xref: /linux/drivers/gpu/drm/radeon/r600.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
1 /*
2  * Copyright 2008 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 #include <linux/slab.h>
29 #include <linux/seq_file.h>
30 #include <linux/firmware.h>
31 #include <linux/platform_device.h>
32 #include <linux/module.h>
33 #include <drm/drmP.h>
34 #include <drm/radeon_drm.h>
35 #include "radeon.h"
36 #include "radeon_asic.h"
37 #include "radeon_mode.h"
38 #include "r600d.h"
39 #include "atom.h"
40 #include "avivod.h"
41 
42 #define PFP_UCODE_SIZE 576
43 #define PM4_UCODE_SIZE 1792
44 #define RLC_UCODE_SIZE 768
45 #define R700_PFP_UCODE_SIZE 848
46 #define R700_PM4_UCODE_SIZE 1360
47 #define R700_RLC_UCODE_SIZE 1024
48 #define EVERGREEN_PFP_UCODE_SIZE 1120
49 #define EVERGREEN_PM4_UCODE_SIZE 1376
50 #define EVERGREEN_RLC_UCODE_SIZE 768
51 #define CAYMAN_RLC_UCODE_SIZE 1024
52 #define ARUBA_RLC_UCODE_SIZE 1536
53 
54 /* Firmware Names */
55 MODULE_FIRMWARE("radeon/R600_pfp.bin");
56 MODULE_FIRMWARE("radeon/R600_me.bin");
57 MODULE_FIRMWARE("radeon/RV610_pfp.bin");
58 MODULE_FIRMWARE("radeon/RV610_me.bin");
59 MODULE_FIRMWARE("radeon/RV630_pfp.bin");
60 MODULE_FIRMWARE("radeon/RV630_me.bin");
61 MODULE_FIRMWARE("radeon/RV620_pfp.bin");
62 MODULE_FIRMWARE("radeon/RV620_me.bin");
63 MODULE_FIRMWARE("radeon/RV635_pfp.bin");
64 MODULE_FIRMWARE("radeon/RV635_me.bin");
65 MODULE_FIRMWARE("radeon/RV670_pfp.bin");
66 MODULE_FIRMWARE("radeon/RV670_me.bin");
67 MODULE_FIRMWARE("radeon/RS780_pfp.bin");
68 MODULE_FIRMWARE("radeon/RS780_me.bin");
69 MODULE_FIRMWARE("radeon/RV770_pfp.bin");
70 MODULE_FIRMWARE("radeon/RV770_me.bin");
71 MODULE_FIRMWARE("radeon/RV730_pfp.bin");
72 MODULE_FIRMWARE("radeon/RV730_me.bin");
73 MODULE_FIRMWARE("radeon/RV710_pfp.bin");
74 MODULE_FIRMWARE("radeon/RV710_me.bin");
75 MODULE_FIRMWARE("radeon/R600_rlc.bin");
76 MODULE_FIRMWARE("radeon/R700_rlc.bin");
77 MODULE_FIRMWARE("radeon/CEDAR_pfp.bin");
78 MODULE_FIRMWARE("radeon/CEDAR_me.bin");
79 MODULE_FIRMWARE("radeon/CEDAR_rlc.bin");
80 MODULE_FIRMWARE("radeon/REDWOOD_pfp.bin");
81 MODULE_FIRMWARE("radeon/REDWOOD_me.bin");
82 MODULE_FIRMWARE("radeon/REDWOOD_rlc.bin");
83 MODULE_FIRMWARE("radeon/JUNIPER_pfp.bin");
84 MODULE_FIRMWARE("radeon/JUNIPER_me.bin");
85 MODULE_FIRMWARE("radeon/JUNIPER_rlc.bin");
86 MODULE_FIRMWARE("radeon/CYPRESS_pfp.bin");
87 MODULE_FIRMWARE("radeon/CYPRESS_me.bin");
88 MODULE_FIRMWARE("radeon/CYPRESS_rlc.bin");
89 MODULE_FIRMWARE("radeon/PALM_pfp.bin");
90 MODULE_FIRMWARE("radeon/PALM_me.bin");
91 MODULE_FIRMWARE("radeon/SUMO_rlc.bin");
92 MODULE_FIRMWARE("radeon/SUMO_pfp.bin");
93 MODULE_FIRMWARE("radeon/SUMO_me.bin");
94 MODULE_FIRMWARE("radeon/SUMO2_pfp.bin");
95 MODULE_FIRMWARE("radeon/SUMO2_me.bin");
96 
97 int r600_debugfs_mc_info_init(struct radeon_device *rdev);
98 
99 /* r600,rv610,rv630,rv620,rv635,rv670 */
100 int r600_mc_wait_for_idle(struct radeon_device *rdev);
101 static void r600_gpu_init(struct radeon_device *rdev);
102 void r600_fini(struct radeon_device *rdev);
103 void r600_irq_disable(struct radeon_device *rdev);
104 static void r600_pcie_gen2_enable(struct radeon_device *rdev);
105 
106 /* get temperature in millidegrees */
107 int rv6xx_get_temp(struct radeon_device *rdev)
108 {
109 	u32 temp = (RREG32(CG_THERMAL_STATUS) & ASIC_T_MASK) >>
110 		ASIC_T_SHIFT;
111 	int actual_temp = temp & 0xff;
112 
113 	if (temp & 0x100)
114 		actual_temp -= 256;
115 
116 	return actual_temp * 1000;
117 }
118 
119 void r600_pm_get_dynpm_state(struct radeon_device *rdev)
120 {
121 	int i;
122 
123 	rdev->pm.dynpm_can_upclock = true;
124 	rdev->pm.dynpm_can_downclock = true;
125 
126 	/* power state array is low to high, default is first */
127 	if ((rdev->flags & RADEON_IS_IGP) || (rdev->family == CHIP_R600)) {
128 		int min_power_state_index = 0;
129 
130 		if (rdev->pm.num_power_states > 2)
131 			min_power_state_index = 1;
132 
133 		switch (rdev->pm.dynpm_planned_action) {
134 		case DYNPM_ACTION_MINIMUM:
135 			rdev->pm.requested_power_state_index = min_power_state_index;
136 			rdev->pm.requested_clock_mode_index = 0;
137 			rdev->pm.dynpm_can_downclock = false;
138 			break;
139 		case DYNPM_ACTION_DOWNCLOCK:
140 			if (rdev->pm.current_power_state_index == min_power_state_index) {
141 				rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
142 				rdev->pm.dynpm_can_downclock = false;
143 			} else {
144 				if (rdev->pm.active_crtc_count > 1) {
145 					for (i = 0; i < rdev->pm.num_power_states; i++) {
146 						if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
147 							continue;
148 						else if (i >= rdev->pm.current_power_state_index) {
149 							rdev->pm.requested_power_state_index =
150 								rdev->pm.current_power_state_index;
151 							break;
152 						} else {
153 							rdev->pm.requested_power_state_index = i;
154 							break;
155 						}
156 					}
157 				} else {
158 					if (rdev->pm.current_power_state_index == 0)
159 						rdev->pm.requested_power_state_index =
160 							rdev->pm.num_power_states - 1;
161 					else
162 						rdev->pm.requested_power_state_index =
163 							rdev->pm.current_power_state_index - 1;
164 				}
165 			}
166 			rdev->pm.requested_clock_mode_index = 0;
167 			/* don't use the power state if crtcs are active and no display flag is set */
168 			if ((rdev->pm.active_crtc_count > 0) &&
169 			    (rdev->pm.power_state[rdev->pm.requested_power_state_index].
170 			     clock_info[rdev->pm.requested_clock_mode_index].flags &
171 			     RADEON_PM_MODE_NO_DISPLAY)) {
172 				rdev->pm.requested_power_state_index++;
173 			}
174 			break;
175 		case DYNPM_ACTION_UPCLOCK:
176 			if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
177 				rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
178 				rdev->pm.dynpm_can_upclock = false;
179 			} else {
180 				if (rdev->pm.active_crtc_count > 1) {
181 					for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
182 						if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
183 							continue;
184 						else if (i <= rdev->pm.current_power_state_index) {
185 							rdev->pm.requested_power_state_index =
186 								rdev->pm.current_power_state_index;
187 							break;
188 						} else {
189 							rdev->pm.requested_power_state_index = i;
190 							break;
191 						}
192 					}
193 				} else
194 					rdev->pm.requested_power_state_index =
195 						rdev->pm.current_power_state_index + 1;
196 			}
197 			rdev->pm.requested_clock_mode_index = 0;
198 			break;
199 		case DYNPM_ACTION_DEFAULT:
200 			rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
201 			rdev->pm.requested_clock_mode_index = 0;
202 			rdev->pm.dynpm_can_upclock = false;
203 			break;
204 		case DYNPM_ACTION_NONE:
205 		default:
206 			DRM_ERROR("Requested mode for not defined action\n");
207 			return;
208 		}
209 	} else {
210 		/* XXX select a power state based on AC/DC, single/dualhead, etc. */
211 		/* for now just select the first power state and switch between clock modes */
212 		/* power state array is low to high, default is first (0) */
213 		if (rdev->pm.active_crtc_count > 1) {
214 			rdev->pm.requested_power_state_index = -1;
215 			/* start at 1 as we don't want the default mode */
216 			for (i = 1; i < rdev->pm.num_power_states; i++) {
217 				if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
218 					continue;
219 				else if ((rdev->pm.power_state[i].type == POWER_STATE_TYPE_PERFORMANCE) ||
220 					 (rdev->pm.power_state[i].type == POWER_STATE_TYPE_BATTERY)) {
221 					rdev->pm.requested_power_state_index = i;
222 					break;
223 				}
224 			}
225 			/* if nothing selected, grab the default state. */
226 			if (rdev->pm.requested_power_state_index == -1)
227 				rdev->pm.requested_power_state_index = 0;
228 		} else
229 			rdev->pm.requested_power_state_index = 1;
230 
231 		switch (rdev->pm.dynpm_planned_action) {
232 		case DYNPM_ACTION_MINIMUM:
233 			rdev->pm.requested_clock_mode_index = 0;
234 			rdev->pm.dynpm_can_downclock = false;
235 			break;
236 		case DYNPM_ACTION_DOWNCLOCK:
237 			if (rdev->pm.requested_power_state_index == rdev->pm.current_power_state_index) {
238 				if (rdev->pm.current_clock_mode_index == 0) {
239 					rdev->pm.requested_clock_mode_index = 0;
240 					rdev->pm.dynpm_can_downclock = false;
241 				} else
242 					rdev->pm.requested_clock_mode_index =
243 						rdev->pm.current_clock_mode_index - 1;
244 			} else {
245 				rdev->pm.requested_clock_mode_index = 0;
246 				rdev->pm.dynpm_can_downclock = false;
247 			}
248 			/* don't use the power state if crtcs are active and no display flag is set */
249 			if ((rdev->pm.active_crtc_count > 0) &&
250 			    (rdev->pm.power_state[rdev->pm.requested_power_state_index].
