev);
2402 	if (r)
2403 		return r;
2404 	/* Setup L2 cache */
2405 	WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
2406 				ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2407 				EFFECTIVE_L2_QUEUE_SIZE(7));
2408 	WREG32(VM_L2_CNTL2, 0);
2409 	WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
2410 	/* Setup TLB control */
2411 	tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
2412 		SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2413 		SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU |
2414 		EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
2415 	if (rdev->flags & RADEON_IS_IGP) {
2416 		WREG32(FUS_MC_VM_MD_L1_TLB0_CNTL, tmp);
2417 		WREG32(FUS_MC_VM_MD_L1_TLB1_CNTL, tmp);
2418 		WREG32(FUS_MC_VM_MD_L1_TLB2_CNTL, tmp);
2419 	} else {
2420 		WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
2421 		WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
2422 		WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
2423 		if ((rdev->family == CHIP_JUNIPER) ||
2424 		    (rdev->family == CHIP_CYPRESS) ||
2425 		    (rdev->family == CHIP_HEMLOCK) ||
2426 		    (rdev->family == CHIP_BARTS))
2427 			WREG32(MC_VM_MD_L1_TLB3_CNTL, tmp);
2428 	}
2429 	WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
2430 	WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
2431 	WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
2432 	WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
2433 	WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
2434 	WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
2435 	WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
2436 	WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
2437 				RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
2438 	WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
2439 			(u32)(rdev->dummy_page.addr >> 12));
2440 	WREG32(VM_CONTEXT1_CNTL, 0);
2441 
2442 	evergreen_pcie_gart_tlb_flush(rdev);
2443 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
2444 		 (unsigned)(rdev->mc.gtt_size >> 20),
2445 		 (unsigned long long)rdev->gart.table_addr);
2446 	rdev->gart.ready = true;
2447 	return 0;
2448 }
2449 
2450 static void evergreen_pcie_gart_disable(struct radeon_device *rdev)
2451 {
2452 	u32 tmp;
2453 
2454 	/* Disable all tables */
2455 	WREG32(VM_CONTEXT0_CNTL, 0);
2456 	WREG32(VM_CONTEXT1_CNTL, 0);
2457 
2458 	/* Setup L2 cache */
2459 	WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
2460 				EFFECTIVE_L2_QUEUE_SIZE(7));
2461 	WREG32(VM_L2_CNTL2, 0);
2462 	WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
2463 	/* Setup TLB control */
2464 	tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
2465 	WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
2466 	WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
2467 	WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
2468 	WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
2469 	WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
2470 	WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
2471 	WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
2472 	radeon_gart_table_vram_unpin(rdev);
2473 }
2474 
2475 static void evergreen_pcie_gart_fini(struct radeon_device *rdev)
2476 {
2477 	evergreen_pcie_gart_disable(rdev);
2478 	radeon_gart_table_vram_free(rdev);
2479 	radeon_gart_fini(rdev);
2480 }
2481 
2482 
2483 static void evergreen_agp_enable(struct radeon_device *rdev)
2484 {
2485 	u32 tmp;
2486 
2487 	/* Setup L2 cache */
2488 	WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
2489 				ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2490 				EFFECTIVE_L2_QUEUE_SIZE(7));
2491 	WREG32(VM_L2_CNTL2, 0);
2492 	WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
2493 	/* Setup TLB control */
2494 	tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
2495 		SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2496 		SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU |
2497 		EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
2498 	WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
2499 	WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
2500 	WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
2501 	WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
2502 	WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
2503 	WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
2504 	WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
2505 	WREG32(VM_CONTEXT0_CNTL, 0);
2506 	WREG32(VM_CONTEXT1_CNTL, 0);
2507 }
2508 
2509 static const unsigned ni_dig_offsets[] =
2510 {
2511 	NI_DIG0_REGISTER_OFFSET,
2512 	NI_DIG1_REGISTER_OFFSET,
2513 	NI_DIG2_REGISTER_OFFSET,
2514 	NI_DIG3_REGISTER_OFFSET,
2515 	NI_DIG4_REGISTER_OFFSET,
2516 	NI_DIG5_REGISTER_OFFSET
2517 };
2518 
2519 static const unsigned ni_tx_offsets[] =
2520 {
2521 	NI_DCIO_UNIPHY0_UNIPHY_TX_CONTROL1,
2522 	NI_DCIO_UNIPHY1_UNIPHY_TX_CONTROL1,
2523 	NI_DCIO_UNIPHY2_UNIPHY_TX_CONTROL1,
2524 	NI_DCIO_UNIPHY3_UNIPHY_TX_CONTROL1,
2525 	NI_DCIO_UNIPHY4_UNIPHY_TX_CONTROL1,
2526 	NI_DCIO_UNIPHY5_UNIPHY_TX_CONTROL1
2527 };
2528 
2529 static const unsigned evergreen_dp_offsets[] =
2530 {
2531 	EVERGREEN_DP0_REGISTER_OFFSET,
2532 	EVERGREEN_DP1_REGISTER_OFFSET,
2533 	EVERGREEN_DP2_REGISTER_OFFSET,
2534 	EVERGREEN_DP3_REGISTER_OFFSET,
2535 	EVERGREEN_DP4_REGISTER_OFFSET,
2536 	EVERGREEN_DP5_REGISTER_OFFSET
2537 };
2538 
2539 static const unsigned evergreen_disp_int_status[] =
2540 {
2541 	DISP_INTERRUPT_STATUS,
2542 	DISP_INTERRUPT_STATUS_CONTINUE,
2543 	DISP_INTERRUPT_STATUS_CONTINUE2,
2544 	DISP_INTERRUPT_STATUS_CONTINUE3,
2545 	DISP_INTERRUPT_STATUS_CONTINUE4,
2546 	DISP_INTERRUPT_STATUS_CONTINUE5
2547 };
2548 
2549 /*
2550  * Assumption is that EVERGREEN_CRTC_MASTER_EN enable for requested crtc
2551  * We go from crtc to connector and it is not relible  since it
2552  * should be an opposite direction .If crtc is enable then
2553  * find the dig_fe which selects this crtc and insure that it enable.
2554  * if such dig_fe is found then find dig_be which selects found dig_be and
2555  * insure that it enable and in DP_SST mode.
2556  * if UNIPHY_PLL_CONTROL1.enable then we should disconnect timing
2557  * from dp symbols clocks .
2558  */
2559 static bool evergreen_is_dp_sst_stream_enabled(struct radeon_device *rdev,
2560 					       unsigned crtc_id, unsigned *ret_dig_fe)
2561 {
2562 	unsigned i;
2563 	unsigned dig_fe;
2564 	unsigned dig_be;
2565 	unsigned dig_en_be;
2566 	unsigned uniphy_pll;
2567 	unsigned digs_fe_selected;
2568 	unsigned dig_be_mode;
2569 	unsigned dig_fe_mask;
2570 	bool is_enabled = false;
2571 	bool found_crtc = false;
2572 
2573 	/* loop through all running dig_fe to find selected crtc */
2574 	for (i = 0; i < ARRAY_SIZE(ni_dig_offsets); i++) {
2575 		dig_fe = RREG32(NI_DIG_FE_CNTL + ni_dig_offsets[i]);
2576 		if (dig_fe & NI_DIG_FE_CNTL_SYMCLK_FE_ON &&
2577 		    crtc_id == NI_DIG_FE_CNTL_SOURCE_SELECT(dig_fe)) {
2578 			/* found running pipe */
2579 			found_crtc = true;
2580 			dig_fe_mask = 1 << i;
2581 			dig_fe = i;
2582 			break;
2583 		}
2584 	}
2585 
2586 	if (found_crtc) {
2587 		/* loop through all running dig_be to find selected dig_fe */
2588 		for (i = 0; i < ARRAY_SIZE(ni_dig_offsets); i++) {
2589 			dig_be = RREG32(NI_DIG_BE_CNTL + ni_dig_offsets[i]);
2590 			/* if dig_fe_selected by dig_be? */
2591 			digs_fe_selected = NI_DIG_BE_CNTL_FE_SOURCE_SELECT(dig_be);
2592 			dig_be_mode = NI_DIG_FE_CNTL_MODE(dig_be);
2593 			if (dig_fe_mask &  digs_fe_selected &&
2594 			    /* if dig_be in sst mode? */
2595 			    dig_be_mode == NI_DIG_BE_DPSST) {
2596 				dig_en_be = RREG32(NI_DIG_BE_EN_CNTL +
2597 						   ni_dig_offsets[i]);
2598 				uniphy_pll = RREG32(NI_DCIO_UNIPHY0_PLL_CONTROL1 +
2599 						    ni_tx_offsets[i]);
2600 				/* dig_be enable and tx is running */
2601 				if (dig_en_be & NI_DIG_BE_EN_CNTL_ENABLE &&
2602 				    dig_en_be & NI_DIG_BE_EN_CNTL_SYMBCLK_ON &&
2603 				    uniphy_pll & NI_DCIO_UNIPHY0_PLL_CONTROL1_ENABLE) {
2604 					is_enabled = true;
2605 					*ret_dig_fe = dig_fe;
2606 					break;
2607 				}
2608 			}
2609 		}
2610 	}
2611 
2612 	return is_enabled;
2613 }
2614 
2615 /*
2616  * Blank dig when in dp sst mode
2617  * Dig ignores crtc timing
2618  */
2619 static void evergreen_blank_dp_output(struct radeon_device *rdev,
2620 				      unsigned dig_fe)
2621 {
2622 	unsigned stream_ctrl;
2623 	unsigned fifo_ctrl;
2624 	unsigned counter = 0;
2625 
2626 	if (dig_fe >= ARRAY_SIZE(evergreen_dp_offsets)) {
2627 		DRM_ERROR("invalid dig_fe %d\n", dig_fe);
2628 		return;
2629 	}
2630 
2631 	stream_ctrl = RREG32(EVERGREEN_DP_VID_STREAM_CNTL +
2632 			     evergreen_dp_offsets[dig_fe]);
2633 	if (!(stream_ctrl & EVERGREEN_DP_VID_STREAM_CNTL_ENABLE)) {
2634 		DRM_ERROR("dig %d , should be enable\n", dig_fe);
2635 		return;
2636 	}
2637 
2638 	stream_ctrl &=~EVERGREEN_DP_VID_STREAM_CNTL_ENABLE;
2639 	WREG32(EVERGREEN_DP_VID_STREAM_CNTL +
2640 	       evergreen_dp_offsets[dig_fe], stream_ctrl);
2641 
2642 	stream_ctrl = RREG32(EVERGREEN_DP_VID_STREAM_CNTL +
2643 			     evergreen_dp_offsets[dig_fe]);
2644 	while (counter < 32 && stream_ctrl & EVERGREEN_DP_VID_STREAM_STATUS) {
2645 		msleep(1);
2646 		counter++;
2647 		stream_ctrl = RREG32(EVERGREEN_DP_VID_STREAM_CNTL +
2648 				     evergreen_dp_offsets[dig_fe]);
2649 	}
2650 	if (counter >= 32 )
2651 		DRM_ERROR("counter exceeds %d\n", counter);
2652 
2653 	fifo_ctrl = RREG32(EVERGREEN_DP_STEER_FIFO + evergreen_dp_offsets[dig_fe]);
2654 	fifo_ctrl |= EVERGREEN_DP_STEER_FIFO_RESET;
2655 	WREG32(EVERGREEN_DP_STEER_FIFO + evergreen_dp_offsets[dig_fe], fifo_ctrl);
2656 
2657 }
2658 
2659 void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save)
2660 {
2661 	u32 crtc_enabled, tmp, frame_count, blackout;
2662 	int i, j;
2663 	unsigned dig_fe;
2664 
2665 	if (!ASIC_IS_NODCE(rdev)) {
2666 		save->vga_render_control = RREG32(VGA_RENDER_CONTROL);
2667 		save->vga_hdp_control = RREG32(VGA_HDP_CONTROL);
2668 
2669 		/* disable VGA render */
2670 		WREG32(VGA_RENDER_CONTROL, 0);
2671 	}
2672 	/* blank the display controllers */
2673 	for (i = 0; i < rdev->num_crtc; i++) {
2674 		crtc_enabled = RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]) & EVERGREEN_CRTC_MASTER_EN;
2675 		if (crtc_enabled) {
2676 			save->crtc_enabled[i] = true;
2677 			if (ASIC_IS_DCE6(rdev)) {
2678 				tmp = RREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i]);
2679 				if (!(tmp & EVERGREEN_CRTC_BLANK_DATA_EN)) {
2680 					radeon_wait_for_vblank(rdev, i);
2681 					WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 1);
2682 					tmp |= EVERGREEN_CRTC_BLANK_DATA_EN;
2683 					WREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i], tmp);
2684 					WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 0);
2685 				}
2686 			} else {
2687 				tmp = RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]);
2688 				if (!(tmp & EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE)) {
2689 					radeon_wait_for_vblank(rdev, i);
2690 					WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 1);
2691 					tmp |= EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
2692 					WREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i], tmp);
2693 					WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 0);
2694 				}
2695 			}
2696 			/* wait for the next frame */
2697 			frame_count = radeon_get_vblank_counter(rdev, i);
2698 			for (j = 0; j < rdev->usec_timeout; j++) {
2699 				if (radeon_get_vblank_counter(rdev, i) != frame_count)
2700 					break;
2701 				udelay(1);
2702 			}
2703 			/*we should disable dig if it drives dp sst*/
2704 			/*but we are in radeon_device_init and the topology is unknown*/
2705 			/*and it is available after radeon_modeset_init*/
2706 			/*the following method radeon_atom_encoder_dpms_dig*/
2707 			/*does the job if we initialize it properly*/
2708 			/*for now we do it this manually*/
2709 			/**/
2710 			if (ASIC_IS_DCE5(rdev) &&
2711 			    evergreen_is_dp_sst_stream_enabled(rdev, i ,&dig_fe))
2712 				evergreen_blank_dp_output(rdev, dig_fe);
2713 			/*we could remove 6 lines below*/
2714 			/* XXX this is a hack to avoid strange behavior with EFI on certain systems */
2715 			WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 1);
2716 			tmp = RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]);
2717 			tmp &= ~EVERGREEN_CRTC_MASTER_EN;
2718 			WREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i], tmp);
2719 			WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 0);
2720 			save->crtc_enabled[i] = false;
2721 			/* ***** */
2722 		} else {
2723 			save->crtc_enabled[i] = false;
2724 		}
2725 	}
2726 
2727 	radeon_mc_wait_for_idle(rdev);
2728 
2729 	blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
2730 	if ((blackout & BLACKOUT_MODE_MASK) != 1) {
2731 		/* Block CPU access */
2732 		WREG32(BIF_FB_EN, 0);
2733 		/* blackout the MC */
2734 		blackout &= ~BLACKOUT_MODE_MASK;
2735 		WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
2736 	}
2737 	/* wait for the MC to settle */
2738 	udelay(100);
2739 
2740 	/* lock double buffered regs */
2741 	for (i = 0; i < rdev->num_crtc; i++) {
2742 		if (save->crtc_enabled[i]) {
2743 			tmp = RREG32(EVERGREEN_GRPH_UPDATE + crtc_offsets[i]);
2744 			if (!(tmp & EVERGREEN_GRPH_UPDATE_LOCK)) {
2745 				tmp |= EVERGREEN_GRPH_UPDATE_LOCK;
2746 				WREG32(EVERGREEN_GRPH_UPDATE + crtc_offsets[i], tmp);
2747 			}
2748 			tmp = RREG32(EVERGREEN_MASTER_UPDATE_LOCK + crtc_offsets[i]);
2749 			if (!(tmp & 1)) {
2750 				tmp |= 1;
2751 				WREG32(EVERGREEN_MASTER_UPDATE_LOCK + crtc_offsets[i], tmp);
2752 			}
2753 		}
2754 	}
2755 }
2756 
2757 void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save)