251 			     clock_info[rdev->pm.requested_clock_mode_index].flags &
252 			     RADEON_PM_MODE_NO_DISPLAY)) {
253 				rdev->pm.requested_clock_mode_index++;
254 			}
255 			break;
256 		case DYNPM_ACTION_UPCLOCK:
257 			if (rdev->pm.requested_power_state_index == rdev->pm.current_power_state_index) {
258 				if (rdev->pm.current_clock_mode_index ==
259 				    (rdev->pm.power_state[rdev->pm.requested_power_state_index].num_clock_modes - 1)) {
260 					rdev->pm.requested_clock_mode_index = rdev->pm.current_clock_mode_index;
261 					rdev->pm.dynpm_can_upclock = false;
262 				} else
263 					rdev->pm.requested_clock_mode_index =
264 						rdev->pm.current_clock_mode_index + 1;
265 			} else {
266 				rdev->pm.requested_clock_mode_index =
267 					rdev->pm.power_state[rdev->pm.requested_power_state_index].num_clock_modes - 1;
268 				rdev->pm.dynpm_can_upclock = false;
269 			}
270 			break;
271 		case DYNPM_ACTION_DEFAULT:
272 			rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
273 			rdev->pm.requested_clock_mode_index = 0;
274 			rdev->pm.dynpm_can_upclock = false;
275 			break;
276 		case DYNPM_ACTION_NONE:
277 		default:
278 			DRM_ERROR("Requested mode for not defined action\n");
279 			return;
280 		}
281 	}
282 
283 	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
284 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
285 		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
286 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
287 		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
288 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
289 		  pcie_lanes);
290 }
291 
292 void rs780_pm_init_profile(struct radeon_device *rdev)
293 {
294 	if (rdev->pm.num_power_states == 2) {
295 		/* default */
296 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
297 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
298 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
299 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
300 		/* low sh */
301 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
302 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
303 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
304 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
305 		/* mid sh */
306 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
307 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
308 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
309 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
310 		/* high sh */
311 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
312 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = 1;
313 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
314 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
315 		/* low mh */
316 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
317 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = 0;
318 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
319 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
320 		/* mid mh */
321 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
322 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = 0;
323 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
324 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
325 		/* high mh */
326 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
327 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = 1;
328 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
329 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
330 	} else if (rdev->pm.num_power_states == 3) {
331 		/* default */
332 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
333 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
334 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
335 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
336 		/* low sh */
337 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 1;
338 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 1;
339 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
340 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
341 		/* mid sh */
342 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 1;
343 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 1;
344 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
345 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
346 		/* high sh */
347 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 1;
348 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = 2;
349 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
350 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
351 		/* low mh */
352 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 1;
353 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = 1;
354 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
355 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
356 		/* mid mh */
357 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 1;
358 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = 1;
359 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
360 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
361 		/* high mh */
362 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 1;
363 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = 2;
364 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
365 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
366 	} else {
367 		/* default */
368 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
369 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
370 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
371 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
372 		/* low sh */
373 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 2;
374 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 2;
375 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
376 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
377 		/* mid sh */
378 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 2;
379 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 2;
380 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
381 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
382 		/* high sh */
383 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 2;
384 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = 3;
385 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
386 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
387 		/* low mh */
388 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 2;
389 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = 0;
390 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
391 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
392 		/* mid mh */
393 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 2;
394 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = 0;
395 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
396 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
397 		/* high mh */
398 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 2;
399 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = 3;
400 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
401 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
402 	}
403 }
404 
405 void r600_pm_init_profile(struct radeon_device *rdev)
406 {
407 	int idx;
408 
409 	if (rdev->family == CHIP_R600) {
410 		/* XXX */
411 		/* default */
412 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
413 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
414 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
415 		rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
416 		/* low sh */
417 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
418 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
419 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
420 		rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
421 		/* mid sh */
422 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
423 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
424 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
425 		rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
426 		/* high sh */
427 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
428 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
429 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
430 		rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
431 		/* low mh */
432 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
433 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
434 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
435 		rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
436 		/* mid mh */
437 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
438 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
439 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
440 		rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
441 		/* high mh */
442 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
443 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
444 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
445 		rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
446 	} else {
447 		if (rdev->pm.num_power_states < 4) {
448 			/* default */
449 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
450 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
451 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
452 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 2;
453 			/* low sh */
454 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 1;
455 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 1;
456 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
457 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
458 			/* mid sh */
459 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 1;
460 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 1;
461 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
462 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 1;
463 			/* high sh */
464 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 1;
465 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = 1;
466 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
467 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 2;
468 			/* low mh */
469 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 2;
470 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = 2;
471 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
472 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
473 			/* low mh */
474 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 2;
475 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = 2;
476 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
477 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 1;
478 			/* high mh */
479 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 2;
480 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = 2;
481 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
482 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 2;
483 		} else {
484 			/* default */
485 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
486 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
487 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
488 			rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 2;
489 			/* low sh */
490 			if (rdev->flags & RADEON_IS_MOBILITY)
491 				idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_BATTERY, 0);
492 			else
493 				idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
494 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = idx;
495 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = idx;
496 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
497 			rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
498 			/* mid sh */
499 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = idx;
500 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = idx;
501 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
502 			rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 1;
503 			/* high sh */
504 			idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
505 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = idx;
506 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = idx;
507 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
508 			rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 2;
509 			/* low mh */
510 			if (rdev->flags & RADEON_IS_MOBILITY)
511 				idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_BATTERY, 1);
512 			else
513 				idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 1);
514 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = idx;
515 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = idx;
516 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
517 			rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
518 			/* mid mh */
519 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = idx;
520 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = idx;
521 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
522 			rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 1;
523 			/* high mh */
524 			idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 1);
525 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = idx;
526 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = idx;
527 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
528 			rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 2;
529 		}
530 	}
531 }
532 
533 void r600_pm_misc(struct radeon_device *rdev)
534 {
535 	int req_ps_idx = rdev->pm.requested_power_state_index;
536 	int req_cm_idx = rdev->pm.requested_clock_mode_index;
537 	struct radeon_power_state *ps = &rdev->pm.power_state[req_ps_idx];
538 	struct radeon_voltage *voltage = &ps->clock_info[req_cm_idx].voltage;
539 
540 	if ((voltage->type == VOLTAGE_SW) && voltage->voltage) {
541 		/* 0xff01 is a flag rather then an actual voltage */
542 		if (voltage->voltage == 0xff01)
543 			return;
544 		if (voltage->voltage != rdev->pm.current_vddc) {
545 			radeon_atom_set_voltage(rdev, voltage->voltage, SET_VOLTAGE_TYPE_ASIC_VDDC);
546 			rdev->pm.current_vddc = voltage->voltage;
547 			DRM_DEBUG_DRIVER("Setting: v: %d\n", voltage->voltage);
548 		}
549 	}
550 }
551 
552 bool r600_gui_idle(struct radeon_device *rdev)
553 {
554 	if (RREG32(GRBM_STATUS) & GUI_ACTIVE)
555 		return false;
556 	else
557 		return true;
558 }
559 
560 /* hpd for digital panel detect/disconnect */
561 bool r600_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
562 {
563 	bool connected = false;
564 
565 	if (ASIC_IS_DCE3(rdev)) {
566 		switch (hpd) {
567 		case RADEON_HPD_1:
568 			if (RREG32(DC_HPD1_INT_STATUS) & DC_HPDx_SENSE)
569 				connected = true;
570 			break;
571 		case RADEON_HPD_2:
572 			if (RREG32(DC_HPD2_INT_STATUS) & DC_HPDx_SENSE)
573 				connected = true;
574 			break;
575 		case RADEON_HPD_3:
576 			if (RREG32(DC_HPD3_INT_STATUS) & DC_HPDx_SENSE)
577 				connected = true;
578 			break;
579 		case RADEON_HPD_4:
580 			if (RREG32(DC_HPD4_INT_STATUS) & DC_HPDx_SENSE)
581 				connected = true;
582 			break;
583 			/* DCE 3.2 */
584 		case RADEON_HPD_5:
585 			if (RREG32(DC_HPD5_INT_STATUS) & DC_HPDx_SENSE)
586 				connected = true;
587 			break;
588 		case RADEON_HPD_6:
589 			if (RREG32(DC_HPD6_INT_STATUS) & DC_HPDx_SENSE)
590 				connected = true;
591 			break;
592 		default:
593 			break;
594 		}
595 	} else {
596 		switch (hpd) {
597 		case RADEON_HPD_1:
598 			if (RREG32(DC_HOT_PLUG_DETECT1_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
599 				connected = true;
600 			break;
601 		case RADEON_HPD_2:
602 			if (RREG32(DC_HOT_PLUG_DETECT2_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
603 				connected = true;
604 			break;
605 		case RADEON_HPD_3:
606 			if (RREG32(DC_HOT_PLUG_DETECT3_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
607 				connected = true;
608 			break;
609 		default:
610 			break;
611 		}
612 	}
613 	return connected;
614 }
615 
616 void r600_hpd_set_polarity(struct radeon_device *rdev,
617 			   enum radeon_hpd_id hpd)
618 {
619 	u32 tmp;
620 	bool connected = r600_hpd_sense(rdev, hpd);
621 
622 	if (ASIC_IS_DCE3(rdev)) {
623 		switch (hpd) {
624 		case RADEON_HPD_1:
625 			tmp = RREG32(DC_HPD1_INT_CONTROL);
626 			if (connected)
627 				tmp &= ~DC_HPDx_INT_POLARITY;
628 			else
629 				tmp |= DC_HPDx_INT_POLARITY;
630 			WREG32(DC_HPD1_INT_CONTROL, tmp);
631 			break;
632 		case RADEON_HPD_2:
633 			tmp = RREG32(DC_HPD2_INT_CONTROL);
634 			if (connected)
635 				tmp &= ~DC_HPDx_INT_POLARITY;
636 			else
637 				tmp |= DC_HPDx_INT_POLARITY;
638 			WREG32(DC_HPD2_INT_CONTROL, tmp);
639 			break;
640 		case RADEON_HPD_3:
641 			tmp = RREG32(DC_HPD3_INT_CONTROL);
642 			if (connected)
643 				tmp &= ~DC_HPDx_INT_POLARITY;
644 			else
645 				tmp |= DC_HPDx_INT_POLARITY;
646 			WREG32(DC_HPD3_INT_CONTROL, tmp);
647 			break;
648 		case RADEON_HPD_4:
649 			tmp = RREG32(DC_HPD4_INT_CONTROL);
650 			if (connected)
651 				tmp &= ~DC_HPDx_INT_POLARITY;
652 			else
653 				tmp |= DC_HPDx_INT_POLARITY;
654 			WREG32(DC_HPD4_INT_CONTROL, tmp);
655 			break;
656 		case RADEON_HPD_5:
657 			tmp = RREG32(DC_HPD5_INT_CONTROL);
658 			if (connected)
659 				tmp &= ~DC_HPDx_INT_POLARITY;
660 			else
661 				tmp |= DC_HPDx_INT_POLARITY;
662 			WREG32(DC_HPD5_INT_CONTROL, tmp);
663 			break;
664 			/* DCE 3.2 */
665 		case RADEON_HPD_6:
666 			tmp = RREG32(DC_HPD6_INT_CONTROL);
667 			if (connected)
668 				tmp &= ~DC_HPDx_INT_POLARITY;
669 			else
670 				tmp |= DC_HPDx_INT_POLARITY;
671 			WREG32(DC_HPD6_INT_CONTROL, tmp);
672 			break;
673 		default:
674 			break;
675 		}
676 	} else {
677 		switch (hpd) {
678 		case RADEON_HPD_1:
679 			tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL);
680 			if (connected)
681 				tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
682 			else
683 				tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
684 			WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
685 			break;
686 		case RADEON_HPD_2:
687 			tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL);
688 			if (connected)
689 				tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
690 			else
691 				tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
692 			WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
693 			break;
694 		case RADEON_HPD_3:
695 			tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL);
696 			if (connected)
697 				tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
698 			else
699 				tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
700 			WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
701 			break;
702 		default:
703 			break;
704 		}
705 	}
706 }
707 
708 void r600_hpd_init(struct radeon_device *rdev)
709 {
710 	struct drm_device *dev = rdev->ddev;
711 	struct drm_connector *connector;
712 	unsigned enable = 0;
713 
714 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
715 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
716 
717 		if (connector->connector_type == DRM_MODE_CONNECTOR_eDP ||
718 		    connector->connector_type == DRM_MODE_CONNECTOR_LVDS) {
719 			/* don't try to enable hpd on eDP or LVDS avoid breaking the
720 			 * aux dp channel on imac and help (but not completely fix)
721 			 * https://bugzilla.redhat.com/show_bug.cgi?id=726143
722 			 */
723 			continue;
724 		}
725 		if (ASIC_IS_DCE3(rdev)) {
726 			u32 tmp = DC_HPDx_CONNECTION_TIMER(0x9c4) | DC_HPDx_RX_INT_TIMER(0xfa);
727 			if (ASIC_IS_DCE32(rdev))
728 				tmp |= DC_HPDx_EN;
729 
730 			switch (radeon_connector->hpd.hpd) {
731 			case RADEON_HPD_1:
732 				WREG32(DC_HPD1_CONTROL, tmp);
733 				break;
734 			case RADEON_HPD_2:
735 				WREG32(DC_HPD2_CONTROL, tmp);
736 				break;
737 			case RADEON_HPD_3:
738 				WREG32(DC_HPD3_CONTROL, tmp);
739 				break;
740 			case RADEON_HPD_4:
741 				WREG32(DC_HPD4_CONTROL, tmp);
742 				break;
743 				/* DCE 3.2 */
744 			case RADEON_HPD_5:
745 				WREG32(DC_HPD5_CONTROL, tmp);
746 				break;
747 			case RADEON_HPD_6:
748 				WREG32(DC_HPD6_CONTROL, tmp);
749 				break;
750 			default:
751 				break;
752 			}
753 		} else {
754 			switch (radeon_connector->hpd.hpd) {
755 			case RADEON_HPD_1:
756 				WREG32(DC_HOT_PLUG_DETECT1_CONTROL, DC_HOT_PLUG_DETECTx_EN);
757 				break;
758 			case RADEON_HPD_2:
759 				WREG32(DC_HOT_PLUG_DETECT2_CONTROL, DC_HOT_PLUG_DETECTx_EN);
760 				break;
761 			case RADEON_HPD_3:
762 				WREG32(DC_HOT_PLUG_DETECT3_CONTROL, DC_HOT_PLUG_DETECTx_EN);
763 				break;
764 			default:
765 				break;
766 			}
767 		}
768 		enable |= 1 << radeon_connector->hpd.hpd;
769 		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
770 	}
771 	radeon_irq_kms_enable_hpd(rdev, enable);
772 }
773 
774 void r600_hpd_fini(struct radeon_device *rdev)
775 {
776 	struct drm_device *dev = rdev->ddev;
777 	struct drm_connector *connector;
778 	unsigned disable = 0;
779 
780 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
781 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
782 		if (ASIC_IS_DCE3(rdev)) {
783 			switch (radeon_connector->hpd.hpd) {
784 			case RADEON_HPD_1:
785 				WREG32(DC_HPD1_CONTROL, 0);
786 				break;
787 			case RADEON_HPD_2:
788 				WREG32(DC_HPD2_CONTROL, 0);
789 				break;
790 			case RADEON_HPD_3:
791 				WREG32(DC_HPD3_CONTROL, 0);
792 				break;
793 			case RADEON_HPD_4:
794 				WREG32(DC_HPD4_CONTROL, 0);
795 				break;
796 				/* DCE 3.2 */
797 			case RADEON_HPD_5:
798 				WREG32(DC_HPD5_CONTROL, 0);
799 				break;
800 			case RADEON_HPD_6:
801 				WREG32(DC_HPD6_CONTROL, 0);
802 				break;
803 			default:
804 				break;
805 			}
806 		} else {
807 			switch (radeon_connector->hpd.hpd) {
808 			case RADEON_HPD_1:
809 				WREG32(DC_HOT_PLUG_DETECT1_CONTROL, 0);
810 				break;
811 			case RADEON_HPD_2:
812 				WREG32(DC_HOT_PLUG_DETECT2_CONTROL, 0);
813 				break;
814 			case RADEON_HPD_3:
815 				WREG32(DC_HOT_PLUG_DETECT3_CONTROL, 0);
816 				break;
817 			default:
818 				break;
819 			}
820 		}
821 		disable |= 1 << radeon_connector->hpd.hpd;
822 	}
823 	radeon_irq_kms_disable_hpd(rdev, disable);
824 }
825 
826 /*
827  * R600 PCIE GART
828  */
829 void r600_pcie_gart_tlb_flush(struct radeon_device *rdev)
830 {
831 	unsigned i;
832 	u32 tmp;
833 
834 	/* flush hdp cache so updates hit vram */
835 	if ((rdev->family >= CHIP_RV770) && (rdev->family <= CHIP_RV740) &&
836 	    !(rdev->flags & RADEON_IS_AGP)) {
837 		void __iomem *ptr = (void *)rdev->gart.ptr;
838 		u32 tmp;
839 
840 		/* r7xx hw bug.  write to HDP_DEBUG1 followed by fb read
841 		 * rather than write to HDP_REG_COHERENCY_FLUSH_CNTL
842 		 * This seems to cause problems on some AGP cards. Just use the old
843 		 * method for them.