2758 {
2759 	u32 tmp, frame_count;
2760 	int i, j;
2761 
2762 	/* update crtc base addresses */
2763 	for (i = 0; i < rdev->num_crtc; i++) {
2764 		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + crtc_offsets[i],
2765 		       upper_32_bits(rdev->mc.vram_start));
2766 		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + crtc_offsets[i],
2767 		       upper_32_bits(rdev->mc.vram_start));
2768 		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + crtc_offsets[i],
2769 		       (u32)rdev->mc.vram_start);
2770 		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + crtc_offsets[i],
2771 		       (u32)rdev->mc.vram_start);
2772 	}
2773 
2774 	if (!ASIC_IS_NODCE(rdev)) {
2775 		WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS_HIGH, upper_32_bits(rdev->mc.vram_start));
2776 		WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS, (u32)rdev->mc.vram_start);
2777 	}
2778 
2779 	/* unlock regs and wait for update */
2780 	for (i = 0; i < rdev->num_crtc; i++) {
2781 		if (save->crtc_enabled[i]) {
2782 			tmp = RREG32(EVERGREEN_MASTER_UPDATE_MODE + crtc_offsets[i]);
2783 			if ((tmp & 0x7) != 0) {
2784 				tmp &= ~0x7;
2785 				WREG32(EVERGREEN_MASTER_UPDATE_MODE + crtc_offsets[i], tmp);
2786 			}
2787 			tmp = RREG32(EVERGREEN_GRPH_UPDATE + crtc_offsets[i]);
2788 			if (tmp & EVERGREEN_GRPH_UPDATE_LOCK) {
2789 				tmp &= ~EVERGREEN_GRPH_UPDATE_LOCK;
2790 				WREG32(EVERGREEN_GRPH_UPDATE + crtc_offsets[i], tmp);
2791 			}
2792 			tmp = RREG32(EVERGREEN_MASTER_UPDATE_LOCK + crtc_offsets[i]);
2793 			if (tmp & 1) {
2794 				tmp &= ~1;
2795 				WREG32(EVERGREEN_MASTER_UPDATE_LOCK + crtc_offsets[i], tmp);
2796 			}
2797 			for (j = 0; j < rdev->usec_timeout; j++) {
2798 				tmp = RREG32(EVERGREEN_GRPH_UPDATE + crtc_offsets[i]);
2799 				if ((tmp & EVERGREEN_GRPH_SURFACE_UPDATE_PENDING) == 0)
2800 					break;
2801 				udelay(1);
2802 			}
2803 		}
2804 	}
2805 
2806 	/* unblackout the MC */
2807 	tmp = RREG32(MC_SHARED_BLACKOUT_CNTL);
2808 	tmp &= ~BLACKOUT_MODE_MASK;
2809 	WREG32(MC_SHARED_BLACKOUT_CNTL, tmp);
2810 	/* allow CPU access */
2811 	WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
2812 
2813 	for (i = 0; i < rdev->num_crtc; i++) {
2814 		if (save->crtc_enabled[i]) {
2815 			if (ASIC_IS_DCE6(rdev)) {
2816 				tmp = RREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i]);
2817 				tmp &= ~EVERGREEN_CRTC_BLANK_DATA_EN;
2818 				WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 1);
2819 				WREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i], tmp);
2820 				WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 0);
2821 			} else {
2822 				tmp = RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]);
2823 				tmp &= ~EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
2824 				WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 1);
2825 				WREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i], tmp);
2826 				WREG32(EVERGREEN_CRTC_UPDATE_LOCK + crtc_offsets[i], 0);
2827 			}
2828 			/* wait for the next frame */
2829 			frame_count = radeon_get_vblank_counter(rdev, i);
2830 			for (j = 0; j < rdev->usec_timeout; j++) {
2831 				if (radeon_get_vblank_counter(rdev, i) != frame_count)
2832 					break;
2833 				udelay(1);
2834 			}
2835 		}
2836 	}
2837 	if (!ASIC_IS_NODCE(rdev)) {
2838 		/* Unlock vga access */
2839 		WREG32(VGA_HDP_CONTROL, save->vga_hdp_control);
2840 		mdelay(1);
2841 		WREG32(VGA_RENDER_CONTROL, save->vga_render_control);
2842 	}
2843 }
2844 
2845 void evergreen_mc_program(struct radeon_device *rdev)
2846 {
2847 	struct evergreen_mc_save save;
2848 	u32 tmp;
2849 	int i, j;
2850 
2851 	/* Initialize HDP */
2852 	for (i = 0, j = 0; i < 32; i++, j += 0x18) {
2853 		WREG32((0x2c14 + j), 0x00000000);
2854 		WREG32((0x2c18 + j), 0x00000000);
2855 		WREG32((0x2c1c + j), 0x00000000);
2856 		WREG32((0x2c20 + j), 0x00000000);
2857 		WREG32((0x2c24 + j), 0x00000000);
2858 	}
2859 	WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
2860 
2861 	evergreen_mc_stop(rdev, &save);
2862 	if (evergreen_mc_wait_for_idle(rdev)) {
2863 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2864 	}
2865 	/* Lockout access through VGA aperture*/
2866 	WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
2867 	/* Update configuration */
2868 	if (rdev->flags & RADEON_IS_AGP) {
2869 		if (rdev->mc.vram_start < rdev->mc.gtt_start) {
2870 			/* VRAM before AGP */
2871 			WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
2872 				rdev->mc.vram_start >> 12);
2873 			WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
2874 				rdev->mc.gtt_end >> 12);
2875 		} else {
2876 			/* VRAM after AGP */
2877 			WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
2878 				rdev->mc.gtt_start >> 12);
2879 			WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
2880 				rdev->mc.vram_end >> 12);
2881 		}
2882 	} else {
2883 		WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
2884 			rdev->mc.vram_start >> 12);
2885 		WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
2886 			rdev->mc.vram_end >> 12);
2887 	}
2888 	WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, rdev->vram_scratch.gpu_addr >> 12);
2889 	/* llano/ontario only */
2890 	if ((rdev->family == CHIP_PALM) ||
2891 	    (rdev->family == CHIP_SUMO) ||
2892 	    (rdev->family == CHIP_SUMO2)) {
2893 		tmp = RREG32(MC_FUS_VM_FB_OFFSET) & 0x000FFFFF;
2894 		tmp |= ((rdev->mc.vram_end >> 20) & 0xF) << 24;
2895 		tmp |= ((rdev->mc.vram_start >> 20) & 0xF) << 20;
2896 		WREG32(MC_FUS_VM_FB_OFFSET, tmp);
2897 	}
2898 	tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
2899 	tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
2900 	WREG32(MC_VM_FB_LOCATION, tmp);
2901 	WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
2902 	WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
2903 	WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
2904 	if (rdev->flags & RADEON_IS_AGP) {
2905 		WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 16);
2906 		WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 16);
2907 		WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
2908 	} else {
2909 		WREG32(MC_VM_AGP_BASE, 0);
2910 		WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
2911 		WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
2912 	}
2913 	if (evergreen_mc_wait_for_idle(rdev)) {
2914 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2915 	}
2916 	evergreen_mc_resume(rdev, &save);
2917 	/* we need to own VRAM, so turn off the VGA renderer here
2918 	 * to stop it overwriting our objects */
2919 	rv515_vga_render_disable(rdev);
2920 }
2921 
2922 /*
2923  * CP.
2924  */
2925 void evergreen_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
2926 {
2927 	struct radeon_ring *ring = &rdev->ring[ib->ring];
2928 	u32 next_rptr;
2929 
2930 	/* set to DX10/11 mode */
2931 	radeon_ring_write(ring, PACKET3(PACKET3_MODE_CONTROL, 0));
2932 	radeon_ring_write(ring, 1);
2933 
2934 	if (ring->rptr_save_reg) {
2935 		next_rptr = ring->wptr + 3 + 4;
2936 		radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2937 		radeon_ring_write(ring, ((ring->rptr_save_reg -
2938 					  PACKET3_SET_CONFIG_REG_START) >> 2));
2939 		radeon_ring_write(ring, next_rptr);
2940 	} else if (rdev->wb.enabled) {
2941 		next_rptr = ring->wptr + 5 + 4;
2942 		radeon_ring_write(ring, PACKET3(PACKET3_MEM_WRITE, 3));
2943 		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
2944 		radeon_ring_write(ring, (upper_32_bits(ring->next_rptr_gpu_addr) & 0xff) | (1 << 18));
2945 		radeon_ring_write(ring, next_rptr);
2946 		radeon_ring_write(ring, 0);
2947 	}
2948 
2949 	radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2950 	radeon_ring_write(ring,
2951 #ifdef __BIG_ENDIAN
2952 			  (2 << 0) |
2953 #endif
2954 			  (ib->gpu_addr & 0xFFFFFFFC));
2955 	radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFF);
2956 	radeon_ring_write(ring, ib->length_dw);
2957 }
2958 
2959 
2960 static int evergreen_cp_load_microcode(struct radeon_device *rdev)
2961 {
2962 	const __be32 *fw_data;
2963 	int i;
2964 
2965 	if (!rdev->me_fw || !rdev->pfp_fw)
2966 		return -EINVAL;
2967 
2968 	r700_cp_stop(rdev);
2969 	WREG32(CP_RB_CNTL,
2970 #ifdef __BIG_ENDIAN
2971 	       BUF_SWAP_32BIT |
2972 #endif
2973 	       RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
2974 
2975 	fw_data = (const __be32 *)rdev->pfp_fw->data;
2976 	WREG32(CP_PFP_UCODE_ADDR, 0);
2977 	for (i = 0; i < EVERGREEN_PFP_UCODE_SIZE; i++)
2978 		WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
2979 	WREG32(CP_PFP_UCODE_ADDR, 0);
2980 
2981 	fw_data = (const __be32 *)rdev->me_fw->data;
2982 	WREG32(CP_ME_RAM_WADDR, 0);
2983 	for (i = 0; i < EVERGREEN_PM4_UCODE_SIZE; i++)
2984 		WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
2985 
2986 	WREG32(CP_PFP_UCODE_ADDR, 0);
2987 	WREG32(CP_ME_RAM_WADDR, 0);
2988 	WREG32(CP_ME_RAM_RADDR, 0);
2989 	return 0;
2990 }
2991 
2992 static int evergreen_cp_start(struct radeon_device *rdev)
2993 {
2994 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2995 	int r, i;
2996 	uint32_t cp_me;
2997 
2998 	r = radeon_ring_lock(rdev, ring, 7);
2999 	if (r) {
3000 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3001 		return r;
3002 	}
3003 	radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
3004 	radeon_ring_write(ring, 0x1);
3005 	radeon_ring_write(ring, 0x0);
3006 	radeon_ring_write(ring, rdev->config.evergreen.max_hw_contexts - 1);
3007 	radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
3008 	radeon_ring_write(ring, 0);
3009 	radeon_ring_write(ring, 0);
3010 	radeon_ring_unlock_commit(rdev, ring, false);
3011 
3012 	cp_me = 0xff;
3013 	WREG32(CP_ME_CNTL, cp_me);
3014 
3015 	r = radeon_ring_lock(rdev, ring, evergreen_default_size + 19);
3016 	if (r) {
3017 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3018 		return r;
3019 	}
3020 
3021 	/* setup clear context state */
3022 	radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3023 	radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3024 
3025 	for (i = 0; i < evergreen_default_size; i++)
3026 		radeon_ring_write(ring, evergreen_default_state[i]);
3027 
3028 	radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3029 	radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3030 
3031 	/* set clear context state */
3032 	radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3033 	radeon_ring_write(ring, 0);
3034 
3035 	/* SQ_VTX_BASE_VTX_LOC */
3036 	radeon_ring_write(ring, 0xc0026f00);
3037 	radeon_ring_write(ring, 0x00000000);
3038 	radeon_ring_write(ring, 0x00000000);
3039 	radeon_ring_write(ring, 0x00000000);
3040 
3041 	/* Clear consts */
3042 	radeon_ring_write(ring, 0xc0036f00);
3043 	radeon_ring_write(ring, 0x00000bc4);
3044 	radeon_ring_write(ring, 0xffffffff);
3045 	radeon_ring_write(ring, 0xffffffff);
3046 	radeon_ring_write(ring, 0xffffffff);
3047 
3048 	radeon_ring_write(ring, 0xc0026900);
3049 	radeon_ring_write(ring, 0x00000316);
3050 	radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
3051 	radeon_ring_write(ring, 0x00000010); /*  */
3052 
3053 	radeon_ring_unlock_commit(rdev, ring, false);
3054 
3055 	return 0;
3056 }
3057 
3058 static int evergreen_cp_resume(struct radeon_device *rdev)
3059 {
3060 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3061 	u32 tmp;
3062 	u32 rb_bufsz;
3063 	int r;
3064 
3065 	/* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
3066 	WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
3067 				 SOFT_RESET_PA |
3068 				 SOFT_RESET_SH |
3069 				 SOFT_RESET_VGT |
3070 				 SOFT_RESET_SPI |
3071 				 SOFT_RESET_SX));
3072 	RREG32(GRBM_SOFT_RESET);
3073 	mdelay(15);
3074 	WREG32(GRBM_SOFT_RESET, 0);
3075 	RREG32(GRBM_SOFT_RESET);
3076 
3077 	/* Set ring buffer size */
3078 	rb_bufsz = order_base_2(ring->ring_size / 8);
3079 	tmp = (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
3080 #ifdef __BIG_ENDIAN
3081 	tmp |= BUF_SWAP_32BIT;
3082 #endif
3083 	WREG32(CP_RB_CNTL, tmp);
3084 	WREG32(CP_SEM_WAIT_TIMER, 0x0);
3085 	WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
3086 
3087 	/* Set the write pointer delay */
3088 	WREG32(CP_RB_WPTR_DELAY, 0);
3089 
3090 	/* Initialize the ring buffer's read and write pointers */
3091 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
3092 	WREG32(CP_RB_RPTR_WR, 0);
3093 	ring->wptr = 0;
3094 	WREG32(CP_RB_WPTR, ring->wptr);
3095 
3096 	/* set the wb address whether it's enabled or not */
3097 	WREG32(CP_RB_RPTR_ADDR,
3098 	       ((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC));
3099 	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
3100 	WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
3101 
3102 	if (rdev->wb.enabled)
3103 		WREG32(SCRATCH_UMSK, 0xff);
3104 	else {
3105 		tmp |= RB_NO_UPDATE;
3106 		WREG32(SCRATCH_UMSK, 0);
3107 	}
3108 
3109 	mdelay(1);
3110 	WREG32(CP_RB_CNTL, tmp);
3111 
3112 	WREG32(CP_RB_BASE, ring->gpu_addr >> 8);
3113 	WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
3114 
3115 	evergreen_cp_start(rdev);
3116 	ring->ready = true;
3117 	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
3118 	if (r) {
3119 		ring->ready = false;
3120 		return r;
3121 	}
3122 	return 0;
3123 }
3124 
3125 /*
3126  * Core functions
3127  */
3128 static void evergreen_gpu_init(struct radeon_device *rdev)
3129 {
3130 	u32 gb_addr_config;
3131 	u32 mc_arb_ramcfg;
3132 	u32 sx_debug_1;
3133 	u32 smx_dc_ctl0;
3134 	u32 sq_config;
3135 	u32 sq_lds_resource_mgmt;
3136 	u32 sq_gpr_resource_mgmt_1;
3137 	u32 sq_gpr_resource_mgmt_2;
3138 	u32 sq_gpr_resource_mgmt_3;
3139 	u32 sq_thread_resource_mgmt;
3140 	u32 sq_thread_resource_mgmt_2;
3141 	u32 sq_stack_resource_mgmt_1;
3142 	u32 sq_stack_resource_mgmt_2;