844 		 */
845 		WREG32(HDP_DEBUG1, 0);
846 		tmp = readl((void __iomem *)ptr);
847 	} else
848 		WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
849 
850 	WREG32(VM_CONTEXT0_INVALIDATION_LOW_ADDR, rdev->mc.gtt_start >> 12);
851 	WREG32(VM_CONTEXT0_INVALIDATION_HIGH_ADDR, (rdev->mc.gtt_end - 1) >> 12);
852 	WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
853 	for (i = 0; i < rdev->usec_timeout; i++) {
854 		/* read MC_STATUS */
855 		tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
856 		tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
857 		if (tmp == 2) {
858 			printk(KERN_WARNING "[drm] r600 flush TLB failed\n");
859 			return;
860 		}
861 		if (tmp) {
862 			return;
863 		}
864 		udelay(1);
865 	}
866 }
867 
868 int r600_pcie_gart_init(struct radeon_device *rdev)
869 {
870 	int r;
871 
872 	if (rdev->gart.robj) {
873 		WARN(1, "R600 PCIE GART already initialized\n");
874 		return 0;
875 	}
876 	/* Initialize common gart structure */
877 	r = radeon_gart_init(rdev);
878 	if (r)
879 		return r;
880 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 8;
881 	return radeon_gart_table_vram_alloc(rdev);
882 }
883 
884 static int r600_pcie_gart_enable(struct radeon_device *rdev)
885 {
886 	u32 tmp;
887 	int r, i;
888 
889 	if (rdev->gart.robj == NULL) {
890 		dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
891 		return -EINVAL;
892 	}
893 	r = radeon_gart_table_vram_pin(rdev);
894 	if (r)
895 		return r;
896 	radeon_gart_restore(rdev);
897 
898 	/* Setup L2 cache */
899 	WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
900 				ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
901 				EFFECTIVE_L2_QUEUE_SIZE(7));
902 	WREG32(VM_L2_CNTL2, 0);
903 	WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
904 	/* Setup TLB control */
905 	tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
906 		SYSTEM_ACCESS_MODE_NOT_IN_SYS |
907 		EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
908 		ENABLE_WAIT_L2_QUERY;
909 	WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
910 	WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
911 	WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
912 	WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
913 	WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
914 	WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
915 	WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
916 	WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
917 	WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
918 	WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
919 	WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
920 	WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
921 	WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
922 	WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
923 	WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
924 	WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
925 	WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
926 	WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
927 				RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
928 	WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
929 			(u32)(rdev->dummy_page.addr >> 12));
930 	for (i = 1; i < 7; i++)
931 		WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
932 
933 	r600_pcie_gart_tlb_flush(rdev);
934 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
935 		 (unsigned)(rdev->mc.gtt_size >> 20),
936 		 (unsigned long long)rdev->gart.table_addr);
937 	rdev->gart.ready = true;
938 	return 0;
939 }
940 
941 static void r600_pcie_gart_disable(struct radeon_device *rdev)
942 {
943 	u32 tmp;
944 	int i;
945 
946 	/* Disable all tables */
947 	for (i = 0; i < 7; i++)
948 		WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
949 
950 	/* Disable L2 cache */
951 	WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
952 				EFFECTIVE_L2_QUEUE_SIZE(7));
953 	WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
954 	/* Setup L1 TLB control */
955 	tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
956 		ENABLE_WAIT_L2_QUERY;
957 	WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
958 	WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
959 	WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
960 	WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
961 	WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
962 	WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
963 	WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
964 	WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
965 	WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp);
966 	WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp);
967 	WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
968 	WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
969 	WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp);
970 	WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
971 	radeon_gart_table_vram_unpin(rdev);
972 }
973 
974 static void r600_pcie_gart_fini(struct radeon_device *rdev)
975 {
976 	radeon_gart_fini(rdev);
977 	r600_pcie_gart_disable(rdev);
978 	radeon_gart_table_vram_free(rdev);
979 }
980 
981 static void r600_agp_enable(struct radeon_device *rdev)
982 {
983 	u32 tmp;
984 	int i;
985 
986 	/* Setup L2 cache */
987 	WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
988 				ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
989 				EFFECTIVE_L2_QUEUE_SIZE(7));
990 	WREG32(VM_L2_CNTL2, 0);
991 	WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
992 	/* Setup TLB control */
993 	tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
994 		SYSTEM_ACCESS_MODE_NOT_IN_SYS |
995 		EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
996 		ENABLE_WAIT_L2_QUERY;
997 	WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
998 	WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
999 	WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
1000 	WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
1001 	WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
1002 	WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
1003 	WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
1004 	WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
1005 	WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
1006 	WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
1007 	WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
1008 	WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
1009 	WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
1010 	WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
1011 	for (i = 0; i < 7; i++)
1012 		WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
1013 }
1014 
1015 int r600_mc_wait_for_idle(struct radeon_device *rdev)
1016 {
1017 	unsigned i;
1018 	u32 tmp;
1019 
1020 	for (i = 0; i < rdev->usec_timeout; i++) {
1021 		/* read MC_STATUS */
1022 		tmp = RREG32(R_000E50_SRBM_STATUS) & 0x3F00;
1023 		if (!tmp)
1024 			return 0;
1025 		udelay(1);
1026 	}
1027 	return -1;
1028 }
1029 
1030 static void r600_mc_program(struct radeon_device *rdev)
1031 {
1032 	struct rv515_mc_save save;
1033 	u32 tmp;
1034 	int i, j;
1035 
1036 	/* Initialize HDP */
1037 	for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1038 		WREG32((0x2c14 + j), 0x00000000);
1039 		WREG32((0x2c18 + j), 0x00000000);
1040 		WREG32((0x2c1c + j), 0x00000000);
1041 		WREG32((0x2c20 + j), 0x00000000);
1042 		WREG32((0x2c24 + j), 0x00000000);
1043 	}
1044 	WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
1045 
1046 	rv515_mc_stop(rdev, &save);
1047 	if (r600_mc_wait_for_idle(rdev)) {
1048 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1049 	}
1050 	/* Lockout access through VGA aperture (doesn't exist before R600) */
1051 	WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
1052 	/* Update configuration */
1053 	if (rdev->flags & RADEON_IS_AGP) {
1054 		if (rdev->mc.vram_start < rdev->mc.gtt_start) {
1055 			/* VRAM before AGP */
1056 			WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1057 				rdev->mc.vram_start >> 12);
1058 			WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1059 				rdev->mc.gtt_end >> 12);
1060 		} else {
1061 			/* VRAM after AGP */
1062 			WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1063 				rdev->mc.gtt_start >> 12);
1064 			WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1065 				rdev->mc.vram_end >> 12);
1066 		}
1067 	} else {
1068 		WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, rdev->mc.vram_start >> 12);
1069 		WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, rdev->mc.vram_end >> 12);
1070 	}
1071 	WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, rdev->vram_scratch.gpu_addr >> 12);
1072 	tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
1073 	tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
1074 	WREG32(MC_VM_FB_LOCATION, tmp);
1075 	WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
1076 	WREG32(HDP_NONSURFACE_INFO, (2 << 7));
1077 	WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
1078 	if (rdev->flags & RADEON_IS_AGP) {
1079 		WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 22);
1080 		WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 22);
1081 		WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
1082 	} else {
1083 		WREG32(MC_VM_AGP_BASE, 0);
1084 		WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
1085 		WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
1086 	}
1087 	if (r600_mc_wait_for_idle(rdev)) {
1088 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1089 	}
1090 	rv515_mc_resume(rdev, &save);
1091 	/* we need to own VRAM, so turn off the VGA renderer here
1092 	 * to stop it overwriting our objects */
1093 	rv515_vga_render_disable(rdev);
1094 }
1095 
1096 /**
1097  * r600_vram_gtt_location - try to find VRAM & GTT location
1098  * @rdev: radeon device structure holding all necessary informations
1099  * @mc: memory controller structure holding memory informations
1100  *
1101  * Function will place try to place VRAM at same place as in CPU (PCI)
1102  * address space as some GPU seems to have issue when we reprogram at
1103  * different address space.
1104  *
1105  * If there is not enough space to fit the unvisible VRAM after the
1106  * aperture then we limit the VRAM size to the aperture.
1107  *
1108  * If we are using AGP then place VRAM adjacent to AGP aperture are we need
1109  * them to be in one from GPU point of view so that we can program GPU to
1110  * catch access outside them (weird GPU policy see ??).
1111  *
1112  * This function will never fails, worst case are limiting VRAM or GTT.
1113  *
1114  * Note: GTT start, end, size should be initialized before calling this
1115  * function on AGP platform.
1116  */
1117 static void r600_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
1118 {
1119 	u64 size_bf, size_af;
1120 
1121 	if (mc->mc_vram_size > 0xE0000000) {
1122 		/* leave room for at least 512M GTT */
1123 		dev_warn(rdev->dev, "limiting VRAM\n");
1124 		mc->real_vram_size = 0xE0000000;
1125 		mc->mc_vram_size = 0xE0000000;
1126 	}
1127 	if (rdev->flags & RADEON_IS_AGP) {
1128 		size_bf = mc->gtt_start;
1129 		size_af = 0xFFFFFFFF - mc->gtt_end;
1130 		if (size_bf > size_af) {
1131 			if (mc->mc_vram_size > size_bf) {
1132 				dev_warn(rdev->dev, "limiting VRAM\n");
1133 				mc->real_vram_size = size_bf;
1134 				mc->mc_vram_size = size_bf;
1135 			}
1136 			mc->vram_start = mc->gtt_start - mc->mc_vram_size;
1137 		} else {
1138 			if (mc->mc_vram_size > size_af) {
1139 				dev_warn(rdev->dev, "limiting VRAM\n");
1140 				mc->real_vram_size = size_af;
1141 				mc->mc_vram_size = size_af;
1142 			}
1143 			mc->vram_start = mc->gtt_end + 1;
1144 		}
1145 		mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
1146 		dev_info(rdev->dev, "VRAM: %lluM 0x%08llX - 0x%08llX (%lluM used)\n",
1147 				mc->mc_vram_size >> 20, mc->vram_start,
1148 				mc->vram_end, mc->real_vram_size >> 20);
1149 	} else {
1150 		u64 base = 0;
1151 		if (rdev->flags & RADEON_IS_IGP) {
1152 			base = RREG32(MC_VM_FB_LOCATION) & 0xFFFF;
1153 			base <<= 24;
1154 		}
1155 		radeon_vram_location(rdev, &rdev->mc, base);
1156 		rdev->mc.gtt_base_align = 0;
1157 		radeon_gtt_location(rdev, mc);
1158 	}
1159 }
1160 
1161 static int r600_mc_init(struct radeon_device *rdev)
1162 {
1163 	u32 tmp;
1164 	int chansize, numchan;
1165 
1166 	/* Get VRAM informations */
1167 	rdev->mc.vram_is_ddr = true;
1168 	tmp = RREG32(RAMCFG);
1169 	if (tmp & CHANSIZE_OVERRIDE) {
1170 		chansize = 16;
1171 	} else if (tmp & CHANSIZE_MASK) {
1172 		chansize = 64;
1173 	} else {
1174 		chansize = 32;
1175 	}
1176 	tmp = RREG32(CHMAP);
1177 	switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
1178 	case 0:
1179 	default:
1180 		numchan = 1;
1181 		break;
1182 	case 1:
1183 		numchan = 2;
1184 		break;
1185 	case 2:
1186 		numchan = 4;
1187 		break;
1188 	case 3:
1189 		numchan = 8;
1190 		break;
1191 	}
1192 	rdev->mc.vram_width = numchan * chansize;
1193 	/* Could aper size report 0 ? */
1194 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
1195 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
1196 	/* Setup GPU memory space */
1197 	rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
1198 	rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
1199 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
1200 	r600_vram_gtt_location(rdev, &rdev->mc);
1201 
1202 	if (rdev->flags & RADEON_IS_IGP) {
1203 		rs690_pm_info(rdev);
1204 		rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
1205 	}
1206 	radeon_update_bandwidth_info(rdev);
1207 	return 0;
1208 }
1209 
1210 int r600_vram_scratch_init(struct radeon_device *rdev)
1211 {
1212 	int r;
1213 
1214 	if (rdev->vram_scratch.robj == NULL) {
1215 		r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE,
1216 				     PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
1217 				     NULL, &rdev->vram_scratch.robj);
1218 		if (r) {
1219 			return r;
1220 		}
1221 	}
1222 
1223 	r = radeon_bo_reserve(rdev->vram_scratch.robj, false);
1224 	if (unlikely(r != 0))
1225 		return r;
1226 	r = radeon_bo_pin(rdev->vram_scratch.robj,
1227 			  RADEON_GEM_DOMAIN_VRAM, &rdev->vram_scratch.gpu_addr);
1228 	if (r) {
1229 		radeon_bo_unreserve(rdev->vram_scratch.robj);
1230 		return r;
1231 	}
1232 	r = radeon_bo_kmap(rdev->vram_scratch.robj,
1233 				(void **)&rdev->vram_scratch.ptr);
1234 	if (r)
1235 		radeon_bo_unpin(rdev->vram_scratch.robj);
1236 	radeon_bo_unreserve(rdev->vram_scratch.robj);
1237 
1238 	return r;
1239 }
1240 
1241 void r600_vram_scratch_fini(struct radeon_device *rdev)
1242 {
1243 	int r;
1244 
1245 	if (rdev->vram_scratch.robj == NULL) {
1246 		return;
1247 	}
1248 	r = radeon_bo_reserve(rdev->vram_scratch.robj, false);
1249 	if (likely(r == 0)) {
1250 		radeon_bo_kunmap(rdev->vram_scratch.robj);
1251 		radeon_bo_unpin(rdev->vram_scratch.robj);
1252 		radeon_bo_unreserve(rdev->vram_scratch.robj);
1253 	}
1254 	radeon_bo_unref(&rdev->vram_scratch.robj);
1255 }
1256 
1257 /* We doesn't check that the GPU really needs a reset we simply do the
1258  * reset, it's up to the caller to determine if the GPU needs one. We
1259  * might add an helper function to check that.