3143 	u32 sq_stack_resource_mgmt_3;
3144 	u32 vgt_cache_invalidation;
3145 	u32 hdp_host_path_cntl, tmp;
3146 	u32 disabled_rb_mask;
3147 	int i, j, ps_thread_count;
3148 
3149 	switch (rdev->family) {
3150 	case CHIP_CYPRESS:
3151 	case CHIP_HEMLOCK:
3152 		rdev->config.evergreen.num_ses = 2;
3153 		rdev->config.evergreen.max_pipes = 4;
3154 		rdev->config.evergreen.max_tile_pipes = 8;
3155 		rdev->config.evergreen.max_simds = 10;
3156 		rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
3157 		rdev->config.evergreen.max_gprs = 256;
3158 		rdev->config.evergreen.max_threads = 248;
3159 		rdev->config.evergreen.max_gs_threads = 32;
3160 		rdev->config.evergreen.max_stack_entries = 512;
3161 		rdev->config.evergreen.sx_num_of_sets = 4;
3162 		rdev->config.evergreen.sx_max_export_size = 256;
3163 		rdev->config.evergreen.sx_max_export_pos_size = 64;
3164 		rdev->config.evergreen.sx_max_export_smx_size = 192;
3165 		rdev->config.evergreen.max_hw_contexts = 8;
3166 		rdev->config.evergreen.sq_num_cf_insts = 2;
3167 
3168 		rdev->config.evergreen.sc_prim_fifo_size = 0x100;
3169 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3170 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3171 		gb_addr_config = CYPRESS_GB_ADDR_CONFIG_GOLDEN;
3172 		break;
3173 	case CHIP_JUNIPER:
3174 		rdev->config.evergreen.num_ses = 1;
3175 		rdev->config.evergreen.max_pipes = 4;
3176 		rdev->config.evergreen.max_tile_pipes = 4;
3177 		rdev->config.evergreen.max_simds = 10;
3178 		rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
3179 		rdev->config.evergreen.max_gprs = 256;
3180 		rdev->config.evergreen.max_threads = 248;
3181 		rdev->config.evergreen.max_gs_threads = 32;
3182 		rdev->config.evergreen.max_stack_entries = 512;
3183 		rdev->config.evergreen.sx_num_of_sets = 4;
3184 		rdev->config.evergreen.sx_max_export_size = 256;
3185 		rdev->config.evergreen.sx_max_export_pos_size = 64;
3186 		rdev->config.evergreen.sx_max_export_smx_size = 192;
3187 		rdev->config.evergreen.max_hw_contexts = 8;
3188 		rdev->config.evergreen.sq_num_cf_insts = 2;
3189 
3190 		rdev->config.evergreen.sc_prim_fifo_size = 0x100;
3191 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3192 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3193 		gb_addr_config = JUNIPER_GB_ADDR_CONFIG_GOLDEN;
3194 		break;
3195 	case CHIP_REDWOOD:
3196 		rdev->config.evergreen.num_ses = 1;
3197 		rdev->config.evergreen.max_pipes = 4;
3198 		rdev->config.evergreen.max_tile_pipes = 4;
3199 		rdev->config.evergreen.max_simds = 5;
3200 		rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
3201 		rdev->config.evergreen.max_gprs = 256;
3202 		rdev->config.evergreen.max_threads = 248;
3203 		rdev->config.evergreen.max_gs_threads = 32;
3204 		rdev->config.evergreen.max_stack_entries = 256;
3205 		rdev->config.evergreen.sx_num_of_sets = 4;
3206 		rdev->config.evergreen.sx_max_export_size = 256;
3207 		rdev->config.evergreen.sx_max_export_pos_size = 64;
3208 		rdev->config.evergreen.sx_max_export_smx_size = 192;
3209 		rdev->config.evergreen.max_hw_contexts = 8;
3210 		rdev->config.evergreen.sq_num_cf_insts = 2;
3211 
3212 		rdev->config.evergreen.sc_prim_fifo_size = 0x100;
3213 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3214 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3215 		gb_addr_config = REDWOOD_GB_ADDR_CONFIG_GOLDEN;
3216 		break;
3217 	case CHIP_CEDAR:
3218 	default:
3219 		rdev->config.evergreen.num_ses = 1;
3220 		rdev->config.evergreen.max_pipes = 2;
3221 		rdev->config.evergreen.max_tile_pipes = 2;
3222 		rdev->config.evergreen.max_simds = 2;
3223 		rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
3224 		rdev->config.evergreen.max_gprs = 256;
3225 		rdev->config.evergreen.max_threads = 192;
3226 		rdev->config.evergreen.max_gs_threads = 16;
3227 		rdev->config.evergreen.max_stack_entries = 256;
3228 		rdev->config.evergreen.sx_num_of_sets = 4;
3229 		rdev->config.evergreen.sx_max_export_size = 128;
3230 		rdev->config.evergreen.sx_max_export_pos_size = 32;
3231 		rdev->config.evergreen.sx_max_export_smx_size = 96;
3232 		rdev->config.evergreen.max_hw_contexts = 4;
3233 		rdev->config.evergreen.sq_num_cf_insts = 1;
3234 
3235 		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
3236 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3237 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3238 		gb_addr_config = CEDAR_GB_ADDR_CONFIG_GOLDEN;
3239 		break;
3240 	case CHIP_PALM:
3241 		rdev->config.evergreen.num_ses = 1;
3242 		rdev->config.evergreen.max_pipes = 2;
3243 		rdev->config.evergreen.max_tile_pipes = 2;
3244 		rdev->config.evergreen.max_simds = 2;
3245 		rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
3246 		rdev->config.evergreen.max_gprs = 256;
3247 		rdev->config.evergreen.max_threads = 192;
3248 		rdev->config.evergreen.max_gs_threads = 16;
3249 		rdev->config.evergreen.max_stack_entries = 256;
3250 		rdev->config.evergreen.sx_num_of_sets = 4;
3251 		rdev->config.evergreen.sx_max_export_size = 128;
3252 		rdev->config.evergreen.sx_max_export_pos_size = 32;
3253 		rdev->config.evergreen.sx_max_export_smx_size = 96;
3254 		rdev->config.evergreen.max_hw_contexts = 4;
3255 		rdev->config.evergreen.sq_num_cf_insts = 1;
3256 
3257 		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
3258 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3259 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3260 		gb_addr_config = CEDAR_GB_ADDR_CONFIG_GOLDEN;
3261 		break;
3262 	case CHIP_SUMO:
3263 		rdev->config.evergreen.num_ses = 1;
3264 		rdev->config.evergreen.max_pipes = 4;
3265 		rdev->config.evergreen.max_tile_pipes = 4;
3266 		if (rdev->pdev->device == 0x9648)
3267 			rdev->config.evergreen.max_simds = 3;
3268 		else if ((rdev->pdev->device == 0x9647) ||
3269 			 (rdev->pdev->device == 0x964a))
3270 			rdev->config.evergreen.max_simds = 4;
3271 		else
3272 			rdev->config.evergreen.max_simds = 5;
3273 		rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
3274 		rdev->config.evergreen.max_gprs = 256;
3275 		rdev->config.evergreen.max_threads = 248;
3276 		rdev->config.evergreen.max_gs_threads = 32;
3277 		rdev->config.evergreen.max_stack_entries = 256;
3278 		rdev->config.evergreen.sx_num_of_sets = 4;
3279 		rdev->config.evergreen.sx_max_export_size = 256;
3280 		rdev->config.evergreen.sx_max_export_pos_size = 64;
3281 		rdev->config.evergreen.sx_max_export_smx_size = 192;
3282 		rdev->config.evergreen.max_hw_contexts = 8;
3283 		rdev->config.evergreen.sq_num_cf_insts = 2;
3284 
3285 		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
3286 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3287 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3288 		gb_addr_config = SUMO_GB_ADDR_CONFIG_GOLDEN;
3289 		break;
3290 	case CHIP_SUMO2:
3291 		rdev->config.evergreen.num_ses = 1;
3292 		rdev->config.evergreen.max_pipes = 4;
3293 		rdev->config.evergreen.max_tile_pipes = 4;
3294 		rdev->config.evergreen.max_simds = 2;
3295 		rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
3296 		rdev->config.evergreen.max_gprs = 256;
3297 		rdev->config.evergreen.max_threads = 248;
3298 		rdev->config.evergreen.max_gs_threads = 32;
3299 		rdev->config.evergreen.max_stack_entries = 512;
3300 		rdev->config.evergreen.sx_num_of_sets = 4;
3301 		rdev->config.evergreen.sx_max_export_size = 256;
3302 		rdev->config.evergreen.sx_max_export_pos_size = 64;
3303 		rdev->config.evergreen.sx_max_export_smx_size = 192;
3304 		rdev->config.evergreen.max_hw_contexts = 4;
3305 		rdev->config.evergreen.sq_num_cf_insts = 2;
3306 
3307 		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
3308 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3309 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3310 		gb_addr_config = SUMO2_GB_ADDR_CONFIG_GOLDEN;
3311 		break;
3312 	case CHIP_BARTS:
3313 		rdev->config.evergreen.num_ses = 2;
3314 		rdev->config.evergreen.max_pipes = 4;
3315 		rdev->config.evergreen.max_tile_pipes = 8;
3316 		rdev->config.evergreen.max_simds = 7;
3317 		rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
3318 		rdev->config.evergreen.max_gprs = 256;
3319 		rdev->config.evergreen.max_threads = 248;
3320 		rdev->config.evergreen.max_gs_threads = 32;
3321 		rdev->config.evergreen.max_stack_entries = 512;
3322 		rdev->config.evergreen.sx_num_of_sets = 4;
3323 		rdev->config.evergreen.sx_max_export_size = 256;
3324 		rdev->config.evergreen.sx_max_export_pos_size = 64;
3325 		rdev->config.evergreen.sx_max_export_smx_size = 192;
3326 		rdev->config.evergreen.max_hw_contexts = 8;
3327 		rdev->config.evergreen.sq_num_cf_insts = 2;
3328 
3329 		rdev->config.evergreen.sc_prim_fifo_size = 0x100;
3330 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3331 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3332 		gb_addr_config = BARTS_GB_ADDR_CONFIG_GOLDEN;
3333 		break;
3334 	case CHIP_TURKS:
3335 		rdev->config.evergreen.num_ses = 1;
3336 		rdev->config.evergreen.max_pipes = 4;
3337 		rdev->config.evergreen.max_tile_pipes = 4;
3338 		rdev->config.evergreen.max_simds = 6;
3339 		rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
3340 		rdev->config.evergreen.max_gprs = 256;
3341 		rdev->config.evergreen.max_threads = 248;
3342 		rdev->config.evergreen.max_gs_threads = 32;
3343 		rdev->config.evergreen.max_stack_entries = 256;
3344 		rdev->config.evergreen.sx_num_of_sets = 4;
3345 		rdev->config.evergreen.sx_max_export_size = 256;
3346 		rdev->config.evergreen.sx_max_export_pos_size = 64;
3347 		rdev->config.evergreen.sx_max_export_smx_size = 192;
3348 		rdev->config.evergreen.max_hw_contexts = 8;
3349 		rdev->config.evergreen.sq_num_cf_insts = 2;
3350 
3351 		rdev->config.evergreen.sc_prim_fifo_size = 0x100;
3352 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3353 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3354 		gb_addr_config = TURKS_GB_ADDR_CONFIG_GOLDEN;
3355 		break;
3356 	case CHIP_CAICOS:
3357 		rdev->config.evergreen.num_ses = 1;
3358 		rdev->config.evergreen.max_pipes = 2;
3359 		rdev->config.evergreen.max_tile_pipes = 2;
3360 		rdev->config.evergreen.max_simds = 2;
3361 		rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
3362 		rdev->config.evergreen.max_gprs = 256;
3363 		rdev->config.evergreen.max_threads = 192;
3364 		rdev->config.evergreen.max_gs_threads = 16;
3365 		rdev->config.evergreen.max_stack_entries = 256;
3366 		rdev->config.evergreen.sx_num_of_sets = 4;
3367 		rdev->config.evergreen.sx_max_export_size = 128;
3368 		rdev->config.evergreen.sx_max_export_pos_size = 32;
3369 		rdev->config.evergreen.sx_max_export_smx_size = 96;
3370 		rdev->config.evergreen.max_hw_contexts = 4;
3371 		rdev->config.evergreen.sq_num_cf_insts = 1;
3372 
3373 		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
3374 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
3375 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
3376 		gb_addr_config = CAICOS_GB_ADDR_CONFIG_GOLDEN;
3377 		break;
3378 	}
3379 
3380 	/* Initialize HDP */
3381 	for (i = 0, j = 0; i < 32; i++, j += 0x18) {
3382 		WREG32((0x2c14 + j), 0x00000000);
3383 		WREG32((0x2c18 + j), 0x00000000);
3384 		WREG32((0x2c1c + j), 0x00000000);
3385 		WREG32((0x2c20 + j), 0x00000000);
3386 		WREG32((0x2c24 + j), 0x00000000);
3387 	}
3388 
3389 	WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
3390 	WREG32(SRBM_INT_CNTL, 0x1);
3391 	WREG32(SRBM_INT_ACK, 0x1);
3392 
3393 	evergreen_fix_pci_max_read_req_size(rdev);
3394 
3395 	RREG32(MC_SHARED_CHMAP);
3396 	if ((rdev->family == CHIP_PALM) ||
3397 	    (rdev->family == CHIP_SUMO) ||
3398 	    (rdev->family == CHIP_SUMO2))
3399 		mc_arb_ramcfg = RREG32(FUS_MC_ARB_RAMCFG);
3400 	else
3401 		mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
3402 
3403 	/* setup tiling info dword.  gb_addr_config is not adequate since it does
3404 	 * not have bank info, so create a custom tiling dword.
3405 	 * bits 3:0   num_pipes
3406 	 * bits 7:4   num_banks
3407 	 * bits 11:8  group_size
3408 	 * bits 15:12 row_size
3409 	 */
3410 	rdev->config.evergreen.tile_config = 0;
3411 	switch (rdev->config.evergreen.max_tile_pipes) {
3412 	case 1:
3413 	default:
3414 		rdev->config.evergreen.tile_config |= (0 << 0);
3415 		break;
3416 	case 2:
3417 		rdev->config.evergreen.tile_config |= (1 << 0);
3418 		break;
3419 	case 4:
3420 		rdev->config.evergreen.tile_config |= (2 << 0);
3421 		break;
3422 	case 8:
3423 		rdev->config.evergreen.tile_config |= (3 << 0);
3424 		break;
3425 	}
3426 	/* num banks is 8 on all fusion asics. 0 = 4, 1 = 8, 2 = 16 */
3427 	if (rdev->flags & RADEON_IS_IGP)
3428 		rdev->config.evergreen.tile_config |= 1 << 4;
3429 	else {
3430 		switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
3431 		case 0: /* four banks */
3432 			rdev->config.evergreen.tile_config |= 0 << 4;
3433 			break;
3434 		case 1: /* eight banks */
3435 			rdev->config.evergreen.tile_config |= 1 << 4;
3436 			break;
3437 		case 2: /* sixteen banks */
3438 		default:
3439 			rdev->config.evergreen.tile_config |= 2 << 4;
3440 			break;
3441 		}
3442 	}
3443 	rdev->config.evergreen.tile_config |= 0 << 8;
3444 	rdev->config.evergreen.tile_config |=
3445 		((gb_addr_config & 0x30000000) >> 28) << 12;
3446 
3447 	if ((rdev->family >= CHIP_CEDAR) && (rdev->family <= CHIP_HEMLOCK)) {
3448 		u32 efuse_straps_4;
3449 		u32 efuse_straps_3;
3450 
3451 		efuse_straps_4 = RREG32_RCU(0x204);
3452 		efuse_straps_3 = RREG32_RCU(0x203);
3453 		tmp = (((efuse_straps_4 & 0xf) << 4) |
3454 		      ((efuse_straps_3 & 0xf0000000) >> 28));
3455 	} else {
3456 		tmp = 0;
3457 		for (i = (rdev->config.evergreen.num_ses - 1); i >= 0; i--) {
3458 			u32 rb_disable_bitmap;
3459 
3460 			WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
3461 			WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
3462 			rb_disable_bitmap = (RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000) >> 16;
3463 			tmp <<= 4;
3464 			tmp |= rb_disable_bitmap;
3465 		}
3466 	}
3467 	/* enabled rb are just the one not disabled :) */
3468 	disabled_rb_mask = tmp;
3469 	tmp = 0;
3470 	for (i = 0; i < rdev->config.evergreen.max_backends; i++)
3471 		tmp |= (1 << i);
3472 	/* if all the backends are disabled, fix it up here */
3473 	if ((disabled_rb_mask & tmp) == tmp) {
3474 		for (i = 0; i < rdev->config.evergreen.max_backends; i++)
3475 			disabled_rb_mask &= ~(1 << i);
3476 	}
3477 
3478 	for (i = 0; i < rdev->config.evergreen.num_ses; i++) {
3479 		u32 simd_disable_bitmap;
3480 
3481 		WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
3482 		WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
3483 		simd_disable_bitmap = (RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffff0000) >> 16;
3484 		simd_disable_bitmap |= 0xffffffff << rdev->config.evergreen.max_simds;
3485 		tmp <<= 16;
3486 		tmp |= simd_disable_bitmap;
3487 	}
3488 	rdev->config.evergreen.active_simds = hweight32(~tmp);
3489 
3490 	WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
3491 	WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
3492 
3493 	WREG32(GB_ADDR_CONFIG, gb_addr_config);
3494 	WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
3495 	WREG32(HDP_ADDR_CONFIG, gb_addr_config);
3496 	WREG32(DMA_TILING_CONFIG, gb_addr_config);
3497 	WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config);
3498 	WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config);
3499 	WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config);
3500 
3501 	if ((rdev->config.evergreen.max_backends == 1) &&
3502 	    (rdev->flags & RADEON_IS_IGP)) {
3503 		if ((disabled_rb_mask & 3) == 1) {
3504 			/* RB0 disabled, RB1 enabled */
3505 			tmp = 0x11111111;
3506 		} else {
3507 			/* RB1 disabled, RB0 enabled */
3508 			tmp = 0x00000000;
3509 		}
3510 	} else {
3511 		tmp = gb_addr_config & NUM_PIPES_MASK;
3512 		tmp = r6xx_remap_render_backend(rdev, tmp, rdev->config.evergreen.max_backends,
3513 						EVERGREEN_MAX_BACKENDS, disabled_rb_mask);
3514 	}
3515 	rdev->config.evergreen.backend_map = tmp;
3516 	WREG32(GB_BACKEND_MAP, tmp);
3517 
3518 	WREG32(CGTS_SYS_TCC_DISABLE, 0);
3519 	WREG32(CGTS_TCC_DISABLE, 0);
3520 	WREG32(CGTS_USER_SYS_TCC_DISABLE, 0);
3521 	WREG32(CGTS_USER_TCC_DISABLE, 0);
3522 
3523 	/* set HW defaults for 3D engine */
3524 	WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) |
3525 				     ROQ_IB2_START(0x2b)));
3526 
3527 	WREG32(CP_MEQ_THRESHOLDS, STQ_SPLIT(0x30));
3528 
3529 	WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO |
3530 			     SYNC_GRADIENT |
3531 			     SYNC_WALKER |
3532 			     SYNC_ALIGNER));
3533 
3534 	sx_debug_1 = RREG32(SX_DEBUG_1);
3535 	sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS;
3536 	WREG32(SX_DEBUG_1, sx_debug_1);
3537 
3538 
3539 	smx_dc_ctl0 = RREG32(SMX_DC_CTL0);
3540 	smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff);
3541 	smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.evergreen.sx_num_of_sets);
3542 	WREG32(SMX_DC_CTL0, smx_dc_ctl0);
3543 
3544 	if (rdev->family <= CHIP_SUMO2)
3545 		WREG32(SMX_SAR_CTL0, 0x00010000);
3546 
3547 	WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_size / 4) - 1) |
3548 					POSITION_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_pos_size / 4) - 1) |
3549 					SMX_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_smx_size / 4) - 1)));
3550 
3551 	WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.evergreen.sc_prim_fifo_size) |
3552 				 SC_HIZ_TILE_FIFO_SIZE(rdev->config.evergreen.sc_hiz_tile_fifo_size) |
3553 				 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.evergreen.sc_earlyz_tile_fifo_size)));
3554 
3555 	WREG32(VGT_NUM_INSTANCES, 1);
3556 	WREG32(SPI_CONFIG_CNTL, 0);
3557 	WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
3558 	WREG32(CP_PERFMON_CNTL, 0);
3559 
3560 	WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.evergreen.sq_num_cf_insts) |
3561 				  FETCH_FIFO_HIWATER(0x4) |
3562 				  DONE_FIFO_HIWATER(0xe0) |
3563 				  ALU_UPDATE_FIFO_HIWATER(0x8)));
3564 
3565 	sq_config = RREG32(SQ_CONFIG);
3566 	sq_config &= ~(PS_PRIO(3) |
3567 		       VS_PRIO(3) |
3568 		       GS_PRIO(3) |
3569 		       ES_PRIO(3));
3570 	sq_config |= (VC_ENABLE |
3571 		      EXPORT_SRC_C |
3572 		      PS_PRIO(0) |
3573 		      VS_PRIO(1) |
3574 		      GS_PRIO(2) |
3575 		      ES_PRIO(3));
3576 
3577 	switch (rdev->family) {
3578 	case CHIP_CEDAR:
3579 	case CHIP_PALM:
3580 	case CHIP_SUMO:
3581 	case CHIP_SUMO2:
3582 	case CHIP_CAICOS:
3583 		/* no vertex cache */
3584 		sq_config &= ~VC_ENABLE;
3585 		break;
3586 	default:
3587 		break;
3588 	}
3589 
3590 	sq_lds_resource_mgmt = RREG32(SQ_LDS_RESOURCE_MGMT);
3591 
3592 	sq_gpr_resource_mgmt_1 = NUM_PS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2))* 12 / 32);
3593 	sq_gpr_resource_mgmt_1 |= NUM_VS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 6 / 32);
3594 	sq_gpr_resource_mgmt_1 |= NUM_CLAUSE_TEMP_GPRS(4);
3595 	sq_gpr_resource_mgmt_2 = NUM_GS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 4 / 32);
3596 	sq_gpr_resource_mgmt_2 |= NUM_ES_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 4 / 32);
3597 	sq_gpr_resource_mgmt_3 = NUM_HS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
3598 	sq_gpr_resource_mgmt_3 |= NUM_LS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
3599 
3600 	switch (rdev->family) {
3601 	case CHIP_CEDAR:
3602 	case CHIP_PALM:
3603 	case CHIP_SUMO:
3604 	case CHIP_SUMO2:
3605 		ps_thread_count = 96;
3606 		break;
3607 	default:
3608 		ps_thread_count = 128;
3609 		break;
3610 	}
3611 
3612 	sq_thread_resource_mgmt = NUM_PS_THREADS(ps_thread_count);
3613 	sq_thread_resource_mgmt |= NUM_VS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
3614 	sq_thread_resource_mgmt |= NUM_GS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
3615 	sq_thread_resource_mgmt |= NUM_ES_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
3616 	sq_thread_resource_mgmt_2 = NUM_HS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
3617 	sq_thread_resource_mgmt_2 |= NUM_LS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
3618 
3619 	sq_stack_resource_mgmt_1 = NUM_PS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
3620 	sq_stack_resource_mgmt_1 |= NUM_VS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
3621 	sq_stack_resource_mgmt_2 = NUM_GS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
3622 	sq_stack_resource_mgmt_2 |= NUM_ES_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
3623 	sq_stack_resource_mgmt_3 = NUM_HS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
3624 	sq_stack_resource_mgmt_3 |= NUM_LS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
3625 
3626 	WREG32(SQ_CONFIG, sq_config);
3627 	WREG32(SQ_GPR_RESOURCE_MGMT_1, sq_gpr_resource_mgmt_1);
3628 	WREG32(SQ_GPR_RESOURCE_MGMT_2, sq_gpr_resource_mgmt_2);
3629 	WREG32(SQ_GPR_RESOURCE_MGMT_3, sq_gpr_resource_mgmt_3);
3630 	WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
3631 	WREG32(SQ_THREAD_RESOURCE_MGMT_2, sq_thread_resource_mgmt_2);
3632 	WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
3633 	WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
3634 	WREG32(SQ_STACK_RESOURCE_MGMT_3, sq_stack_resource_mgmt_3);
3635 	WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0);
3636 	WREG32(SQ_LDS_RESOURCE_MGMT, sq_lds_resource_mgmt);
3637 
3638 	WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
3639 					  FORCE_EOV_MAX_REZ_CNT(255)));
3640 
3641 	switch (rdev->family) {
3642 	case CHIP_CEDAR:
3643 	case CHIP_PALM:
3644 	case CHIP_SUMO:
3645 	case CHIP_SUMO2:
3646 	case CHIP_CAICOS:
3647 		vgt_cache_invalidation = CACHE_INVALIDATION(TC_ONLY);
3648 		break;
3649 	default:
3650 		vgt_cache_invalidation = CACHE_INVALIDATION(VC_AND_TC);
3651 		break;
3652 	}
3653 	vgt_cache_invalidation |= AUTO_INVLD_EN(ES_AND_GS_AUTO);
3654 	WREG32(VGT_CACHE_INVALIDATION, vgt_cache_invalidation);
3655 
3656 	WREG32(VGT_GS_VERTEX_REUSE, 16);
3657 	WREG32(PA_SU_LINE_STIPPLE_VALUE, 0);
3658 	WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
3659 
3660 	WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
3661 	WREG32(VGT_OUT_DEALLOC_CNTL, 16);
3662 
3663 	WREG32(CB_PERF_CTR0_SEL_0, 0);
3664 	WREG32(CB_PERF_CTR0_SEL_1, 0);
3665 	WREG32(CB_PERF_CTR1_SEL_0, 0);
3666 	WREG32(CB_PERF_CTR1_SEL_1, 0);
3667 	WREG32(CB_PERF_CTR2_SEL_0, 0);
3668 	WREG32(CB_PERF_CTR2_SEL_1, 0);
3669 	WREG32(CB_PERF_CTR3_SEL_0, 0);
3670 	WREG32(CB_PERF_CTR3_SEL_1, 0);
3671 
3672 	/* clear render buffer base addresses */
3673 	WREG32(CB_COLOR0_BASE, 0);
3674 	WREG32(CB_COLOR1_BASE, 0);
3675 	WREG32(CB_COLOR2_BASE, 0);
3676 	WREG32(CB_COLOR3_BASE, 0);
3677 	WREG32(CB_COLOR4_BASE, 0);
3678 	WREG32(CB_COLOR5_BASE, 0);
3679 	WREG32(CB_COLOR6_BASE, 0);
3680 	WREG32(CB_COLOR7_BASE, 0);
3681 	WREG32(CB_COLOR8_BASE, 0);
3682 	WREG32(CB_COLOR9_BASE, 0);
3683 	WREG32(CB_COLOR10_BASE, 0);
3684 	WREG32(CB_COLOR11_BASE, 0);
3685 
3686 	/* set the shader const cache sizes to 0 */
3687 	for (i = SQ_ALU_CONST_BUFFER_SIZE_PS_0; i < 0x28200; i += 4)
3688 		WREG32(i, 0);
3689 	for (i = SQ_ALU_CONST_BUFFER_SIZE_HS_0; i < 0x29000; i += 4)
3690 		WREG32(i, 0);
3691 
3692 	tmp = RREG32(HDP_MISC_CNTL);
3693 	tmp |= HDP_FLUSH_INVALIDATE_CACHE;
3694 	WREG32(HDP_MISC_CNTL, tmp);
3695 
3696 	hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
3697 	WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
3698 
3699 	WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
3700 
3701 	udelay(50);
3702 
3703 }
3704 
3705 int evergreen_mc_init(struct radeon_device *rdev)
3706 {
3707 	u32 tmp;
3708 	int chansize, numchan;
3709 
3710 	/* Get VRAM informations */
3711 	rdev->mc.vram_is_ddr = true;
3712 	if ((rdev->family == CHIP_PALM) ||
3713 	    (rdev->family == CHIP_SUMO) ||
3714 	    (rdev->family == CHIP_SUMO2))
3715 		tmp = RREG32(FUS_MC_ARB_RAMCFG);
3716 	else
3717 		tmp = RREG32(MC_ARB_RAMCFG);
3718 	if (tmp & CHANSIZE_OVERRIDE) {
3719 		chansize = 16;
3720 	} else if (tmp & CHANSIZE_MASK) {
3721 		chansize = 64;
3722 	} else {
3723 		chansize = 32;
3724 	}
3725 	tmp = RREG32(MC_SHARED_CHMAP);
3726 	switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
3727 	case 0:
3728 	default:
3729 		numchan = 1;
3730 		break;
3731 	case 1:
3732 		numchan = 2;
3733 		break;
3734 	case 2:
3735 		numchan = 4;
3736 		break;
3737 	case 3:
3738 		numchan = 8;
3739 		break;
3740 	}
3741 	rdev->mc.vram_width = numchan * chansize;
3742 	/* Could aper size report 0 ? */
3743 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
3744 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
3745 	/* Setup GPU memory space */
3746 	if ((rdev->family == CHIP_PALM) ||
3747 	    (rdev->family == CHIP_SUMO) ||
3748 	    (rdev->family == CHIP_SUMO2)) {
3749 		/* size in bytes on fusion */
3750 		rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
3751 		rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
3752 	} else {
3753 		/* size in MB on evergreen/cayman/tn */
3754 		rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
3755 		rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024ULL * 1024ULL;
3756 	}
3757 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
3758 	r700_vram_gtt_location(rdev, &rdev->mc);
3759 	radeon_update_bandwidth_info(rdev);
3760 
3761 	return 0;
3762 }
3763 
3764 void evergreen_print_gpu_status_regs(struct radeon_device *rdev)
3765 {
3766 	dev_info(rdev->dev, "  GRBM_STATUS               = 0x%08X\n",
3767 		RREG32(GRBM_STATUS));
3768 	dev_info(rdev->dev, "  GRBM_STATUS_SE0           = 0x%08X\n",
3769 		RREG32(GRBM_STATUS_SE0));
3770 	dev_info(rdev->dev, "  GRBM_STATUS_SE1           = 0x%08X\n",
3771 		RREG32(GRBM_STATUS_SE1));
3772 	dev_info(rdev->dev, "  SRBM_STATUS               = 0x%08X\n",
3773 		RREG32(SRBM_STATUS));
3774 	dev_info(rdev->dev, "  SRBM_STATUS2              = 0x%08X\n",
3775 		RREG32(SRBM_STATUS2));
3776 	dev_info(rdev->dev, "  R_008674_CP_STALLED_STAT1 = 0x%08X\n",
3777 		RREG32(CP_STALLED_STAT1));
3778 	dev_info(rdev->dev, "  R_008678_CP_STALLED_STAT2 = 0x%08X\n",
3779 		RREG32(CP_STALLED_STAT2));
3780 	dev_info(rdev->dev, "  R_00867C_CP_BUSY_STAT     = 0x%08X\n",
3781 		RREG32(CP_BUSY_STAT));
3782 	dev_info(rdev->dev, "  R_008680_CP_STAT          = 0x%08X\n",
3783 		RREG32(CP_STAT));
3784 	dev_info(rdev->dev, "  R_00D034_DMA_STATUS_REG   = 0x%08X\n",
3785 		RREG32(DMA_STATUS_REG));
3786 	if (rdev->family >= CHIP_CAYMAN) {
3787 		dev_info(rdev->dev, "  R_00D834_DMA_STATUS_REG   = 0x%08X\n",
3788 			 RREG32(DMA_STATUS_REG + 0x800));
3789 	}
3790 }
3791 
3792 bool evergreen_is_display_hung(struct radeon_device *rdev)
3793 {
3794 	u32 crtc_hung = 0;
3795 	u32 crtc_status[6];
3796 	u32 i, j, tmp;
3797 
3798 	for (i = 0; i < rdev->num_crtc; i++) {
3799 		if (RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]) & EVERGREEN_CRTC_MASTER_EN) {
3800 			crtc_status[i] = RREG32(EVERGREEN_CRTC_STATUS_HV_COUNT + crtc_offsets[i]);
3801 			crtc_hung |= (1 << i);
3802 		}
3803 	}
3804 
3805 	for (j = 0; j < 10; j++) {
3806 		for (i = 0; i < rdev->num_crtc; i++) {
3807 			if (crtc_hung & (1 << i)) {
3808 				tmp = RREG32(EVERGREEN_CRTC_STATUS_HV_COUNT + crtc_offsets[i]);
3809 				if (tmp != crtc_status[i])
3810 					crtc_hung &= ~(1 << i);
3811 			}
3812 		}
3813 		if (crtc_hung == 0)
3814 			return false;
3815 		udelay(100);
3816 	}
3817 
3818 	return true;
3819 }
3820 
3821 u32 evergreen_gpu_check_soft_reset(struct radeon_device *rdev)
3822 {
3823 	u32 reset_mask = 0;
3824 	u32 tmp;
3825 
3826 	/* GRBM_STATUS */