1260  */
1261 static int r600_gpu_soft_reset(struct radeon_device *rdev)
1262 {
1263 	struct rv515_mc_save save;
1264 	u32 grbm_busy_mask = S_008010_VC_BUSY(1) | S_008010_VGT_BUSY_NO_DMA(1) |
1265 				S_008010_VGT_BUSY(1) | S_008010_TA03_BUSY(1) |
1266 				S_008010_TC_BUSY(1) | S_008010_SX_BUSY(1) |
1267 				S_008010_SH_BUSY(1) | S_008010_SPI03_BUSY(1) |
1268 				S_008010_SMX_BUSY(1) | S_008010_SC_BUSY(1) |
1269 				S_008010_PA_BUSY(1) | S_008010_DB03_BUSY(1) |
1270 				S_008010_CR_BUSY(1) | S_008010_CB03_BUSY(1) |
1271 				S_008010_GUI_ACTIVE(1);
1272 	u32 grbm2_busy_mask = S_008014_SPI0_BUSY(1) | S_008014_SPI1_BUSY(1) |
1273 			S_008014_SPI2_BUSY(1) | S_008014_SPI3_BUSY(1) |
1274 			S_008014_TA0_BUSY(1) | S_008014_TA1_BUSY(1) |
1275 			S_008014_TA2_BUSY(1) | S_008014_TA3_BUSY(1) |
1276 			S_008014_DB0_BUSY(1) | S_008014_DB1_BUSY(1) |
1277 			S_008014_DB2_BUSY(1) | S_008014_DB3_BUSY(1) |
1278 			S_008014_CB0_BUSY(1) | S_008014_CB1_BUSY(1) |
1279 			S_008014_CB2_BUSY(1) | S_008014_CB3_BUSY(1);
1280 	u32 tmp;
1281 
1282 	if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE))
1283 		return 0;
1284 
1285 	dev_info(rdev->dev, "GPU softreset \n");
1286 	dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
1287 		RREG32(R_008010_GRBM_STATUS));
1288 	dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
1289 		RREG32(R_008014_GRBM_STATUS2));
1290 	dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
1291 		RREG32(R_000E50_SRBM_STATUS));
1292 	dev_info(rdev->dev, "  R_008674_CP_STALLED_STAT1 = 0x%08X\n",
1293 		RREG32(CP_STALLED_STAT1));
1294 	dev_info(rdev->dev, "  R_008678_CP_STALLED_STAT2 = 0x%08X\n",
1295 		RREG32(CP_STALLED_STAT2));
1296 	dev_info(rdev->dev, "  R_00867C_CP_BUSY_STAT     = 0x%08X\n",
1297 		RREG32(CP_BUSY_STAT));
1298 	dev_info(rdev->dev, "  R_008680_CP_STAT          = 0x%08X\n",
1299 		RREG32(CP_STAT));
1300 	rv515_mc_stop(rdev, &save);
1301 	if (r600_mc_wait_for_idle(rdev)) {
1302 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1303 	}
1304 	/* Disable CP parsing/prefetching */
1305 	WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
1306 	/* Check if any of the rendering block is busy and reset it */
1307 	if ((RREG32(R_008010_GRBM_STATUS) & grbm_busy_mask) ||
1308 	    (RREG32(R_008014_GRBM_STATUS2) & grbm2_busy_mask)) {
1309 		tmp = S_008020_SOFT_RESET_CR(1) |
1310 			S_008020_SOFT_RESET_DB(1) |
1311 			S_008020_SOFT_RESET_CB(1) |
1312 			S_008020_SOFT_RESET_PA(1) |
1313 			S_008020_SOFT_RESET_SC(1) |
1314 			S_008020_SOFT_RESET_SMX(1) |
1315 			S_008020_SOFT_RESET_SPI(1) |
1316 			S_008020_SOFT_RESET_SX(1) |
1317 			S_008020_SOFT_RESET_SH(1) |
1318 			S_008020_SOFT_RESET_TC(1) |
1319 			S_008020_SOFT_RESET_TA(1) |
1320 			S_008020_SOFT_RESET_VC(1) |
1321 			S_008020_SOFT_RESET_VGT(1);
1322 		dev_info(rdev->dev, "  R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
1323 		WREG32(R_008020_GRBM_SOFT_RESET, tmp);
1324 		RREG32(R_008020_GRBM_SOFT_RESET);
1325 		mdelay(15);
1326 		WREG32(R_008020_GRBM_SOFT_RESET, 0);
1327 	}
1328 	/* Reset CP (we always reset CP) */
1329 	tmp = S_008020_SOFT_RESET_CP(1);
1330 	dev_info(rdev->dev, "R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
1331 	WREG32(R_008020_GRBM_SOFT_RESET, tmp);
1332 	RREG32(R_008020_GRBM_SOFT_RESET);
1333 	mdelay(15);
1334 	WREG32(R_008020_GRBM_SOFT_RESET, 0);
1335 	/* Wait a little for things to settle down */
1336 	mdelay(1);
1337 	dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
1338 		RREG32(R_008010_GRBM_STATUS));
1339 	dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
1340 		RREG32(R_008014_GRBM_STATUS2));
1341 	dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
1342 		RREG32(R_000E50_SRBM_STATUS));
1343 	dev_info(rdev->dev, "  R_008674_CP_STALLED_STAT1 = 0x%08X\n",
1344 		RREG32(CP_STALLED_STAT1));
1345 	dev_info(rdev->dev, "  R_008678_CP_STALLED_STAT2 = 0x%08X\n",
1346 		RREG32(CP_STALLED_STAT2));
1347 	dev_info(rdev->dev, "  R_00867C_CP_BUSY_STAT     = 0x%08X\n",
1348 		RREG32(CP_BUSY_STAT));
1349 	dev_info(rdev->dev, "  R_008680_CP_STAT          = 0x%08X\n",
1350 		RREG32(CP_STAT));
1351 	rv515_mc_resume(rdev, &save);
1352 	return 0;
1353 }
1354 
1355 bool r600_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
1356 {
1357 	u32 srbm_status;
1358 	u32 grbm_status;
1359 	u32 grbm_status2;
1360 
1361 	srbm_status = RREG32(R_000E50_SRBM_STATUS);
1362 	grbm_status = RREG32(R_008010_GRBM_STATUS);
1363 	grbm_status2 = RREG32(R_008014_GRBM_STATUS2);
1364 	if (!G_008010_GUI_ACTIVE(grbm_status)) {
1365 		radeon_ring_lockup_update(ring);
1366 		return false;
1367 	}
1368 	/* force CP activities */
1369 	radeon_ring_force_activity(rdev, ring);
1370 	return radeon_ring_test_lockup(rdev, ring);
1371 }
1372 
1373 /**
1374  * r600_dma_is_lockup - Check if the DMA engine is locked up
1375  *
1376  * @rdev: radeon_device pointer
1377  * @ring: radeon_ring structure holding ring information
1378  *
1379  * Check if the async DMA engine is locked up (r6xx-evergreen).
1380  * Returns true if the engine appears to be locked up, false if not.
1381  */
1382 bool r600_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
1383 {
1384 	u32 dma_status_reg;
1385 
1386 	dma_status_reg = RREG32(DMA_STATUS_REG);
1387 	if (dma_status_reg & DMA_IDLE) {
1388 		radeon_ring_lockup_update(ring);
1389 		return false;
1390 	}
1391 	/* force ring activities */
1392 	radeon_ring_force_activity(rdev, ring);
1393 	return radeon_ring_test_lockup(rdev, ring);
1394 }
1395 
1396 int r600_asic_reset(struct radeon_device *rdev)
1397 {
1398 	return r600_gpu_soft_reset(rdev);
1399 }
1400 
1401 u32 r6xx_remap_render_backend(struct radeon_device *rdev,
1402 			      u32 tiling_pipe_num,
1403 			      u32 max_rb_num,
1404 			      u32 total_max_rb_num,
1405 			      u32 disabled_rb_mask)
1406 {
1407 	u32 rendering_pipe_num, rb_num_width, req_rb_num;
1408 	u32 pipe_rb_ratio, pipe_rb_remain;
1409 	u32 data = 0, mask = 1 << (max_rb_num - 1);
1410 	unsigned i, j;
1411 
1412 	/* mask out the RBs that don't exist on that asic */
1413 	disabled_rb_mask |= (0xff << max_rb_num) & 0xff;
1414 
1415 	rendering_pipe_num = 1 << tiling_pipe_num;
1416 	req_rb_num = total_max_rb_num - r600_count_pipe_bits(disabled_rb_mask);
1417 	BUG_ON(rendering_pipe_num < req_rb_num);
1418 
1419 	pipe_rb_ratio = rendering_pipe_num / req_rb_num;
1420 	pipe_rb_remain = rendering_pipe_num - pipe_rb_ratio * req_rb_num;
1421 
1422 	if (rdev->family <= CHIP_RV740) {
1423 		/* r6xx/r7xx */
1424 		rb_num_width = 2;
1425 	} else {
1426 		/* eg+ */
1427 		rb_num_width = 4;
1428 	}
1429 
1430 	for (i = 0; i < max_rb_num; i++) {
1431 		if (!(mask & disabled_rb_mask)) {
1432 			for (j = 0; j < pipe_rb_ratio; j++) {
1433 				data <<= rb_num_width;
1434 				data |= max_rb_num - i - 1;
1435 			}
1436 			if (pipe_rb_remain) {
1437 				data <<= rb_num_width;
1438 				data |= max_rb_num - i - 1;
1439 				pipe_rb_remain--;
1440 			}
1441 		}
1442 		mask >>= 1;
1443 	}
1444 
1445 	return data;
1446 }
1447 
1448 int r600_count_pipe_bits(uint32_t val)
1449 {
1450 	return hweight32(val);
1451 }
1452 
1453 static void r600_gpu_init(struct radeon_device *rdev)
1454 {
1455 	u32 tiling_config;
1456 	u32 ramcfg;
1457 	u32 cc_rb_backend_disable;
1458 	u32 cc_gc_shader_pipe_config;
1459 	u32 tmp;
1460 	int i, j;
1461 	u32 sq_config;
1462 	u32 sq_gpr_resource_mgmt_1 = 0;
1463 	u32 sq_gpr_resource_mgmt_2 = 0;
1464 	u32 sq_thread_resource_mgmt = 0;
1465 	u32 sq_stack_resource_mgmt_1 = 0;
1466 	u32 sq_stack_resource_mgmt_2 = 0;
1467 	u32 disabled_rb_mask;
1468 
1469 	rdev->config.r600.tiling_group_size = 256;
1470 	switch (rdev->family) {
1471 	case CHIP_R600:
1472 		rdev->config.r600.max_pipes = 4;
1473 		rdev->config.r600.max_tile_pipes = 8;
1474 		rdev->config.r600.max_simds = 4;
1475 		rdev->config.r600.max_backends = 4;
1476 		rdev->config.r600.max_gprs = 256;
1477 		rdev->config.r600.max_threads = 192;
1478 		rdev->config.r600.max_stack_entries = 256;
1479 		rdev->config.r600.max_hw_contexts = 8;
1480 		rdev->config.r600.max_gs_threads = 16;
1481 		rdev->config.r600.sx_max_export_size = 128;
1482 		rdev->config.r600.sx_max_export_pos_size = 16;
1483 		rdev->config.r600.sx_max_export_smx_size = 128;
1484 		rdev->config.r600.sq_num_cf_insts = 2;
1485 		break;
1486 	case CHIP_RV630:
1487 	case CHIP_RV635:
1488 		rdev->config.r600.max_pipes = 2;
1489 		rdev->config.r600.max_tile_pipes = 2;
1490 		rdev->config.r600.max_simds = 3;
1491 		rdev->config.r600.max_backends = 1;
1492 		rdev->config.r600.max_gprs = 128;
1493 		rdev->config.r600.max_threads = 192;
1494 		rdev->config.r600.max_stack_entries = 128;
1495 		rdev->config.r600.max_hw_contexts = 8;
1496 		rdev->config.r600.max_gs_threads = 4;
1497 		rdev->config.r600.sx_max_export_size = 128;
1498 		rdev->config.r600.sx_max_export_pos_size = 16;
1499 		rdev->config.r600.sx_max_export_smx_size = 128;
1500 		rdev->config.r600.sq_num_cf_insts = 2;
1501 		break;
1502 	case CHIP_RV610:
1503 	case CHIP_RV620:
1504 	case CHIP_RS780:
1505 	case CHIP_RS880:
1506 		rdev->config.r600.max_pipes = 1;
1507 		rdev->config.r600.max_tile_pipes = 1;
1508 		rdev->config.r600.max_simds = 2;
1509 		rdev->config.r600.max_backends = 1;
1510 		rdev->config.r600.max_gprs = 128;
1511 		rdev->config.r600.max_threads = 192;
1512 		rdev->config.r600.max_stack_entries = 128;
1513 		rdev->config.r600.max_hw_contexts = 4;
1514 		rdev->config.r600.max_gs_threads = 4;
1515 		rdev->config.r600.sx_max_export_size = 128;
1516 		rdev->config.r600.sx_max_export_pos_size = 16;
1517 		rdev->config.r600.sx_max_export_smx_size = 128;