3827 	tmp = RREG32(GRBM_STATUS);
3828 	if (tmp & (PA_BUSY | SC_BUSY |
3829 		   SH_BUSY | SX_BUSY |
3830 		   TA_BUSY | VGT_BUSY |
3831 		   DB_BUSY | CB_BUSY |
3832 		   SPI_BUSY | VGT_BUSY_NO_DMA))
3833 		reset_mask |= RADEON_RESET_GFX;
3834 
3835 	if (tmp & (CF_RQ_PENDING | PF_RQ_PENDING |
3836 		   CP_BUSY | CP_COHERENCY_BUSY))
3837 		reset_mask |= RADEON_RESET_CP;
3838 
3839 	if (tmp & GRBM_EE_BUSY)
3840 		reset_mask |= RADEON_RESET_GRBM | RADEON_RESET_GFX | RADEON_RESET_CP;
3841 
3842 	/* DMA_STATUS_REG */
3843 	tmp = RREG32(DMA_STATUS_REG);
3844 	if (!(tmp & DMA_IDLE))
3845 		reset_mask |= RADEON_RESET_DMA;
3846 
3847 	/* SRBM_STATUS2 */
3848 	tmp = RREG32(SRBM_STATUS2);
3849 	if (tmp & DMA_BUSY)
3850 		reset_mask |= RADEON_RESET_DMA;
3851 
3852 	/* SRBM_STATUS */
3853 	tmp = RREG32(SRBM_STATUS);
3854 	if (tmp & (RLC_RQ_PENDING | RLC_BUSY))
3855 		reset_mask |= RADEON_RESET_RLC;
3856 
3857 	if (tmp & IH_BUSY)
3858 		reset_mask |= RADEON_RESET_IH;
3859 
3860 	if (tmp & SEM_BUSY)
3861 		reset_mask |= RADEON_RESET_SEM;
3862 
3863 	if (tmp & GRBM_RQ_PENDING)
3864 		reset_mask |= RADEON_RESET_GRBM;
3865 
3866 	if (tmp & VMC_BUSY)
3867 		reset_mask |= RADEON_RESET_VMC;
3868 
3869 	if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
3870 		   MCC_BUSY | MCD_BUSY))
3871 		reset_mask |= RADEON_RESET_MC;
3872 
3873 	if (evergreen_is_display_hung(rdev))
3874 		reset_mask |= RADEON_RESET_DISPLAY;
3875 
3876 	/* VM_L2_STATUS */
3877 	tmp = RREG32(VM_L2_STATUS);
3878 	if (tmp & L2_BUSY)
3879 		reset_mask |= RADEON_RESET_VMC;
3880 
3881 	/* Skip MC reset as it's mostly likely not hung, just busy */
3882 	if (reset_mask & RADEON_RESET_MC) {
3883 		DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask);
3884 		reset_mask &= ~RADEON_RESET_MC;
3885 	}
3886 
3887 	return reset_mask;
3888 }
3889 
3890 static void evergreen_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
3891 {
3892 	struct evergreen_mc_save save;
3893 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
3894 	u32 tmp;
3895 
3896 	if (reset_mask == 0)
3897 		return;
3898 
3899 	dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
3900 
3901 	evergreen_print_gpu_status_regs(rdev);
3902 
3903 	/* Disable CP parsing/prefetching */
3904 	WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT);
3905 
3906 	if (reset_mask & RADEON_RESET_DMA) {
3907 		/* Disable DMA */
3908 		tmp = RREG32(DMA_RB_CNTL);
3909 		tmp &= ~DMA_RB_ENABLE;
3910 		WREG32(DMA_RB_CNTL, tmp);
3911 	}
3912 
3913 	udelay(50);
3914 
3915 	evergreen_mc_stop(rdev, &save);
3916 	if (evergreen_mc_wait_for_idle(rdev)) {
3917 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
3918 	}
3919 
3920 	if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE)) {
3921 		grbm_soft_reset |= SOFT_RESET_DB |
3922 			SOFT_RESET_CB |
3923 			SOFT_RESET_PA |
3924 			SOFT_RESET_SC |
3925 			SOFT_RESET_SPI |
3926 			SOFT_RESET_SX |
3927 			SOFT_RESET_SH |
3928 			SOFT_RESET_TC |
3929 			SOFT_RESET_TA |
3930 			SOFT_RESET_VC |
3931 			SOFT_RESET_VGT;
3932 	}
3933 
3934 	if (reset_mask & RADEON_RESET_CP) {
3935 		grbm_soft_reset |= SOFT_RESET_CP |
3936 			SOFT_RESET_VGT;
3937 
3938 		srbm_soft_reset |= SOFT_RESET_GRBM;
3939 	}
3940 
3941 	if (reset_mask & RADEON_RESET_DMA)
3942 		srbm_soft_reset |= SOFT_RESET_DMA;
3943 
3944 	if (reset_mask & RADEON_RESET_DISPLAY)
3945 		srbm_soft_reset |= SOFT_RESET_DC;
3946 
3947 	if (reset_mask & RADEON_RESET_RLC)
3948 		srbm_soft_reset |= SOFT_RESET_RLC;
3949 
3950 	if (reset_mask & RADEON_RESET_SEM)
3951 		srbm_soft_reset |= SOFT_RESET_SEM;
3952 
3953 	if (reset_mask & RADEON_RESET_IH)
3954 		srbm_soft_reset |= SOFT_RESET_IH;
3955 
3956 	if (reset_mask & RADEON_RESET_GRBM)
3957 		srbm_soft_reset |= SOFT_RESET_GRBM;
3958 
3959 	if (reset_mask & RADEON_RESET_VMC)
3960 		srbm_soft_reset |= SOFT_RESET_VMC;
3961 
3962 	if (!(rdev->flags & RADEON_IS_IGP)) {
3963 		if (reset_mask & RADEON_RESET_MC)
3964 			srbm_soft_reset |= SOFT_RESET_MC;
3965 	}
3966 
3967 	if (grbm_soft_reset) {
3968 		tmp = RREG32(GRBM_SOFT_RESET);
3969 		tmp |= grbm_soft_reset;
3970 		dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
3971 		WREG32(GRBM_SOFT_RESET, tmp);
3972 		tmp = RREG32(GRBM_SOFT_RESET);
3973 
3974 		udelay(50);
3975 
3976 		tmp &= ~grbm_soft_reset;
3977 		WREG32(GRBM_SOFT_RESET, tmp);
3978 		tmp = RREG32(GRBM_SOFT_RESET);
3979 	}
3980 
3981 	if (srbm_soft_reset) {
3982 		tmp = RREG32(SRBM_SOFT_RESET);
3983 		tmp |= srbm_soft_reset;
3984 		dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
3985 		WREG32(SRBM_SOFT_RESET, tmp);
3986 		tmp = RREG32(SRBM_SOFT_RESET);
3987 
3988 		udelay(50);
3989 
3990 		tmp &= ~srbm_soft_reset;
3991 		WREG32(SRBM_SOFT_RESET, tmp);
3992 		tmp = RREG32(SRBM_SOFT_RESET);
3993 	}
3994 
3995 	/* Wait a little for things to settle down */
3996 	udelay(50);
3997 
3998 	evergreen_mc_resume(rdev, &save);
3999 	udelay(50);
4000 
4001 	evergreen_print_gpu_status_regs(rdev);
4002 }
4003 
4004 void evergreen_gpu_pci_config_reset(struct radeon_device *rdev)
4005 {
4006 	struct evergreen_mc_save save;
4007 	u32 tmp, i;
4008 
4009 	dev_info(rdev->dev, "GPU pci config reset\n");
4010 
4011 	/* disable dpm? */
4012 
4013 	/* Disable CP parsing/prefetching */
4014 	WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT);
4015 	udelay(50);
4016 	/* Disable DMA */
4017 	tmp = RREG32(DMA_RB_CNTL);
4018 	tmp &= ~DMA_RB_ENABLE;
4019 	WREG32(DMA_RB_CNTL, tmp);
4020 	/* XXX other engines? */
4021 
4022 	/* halt the rlc */
4023 	r600_rlc_stop(rdev);
4024 
4025 	udelay(50);
4026 
4027 	/* set mclk/sclk to bypass */
4028 	rv770_set_clk_bypass_mode(rdev);
4029 	/* disable BM */
4030 	pci_clear_master(rdev->pdev);
4031 	/* disable mem access */
4032 	evergreen_mc_stop(rdev, &save);
4033 	if (evergreen_mc_wait_for_idle(rdev)) {
4034 		dev_warn(rdev->dev, "Wait for MC idle timed out !\n");
4035 	}
4036 	/* reset */
4037 	radeon_pci_config_reset(rdev);
4038 	/* wait for asic to come out of reset */
4039 	for (i = 0; i < rdev->usec_timeout; i++) {
4040 		if (RREG32(CONFIG_MEMSIZE) != 0xffffffff)
4041 			break;
4042 		udelay(1);
4043 	}
4044 }
4045 
4046 int evergreen_asic_reset(struct radeon_device *rdev, bool hard)
4047 {
4048 	u32 reset_mask;
4049 
4050 	if (hard) {
4051 		evergreen_gpu_pci_config_reset(rdev);
4052 		return 0;
4053 	}
4054 
4055 	reset_mask = evergreen_gpu_check_soft_reset(rdev);
4056 
4057 	if (reset_mask)
4058 		r600_set_bios_scratch_engine_hung(rdev, true);
4059 
4060 	/* try soft reset */
4061 	evergreen_gpu_soft_reset(rdev, reset_mask);
4062 
4063 	reset_mask = evergreen_gpu_check_soft_reset(rdev);
4064 
4065 	/* try pci config reset */
4066 	if (reset_mask && radeon_hard_reset)
4067 		evergreen_gpu_pci_config_reset(rdev);
4068 
4069 	reset_mask = evergreen_gpu_check_soft_reset(rdev);
4070 
4071 	if (!reset_mask)
4072 		r600_set_bios_scratch_engine_hung(rdev, false);
4073 
4074 	return 0;
4075 }
4076 
4077 /**
4078  * evergreen_gfx_is_lockup - Check if the GFX engine is locked up
4079  *
4080  * @rdev: radeon_device pointer
4081  * @ring: radeon_ring structure holding ring information
4082  *
4083  * Check if the GFX engine is locked up.
4084  * Returns true if the engine appears to be locked up, false if not.
4085  */
4086 bool evergreen_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
4087 {
4088 	u32 reset_mask = evergreen_gpu_check_soft_reset(rdev);
4089 
4090 	if (!(reset_mask & (RADEON_RESET_GFX |
4091 			    RADEON_RESET_COMPUTE |
4092 			    RADEON_RESET_CP))) {
4093 		radeon_ring_lockup_update(rdev, ring);
4094 		return false;
4095 	}
4096 	return radeon_ring_test_lockup(rdev, ring);
4097 }
4098 
4099 /*
4100  * RLC
4101  */
4102 #define RLC_SAVE_RESTORE_LIST_END_MARKER    0x00000000
4103 #define RLC_CLEAR_STATE_END_MARKER          0x00000001
4104 
4105 void sumo_rlc_fini(struct radeon_device *rdev)
4106 {
4107 	int r;
4108 
4109 	/* save restore block */
4110 	if (rdev->rlc.save_restore_obj) {
4111 		r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
4112 		if (unlikely(r != 0))
4113 			dev_warn(rdev->dev, "(%d) reserve RLC sr bo failed\n", r);
4114 		radeon_bo_unpin(rdev->rlc.save_restore_obj);
4115 		radeon_bo_unreserve(rdev->rlc.save_restore_obj);
4116 
4117 		radeon_bo_unref(&rdev->rlc.save_restore_obj);
4118 		rdev->rlc.save_restore_obj = NULL;
4119 	}
4120 
4121 	/* clear state block */
4122 	if (rdev->rlc.clear_state_obj) {
4123 		r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
4124 		if (unlikely(r != 0))
4125 			dev_warn(rdev->dev, "(%d) reserve RLC c bo failed\n", r);
4126 		radeon_bo_unpin(rdev->rlc.clear_state_obj);
4127 		radeon_bo_unreserve(rdev->rlc.clear_state_obj);
4128 
4129 		radeon_bo_unref(&rdev->rlc.clear_state_obj);
4130 		rdev->rlc.clear_state_obj = NULL;
4131 	}
4132 
4133 	/* clear state block */
4134 	if (rdev->rlc.cp_table_obj) {
4135 		r = radeon_bo_reserve(rdev->rlc.cp_table_obj, false);
4136 		if (unlikely(r != 0))
4137 			dev_warn(rdev->dev, "(%d) reserve RLC cp table bo failed\n", r);
4138 		radeon_bo_unpin(rdev->rlc.cp_table_obj);
4139 		radeon_bo_unreserve(rdev->rlc.cp_table_obj);
4140 
4141 		radeon_bo_unref(&rdev->rlc.cp_table_obj);
4142 		rdev->rlc.cp_table_obj = NULL;
4143 	}
4144 }
4145 
4146 #define CP_ME_TABLE_SIZE    96
4147 
4148 int sumo_rlc_init(struct radeon_device *rdev)
4149 {
4150 	const u32 *src_ptr;
4151 	volatile u32 *dst_ptr;
4152 	u32 dws, data, i, j, k, reg_num;
4153 	u32 reg_list_num, reg_list_hdr_blk_index, reg_list_blk_index = 0;
4154 	u64 reg_list_mc_addr;
4155 	const struct cs_section_def *cs_data;
4156 	int r;
4157 
4158 	src_ptr = rdev->rlc.reg_list;
4159 	dws = rdev->rlc.reg_list_size;
4160 	if (rdev->family >= CHIP_BONAIRE) {
4161 		dws += (5 * 16) + 48 + 48 + 64;
4162 	}
4163 	cs_data = rdev->rlc.cs_data;
4164 
4165 	if (src_ptr) {
4166 		/* save restore block */
4167 		if (rdev->rlc.save_restore_obj == NULL) {
4168 			r = radeon_bo_create(rdev, dws * 4, PAGE_SIZE, true,
4169 					     RADEON_GEM_DOMAIN_VRAM, 0, NULL,
4170 					     NULL, &rdev->rlc.save_restore_obj);
4171 			if (r) {
4172 				dev_warn(rdev->dev, "(%d) create RLC sr bo failed\n", r);
4173 				return r;
4174 			}
4175 		}
4176 
4177 		r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
4178 		if (unlikely(r != 0)) {
4179 			sumo_rlc_fini(rdev);
4180 			return r;
4181 		}
4182 		r = radeon_bo_pin(rdev->rlc.save_restore_obj, RADEON_GEM_DOMAIN_VRAM,
4183 				  &rdev->rlc.save_restore_gpu_addr);
4184 		if (r) {
4185 			radeon_bo_unreserve(rdev->rlc.save_restore_obj);
4186 			dev_warn(rdev->dev, "(%d) pin RLC sr bo failed\n", r);
4187 			sumo_rlc_fini(rdev);
4188 			return r;
4189 		}
4190 
4191 		r = radeon_bo_kmap(rdev->rlc.save_restore_obj, (void **)&rdev->rlc.sr_ptr);
4192 		if (r) {
4193 			dev_warn(rdev->dev, "(%d) map RLC sr bo failed\n", r);
4194 			sumo_rlc_fini(rdev);
4195 			return r;
4196 		}
4197 		/* write the sr buffer */
4198 		dst_ptr = rdev->rlc.sr_ptr;
4199 		if (rdev->family >= CHIP_TAHITI) {
4200 			/* SI */
4201 			for (i = 0; i < rdev->rlc.reg_list_size; i++)
4202 				dst_ptr[i] = cpu_to_le32(src_ptr[i]);
4203 		} else {
4204 			/* ON/LN/TN */
4205 			/* format:
4206 			 * dw0: (reg2 << 16) | reg1
4207 			 * dw1: reg1 save space
4208 			 * dw2: reg2 save space
4209 			 */
4210 			for (i = 0; i < dws; i++) {
4211 				data = src_ptr[i] >> 2;
4212 				i++;
4213 				if (i < dws)
4214 					data |= (src_ptr[i] >> 2) << 16;
4215 				j = (((i - 1) * 3) / 2);
4216 				dst_ptr[j] = cpu_to_le32(data);
4217 			}
4218 			j = ((i * 3) / 2);
4219 			dst_ptr[j] = cpu_to_le32(RLC_SAVE_RESTORE_LIST_END_MARKER);
4220 		}
4221 		radeon_bo_kunmap(rdev->rlc.save_restore_obj);
4222 		radeon_bo_unreserve(rdev->rlc.save_restore_obj);
4223 	}
4224 
4225 	if (cs_data) {
4226 		/* clear state block */
4227 		if (rdev->family >= CHIP_BONAIRE) {
4228 			rdev->rlc.clear_state_size = dws = cik_get_csb_size(rdev);
4229 		} else if (rdev->family >= CHIP_TAHITI) {
4230 			rdev->rlc.clear_state_size = si_get_csb_size(rdev);
4231 			dws = rdev->rlc.clear_state_size + (256 / 4);
4232 		} else {
4233 			reg_list_num = 0;
4234 			dws = 0;
4235 			for (i = 0; cs_data[i].section != NULL; i++) {
4236 				for (j = 0; cs_data[i].section[j].extent != NULL; j++) {
4237 					reg_list_num++;
4238 					dws += cs_data[i].section[j].reg_count;
4239 				}
4240 			}
4241 			reg_list_blk_index = (3 * reg_list_num + 2);
4242 			dws += reg_list_blk_index;
4243 			rdev->rlc.clear_state_size = dws;
4244 		}
4245 
4246 		if (rdev->rlc.clear_state_obj == NULL) {
4247 			r = radeon_bo_create(rdev, dws * 4, PAGE_SIZE, true,
4248 					     RADEON_GEM_DOMAIN_VRAM, 0, NULL,
4249 					     NULL, &rdev->rlc.clear_state_obj);
4250 			if (r) {
4251 				dev_warn(rdev->dev, "(%d) create RLC c bo failed\n", r);
4252 				sumo_rlc_fini(rdev);
4253 				return r;
4254 			}
4255 		}
4256 		r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
4257 		if (unlikely(r != 0)) {
4258 			sumo_rlc_fini(rdev);
4259 			return r;
4260 		}
4261 		r = radeon_bo_pin(rdev->rlc.clear_state_obj, RADEON_GEM_DOMAIN_VRAM,
4262 				  &rdev->rlc.clear_state_gpu_addr);
4263 		if (r) {
4264 			radeon_bo_unreserve(rdev->rlc.clear_state_obj);
4265 			dev_warn(rdev->dev, "(%d) pin RLC c bo failed\n", r);
4266 			sumo_rlc_fini(rdev);
4267 			return r;
4268 		}
4269 
4270 		r = radeon_bo_kmap(rdev->rlc.clear_state_obj, (void **)&rdev->rlc.cs_ptr);
4271 		if (r) {
4272 			dev_warn(rdev->dev, "(%d) map RLC c bo failed\n", r);
4273 			sumo_rlc_fini(rdev);
4274 			return r;
4275 		}
4276 		/* set up the cs buffer */
4277 		dst_ptr = rdev->rlc.cs_ptr;
4278 		if (rdev->family >= CHIP_BONAIRE) {
4279 			cik_get_csb_buffer(rdev, dst_ptr);
4280 		} else if (rdev->family >= CHIP_TAHITI) {
4281 			reg_list_mc_addr = rdev->rlc.clear_state_gpu_addr + 256;