1518 		rdev->config.r600.sq_num_cf_insts = 1;
1519 		break;
1520 	case CHIP_RV670:
1521 		rdev->config.r600.max_pipes = 4;
1522 		rdev->config.r600.max_tile_pipes = 4;
1523 		rdev->config.r600.max_simds = 4;
1524 		rdev->config.r600.max_backends = 4;
1525 		rdev->config.r600.max_gprs = 192;
1526 		rdev->config.r600.max_threads = 192;
1527 		rdev->config.r600.max_stack_entries = 256;
1528 		rdev->config.r600.max_hw_contexts = 8;
1529 		rdev->config.r600.max_gs_threads = 16;
1530 		rdev->config.r600.sx_max_export_size = 128;
1531 		rdev->config.r600.sx_max_export_pos_size = 16;
1532 		rdev->config.r600.sx_max_export_smx_size = 128;
1533 		rdev->config.r600.sq_num_cf_insts = 2;
1534 		break;
1535 	default:
1536 		break;
1537 	}
1538 
1539 	/* Initialize HDP */
1540 	for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1541 		WREG32((0x2c14 + j), 0x00000000);
1542 		WREG32((0x2c18 + j), 0x00000000);
1543 		WREG32((0x2c1c + j), 0x00000000);
1544 		WREG32((0x2c20 + j), 0x00000000);
1545 		WREG32((0x2c24 + j), 0x00000000);
1546 	}
1547 
1548 	WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1549 
1550 	/* Setup tiling */
1551 	tiling_config = 0;
1552 	ramcfg = RREG32(RAMCFG);
1553 	switch (rdev->config.r600.max_tile_pipes) {
1554 	case 1:
1555 		tiling_config |= PIPE_TILING(0);
1556 		break;
1557 	case 2:
1558 		tiling_config |= PIPE_TILING(1);
1559 		break;
1560 	case 4:
1561 		tiling_config |= PIPE_TILING(2);
1562 		break;
1563 	case 8:
1564 		tiling_config |= PIPE_TILING(3);
1565 		break;
1566 	default:
1567 		break;
1568 	}
1569 	rdev->config.r600.tiling_npipes = rdev->config.r600.max_tile_pipes;
1570 	rdev->config.r600.tiling_nbanks = 4 << ((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
1571 	tiling_config |= BANK_TILING((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
1572 	tiling_config |= GROUP_SIZE((ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT);
1573 
1574 	tmp = (ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT;
1575 	if (tmp > 3) {
1576 		tiling_config |= ROW_TILING(3);
1577 		tiling_config |= SAMPLE_SPLIT(3);
1578 	} else {
1579 		tiling_config |= ROW_TILING(tmp);
1580 		tiling_config |= SAMPLE_SPLIT(tmp);
1581 	}
1582 	tiling_config |= BANK_SWAPS(1);
1583 
1584 	cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000;
1585 	tmp = R6XX_MAX_BACKENDS -
1586 		r600_count_pipe_bits((cc_rb_backend_disable >> 16) & R6XX_MAX_BACKENDS_MASK);
1587 	if (tmp < rdev->config.r600.max_backends) {
1588 		rdev->config.r600.max_backends = tmp;
1589 	}
1590 
1591 	cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0x00ffff00;
1592 	tmp = R6XX_MAX_PIPES -
1593 		r600_count_pipe_bits((cc_gc_shader_pipe_config >> 8) & R6XX_MAX_PIPES_MASK);
1594 	if (tmp < rdev->config.r600.max_pipes) {
1595 		rdev->config.r600.max_pipes = tmp;
1596 	}
1597 	tmp = R6XX_MAX_SIMDS -
1598 		r600_count_pipe_bits((cc_gc_shader_pipe_config >> 16) & R6XX_MAX_SIMDS_MASK);
1599 	if (tmp < rdev->config.r600.max_simds) {
1600 		rdev->config.r600.max_simds = tmp;
1601 	}
1602 
1603 	disabled_rb_mask = (RREG32(CC_RB_BACKEND_DISABLE) >> 16) & R6XX_MAX_BACKENDS_MASK;
1604 	tmp = (tiling_config & PIPE_TILING__MASK) >> PIPE_TILING__SHIFT;
1605 	tmp = r6xx_remap_render_backend(rdev, tmp, rdev->config.r600.max_backends,
1606 					R6XX_MAX_BACKENDS, disabled_rb_mask);
1607 	tiling_config |= tmp << 16;
1608 	rdev->config.r600.backend_map = tmp;
1609 
1610 	rdev->config.r600.tile_config = tiling_config;
1611 	WREG32(GB_TILING_CONFIG, tiling_config);
1612 	WREG32(DCP_TILING_CONFIG, tiling_config & 0xffff);
1613 	WREG32(HDP_TILING_CONFIG, tiling_config & 0xffff);
1614 	WREG32(DMA_TILING_CONFIG, tiling_config & 0xffff);
1615 
1616 	tmp = R6XX_MAX_PIPES - r600_count_pipe_bits((cc_gc_shader_pipe_config & INACTIVE_QD_PIPES_MASK) >> 8);
1617 	WREG32(VGT_OUT_DEALLOC_CNTL, (tmp * 4) & DEALLOC_DIST_MASK);
1618 	WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((tmp * 4) - 2) & VTX_REUSE_DEPTH_MASK);
1619 
1620 	/* Setup some CP states */
1621 	WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | ROQ_IB2_START(0x2b)));
1622 	WREG32(CP_MEQ_THRESHOLDS, (MEQ_END(0x40) | ROQ_END(0x40)));
1623 
1624 	WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO | SYNC_GRADIENT |
1625 			     SYNC_WALKER | SYNC_ALIGNER));
1626 	/* Setup various GPU states */
1627 	if (rdev->family == CHIP_RV670)
1628 		WREG32(ARB_GDEC_RD_CNTL, 0x00000021);
1629 
1630 	tmp = RREG32(SX_DEBUG_1);
1631 	tmp |= SMX_EVENT_RELEASE;
1632 	if ((rdev->family > CHIP_R600))
1633 		tmp |= ENABLE_NEW_SMX_ADDRESS;
1634 	WREG32(SX_DEBUG_1, tmp);
1635 
1636 	if (((rdev->family) == CHIP_R600) ||
1637 	    ((rdev->family) == CHIP_RV630) ||
1638 	    ((rdev->family) == CHIP_RV610) ||
1639 	    ((rdev->family) == CHIP_RV620) ||
1640 	    ((rdev->family) == CHIP_RS780) ||
1641 	    ((rdev->family) == CHIP_RS880)) {
1642 		WREG32(DB_DEBUG, PREZ_MUST_WAIT_FOR_POSTZ_DONE);
1643 	} else {
1644 		WREG32(DB_DEBUG, 0);
1645 	}
1646 	WREG32(DB_WATERMARKS, (DEPTH_FREE(4) | DEPTH_CACHELINE_FREE(16) |
1647 			       DEPTH_FLUSH(16) | DEPTH_PENDING_FREE(4)));
1648 
1649 	WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1650 	WREG32(VGT_NUM_INSTANCES, 0);
1651 
1652 	WREG32(SPI_CONFIG_CNTL, GPR_WRITE_PRIORITY(0));
1653 	WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(0));
1654 
1655 	tmp = RREG32(SQ_MS_FIFO_SIZES);
1656 	if (((rdev->family) == CHIP_RV610) ||
1657 	    ((rdev->family) == CHIP_RV620) ||
1658 	    ((rdev->family) == CHIP_RS780) ||
1659 	    ((rdev->family) == CHIP_RS880)) {
1660 		tmp = (CACHE_FIFO_SIZE(0xa) |
1661 		       FETCH_FIFO_HIWATER(0xa) |
1662 		       DONE_FIFO_HIWATER(0xe0) |
1663 		       ALU_UPDATE_FIFO_HIWATER(0x8));
1664 	} else if (((rdev->family) == CHIP_R600) ||
1665 		   ((rdev->family) == CHIP_RV630)) {
1666 		tmp &= ~DONE_FIFO_HIWATER(0xff);
1667 		tmp |= DONE_FIFO_HIWATER(0x4);
1668 	}
1669 	WREG32(SQ_MS_FIFO_SIZES, tmp);
1670 
1671 	/* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
1672 	 * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
1673 	 */
1674 	sq_config = RREG32(SQ_CONFIG);
1675 	sq_config &= ~(PS_PRIO(3) |
1676 		       VS_PRIO(3) |
1677 		       GS_PRIO(3) |
1678 		       ES_PRIO(3));
1679 	sq_config |= (DX9_CONSTS |
1680 		      VC_ENABLE |
1681 		      PS_PRIO(0) |
1682 		      VS_PRIO(1) |
1683 		      GS_PRIO(2) |
1684 		      ES_PRIO(3));
1685 
1686 	if ((rdev->family) == CHIP_R600) {
1687 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(124) |
1688 					  NUM_VS_GPRS(124) |
1689 					  NUM_CLAUSE_TEMP_GPRS(4));
1690 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(0) |
1691 					  NUM_ES_GPRS(0));
1692 		sq_thread_resource_mgmt = (NUM_PS_THREADS(136) |
1693 					   NUM_VS_THREADS(48) |
1694 					   NUM_GS_THREADS(4) |
1695 					   NUM_ES_THREADS(4));
1696 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(128) |
1697 					    NUM_VS_STACK_ENTRIES(128));
1698 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(0) |
1699 					    NUM_ES_STACK_ENTRIES(0));
1700 	} else if (((rdev->family) == CHIP_RV610) ||
1701 		   ((rdev->family) == CHIP_RV620) ||
1702 		   ((rdev->family) == CHIP_RS780) ||
1703 		   ((rdev->family) == CHIP_RS880)) {
1704 		/* no vertex cache */
1705 		sq_config &= ~VC_ENABLE;
1706 
1707 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1708 					  NUM_VS_GPRS(44) |
1709 					  NUM_CLAUSE_TEMP_GPRS(2));
1710 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1711 					  NUM_ES_GPRS(17));
1712 		sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1713 					   NUM_VS_THREADS(78) |
1714 					   NUM_GS_THREADS(4) |
1715 					   NUM_ES_THREADS(31));
1716 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
1717 					    NUM_VS_STACK_ENTRIES(40));
1718 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
1719 					    NUM_ES_STACK_ENTRIES(16));
1720 	} else if (((rdev->family) == CHIP_RV630) ||
1721 		   ((rdev->family) == CHIP_RV635)) {
1722 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1723 					  NUM_VS_GPRS(44) |
1724 					  NUM_CLAUSE_TEMP_GPRS(2));
1725 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(18) |
1726 					  NUM_ES_GPRS(18));
1727 		sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1728 					   NUM_VS_THREADS(78) |
1729 					   NUM_GS_THREADS(4) |
1730 					   NUM_ES_THREADS(31));
1731 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
1732 					    NUM_VS_STACK_ENTRIES(40));
1733 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
1734 					    NUM_ES_STACK_ENTRIES(16));
1735 	} else if ((rdev->family) == CHIP_RV670) {
1736 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1737 					  NUM_VS_GPRS(44) |
1738 					  NUM_CLAUSE_TEMP_GPRS(2));
1739 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1740 					  NUM_ES_GPRS(17));
1741 		sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1742 					   NUM_VS_THREADS(78) |
1743 					   NUM_GS_THREADS(4) |
1744 					   NUM_ES_THREADS(31));
1745 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(64) |
1746 					    NUM_VS_STACK_ENTRIES(64));
1747 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(64) |
1748 					    NUM_ES_STACK_ENTRIES(64));
1749 	}
1750 
1751 	WREG32(SQ_CONFIG, sq_config);
1752 	WREG32(SQ_GPR_RESOURCE_MGMT_1,  sq_gpr_resource_mgmt_1);
1753 	WREG32(SQ_GPR_RESOURCE_MGMT_2,  sq_gpr_resource_mgmt_2);
1754 	WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1755 	WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
1756 	WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
1757 
1758 	if (((rdev->family) == CHIP_RV610) ||
1759 	    ((rdev->family) == CHIP_RV620) ||
1760 	    ((rdev->family) == CHIP_RS780) ||
1761 	    ((rdev->family) == CHIP_RS880)) {
1762 		WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(TC_ONLY));
1763 	} else {
1764 		WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC));
1765 	}
1766 
1767 	/* More default values. 2D/3D driver should adjust as needed */
1768 	WREG32(PA_SC_AA_SAMPLE_LOCS_2S, (S0_X(0xc) | S0_Y(0x4) |
1769 					 S1_X(0x4) | S1_Y(0xc)));
1770 	WREG32(PA_SC_AA_SAMPLE_LOCS_4S, (S0_X(0xe) | S0_Y(0xe) |
1771 					 S1_X(0x2) | S1_Y(0x2) |
1772 					 S2_X(0xa) | S2_Y(0x6) |
1773 					 S3_X(0x6) | S3_Y(0xa)));
1774 	WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD0, (S0_X(0xe) | S0_Y(0xb) |
1775 					     S1_X(0x4) | S1_Y(0xc) |
1776 					     S2_X(0x1) | S2_Y(0x6) |
1777 					     S3_X(0xa) | S3_Y(0xe)));
1778 	WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD1, (S4_X(0x6) | S4_Y(0x1) |
1779 					     S5_X(0x0) | S5_Y(0x0) |
1780 					     S6_X(0xb) | S6_Y(0x4) |
1781 					     S7_X(0x7) | S7_Y(0x8)));
1782 
1783 	WREG32(VGT_STRMOUT_EN, 0);
1784 	tmp = rdev->config.r600.max_pipes * 16;
1785 	switch (rdev->family) {
1786 	case CHIP_RV610:
1787 	case CHIP_RV620:
1788 	case CHIP_RS780:
1789 	case CHIP_RS880:
1790 		tmp += 32;
1791 		break;
1792 	case CHIP_RV670:
1793 		tmp += 128;
1794 		break;
1795 	default:
1796 		break;
1797 	}
1798 	if (tmp > 256) {
1799 		tmp = 256;
1800 	}
1801 	WREG32(VGT_ES_PER_GS, 128);
1802 	WREG32(VGT_GS_PER_ES, tmp);
1803 	WREG32(VGT_GS_PER_VS, 2);
1804 	WREG32(VGT_GS_VERTEX_REUSE, 16);
1805 
1806 	/* more default values. 2D/3D driver should adjust as needed */
1807 	WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1808 	WREG32(VGT_STRMOUT_EN, 0);
1809 	WREG32(SX_MISC, 0);
1810 	WREG32(PA_SC_MODE_CNTL, 0);
1811 	WREG32(PA_SC_AA_CONFIG, 0);
1812 	WREG32(PA_SC_LINE_STIPPLE, 0);
1813 	WREG32(SPI_INPUT_Z, 0);
1814 	WREG32(SPI_PS_IN_CONTROL_0, NUM_INTERP(2));
1815 	WREG32(CB_COLOR7_FRAG, 0);
1816 
1817 	/* Clear render buffer base addresses */
1818 	WREG32(CB_COLOR0_BASE, 0);
1819 	WREG32(CB_COLOR1_BASE, 0);
1820 	WREG32(CB_COLOR2_BASE, 0);
1821 	WREG32(CB_COLOR3_BASE, 0);
1822 	WREG32(CB_COLOR4_BASE, 0);
1823 	WREG32(CB_COLOR5_BASE, 0);
1824 	WREG32(CB_COLOR6_BASE, 0);
1825 	WREG32(CB_COLOR7_BASE, 0);
1826 	WREG32(CB_COLOR7_FRAG, 0);
1827 
1828 	switch (rdev->family) {
1829 	case CHIP_RV610:
1830 	case CHIP_RV620:
1831 	case CHIP_RS780:
1832 	case CHIP_RS880:
1833 		tmp = TC_L2_SIZE(8);
1834 		break;
1835 	case CHIP_RV630:
1836 	case CHIP_RV635:
1837 		tmp = TC_L2_SIZE(4);
1838 		break;
1839 	case CHIP_R600:
1840 		tmp = TC_L2_SIZE(0) | L2_DISABLE_LATE_HIT;
1841 		break;
1842 	default:
1843 		tmp = TC_L2_SIZE(0);
1844 		break;
1845 	}
1846 	WREG32(TC_CNTL, tmp);
1847 
1848 	tmp = RREG32(HDP_HOST_PATH_CNTL);
1849 	WREG32(HDP_HOST_PATH_CNTL, tmp);
1850 
1851 	tmp = RREG32(ARB_POP);
1852 	tmp |= ENABLE_TC128;
1853 	WREG32(ARB_POP, tmp);
1854 
1855 	WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1856 	WREG32(PA_CL_ENHANCE, (CLIP_VTX_REORDER_ENA |
1857 			       NUM_CLIP_SEQ(3)));
1858 	WREG32(PA_SC_ENHANCE, FORCE_EOV_MAX_CLK_CNT(4095));
1859 	WREG32(VC_ENHANCE, 0);
1860 }
1861 
1862 
1863 /*
1864  * Indirect registers accessor
1865  */
1866 u32 r600_pciep_rreg(struct radeon_device *rdev, u32 reg)
1867 {
1868 	u32 r;
1869 
1870 	WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1871 	(void)RREG32(PCIE_PORT_INDEX);
1872 	r = RREG32(PCIE_PORT_DATA);
1873 	return r;
1874 }
1875 
1876 void r600_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
1877 {
1878 	WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1879 	(void)RREG32(PCIE_PORT_INDEX);
1880 	WREG32(PCIE_PORT_DATA, (v));
1881 	(void)RREG32(PCIE_PORT_DATA);
1882 }
1883 
1884 /*
1885  * CP & Ring
1886  */
1887 void r600_cp_stop(struct radeon_device *rdev)
1888 {
1889 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1890 	WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
1891 	WREG32(SCRATCH_UMSK, 0);
1892 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1893 }
1894 
1895 int r600_init_microcode(struct radeon_device *rdev)
1896 {
1897 	struct platform_device *pdev;
1898 	const char *chip_name;
1899 	const char *rlc_chip_name;
1900 	size_t pfp_req_size, me_req_size, rlc_req_size;
1901 	char fw_name[30];
1902 	int err;
1903 
1904 	DRM_DEBUG("\n");
1905 
1906 	pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
1907 	err = IS_ERR(pdev);
1908 	if (err) {
1909 		printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
1910 		return -EINVAL;
1911 	}
1912 
1913 	switch (rdev->family) {
1914 	case CHIP_R600:
1915 		chip_name = "R600";
1916 		rlc_chip_name = "R600";
1917 		break;
1918 	case CHIP_RV610:
1919 		chip_name = "RV610";
1920 		rlc_chip_name = "R600";
1921 		break;
1922 	case CHIP_RV630:
1923 		chip_name = "RV630";
1924 		rlc_chip_name = "R600";
1925 		break;
1926 	case CHIP_RV620:
1927 		chip_name = "RV620";
1928 		rlc_chip_name = "R600";
1929 		break;
1930 	case CHIP_RV635:
1931 		chip_name = "RV635";
1932 		rlc_chip_name = "R600";
1933 		break;
1934 	case CHIP_RV670:
1935 		chip_name = "RV670";
1936 		rlc_chip_name = "R600";
1937 		break;
1938 	case CHIP_RS780:
1939 	case CHIP_RS880:
1940 		chip_name = "RS780";
1941 		rlc_chip_name = "R600";
1942 		break;
1943 	case CHIP_RV770:
1944 		chip_name = "RV770";
1945 		rlc_chip_name = "R700";
1946 		break;
1947 	case CHIP_RV730:
1948 	case CHIP_RV740:
1949 		chip_name = "RV730";
1950 		rlc_chip_name = "R700";
1951 		break;
1952 	case CHIP_RV710:
1953 		chip_name = "RV710";
1954 		rlc_chip_name = "R700";
1955 		break;
1956 	case CHIP_CEDAR:
1957 		chip_name = "CEDAR";
1958 		rlc_chip_name = "CEDAR";
1959 		break;
1960 	case CHIP_REDWOOD:
1961 		chip_name = "REDWOOD";
1962 		rlc_chip_name = "REDWOOD";
1963 		break;
1964 	case CHIP_JUNIPER:
1965 		chip_name = "JUNIPER";
1966 		rlc_chip_name = "JUNIPER";
1967 		break;
1968 	case CHIP_CYPRESS:
1969 	case CHIP_HEMLOCK:
1970 		chip_name = "CYPRESS";
1971 		rlc_chip_name = "CYPRESS";
1972 		break;
1973 	case CHIP_PALM:
1974 		chip_name = "PALM";
1975 		rlc_chip_name = "SUMO";
1976 		break;
1977 	case CHIP_SUMO:
1978 		chip_name = "SUMO";
1979 		rlc_chip_name = "SUMO";
1980 		break;
1981 	case CHIP_SUMO2:
1982 		chip_name = "SUMO2";
1983 		rlc_chip_name = "SUMO";
1984 		break;
1985 	default: BUG();
1986 	}
1987 
1988 	if (rdev->family >= CHIP_CEDAR) {
1989 		pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
1990 		me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
1991 		rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
1992 	} else if (rdev->family >= CHIP_RV770) {
1993 		pfp_req_size = R700_PFP_UCODE_SIZE * 4;
1994 		me_req_size = R700_PM4_UCODE_SIZE * 4;
1995 		rlc_req_size = R700_RLC_UCODE_SIZE * 4;
1996 	} else {
1997 		pfp_req_size = PFP_UCODE_SIZE * 4;
1998 		me_req_size = PM4_UCODE_SIZE * 12;
1999 		rlc_req_size = RLC_UCODE_SIZE * 4;
2000 	}
2001 
2002 	DRM_INFO("Loading %s Microcode\n", chip_name);
2003 
2004 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
2005 	err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
2006 	if (err)
2007 		goto out;
2008 	if (rdev->pfp_fw->size != pfp_req_size) {
2009 		printk(KERN_ERR
2010 		       "r600_cp: Bogus length %zu in firmware \"%s\"\n",
2011 		       rdev->pfp_fw->size, fw_name);
2012 		err = -EINVAL;
2013 		goto out;
2014 	}
2015 
2016 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
2017 	err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
2018 	if (err)
2019 		goto out;
2020 	if (rdev->me_fw->size != me_req_size) {
2021 		printk(KERN_ERR
2022 		       "r600_cp: Bogus length %zu in firmware \"%s\"\n",
2023 		       rdev->me_fw->size, fw_name);
2024 		err = -EINVAL;
2025 	}
2026 
2027 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
2028 	err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
2029 	if (err)
2030 		goto out;
2031 	if (rdev->rlc_fw->size != rlc_req_size) {
2032 		printk(KERN_ERR
2033 		       "r600_rlc: Bogus length %zu in firmware \"%s\"\n",
2034 		       rdev->rlc_fw->size, fw_name);
2035 		err = -EINVAL;
2036 	}
2037 
2038 out:
2039 	platform_device_unregister(pdev);
2040 
2041 	if (err) {
2042 		if (err != -EINVAL)
2043 			printk(KERN_ERR
2044 			       "r600_cp: Failed to load firmware \"%s\"\n",
2045 			       fw_name);
2046 		release_firmware(rdev->pfp_fw);
2047 		rdev->pfp_fw = NULL;
2048 		release_firmware(rdev->me_fw);
2049 		rdev->me_fw = NULL;
2050 		release_firmware(rdev->rlc_fw);
2051 		rdev->rlc_fw = NULL;
2052 	}
2053 	return err;
2054 }
2055 
2056 static int r600_cp_load_microcode(struct radeon_device *rdev)
2057 {
2058 	const __be32 *fw_data;
2059 	int i;
2060 
2061 	if (!rdev->me_fw || !rdev->pfp_fw)
2062 		return -EINVAL;
2063 
2064 	r600_cp_stop(rdev);
2065 
2066 	WREG32(CP_RB_CNTL,
2067 #ifdef __BIG_ENDIAN
2068 	       BUF_SWAP_32BIT |
2069 #endif
2070 	       RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
2071 
2072 	/* Reset cp */
2073 	WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
2074 	RREG32(GRBM_SOFT_RESET);
2075 	mdelay(15);
2076 	WREG32(GRBM_SOFT_RESET, 0);
2077 
2078 	WREG32(CP_ME_RAM_WADDR, 0);
2079 
2080 	fw_data = (const __be32 *)rdev->me_fw->data;
2081 	WREG32(CP_ME_RAM_WADDR, 0);
2082 	for (i = 0; i < PM4_UCODE_SIZE * 3; i++)
2083 		WREG32(CP_ME_RAM_DATA,
2084 		       be32_to_cpup(fw_data++));
2085 
2086 	fw_data = (const __be32 *)rdev->pfp_fw->data;
2087 	WREG32(CP_PFP_UCODE_ADDR, 0);
2088 	for (i = 0; i < PFP_UCODE_SIZE; i++)