4282 			dst_ptr[0] = cpu_to_le32(upper_32_bits(reg_list_mc_addr));
4283 			dst_ptr[1] = cpu_to_le32(lower_32_bits(reg_list_mc_addr));
4284 			dst_ptr[2] = cpu_to_le32(rdev->rlc.clear_state_size);
4285 			si_get_csb_buffer(rdev, &dst_ptr[(256/4)]);
4286 		} else {
4287 			reg_list_hdr_blk_index = 0;
4288 			reg_list_mc_addr = rdev->rlc.clear_state_gpu_addr + (reg_list_blk_index * 4);
4289 			data = upper_32_bits(reg_list_mc_addr);
4290 			dst_ptr[reg_list_hdr_blk_index] = cpu_to_le32(data);
4291 			reg_list_hdr_blk_index++;
4292 			for (i = 0; cs_data[i].section != NULL; i++) {
4293 				for (j = 0; cs_data[i].section[j].extent != NULL; j++) {
4294 					reg_num = cs_data[i].section[j].reg_count;
4295 					data = reg_list_mc_addr & 0xffffffff;
4296 					dst_ptr[reg_list_hdr_blk_index] = cpu_to_le32(data);
4297 					reg_list_hdr_blk_index++;
4298 
4299 					data = (cs_data[i].section[j].reg_index * 4) & 0xffffffff;
4300 					dst_ptr[reg_list_hdr_blk_index] = cpu_to_le32(data);
4301 					reg_list_hdr_blk_index++;
4302 
4303 					data = 0x08000000 | (reg_num * 4);
4304 					dst_ptr[reg_list_hdr_blk_index] = cpu_to_le32(data);
4305 					reg_list_hdr_blk_index++;
4306 
4307 					for (k = 0; k < reg_num; k++) {
4308 						data = cs_data[i].section[j].extent[k];
4309 						dst_ptr[reg_list_blk_index + k] = cpu_to_le32(data);
4310 					}
4311 					reg_list_mc_addr += reg_num * 4;
4312 					reg_list_blk_index += reg_num;
4313 				}
4314 			}
4315 			dst_ptr[reg_list_hdr_blk_index] = cpu_to_le32(RLC_CLEAR_STATE_END_MARKER);
4316 		}
4317 		radeon_bo_kunmap(rdev->rlc.clear_state_obj);
4318 		radeon_bo_unreserve(rdev->rlc.clear_state_obj);
4319 	}
4320 
4321 	if (rdev->rlc.cp_table_size) {
4322 		if (rdev->rlc.cp_table_obj == NULL) {
4323 			r = radeon_bo_create(rdev, rdev->rlc.cp_table_size,
4324 					     PAGE_SIZE, true,
4325 					     RADEON_GEM_DOMAIN_VRAM, 0, NULL,
4326 					     NULL, &rdev->rlc.cp_table_obj);
4327 			if (r) {
4328 				dev_warn(rdev->dev, "(%d) create RLC cp table bo failed\n", r);
4329 				sumo_rlc_fini(rdev);
4330 				return r;
4331 			}
4332 		}
4333 
4334 		r = radeon_bo_reserve(rdev->rlc.cp_table_obj, false);
4335 		if (unlikely(r != 0)) {
4336 			dev_warn(rdev->dev, "(%d) reserve RLC cp table bo failed\n", r);
4337 			sumo_rlc_fini(rdev);
4338 			return r;
4339 		}
4340 		r = radeon_bo_pin(rdev->rlc.cp_table_obj, RADEON_GEM_DOMAIN_VRAM,
4341 				  &rdev->rlc.cp_table_gpu_addr);
4342 		if (r) {
4343 			radeon_bo_unreserve(rdev->rlc.cp_table_obj);
4344 			dev_warn(rdev->dev, "(%d) pin RLC cp_table bo failed\n", r);
4345 			sumo_rlc_fini(rdev);
4346 			return r;
4347 		}
4348 		r = radeon_bo_kmap(rdev->rlc.cp_table_obj, (void **)&rdev->rlc.cp_table_ptr);
4349 		if (r) {
4350 			dev_warn(rdev->dev, "(%d) map RLC cp table bo failed\n", r);
4351 			sumo_rlc_fini(rdev);
4352 			return r;
4353 		}
4354 
4355 		cik_init_cp_pg_table(rdev);
4356 
4357 		radeon_bo_kunmap(rdev->rlc.cp_table_obj);
4358 		radeon_bo_unreserve(rdev->rlc.cp_table_obj);
4359 
4360 	}
4361 
4362 	return 0;
4363 }
4364 
4365 static void evergreen_rlc_start(struct radeon_device *rdev)
4366 {
4367 	u32 mask = RLC_ENABLE;
4368 
4369 	if (rdev->flags & RADEON_IS_IGP) {
4370 		mask |= GFX_POWER_GATING_ENABLE | GFX_POWER_GATING_SRC;
4371 	}
4372 
4373 	WREG32(RLC_CNTL, mask);
4374 }
4375 
4376 int evergreen_rlc_resume(struct radeon_device *rdev)
4377 {
4378 	u32 i;
4379 	const __be32 *fw_data;
4380 
4381 	if (!rdev->rlc_fw)
4382 		return -EINVAL;
4383 
4384 	r600_rlc_stop(rdev);
4385 
4386 	WREG32(RLC_HB_CNTL, 0);
4387 
4388 	if (rdev->flags & RADEON_IS_IGP) {
4389 		if (rdev->family == CHIP_ARUBA) {
4390 			u32 always_on_bitmap =
4391 				3 | (3 << (16 * rdev->config.cayman.max_shader_engines));
4392 			/* find out the number of active simds */
4393 			u32 tmp = (RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffff0000) >> 16;
4394 			tmp |= 0xffffffff << rdev->config.cayman.max_simds_per_se;
4395 			tmp = hweight32(~tmp);
4396 			if (tmp == rdev->config.cayman.max_simds_per_se) {
4397 				WREG32(TN_RLC_LB_ALWAYS_ACTIVE_SIMD_MASK, always_on_bitmap);
4398 				WREG32(TN_RLC_LB_PARAMS, 0x00601004);
4399 				WREG32(TN_RLC_LB_INIT_SIMD_MASK, 0xffffffff);
4400 				WREG32(TN_RLC_LB_CNTR_INIT, 0x00000000);
4401 				WREG32(TN_RLC_LB_CNTR_MAX, 0x00002000);
4402 			}
4403 		} else {
4404 			WREG32(RLC_HB_WPTR_LSB_ADDR, 0);
4405 			WREG32(RLC_HB_WPTR_MSB_ADDR, 0);
4406 		}
4407 		WREG32(TN_RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
4408 		WREG32(TN_RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8);
4409 	} else {
4410 		WREG32(RLC_HB_BASE, 0);
4411 		WREG32(RLC_HB_RPTR, 0);
4412 		WREG32(RLC_HB_WPTR, 0);
4413 		WREG32(RLC_HB_WPTR_LSB_ADDR, 0);
4414 		WREG32(RLC_HB_WPTR_MSB_ADDR, 0);
4415 	}
4416 	WREG32(RLC_MC_CNTL, 0);
4417 	WREG32(RLC_UCODE_CNTL, 0);
4418 
4419 	fw_data = (const __be32 *)rdev->rlc_fw->data;
4420 	if (rdev->family >= CHIP_ARUBA) {
4421 		for (i = 0; i < ARUBA_RLC_UCODE_SIZE; i++) {
4422 			WREG32(RLC_UCODE_ADDR, i);
4423 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
4424 		}
4425 	} else if (rdev->family >= CHIP_CAYMAN) {
4426 		for (i = 0; i < CAYMAN_RLC_UCODE_SIZE; i++) {
4427 			WREG32(RLC_UCODE_ADDR, i);
4428 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
4429 		}
4430 	} else {
4431 		for (i = 0; i < EVERGREEN_RLC_UCODE_SIZE; i++) {
4432 			WREG32(RLC_UCODE_ADDR, i);
4433 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
4434 		}
4435 	}
4436 	WREG32(RLC_UCODE_ADDR, 0);
4437 
4438 	evergreen_rlc_start(rdev);
4439 
4440 	return 0;
4441 }
4442 
4443 /* Interrupts */
4444 
4445 u32 evergreen_get_vblank_counter(struct radeon_device *rdev, int crtc)
4446 {
4447 	if (crtc >= rdev->num_crtc)
4448 		return 0;
4449 	else
4450 		return RREG32(CRTC_STATUS_FRAME_COUNT + crtc_offsets[crtc]);
4451 }
4452 
4453 void evergreen_disable_interrupt_state(struct radeon_device *rdev)
4454 {
4455 	int i;
4456 	u32 tmp;
4457 
4458 	if (rdev->family >= CHIP_CAYMAN) {
4459 		cayman_cp_int_cntl_setup(rdev, 0,
4460 					 CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
4461 		cayman_cp_int_cntl_setup(rdev, 1, 0);
4462 		cayman_cp_int_cntl_setup(rdev, 2, 0);
4463 		tmp = RREG32(CAYMAN_DMA1_CNTL) & ~TRAP_ENABLE;
4464 		WREG32(CAYMAN_DMA1_CNTL, tmp);
4465 	} else
4466 		WREG32(CP_INT_CNTL, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
4467 	tmp = RREG32(DMA_CNTL) & ~TRAP_ENABLE;
4468 	WREG32(DMA_CNTL, tmp);
4469 	WREG32(GRBM_INT_CNTL, 0);
4470 	WREG32(SRBM_INT_CNTL, 0);
4471 	for (i = 0; i < rdev->num_crtc; i++)
4472 		WREG32(INT_MASK + crtc_offsets[i], 0);
4473 	for (i = 0; i < rdev->num_crtc; i++)
4474 		WREG32(GRPH_INT_CONTROL + crtc_offsets[i], 0);
4475 
4476 	/* only one DAC on DCE5 */
4477 	if (!ASIC_IS_DCE5(rdev))
4478 		WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
4479 	WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
4480 
4481 	for (i = 0; i < 6; i++)
4482 		WREG32_AND(DC_HPDx_INT_CONTROL(i), DC_HPDx_INT_POLARITY);
4483 }
4484 
4485 /* Note that the order we write back regs here is important */
4486 int evergreen_irq_set(struct radeon_device *rdev)
4487 {
4488 	int i;
4489 	u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
4490 	u32 cp_int_cntl1 = 0, cp_int_cntl2 = 0;
4491 	u32 grbm_int_cntl = 0;
4492 	u32 dma_cntl, dma_cntl1 = 0;
4493 	u32 thermal_int = 0;
4494 
4495 	if (!rdev->irq.installed) {
4496 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
4497 		return -EINVAL;
4498 	}
4499 	/* don't enable anything if the ih is disabled */
4500 	if (!rdev->ih.enabled) {
4501 		r600_disable_interrupts(rdev);
4502 		/* force the active interrupt state to all disabled */
4503 		evergreen_disable_interrupt_state(rdev);
4504 		return 0;
4505 	}
4506 
4507 	if (rdev->family == CHIP_ARUBA)
4508 		thermal_int = RREG32(TN_CG_THERMAL_INT_CTRL) &
4509 			~(THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW);
4510 	else
4511 		thermal_int = RREG32(CG_THERMAL_INT) &
4512 			~(THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW);
4513 
4514 	dma_cntl = RREG32(DMA_CNTL) & ~TRAP_ENABLE;
4515 
4516 	if (rdev->family >= CHIP_CAYMAN) {
4517 		/* enable CP interrupts on all rings */
4518 		if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
4519 			DRM_DEBUG("evergreen_irq_set: sw int gfx\n");
4520 			cp_int_cntl |= TIME_STAMP_INT_ENABLE;
4521 		}
4522 		if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
4523 			DRM_DEBUG("evergreen_irq_set: sw int cp1\n");
4524 			cp_int_cntl1 |= TIME_STAMP_INT_ENABLE;
4525 		}
4526 		if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
4527 			DRM_DEBUG("evergreen_irq_set: sw int cp2\n");
4528 			cp_int_cntl2 |= TIME_STAMP_INT_ENABLE;
4529 		}
4530 	} else {
4531 		if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
4532 			DRM_DEBUG("evergreen_irq_set: sw int gfx\n");
4533 			cp_int_cntl |= RB_INT_ENABLE;
4534 			cp_int_cntl |= TIME_STAMP_INT_ENABLE;
4535 		}
4536 	}
4537 
4538 	if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
4539 		DRM_DEBUG("r600_irq_set: sw int dma\n");
4540 		dma_cntl |= TRAP_ENABLE;
4541 	}
4542 
4543 	if (rdev->family >= CHIP_CAYMAN) {
4544 		dma_cntl1 = RREG32(CAYMAN_DMA1_CNTL) & ~TRAP_ENABLE;
4545 		if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
4546 			DRM_DEBUG("r600_irq_set: sw int dma1\n");
4547 			dma_cntl1 |= TRAP_ENABLE;
4548 		}
4549 	}
4550 
4551 	if (rdev->irq.dpm_thermal) {
4552 		DRM_DEBUG("dpm thermal\n");
4553 		thermal_int |= THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW;
4554 	}
4555 
4556 	if (rdev->family >= CHIP_CAYMAN) {
4557 		cayman_cp_int_cntl_setup(rdev, 0, cp_int_cntl);
4558 		cayman_cp_int_cntl_setup(rdev, 1, cp_int_cntl1);
4559 		cayman_cp_int_cntl_setup(rdev, 2, cp_int_cntl2);
4560 	} else
4561 		WREG32(CP_INT_CNTL, cp_int_cntl);
4562 
4563 	WREG32(DMA_CNTL, dma_cntl);
4564 
4565 	if (rdev->family >= CHIP_CAYMAN)
4566 		WREG32(CAYMAN_DMA1_CNTL, dma_cntl1);
4567 
4568 	WREG32(GRBM_INT_CNTL, grbm_int_cntl);
4569 
4570 	for (i = 0; i < rdev->num_crtc; i++) {
4571 		radeon_irq_kms_set_irq_n_enabled(
4572 		    rdev, INT_MASK + crtc_offsets[i],
4573 		    VBLANK_INT_MASK,
4574 		    rdev->irq.crtc_vblank_int[i] ||
4575 		    atomic_read(&rdev->irq.pflip[i]), "vblank", i);
4576 	}
4577 
4578 	for (i = 0; i < rdev->num_crtc; i++)
4579 		WREG32(GRPH_INT_CONTROL + crtc_offsets[i], GRPH_PFLIP_INT_MASK);
4580 
4581 	for (i = 0; i < 6; i++) {
4582 		radeon_irq_kms_set_irq_n_enabled(
4583 		    rdev, DC_HPDx_INT_CONTROL(i),
4584 		    DC_HPDx_INT_EN | DC_HPDx_RX_INT_EN,
4585 		    rdev->irq.hpd[i], "HPD", i);
4586 	}
4587 
4588 	if (rdev->family == CHIP_ARUBA)
4589 		WREG32(TN_CG_THERMAL_INT_CTRL, thermal_int);
4590 	else
4591 		WREG32(CG_THERMAL_INT, thermal_int);
4592 
4593 	for (i = 0; i < 6; i++) {
4594 		radeon_irq_kms_set_irq_n_enabled(
4595 		    rdev, AFMT_AUDIO_PACKET_CONTROL + crtc_offsets[i],
4596 		    AFMT_AZ_FORMAT_WTRIG_MASK,
4597 		    rdev->irq.afmt[i], "HDMI", i);
4598 	}
4599 
4600 	/* posting read */
4601 	RREG32(SRBM_STATUS);
4602 
4603 	return 0;
4604 }
4605 
4606 /* Note that the order we write back regs here is important */
4607 static void evergreen_irq_ack(struct radeon_device *rdev)
4608 {
4609 	int i, j;
4610 	u32 *grph_int = rdev->irq.stat_regs.evergreen.grph_int;
4611 	u32 *disp_int = rdev->irq.stat_regs.evergreen.disp_int;
4612 	u32 *afmt_status = rdev->irq.stat_regs.evergreen.afmt_status;
4613 
4614 	for (i = 0; i < 6; i++) {
4615 		disp_int[i] = RREG32(evergreen_disp_int_status[i]);
4616 		afmt_status[i] = RREG32(AFMT_STATUS + crtc_offsets[i]);
4617 		if (i < rdev->num_crtc)
4618 			grph_int[i] = RREG32(GRPH_INT_STATUS + crtc_offsets[i]);
4619 	}
4620 
4621 	/* We write back each interrupt register in pairs of two */
4622 	for (i = 0; i < rdev->num_crtc; i += 2) {
4623 		for (j = i; j < (i + 2); j++) {
4624 			if (grph_int[j] & GRPH_PFLIP_INT_OCCURRED)
4625 				WREG32(GRPH_INT_STATUS + crtc_offsets[j],
4626 				       GRPH_PFLIP_INT_CLEAR);
4627 		}
4628 
4629 		for (j = i; j < (i + 2); j++) {
4630 			if (disp_int[j] & LB_D1_VBLANK_INTERRUPT)
4631 				WREG32(VBLANK_STATUS + crtc_offsets[j],
4632 				       VBLANK_ACK);
4633 			if (disp_int[j] & LB_D1_VLINE_INTERRUPT)
4634 				WREG32(VLINE_STATUS + crtc_offsets[j],
4635 				       VLINE_ACK);
4636 		}
4637 	}
4638 
4639 	for (i = 0; i < 6; i++) {
4640 		if (disp_int[i] & DC_HPD1_INTERRUPT)
4641 			WREG32_OR(DC_HPDx_INT_CONTROL(i), DC_HPDx_INT_ACK);
4642 	}
4643 
4644 	for (i = 0; i < 6; i++) {
4645 		if (disp_int[i] & DC_HPD1_RX_INTERRUPT)
4646 			WREG32_OR(DC_HPDx_INT_CONTROL(i), DC_HPDx_RX_INT_ACK);
4647 	}
4648 
4649 	for (i = 0; i < 6; i++) {
4650 		if (afmt_status[i] & AFMT_AZ_FORMAT_WTRIG)
4651 			WREG32_OR(AFMT_AUDIO_PACKET_CONTROL + crtc_offsets[i],
4652 				  AFMT_AZ_FORMAT_WTRIG_ACK);
4653 	}
4654 }
4655 
4656 static void evergreen_irq_disable(struct radeon_device *rdev)
4657 {
4658 	r600_disable_interrupts(rdev);
4659 	/* Wait and acknowledge irq */
4660 	mdelay(1);
4661 	evergreen_irq_ack(rdev);
4662 	evergreen_disable_interrupt_state(rdev);
4663 }
4664 
4665 void evergreen_irq_suspend(struct radeon_device *rdev)
4666 {
4667 	evergreen_irq_disable(rdev);
4668 	r600_rlc_stop(rdev);
4669 }
4670 
4671 static u32 evergreen_get_ih_wptr(struct radeon_device *rdev)
4672 {
4673 	u32 wptr, tmp;
4674 
4675 	if (rdev->wb.enabled)
4676 		wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
4677 	else
4678 		wptr = RREG32(IH_RB_WPTR);
4679 
4680 	if (wptr & RB_OVERFLOW) {
4681 		wptr &= ~RB_OVERFLOW;
4682 		/* When a ring buffer overflow happen start parsing interrupt
4683 		 * from the last not overwritten vector (wptr + 16). Hopefully
4684 		 * this should allow us to catchup.
4685 		 */
4686 		dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n",
4687 			 wptr, rdev->ih.rptr, (wptr + 16) & rdev->ih.ptr_mask);
4688 		rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
4689 		tmp = RREG32(IH_RB_CNTL);
4690 		tmp |= IH_WPTR_OVERFLOW_CLEAR;
4691 		WREG32(IH_RB_CNTL, tmp);
4692 	}
4693 	return (wptr & rdev->ih.ptr_mask);
4694 }
4695 
4696 int evergreen_irq_process(struct radeon_device *rdev)
4697 {
4698 	u32 *disp_int = rdev->irq.stat_regs.evergreen.disp_int;
4699 	u32 *afmt_status = rdev->irq.stat_regs.evergreen.afmt_status;
4700 	u32 crtc_idx, hpd_idx, afmt_idx;
4701 	u32 mask;
4702 	u32 wptr;
4703 	u32 rptr;
4704 	u32 src_id, src_data;
4705 	u32 ring_index;
4706 	bool queue_hotplug = false;
4707 	bool queue_hdmi = false;
4708 	bool queue_dp = false;
4709 	bool queue_thermal = false;
4710 	u32 status, addr;
4711 	const char *event_name;
4712 
4713 	if (!rdev->ih.enabled || rdev->shutdown)
4714 		return IRQ_NONE;
4715 
4716 	wptr = evergreen_get_ih_wptr(rdev);
4717 
4718 restart_ih:
4719 	/* is somebody else already processing irqs? */
4720 	if (atomic_xchg(&rdev->ih.lock, 1))
4721 		return IRQ_NONE;
4722 
4723 	rptr = rdev->ih.rptr;
4724 	DRM_DEBUG("evergreen_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
4725 
4726 	/* Order reading of wptr vs. reading of IH ring data */
4727 	rmb();
4728 
4729 	/* display interrupts */
4730 	evergreen_irq_ack(rdev);
4731 
4732 	while (rptr != wptr) {
4733 		/* wptr/rptr are in bytes! */
4734 		ring_index = rptr / 4;
4735 		src_id =  le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
4736 		src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
4737 
4738 		switch (src_id) {
4739 		case 1: /* D1 vblank/vline */
4740 		case 2: /* D2 vblank/vline */
4741 		case 3: /* D3 vblank/vline */
4742 		case 4: /* D4 vblank/vline */
4743 		case 5: /* D5 vblank/vline */
4744 		case 6: /* D6 vblank/vline */
4745 			crtc_idx = src_id - 1;
4746 
4747 			if (src_data == 0) { /* vblank */
4748 				mask = LB_D1_VBLANK_INTERRUPT;
4749 				event_name = "vblank";
4750 
4751 				if (rdev->irq.crtc_vblank_int[crtc_idx]) {
4752 					drm_handle_vblank(rdev->ddev, crtc_idx);
4753 					rdev->pm.vblank_sync = true;
4754 					wake_up(&rdev->irq.vblank_queue);
4755 				}
4756 				if (atomic_read(&rdev->irq.pflip[crtc_idx])) {
4757 					radeon_crtc_handle_vblank(rdev,
4758 								  crtc_idx);
4759 				}
4760 
4761 			} else if (src_data == 1) { /* vline */
4762 				mask = LB_D1_VLINE_INTERRUPT;
4763 				event_name = "vline";
4764 			} else {
4765 				DRM_DEBUG("Unhandled interrupt: %d %d\n",
4766 					  src_id, src_data);
4767 				break;
4768 			}
4769 
4770 			if (!(disp_int[crtc_idx] & mask)) {
4771 				DRM_DEBUG("IH: D%d %s - IH event w/o asserted irq bit?\n",
4772 					  crtc_idx + 1, event_name);
4773 			}
4774 
4775 			disp_int[crtc_idx] &= ~mask;
4776 			DRM_DEBUG("IH: D%d %s\n", crtc_idx + 1, event_name);
4777 
4778 			break;
4779 		case 8: /* D1 page flip */
4780 		case 10: /* D2 page flip */
4781 		case 12: /* D3 page flip */
4782 		case 14: /* D4 page flip */
4783 		case 16: /* D5 page flip */
4784 		case 18: /* D6 page flip */
4785 			DRM_DEBUG("IH: D%d flip\n", ((src_id - 8) >> 1) + 1);
4786 			if (radeon_use_pflipirq > 0)
4787 				radeon_crtc_handle_flip(rdev, (src_id - 8) >> 1);
4788 			break;
4789 		case 42: /* HPD hotplug */
4790 			if (src_data <= 5) {
4791 				hpd_idx = src_data;
4792 				mask = DC_HPD1_INTERRUPT;
4793 				queue_hotplug = true;
4794 				event_name = "HPD";
4795 
4796 			} else if (src_data <= 11) {
4797 				hpd_idx = src_data - 6;
4798 				mask = DC_HPD1_RX_INTERRUPT;
4799 				queue_dp = true;
4800 				event_name = "HPD_RX";
4801 
4802 			} else {
4803 				DRM_DEBUG("Unhandled interrupt: %d %d\n",
4804 					  src_id, src_data);
4805 				break;
4806 			}
4807 
4808 			if (!(disp_int[hpd_idx] & mask))
4809 				DRM_DEBUG("IH: IH event w/o asserted irq bit?\n");
4810 
4811 			disp_int[hpd_idx] &= ~mask;
4812 			DRM_DEBUG("IH: %s%d\n", event_name, hpd_idx + 1);
4813 
4814 			break;
4815 		case 44: /* hdmi */
4816 			afmt_idx = src_data;
4817 			if (!(afmt_status[afmt_idx] & AFMT_AZ_FORMAT_WTRIG))
4818 				DRM_DEBUG("IH: IH event w/o asserted irq bit?\n");
4819 
4820 			if (afmt_idx > 5) {
4821 				DRM_ERROR("Unhandled interrupt: %d %d\n",
4822 					  src_id, src_data);
4823 				break;
4824 			}
4825 			afmt_status[afmt_idx] &= ~AFMT_AZ_FORMAT_WTRIG;
4826 			queue_hdmi = true;
4827 			DRM_DEBUG("IH: HDMI%d\n", afmt_idx + 1);
4828 			break;
4829 		case 96:
4830 			DRM_ERROR("SRBM_READ_ERROR: 0x%x\n", RREG32(SRBM_READ_ERROR));
4831 			WREG32(SRBM_INT_ACK, 0x1);
4832 			break;
4833 		case 124: /* UVD */
4834 			DRM_DEBUG("IH: UVD int: 0x%08x\n", src_data);
4835 			radeon_fence_process(rdev, R600_RING_TYPE_UVD_INDEX);
4836 			break;
4837 		case 146:
4838 		case 147:
4839 			addr = RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR);
4840 			status = RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS);
4841 			/* reset addr and status */
4842 			WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
4843 			if (addr == 0x0 && status == 0x0)
4844 				break;
4845 			dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data);
4846 			dev_err(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
4847 				addr);
4848 			dev_err(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
4849 				status);
4850 			cayman_vm_decode_fault(rdev, status, addr);
4851 			break;
4852 		case 176: /* CP_INT in ring buffer */
4853 		case 177: /* CP_INT in IB1 */
4854 		case 178: /* CP_INT in IB2 */
4855 			DRM_DEBUG("IH: CP int: 0x%08x\n", src_data);
4856 			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
4857 			break;
4858 		case 181: /* CP EOP event */
4859 			DRM_DEBUG("IH: CP EOP\n");
4860 			if (rdev->family >= CHIP_CAYMAN) {
4861 				switch (src_data) {
4862 				case 0:
4863 					radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
4864 					break;
4865 				case 1:
4866 					radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4867 					break;
4868 				case 2:
4869 					radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4870 					break;
4871 				}
4872 			} else
4873 				radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
4874 			break;
4875 		case 224: /* DMA trap event */
4876 			DRM_DEBUG("IH: DMA trap\n");
4877 			radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
4878 			break;
4879 		case 230: /* thermal low to high */
4880 			DRM_DEBUG("IH: thermal low to high\n");
4881 			rdev->pm.dpm.thermal.high_to_low = false;
4882 			queue_thermal = true;
4883 			break;
4884 		case 231: /* thermal high to low */
4885 			DRM_DEBUG("IH: thermal high to low\n");
4886 			rdev->pm.dpm.thermal.high_to_low = true;
4887 			queue_thermal = true;
4888 			break;
4889 		case 233: /* GUI IDLE */
4890 			DRM_DEBUG("IH: GUI idle\n");
4891 			break;
4892 		case 244: /* DMA trap event */
4893 			if (rdev->family >= CHIP_CAYMAN) {
4894 				DRM_DEBUG("IH: DMA1 trap\n");
4895 				radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
4896 			}
4897 			break;
4898 		default:
4899 			DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4900 			break;
4901 		}
4902 
4903 		/* wptr/rptr are in bytes! */
4904 		rptr += 16;
4905 		rptr &= rdev->ih.ptr_mask;
4906 		WREG32(IH_RB_RPTR, rptr);
4907 	}
4908 	if (queue_dp)
4909 		schedule_work(&rdev->dp_work);
4910 	if (queue_hotplug)
4911 		schedule_delayed_work(&rdev->hotplug_work, 0);
4912 	if (queue_hdmi)
4913 		schedule_work(&rdev->audio_work);
4914 	if (queue_thermal && rdev->pm.dpm_enabled)
4915 		schedule_work(&rdev->pm.dpm.thermal.work);
4916 	rdev->ih.rptr = rptr;
4917 	atomic_set(&rdev->ih.lock, 0);
4918 
4919 	/* make sure wptr hasn't changed while processing */
4920 	wptr = evergreen_get_ih_wptr(rdev);
4921 	if (wptr != rptr)
4922 		goto restart_ih;
4923 
4924 	return IRQ_HANDLED;
4925 }
4926 
4927 static void evergreen_uvd_init(struct radeon_device *rdev)
4928 {
4929 	int r;
4930 
4931 	if (!rdev->has_uvd)
4932 		return;
4933 
4934 	r = radeon_uvd_init(rdev);
4935 	if (r) {
4936 		dev_err(rdev->dev, "failed UVD (%d) init.\n", r);
4937 		/*
4938 		 * At this point rdev->uvd.vcpu_bo is NULL which trickles down
4939 		 * to early fails uvd_v2_2_resume() and thus nothing happens
4940 		 * there. So it is pointless to try to go through that code
4941 		 * hence why we disable uvd here.
4942 		 */
4943 		rdev->has_uvd = false;
4944 		return;
4945 	}
4946 	rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_obj = NULL;
4947 	r600_ring_init(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX], 4096);
4948 }
4949 
4950 static void evergreen_uvd_start(struct radeon_device *rdev)
4951 {
4952 	int r;
4953 
4954 	if (!rdev->has_uvd)
4955 		return;
4956 
4957 	r = uvd_v2_2_resume(rdev);
4958 	if (r) {
4959 		dev_err(rdev->dev, "failed UVD resume (%d).\n", r);
4960 		goto error;
4961 	}
4962 	r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_UVD_INDEX);
4963 	if (r) {
4964 		dev_err(rdev->dev, "failed initializing UVD fences (%d).\n", r);
4965 		goto error;
4966 	}
4967 	return;
4968 
4969 error:
4970 	rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0;
4971 }
4972 
4973 static void evergreen_uvd_resume(struct radeon_device *rdev)
4974 {
4975 	struct radeon_ring *ring;
4976 	int r;
4977 
4978 	if (!rdev->has_uvd || !rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size)
4979 		return;
4980 
4981 	ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX];
4982 	r = radeon_ring_init(rdev, ring, ring->ring_size, 0, PACKET0(UVD_NO_OP, 0));
4983 	if (r) {
4984 		dev_err(rdev->dev, "failed initializing UVD ring (%d).\n", r);
4985 		return;
4986 	}
4987 	r = uvd_v1_0_init(rdev);
4988 	if (r) {
4989 		dev_err(rdev->dev, "failed initializing UVD (%d).\n", r);
4990 		return;
4991 	}
4992 }
4993 
4994 static int evergreen_startup(struct radeon_device *rdev)
4995 {
4996 	struct radeon_ring *ring;
4997 	int r;
4998 
4999 	/* enable pcie gen2 link */
5000 	evergreen_pcie_gen2_enable(rdev);
5001 	/* enable aspm */
5002 	evergreen_program_aspm(rdev);
5003 
5004 	/* scratch needs to be initialized before MC */
5005 	r = r600_vram_scratch_init(rdev);
5006 	if (r)
5007 		return r;
5008 
5009 	evergreen_mc_program(rdev);
5010 
5011 	if (ASIC_IS_DCE5(rdev) && !rdev->pm.dpm_enabled) {
5012 		r = ni_mc_load_microcode(rdev);
5013 		if (r) {
5014 			DRM_ERROR("Failed to load MC firmware!\n");
5015 			return r;
5016 		}
5017 	}
5018 
5019 	if (rdev->flags & RADEON_IS_AGP) {
5020 		evergreen_agp_enable(rdev);
5021 	} else {
5022 		r = evergreen_pcie_gart_enable(rdev);
5023 		if (r)
5024 			return r;
5025 	}
5026 	evergreen_gpu_init(rdev);
5027 
5028 	/* allocate rlc buffers */
5029 	if (rdev->flags & RADEON_IS_IGP) {
5030 		rdev->rlc.reg_list = sumo_rlc_save_restore_register_list;
5031 		rdev->rlc.reg_list_size =
5032 			(u32)ARRAY_SIZE(sumo_rlc_save_restore_register_list);
5033 		rdev->rlc.cs_data = evergreen_cs_data;
5034 		r = sumo_rlc_init(rdev);
5035 		if (r) {
5036 			DRM_ERROR("Failed to init rlc BOs!\n");
5037 			return r;
5038 		}
5039 	}
5040 
5041 	/* allocate wb buffer */
5042 	r = radeon_wb_init(rdev);
5043 	if (r)
5044 		return r;
5045 
5046 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
5047 	if (r) {
5048 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
5049 		return r;
5050 	}
5051 
5052 	r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
5053 	if (r) {
5054 		dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
5055 		return r;
5056 	}
5057 
5058 	evergreen_uvd_start(rdev);
5059 
5060 	/* Enable IRQ */
5061 	if (!rdev->irq.installed) {
5062 		r = radeon_irq_kms_init(rdev);
5063 		if (r)
5064 			return r;
5065 	}
5066 
5067 	r = r600_irq_init(rdev);
5068 	if (r) {
5069 		DRM_ERROR("radeon: IH init failed (%d).\n", r);
5070 		radeon_irq_kms_fini(rdev);
5071 		return r;
5072 	}
5073 	evergreen_irq_set(rdev);
5074 
5075 	ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
5076 	r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
5077 			     RADEON_CP_PACKET2);
5078 	if (r)
5079 		return r;
5080 
5081 	ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
5082 	r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
5083 			     DMA_PACKET(DMA_PACKET_NOP, 0, 0));
5084 	if (r)
5085 		return r;
5086 
5087 	r = evergreen_cp_load_microcode(rdev);
5088 	if (r)
5089 		return r;
5090 	r = evergreen_cp_resume(rdev);
5091 	if (r)
5092 		return r;
5093 	r = r600_dma_resume(rdev);
5094 	if (r)
5095 		return r;
5096 
5097 	evergreen_uvd_resume(rdev);
5098 
5099 	r = radeon_ib_pool_init(rdev);
5100 	if (r) {
5101 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
5102 		return r;
5103 	}
5104 
5105 	r = radeon_audio_init(rdev);
5106 	if (r) {
5107 		DRM_ERROR("radeon: audio init failed\n");
5108 		return r;
5109 	}
5110 
5111 	return 0;
5112 }
5113 
5114 int evergreen_resume(struct radeon_device *rdev)
5115 {
5116 	int r;
5117 
5118 	/* reset the asic, the gfx blocks are often in a bad state
5119 	 * after the driver is unloaded or after a resume
5120 	 */
5121 	if (radeon_asic_reset(rdev))
5122 		dev_warn(rdev->dev, "GPU reset failed !\n");
5123 	/* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
5124 	 * posting will perform necessary task to bring back GPU into good
5125 	 * shape.