2089 		WREG32(CP_PFP_UCODE_DATA,
2090 		       be32_to_cpup(fw_data++));
2091 
2092 	WREG32(CP_PFP_UCODE_ADDR, 0);
2093 	WREG32(CP_ME_RAM_WADDR, 0);
2094 	WREG32(CP_ME_RAM_RADDR, 0);
2095 	return 0;
2096 }
2097 
2098 int r600_cp_start(struct radeon_device *rdev)
2099 {
2100 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2101 	int r;
2102 	uint32_t cp_me;
2103 
2104 	r = radeon_ring_lock(rdev, ring, 7);
2105 	if (r) {
2106 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
2107 		return r;
2108 	}
2109 	radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
2110 	radeon_ring_write(ring, 0x1);
2111 	if (rdev->family >= CHIP_RV770) {
2112 		radeon_ring_write(ring, 0x0);
2113 		radeon_ring_write(ring, rdev->config.rv770.max_hw_contexts - 1);
2114 	} else {
2115 		radeon_ring_write(ring, 0x3);
2116 		radeon_ring_write(ring, rdev->config.r600.max_hw_contexts - 1);
2117 	}
2118 	radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
2119 	radeon_ring_write(ring, 0);
2120 	radeon_ring_write(ring, 0);
2121 	radeon_ring_unlock_commit(rdev, ring);
2122 
2123 	cp_me = 0xff;
2124 	WREG32(R_0086D8_CP_ME_CNTL, cp_me);
2125 	return 0;
2126 }
2127 
2128 int r600_cp_resume(struct radeon_device *rdev)
2129 {
2130 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2131 	u32 tmp;
2132 	u32 rb_bufsz;
2133 	int r;
2134 
2135 	/* Reset cp */
2136 	WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
2137 	RREG32(GRBM_SOFT_RESET);
2138 	mdelay(15);
2139 	WREG32(GRBM_SOFT_RESET, 0);
2140 
2141 	/* Set ring buffer size */
2142 	rb_bufsz = drm_order(ring->ring_size / 8);
2143 	tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2144 #ifdef __BIG_ENDIAN
2145 	tmp |= BUF_SWAP_32BIT;
2146 #endif
2147 	WREG32(CP_RB_CNTL, tmp);
2148 	WREG32(CP_SEM_WAIT_TIMER, 0x0);
2149 
2150 	/* Set the write pointer delay */
2151 	WREG32(CP_RB_WPTR_DELAY, 0);
2152 
2153 	/* Initialize the ring buffer's read and write pointers */
2154 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
2155 	WREG32(CP_RB_RPTR_WR, 0);
2156 	ring->wptr = 0;
2157 	WREG32(CP_RB_WPTR, ring->wptr);
2158 
2159 	/* set the wb address whether it's enabled or not */
2160 	WREG32(CP_RB_RPTR_ADDR,
2161 	       ((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC));
2162 	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
2163 	WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
2164 
2165 	if (rdev->wb.enabled)
2166 		WREG32(SCRATCH_UMSK, 0xff);
2167 	else {
2168 		tmp |= RB_NO_UPDATE;
2169 		WREG32(SCRATCH_UMSK, 0);
2170 	}
2171 
2172 	mdelay(1);
2173 	WREG32(CP_RB_CNTL, tmp);
2174 
2175 	WREG32(CP_RB_BASE, ring->gpu_addr >> 8);
2176 	WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
2177 
2178 	ring->rptr = RREG32(CP_RB_RPTR);
2179 
2180 	r600_cp_start(rdev);
2181 	ring->ready = true;
2182 	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
2183 	if (r) {
2184 		ring->ready = false;
2185 		return r;
2186 	}
2187 	return 0;
2188 }
2189 
2190 void r600_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size)
2191 {
2192 	u32 rb_bufsz;
2193 	int r;
2194 
2195 	/* Align ring size */
2196 	rb_bufsz = drm_order(ring_size / 8);
2197 	ring_size = (1 << (rb_bufsz + 1)) * 4;
2198 	ring->ring_size = ring_size;
2199 	ring->align_mask = 16 - 1;
2200 
2201 	if (radeon_ring_supports_scratch_reg(rdev, ring)) {
2202 		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
2203 		if (r) {
2204 			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
2205 			ring->rptr_save_reg = 0;
2206 		}
2207 	}
2208 }
2209 
2210 void r600_cp_fini(struct radeon_device *rdev)
2211 {
2212 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2213 	r600_cp_stop(rdev);
2214 	radeon_ring_fini(rdev, ring);
2215 	radeon_scratch_free(rdev, ring->rptr_save_reg);
2216 }
2217 
2218 /*
2219  * DMA
2220  * Starting with R600, the GPU has an asynchronous
2221  * DMA engine.  The programming model is very similar
2222  * to the 3D engine (ring buffer, IBs, etc.), but the
2223  * DMA controller has it's own packet format that is
2224  * different form the PM4 format used by the 3D engine.
2225  * It supports copying data, writing embedded data,
2226  * solid fills, and a number of other things.  It also
2227  * has support for tiling/detiling of buffers.
2228  */
2229 /**
2230  * r600_dma_stop - stop the async dma engine
2231  *
2232  * @rdev: radeon_device pointer
2233  *
2234  * Stop the async dma engine (r6xx-evergreen).
2235  */
2236 void r600_dma_stop(struct radeon_device *rdev)
2237 {
2238 	u32 rb_cntl = RREG32(DMA_RB_CNTL);
2239 
2240 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
2241 
2242 	rb_cntl &= ~DMA_RB_ENABLE;
2243 	WREG32(DMA_RB_CNTL, rb_cntl);
2244 
2245 	rdev->ring[R600_RING_TYPE_DMA_INDEX].ready = false;
2246 }
2247 
2248 /**
2249  * r600_dma_resume - setup and start the async dma engine
2250  *
2251  * @rdev: radeon_device pointer
2252  *
2253  * Set up the DMA ring buffer and enable it. (r6xx-evergreen).
2254  * Returns 0 for success, error for failure.
2255  */
2256 int r600_dma_resume(struct radeon_device *rdev)
2257 {
2258 	struct radeon_ring *ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
2259 	u32 rb_cntl, dma_cntl;
2260 	u32 rb_bufsz;
2261 	int r;
2262 
2263 	/* Reset dma */
2264 	if (rdev->family >= CHIP_RV770)
2265 		WREG32(SRBM_SOFT_RESET, RV770_SOFT_RESET_DMA);
2266 	else
2267 		WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA);
2268 	RREG32(SRBM_SOFT_RESET);
2269 	udelay(50);
2270 	WREG32(SRBM_SOFT_RESET, 0);
2271 
2272 	WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL, 0);
2273 	WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL, 0);
2274 
2275 	/* Set ring buffer size in dwords */
2276 	rb_bufsz = drm_order(ring->ring_size / 4);
2277 	rb_cntl = rb_bufsz << 1;
2278 #ifdef __BIG_ENDIAN
2279 	rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE;
2280 #endif
2281 	WREG32(DMA_RB_CNTL, rb_cntl);
2282 
2283 	/* Initialize the ring buffer's read and write pointers */
2284 	WREG32(DMA_RB_RPTR, 0);
2285 	WREG32(DMA_RB_WPTR, 0);
2286 
2287 	/* set the wb address whether it's enabled or not */
2288 	WREG32(DMA_RB_RPTR_ADDR_HI,
2289 	       upper_32_bits(rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFF);
2290 	WREG32(DMA_RB_RPTR_ADDR_LO,
2291 	       ((rdev->wb.gpu_addr + R600_WB_DMA_RPTR_OFFSET) & 0xFFFFFFFC));
2292 
2293 	if (rdev->wb.enabled)
2294 		rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE;
2295 
2296 	WREG32(DMA_RB_BASE, ring->gpu_addr >> 8);
2297 
2298 	/* enable DMA IBs */
2299 	WREG32(DMA_IB_CNTL, DMA_IB_ENABLE);
2300 
2301 	dma_cntl = RREG32(DMA_CNTL);
2302 	dma_cntl &= ~CTXEMPTY_INT_ENABLE;
2303 	WREG32(DMA_CNTL, dma_cntl);
2304 
2305 	if (rdev->family >= CHIP_RV770)
2306 		WREG32(DMA_MODE, 1);
2307 
2308 	ring->wptr = 0;
2309 	WREG32(DMA_RB_WPTR, ring->wptr << 2);
2310 
2311 	ring->rptr = RREG32(DMA_RB_RPTR) >> 2;
2312 
2313 	WREG32(DMA_RB_CNTL, rb_cntl | DMA_RB_ENABLE);
2314 
2315 	ring->ready = true;
2316 
2317 	r = radeon_ring_test(rdev, R600_RING_TYPE_DMA_INDEX, ring);
2318 	if (r) {
2319 		ring->ready = false;
2320 		return r;
2321 	}
2322 
2323 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
2324 
2325 	return 0;
2326 }
2327 
2328 /**
2329  * r600_dma_fini - tear down the async dma engine
2330  *
2331  * @rdev: radeon_device pointer
2332  *
2333  * Stop the async dma engine and free the ring (r6xx-evergreen).
2334  */
2335 void r600_dma_fini(struct radeon_device *rdev)
2336 {
2337 	r600_dma_stop(rdev);
2338 	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]);
2339 }
2340 
2341 /*
2342  * GPU scratch registers helpers function.
2343  */
2344 void r600_scratch_init(struct radeon_device *rdev)
2345 {
2346 	int i;
2347 
2348 	rdev->scratch.num_reg = 7;
2349 	rdev->scratch.reg_base = SCRATCH_REG0;
2350 	for (i = 0; i < rdev->scratch.num_reg; i++) {
2351 		rdev->scratch.free[i] = true;
2352 		rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
2353 	}
2354 }
2355 
2356 int r600_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
2357 {
2358 	uint32_t scratch;
2359 	uint32_t tmp = 0;
2360 	unsigned i;
2361 	int r;
2362 
2363 	r = radeon_scratch_get(rdev, &scratch);
2364 	if (r) {
2365 		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
2366 		return r;
2367 	}
2368 	WREG32(scratch, 0xCAFEDEAD);
2369 	r = radeon_ring_lock(rdev, ring, 3);
2370 	if (r) {
2371 		DRM_ERROR("radeon: cp failed to lock ring %d (%d).\n", ring->idx, r);
2372 		radeon_scratch_free(rdev, scratch);
2373 		return r;
2374 	}
2375 	radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2376 	radeon_ring_write(ring, ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
2377 	radeon_ring_write(ring, 0xDEADBEEF);
2378 	radeon_ring_unlock_commit(rdev, ring);
2379 	for (i = 0; i < rdev->usec_timeout; i++) {
2380 		tmp = RREG32(scratch);
2381 		if (tmp == 0xDEADBEEF)
2382 			break;
2383 		DRM_UDELAY(1);
2384 	}
2385 	if (i < rdev->usec_timeout) {
2386 		DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
2387 	} else {
2388 		DRM_ERROR("radeon: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
2389 			  ring->idx, scratch, tmp);
2390 		r = -EINVAL;
2391 	}
2392 	radeon_scratch_free(rdev, scratch);
2393 	return r;
2394 }
2395 
2396 /**
2397  * r600_dma_ring_test - simple async dma engine test
2398  *
2399  * @rdev: radeon_device pointer
2400  * @ring: radeon_ring structure holding ring information
2401  *
2402  * Test the DMA engine by writing using it to write an
2403  * value to memory. (r6xx-SI).
2404  * Returns 0 for success, error for failure.