5126 	 */
5127 	/* post card */
5128 	atom_asic_init(rdev->mode_info.atom_context);
5129 
5130 	/* init golden registers */
5131 	evergreen_init_golden_registers(rdev);
5132 
5133 	if (rdev->pm.pm_method == PM_METHOD_DPM)
5134 		radeon_pm_resume(rdev);
5135 
5136 	rdev->accel_working = true;
5137 	r = evergreen_startup(rdev);
5138 	if (r) {
5139 		DRM_ERROR("evergreen startup failed on resume\n");
5140 		rdev->accel_working = false;
5141 		return r;
5142 	}
5143 
5144 	return r;
5145 
5146 }
5147 
5148 int evergreen_suspend(struct radeon_device *rdev)
5149 {
5150 	radeon_pm_suspend(rdev);
5151 	radeon_audio_fini(rdev);
5152 	if (rdev->has_uvd) {
5153 		uvd_v1_0_fini(rdev);
5154 		radeon_uvd_suspend(rdev);
5155 	}
5156 	r700_cp_stop(rdev);
5157 	r600_dma_stop(rdev);
5158 	evergreen_irq_suspend(rdev);
5159 	radeon_wb_disable(rdev);
5160 	evergreen_pcie_gart_disable(rdev);
5161 
5162 	return 0;
5163 }
5164 
5165 /* Plan is to move initialization in that function and use
5166  * helper function so that radeon_device_init pretty much
5167  * do nothing more than calling asic specific function. This
5168  * should also allow to remove a bunch of callback function
5169  * like vram_info.
5170  */
5171 int evergreen_init(struct radeon_device *rdev)
5172 {
5173 	int r;
5174 
5175 	/* Read BIOS */
5176 	if (!radeon_get_bios(rdev)) {
5177 		if (ASIC_IS_AVIVO(rdev))
5178 			return -EINVAL;
5179 	}
5180 	/* Must be an ATOMBIOS */
5181 	if (!rdev->is_atom_bios) {
5182 		dev_err(rdev->dev, "Expecting atombios for evergreen GPU\n");
5183 		return -EINVAL;
5184 	}
5185 	r = radeon_atombios_init(rdev);
5186 	if (r)
5187 		return r;
5188 	/* reset the asic, the gfx blocks are often in a bad state
5189 	 * after the driver is unloaded or after a resume
5190 	 */
5191 	if (radeon_asic_reset(rdev))
5192 		dev_warn(rdev->dev, "GPU reset failed !\n");
5193 	/* Post card if necessary */
5194 	if (!radeon_card_posted(rdev)) {
5195 		if (!rdev->bios) {
5196 			dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
5197 			return -EINVAL;
5198 		}
5199 		DRM_INFO("GPU not posted. posting now...\n");
5200 		atom_asic_init(rdev->mode_info.atom_context);
5201 	}
5202 	/* init golden registers */
5203 	evergreen_init_golden_registers(rdev);
5204 	/* Initialize scratch registers */
5205 	r600_scratch_init(rdev);
5206 	/* Initialize surface registers */
5207 	radeon_surface_init(rdev);
5208 	/* Initialize clocks */
5209 	radeon_get_clock_info(rdev->ddev);
5210 	/* Fence driver */
5211 	r = radeon_fence_driver_init(rdev);
5212 	if (r)
5213 		return r;
5214 	/* initialize AGP */
5215 	if (rdev->flags & RADEON_IS_AGP) {
5216 		r = radeon_agp_init(rdev);
5217 		if (r)
5218 			radeon_agp_disable(rdev);
5219 	}
5220 	/* initialize memory controller */
5221 	r = evergreen_mc_init(rdev);
5222 	if (r)
5223 		return r;
5224 	/* Memory manager */
5225 	r = radeon_bo_init(rdev);
5226 	if (r)
5227 		return r;
5228 
5229 	if (ASIC_IS_DCE5(rdev)) {
5230 		if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw || !rdev->mc_fw) {
5231 			r = ni_init_microcode(rdev);
5232 			if (r) {
5233 				DRM_ERROR("Failed to load firmware!\n");
5234 				return r;
5235 			}
5236 		}
5237 	} else {
5238 		if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
5239 			r = r600_init_microcode(rdev);
5240 			if (r) {
5241 				DRM_ERROR("Failed to load firmware!\n");
5242 				return r;
5243 			}
5244 		}
5245 	}
5246 
5247 	/* Initialize power management */
5248 	radeon_pm_init(rdev);
5249 
5250 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ring_obj = NULL;
5251 	r600_ring_init(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX], 1024 * 1024);
5252 
5253 	rdev->ring[R600_RING_TYPE_DMA_INDEX].ring_obj = NULL;
5254 	r600_ring_init(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX], 64 * 1024);
5255 
5256 	evergreen_uvd_init(rdev);
5257 
5258 	rdev->ih.ring_obj = NULL;
5259 	r600_ih_ring_init(rdev, 64 * 1024);
5260 
5261 	r = r600_pcie_gart_init(rdev);
5262 	if (r)
5263 		return r;
5264 
5265 	rdev->accel_working = true;
5266 	r = evergreen_startup(rdev);
5267 	if (r) {
5268 		dev_err(rdev->dev, "disabling GPU acceleration\n");
5269 		r700_cp_fini(rdev);
5270 		r600_dma_fini(rdev);
5271 		r600_irq_fini(rdev);
5272 		if (rdev->flags & RADEON_IS_IGP)
5273 			sumo_rlc_fini(rdev);
5274 		radeon_wb_fini(rdev);
5275 		radeon_ib_pool_fini(rdev);
5276 		radeon_irq_kms_fini(rdev);
5277 		evergreen_pcie_gart_fini(rdev);
5278 		rdev->accel_working = false;
5279 	}
5280 
5281 	/* Don't start up if the MC ucode is missing on BTC parts.
5282 	 * The default clocks and voltages before the MC ucode
5283 	 * is loaded are not suffient for advanced operations.
5284 	 */
5285 	if (ASIC_IS_DCE5(rdev)) {
5286 		if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) {
5287 			DRM_ERROR("radeon: MC ucode required for NI+.\n");
5288 			return -EINVAL;
5289 		}
5290 	}
5291 
5292 	return 0;
5293 }
5294 
5295 void evergreen_fini(struct radeon_device *rdev)
5296 {
5297 	radeon_pm_fini(rdev);
5298 	radeon_audio_fini(rdev);
5299 	r700_cp_fini(rdev);
5300 	r600_dma_fini(rdev);
5301 	r600_irq_fini(rdev);
5302 	if (rdev->flags & RADEON_IS_IGP)
5303 		sumo_rlc_fini(rdev);
5304 	radeon_wb_fini(rdev);
5305 	radeon_ib_pool_fini(rdev);
5306 	radeon_irq_kms_fini(rdev);
5307 	uvd_v1_0_fini(rdev);
5308 	radeon_uvd_fini(rdev);
5309 	evergreen_pcie_gart_fini(rdev);
5310 	r600_vram_scratch_fini(rdev);
5311 	radeon_gem_fini(rdev);
5312 	radeon_fence_driver_fini(rdev);
5313 	radeon_agp_fini(rdev);
5314 	radeon_bo_fini(rdev);
5315 	radeon_atombios_fini(rdev);
5316 	kfree(rdev->bios);
5317 	rdev->bios = NULL;
5318 }
5319 
5320 void evergreen_pcie_gen2_enable(struct radeon_device *rdev)
5321 {
5322 	u32 link_width_cntl, speed_cntl;
5323 
5324 	if (radeon_pcie_gen2 == 0)
5325 		return;
5326 
5327 	if (rdev->flags & RADEON_IS_IGP)
5328 		return;
5329 
5330 	if (!(rdev->flags & RADEON_IS_PCIE))
5331 		return;
5332 
5333 	/* x2 cards have a special sequence */
5334 	if (ASIC_IS_X2(rdev))
5335 		return;
5336 
5337 	if ((rdev->pdev->bus->max_bus_speed != PCIE_SPEED_5_0GT) &&
5338 		(rdev->pdev->bus->max_bus_speed != PCIE_SPEED_8_0GT))
5339 		return;
5340 
5341 	speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
5342 	if (speed_cntl & LC_CURRENT_DATA_RATE) {
5343 		DRM_INFO("PCIE gen 2 link speeds already enabled\n");
5344 		return;
5345 	}
5346 
5347 	DRM_INFO("enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0\n");
5348 
5349 	if ((speed_cntl & LC_OTHER_SIDE_EVER_SENT_GEN2) ||
5350 	    (speed_cntl & LC_OTHER_SIDE_SUPPORTS_GEN2)) {
5351 
5352 		link_width_cntl = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
5353 		link_width_cntl &= ~LC_UPCONFIGURE_DIS;
5354 		WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
5355 
5356 		speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
5357 		speed_cntl &= ~LC_TARGET_LINK_SPEED_OVERRIDE_EN;
5358 		WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
5359 
5360 		speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
5361 		speed_cntl |= LC_CLR_FAILED_SPD_CHANGE_CNT;
5362 		WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
5363 
5364 		speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
5365 		speed_cntl &= ~LC_CLR_FAILED_SPD_CHANGE_CNT;
5366 		WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
5367 
5368 		speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL);
5369 		speed_cntl |= LC_GEN2_EN_STRAP;
5370 		WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl);
5371 
5372 	} else {
5373 		link_width_cntl = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
5374 		/* XXX: only disable it if gen1 bridge vendor == 0x111d or 0x1106 */
5375 		if (1)
5376 			link_width_cntl |= LC_UPCONFIGURE_DIS;
5377 		else
5378 			link_width_cntl &= ~LC_UPCONFIGURE_DIS;
5379 		WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
5380 	}
5381 }
5382 
5383 void evergreen_program_aspm(struct radeon_device *rdev)
5384 {
5385 	u32 data, orig;
5386 	u32 pcie_lc_cntl, pcie_lc_cntl_old;
5387 	bool disable_l0s, disable_l1 = false, disable_plloff_in_l1 = false;
5388 	/* fusion_platform = true
5389 	 * if the system is a fusion system
5390 	 * (APU or DGPU in a fusion system).
5391 	 * todo: check if the system is a fusion platform.
5392 	 */
5393 	bool fusion_platform = false;
5394 
5395 	if (radeon_aspm == 0)
5396 		return;
5397 
5398 	if (!(rdev->flags & RADEON_IS_PCIE))
5399 		return;
5400 
5401 	switch (rdev->family) {
5402 	case CHIP_CYPRESS:
5403 	case CHIP_HEMLOCK:
5404 	case CHIP_JUNIPER:
5405 	case CHIP_REDWOOD:
5406 	case CHIP_CEDAR:
5407 	case CHIP_SUMO:
5408 	case CHIP_SUMO2:
5409 	case CHIP_PALM:
5410 	case CHIP_ARUBA:
5411 		disable_l0s = true;
5412 		break;
5413 	default:
5414 		disable_l0s = false;
5415 		break;
5416 	}
5417 
5418 	if (rdev->flags & RADEON_IS_IGP)
5419 		fusion_platform = true; /* XXX also dGPUs in a fusion system */
5420 
5421 	data = orig = RREG32_PIF_PHY0(PB0_PIF_PAIRING);
5422 	if (fusion_platform)
5423 		data &= ~MULTI_PIF;
5424 	else
5425 		data |= MULTI_PIF;
5426 	if (data != orig)
5427 		WREG32_PIF_PHY0(PB0_PIF_PAIRING, data);
5428 
5429 	data = orig = RREG32_PIF_PHY1(PB1_PIF_PAIRING);
5430 	if (fusion_platform)
5431 		data &= ~MULTI_PIF;
5432 	else
5433 		data |= MULTI_PIF;
5434 	if (data != orig)
5435 		WREG32_PIF_PHY1(PB1_PIF_PAIRING, data);
5436 
5437 	pcie_lc_cntl = pcie_lc_cntl_old = RREG32_PCIE_PORT(PCIE_LC_CNTL);
5438 	pcie_lc_cntl &= ~(LC_L0S_INACTIVITY_MASK | LC_L1_INACTIVITY_MASK);
5439 	if (!disable_l0s) {
5440 		if (rdev->family >= CHIP_BARTS)
5441 			pcie_lc_cntl |= LC_L0S_INACTIVITY(7);
5442 		else
5443 			pcie_lc_cntl |= LC_L0S_INACTIVITY(3);
5444 	}
5445 
5446 	if (!disable_l1) {
5447 		if (rdev->family >= CHIP_BARTS)
5448 			pcie_lc_cntl |= LC_L1_INACTIVITY(7);
5449 		else
5450 			pcie_lc_cntl |= LC_L1_INACTIVITY(8);
5451 
5452 		if (!disable_plloff_in_l1) {
5453 			data = orig = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0);
5454 			data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
5455 			data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
5456 			if (data != orig)
5457 				WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0, data);
5458 
5459 			data = orig = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1);
5460 			data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
5461 			data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
5462 			if (data != orig)
5463 				WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1, data);
5464 
5465 			data = orig = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0);
5466 			data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK);
5467 			data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7);
5468 			if (data != orig)
5469 				WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0, data);
5470 
5471 			data = orig = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1);
5472 			data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK);
5473 			data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7);
5474 			if (data != orig)
5475 				WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1, data);
5476 
5477 			if (rdev->family >= CHIP_BARTS) {
5478 				data = orig = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0);
5479 				data &= ~PLL_RAMP_UP_TIME_0_MASK;
5480 				data |= PLL_RAMP_UP_TIME_0(4);
5481 				if (data != orig)
5482 					WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0, data);
5483 
5484 				data = orig = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1);
5485 				data &= ~PLL_RAMP_UP_TIME_1_MASK;
5486 				data |= PLL_RAMP_UP_TIME_1(4);
5487 				if (data != orig)
5488 					WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1, data);
5489 
5490 				data = orig = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0);
5491 				data &= ~PLL_RAMP_UP_TIME_0_MASK;
5492 				data |= PLL_RAMP_UP_TIME_0(4);
5493 				if (data != orig)
5494 					WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0, data);
5495 
5496 				data = orig = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1);
5497 				data &= ~PLL_RAMP_UP_TIME_1_MASK;
5498 				data |= PLL_RAMP_UP_TIME_1(4);
5499 				if (data != orig)
5500 					WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1, data);
5501 			}
5502 
5503 			data = orig = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL);
5504 			data &= ~LC_DYN_LANES_PWR_STATE_MASK;
5505 			data |= LC_DYN_LANES_PWR_STATE(3);
5506 			if (data != orig)
5507 				WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data);
5508 
5509 			if (rdev->family >= CHIP_BARTS) {
5510 				data = orig = RREG32_PIF_PHY0(PB0_PIF_CNTL);
5511 				data &= ~LS2_EXIT_TIME_MASK;
5512 				data |= LS2_EXIT_TIME(1);
5513 				if (data != orig)
5514 					WREG32_PIF_PHY0(PB0_PIF_CNTL, data);
5515 
5516 				data = orig = RREG32_PIF_PHY1(PB1_PIF_CNTL);
5517 				data &= ~LS2_EXIT_TIME_MASK;
5518 				data |= LS2_EXIT_TIME(1);
5519 				if (data != orig)
5520 					WREG32_PIF_PHY1(PB1_PIF_CNTL, data);
5521 			}
5522 		}
5523 	}
5524 
5525 	/* evergreen parts only */
5526 	if (rdev->family < CHIP_BARTS)
5527 		pcie_lc_cntl |= LC_PMI_TO_L1_DIS;
5528 
5529 	if (pcie_lc_cntl != pcie_lc_cntl_old)
5530 		WREG32_PCIE_PORT(PCIE_LC_CNTL, pcie_lc_cntl);
5531 }
5532