2405  */
2406 int r600_dma_ring_test(struct radeon_device *rdev,
2407 		       struct radeon_ring *ring)
2408 {
2409 	unsigned i;
2410 	int r;
2411 	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
2412 	u32 tmp;
2413 
2414 	if (!ptr) {
2415 		DRM_ERROR("invalid vram scratch pointer\n");
2416 		return -EINVAL;
2417 	}
2418 
2419 	tmp = 0xCAFEDEAD;
2420 	writel(tmp, ptr);
2421 
2422 	r = radeon_ring_lock(rdev, ring, 4);
2423 	if (r) {
2424 		DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r);
2425 		return r;
2426 	}
2427 	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
2428 	radeon_ring_write(ring, rdev->vram_scratch.gpu_addr & 0xfffffffc);
2429 	radeon_ring_write(ring, upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xff);
2430 	radeon_ring_write(ring, 0xDEADBEEF);
2431 	radeon_ring_unlock_commit(rdev, ring);
2432 
2433 	for (i = 0; i < rdev->usec_timeout; i++) {
2434 		tmp = readl(ptr);
2435 		if (tmp == 0xDEADBEEF)
2436 			break;
2437 		DRM_UDELAY(1);
2438 	}
2439 
2440 	if (i < rdev->usec_timeout) {
2441 		DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i);
2442 	} else {
2443 		DRM_ERROR("radeon: ring %d test failed (0x%08X)\n",
2444 			  ring->idx, tmp);
2445 		r = -EINVAL;
2446 	}
2447 	return r;
2448 }
2449 
2450 /*
2451  * CP fences/semaphores
2452  */
2453 
2454 void r600_fence_ring_emit(struct radeon_device *rdev,
2455 			  struct radeon_fence *fence)
2456 {
2457 	struct radeon_ring *ring = &rdev->ring[fence->ring];
2458 
2459 	if (rdev->wb.use_event) {
2460 		u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
2461 		/* flush read cache over gart */
2462 		radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
2463 		radeon_ring_write(ring, PACKET3_TC_ACTION_ENA |
2464 					PACKET3_VC_ACTION_ENA |
2465 					PACKET3_SH_ACTION_ENA);
2466 		radeon_ring_write(ring, 0xFFFFFFFF);
2467 		radeon_ring_write(ring, 0);
2468 		radeon_ring_write(ring, 10); /* poll interval */
2469 		/* EVENT_WRITE_EOP - flush caches, send int */
2470 		radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
2471 		radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT_TS) | EVENT_INDEX(5));
2472 		radeon_ring_write(ring, addr & 0xffffffff);
2473 		radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
2474 		radeon_ring_write(ring, fence->seq);
2475 		radeon_ring_write(ring, 0);
2476 	} else {
2477 		/* flush read cache over gart */
2478 		radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
2479 		radeon_ring_write(ring, PACKET3_TC_ACTION_ENA |
2480 					PACKET3_VC_ACTION_ENA |
2481 					PACKET3_SH_ACTION_ENA);
2482 		radeon_ring_write(ring, 0xFFFFFFFF);
2483 		radeon_ring_write(ring, 0);
2484 		radeon_ring_write(ring, 10); /* poll interval */
2485 		radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
2486 		radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0));
2487 		/* wait for 3D idle clean */
2488 		radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2489 		radeon_ring_write(ring, (WAIT_UNTIL - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
2490 		radeon_ring_write(ring, WAIT_3D_IDLE_bit | WAIT_3D_IDLECLEAN_bit);
2491 		/* Emit fence sequence & fire IRQ */
2492 		radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2493 		radeon_ring_write(ring, ((rdev->fence_drv[fence->ring].scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
2494 		radeon_ring_write(ring, fence->seq);
2495 		/* CP_INTERRUPT packet 3 no longer exists, use packet 0 */
2496 		radeon_ring_write(ring, PACKET0(CP_INT_STATUS, 0));
2497 		radeon_ring_write(ring, RB_INT_STAT);
2498 	}
2499 }
2500 
2501 void r600_semaphore_ring_emit(struct radeon_device *rdev,
2502 			      struct radeon_ring *ring,
2503 			      struct radeon_semaphore *semaphore,
2504 			      bool emit_wait)
2505 {
2506 	uint64_t addr = semaphore->gpu_addr;
2507 	unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL;
2508 
2509 	if (rdev->family < CHIP_CAYMAN)
2510 		sel |= PACKET3_SEM_WAIT_ON_SIGNAL;
2511 
2512 	radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1));
2513 	radeon_ring_write(ring, addr & 0xffffffff);
2514 	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | sel);
2515 }
2516 
2517 /*
2518  * DMA fences/semaphores
2519  */
2520 
2521 /**
2522  * r600_dma_fence_ring_emit - emit a fence on the DMA ring
2523  *
2524  * @rdev: radeon_device pointer
2525  * @fence: radeon fence object
2526  *
2527  * Add a DMA fence packet to the ring to write
2528  * the fence seq number and DMA trap packet to generate
2529  * an interrupt if needed (r6xx-r7xx).
2530  */
2531 void r600_dma_fence_ring_emit(struct radeon_device *rdev,
2532 			      struct radeon_fence *fence)
2533 {
2534 	struct radeon_ring *ring = &rdev->ring[fence->ring];
2535 	u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
2536 
2537 	/* write the fence */
2538 	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_FENCE, 0, 0, 0));
2539 	radeon_ring_write(ring, addr & 0xfffffffc);
2540 	radeon_ring_write(ring, (upper_32_bits(addr) & 0xff));
2541 	radeon_ring_write(ring, lower_32_bits(fence->seq));
2542 	/* generate an interrupt */
2543 	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_TRAP, 0, 0, 0));
2544 }
2545 
2546 /**
2547  * r600_dma_semaphore_ring_emit - emit a semaphore on the dma ring
2548  *
2549  * @rdev: radeon_device pointer
2550  * @ring: radeon_ring structure holding ring information
2551  * @semaphore: radeon semaphore object
2552  * @emit_wait: wait or signal semaphore
2553  *
2554  * Add a DMA semaphore packet to the ring wait on or signal
2555  * other rings (r6xx-SI).
2556  */
2557 void r600_dma_semaphore_ring_emit(struct radeon_device *rdev,
2558 				  struct radeon_ring *ring,
2559 				  struct radeon_semaphore *semaphore,
2560 				  bool emit_wait)
2561 {
2562 	u64 addr = semaphore->gpu_addr;
2563 	u32 s = emit_wait ? 0 : 1;
2564 
2565 	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SEMAPHORE, 0, s, 0));
2566 	radeon_ring_write(ring, addr & 0xfffffffc);
2567 	radeon_ring_write(ring, upper_32_bits(addr) & 0xff);
2568 }
2569 
2570 int r600_copy_blit(struct radeon_device *rdev,
2571 		   uint64_t src_offset,
2572 		   uint64_t dst_offset,
2573 		   unsigned num_gpu_pages,
2574 		   struct radeon_fence **fence)
2575 {
2576 	struct radeon_semaphore *sem = NULL;
2577 	struct radeon_sa_bo *vb = NULL;
2578 	int r;
2579 
2580 	r = r600_blit_prepare_copy(rdev, num_gpu_pages, fence, &vb, &sem);
2581 	if (r) {
2582 		return r;
2583 	}
2584 	r600_kms_blit_copy(rdev, src_offset, dst_offset, num_gpu_pages, vb);
2585 	r600_blit_done_copy(rdev, fence, vb, sem);
2586 	return 0;
2587 }
2588 
2589 /**
2590  * r600_copy_dma - copy pages using the DMA engine
2591  *
2592  * @rdev: radeon_device pointer
2593  * @src_offset: src GPU address
2594  * @dst_offset: dst GPU address
2595  * @num_gpu_pages: number of GPU pages to xfer
2596  * @fence: radeon fence object
2597  *
2598  * Copy GPU paging using the DMA engine (r6xx-r7xx).
2599  * Used by the radeon ttm implementation to move pages if
2600  * registered as the asic copy callback.
2601  */
2602 int r600_copy_dma(struct radeon_device *rdev,
2603 		  uint64_t src_offset, uint64_t dst_offset,
2604 		  unsigned num_gpu_pages,
2605 		  struct radeon_fence **fence)
2606 {
2607 	struct radeon_semaphore *sem = NULL;
2608 	int ring_index = rdev->asic->copy.dma_ring_index;
2609 	struct radeon_ring *ring = &rdev->ring[ring_index];
2610 	u32 size_in_dw, cur_size_in_dw;
2611 	int i, num_loops;
2612 	int r = 0;
2613 
2614 	r = radeon_semaphore_create(rdev, &sem);
2615 	if (r) {
2616 		DRM_ERROR("radeon: moving bo (%d).\n", r);
2617 		return r;
2618 	}
2619 
2620 	size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4;
2621 	num_loops = DIV_ROUND_UP(size_in_dw, 0xffff);
2622 	r = radeon_ring_lock(rdev, ring, num_loops * 5 + 8);
2623 	if (r) {
2624 		DRM_ERROR("radeon: moving bo (%d).\n", r);
2625 		radeon_semaphore_free(rdev, &sem, NULL);
2626 		return r;
2627 	}
2628 
2629 	if (radeon_fence_need_sync(*fence, ring->idx)) {
2630 		radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
2631 					    ring->idx);
2632 		radeon_fence_note_sync(*fence, ring->idx);
2633 	} else {
2634 		radeon_semaphore_free(rdev, &sem, NULL);
2635 	}
2636 
2637 	for (i = 0; i < num_loops; i++) {
2638 		cur_size_in_dw = size_in_dw;
2639 		if (cur_size_in_dw > 0xFFFF)
2640 			cur_size_in_dw = 0xFFFF;
2641 		size_in_dw -= cur_size_in_dw;
2642 		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw));
2643 		radeon_ring_write(ring, dst_offset & 0xfffffffc);
2644 		radeon_ring_write(ring, src_offset & 0xfffffffc);
2645 		radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
2646 		radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
2647 		src_offset += cur_size_in_dw * 4;
2648 		dst_offset += cur_size_in_dw * 4;
2649 	}
2650 
2651 	r = radeon_fence_emit(rdev, fence, ring->idx);
2652 	if (r) {
2653 		radeon_ring_unlock_undo(rdev, ring);
2654 		return r;
2655 	}
2656 
2657 	radeon_ring_unlock_commit(rdev, ring);
2658 	radeon_semaphore_free(rdev, &sem, *fence);
2659 
2660 	return r;
2661 }
2662 
2663 int r600_set_surface_reg(struct radeon_device *rdev, int reg,
2664 			 uint32_t tiling_flags, uint32_t pitch,
2665 			 uint32_t offset, uint32_t obj_size)
2666 {
2667 	/* FIXME: implement */
2668 	return 0;
2669 }
2670 
2671 void r600_clear_surface_reg(struct radeon_device *rdev, int reg)
2672 {
2673 	/* FIXME: implement */
2674 }
2675 
2676 static int r600_startup(struct radeon_device *rdev)
2677 {
2678 	struct radeon_ring *ring;
2679 	int r;
2680 
2681 	/* enable pcie gen2 link */
2682 	r600_pcie_gen2_enable(rdev);
2683 
2684 	if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
2685 		r = r600_init_microcode(rdev);
2686 		if (r) {
2687 			DRM_ERROR("Failed to load firmware!\n");
2688 			return r;
2689 		}
2690 	}
2691 
2692 	r = r600_vram_scratch_init(rdev);
2693 	if (r)
2694 		return r;
2695 
2696 	r600_mc_program(rdev);
2697 	if (rdev->flags & RADEON_IS_AGP) {
2698 		r600_agp_enable(rdev);
2699 	} else {
2700 		r = r600_pcie_gart_enable(rdev);
2701 		if (r)
2702 			return r;
2703 	}
2704 	r600_gpu_init(rdev);
2705 	r = r600_blit_init(rdev);
2706 	if (r) {
2707 		r600_blit_fini(rdev);
2708 		rdev->asic->copy.copy = NULL;
2709 		dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
2710 	}
2711 
2712 	/* allocate wb buffer */
2713 	r = radeon_wb_init(rdev);
2714 	if (r)
2715 		return r;
2716 
2717 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
2718 	if (r) {
2719 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
2720 		return r;
2721 	}
2722 
2723 	r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
2724 	if (r) {
2725 		dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
2726 		return r;
2727 	}
2728 
2729 	/* Enable IRQ */
2730 	r = r600_irq_init(rdev);
2731 	if (r) {
2732 		DRM_ERROR("radeon: IH init failed (%d).\n", r);
2733 		radeon_irq_kms_fini(rdev);
2734 		return r;
2735 	}
2736 	r600_irq_set(rdev);
2737 
2738 	ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2739 	r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
2740 			     R600_CP_RB_RPTR, R600_CP_RB_WPTR,
2741 			     0, 0xfffff, RADEON_CP_PACKET2);
2742 	if (r)
2743 		return r;
2744 
2745 	ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
2746 	r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
2747 			     DMA_RB_RPTR, DMA_RB_WPTR,
2748 			     2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
2749 	if (r)
2750 		return r;
2751 
2752 	r = r600_cp_load_microcode(rdev);
2753 	if (r)
2754 		return r;
2755 	r = r600_cp_resume(rdev);
2756 	if (r)
2757 		return r;
2758 
2759 	r = r600_dma_resume(rdev);
2760 	if (r)
2761 		return r;
2762 
2763 	r = radeon_ib_pool_init(rdev);
2764 	if (r) {
2765 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
2766 		return r;
2767 	}
2768 
2769 	r = r600_audio_init(rdev);
2770 	if (r) {
2771 		DRM_ERROR("radeon: audio init failed\n");
2772 		return r;
2773 	}
2774 
2775 	return 0;
2776 }
2777 
2778 void r600_vga_set_state(struct radeon_device *rdev, bool state)
2779 {
2780 	uint32_t temp;
2781 
2782 	temp = RREG32(CONFIG_CNTL);
2783 	if (state == false) {
2784 		temp &= ~(1<<0);
2785 		temp |= (1<<1);
2786 	} else {
2787 		temp &= ~(1<<1);
2788 	}
2789 	WREG32(CONFIG_CNTL, temp);
2790 }
2791 
2792 int r600_resume(struct radeon_device *rdev)
2793 {
2794 	int r;
2795 
2796 	/* Do not reset GPU before posting, on r600 hw unlike on r500 hw,
2797 	 * posting will perform necessary task to bring back GPU into good
2798 	 * shape.
2799 	 */
2800 	/* post card */
2801 	atom_asic_init(rdev->mode_info.atom_context);
2802 
2803 	rdev->accel_working = true;
2804 	r = r600_startup(rdev);
2805 	if (r) {
2806 		DRM_ERROR("r600 startup failed on resume\n");
2807 		rdev->accel_working = false;
2808 		return r;
2809 	}
2810 
2811 	return r;
2812 }
2813 
2814 int r600_suspend(struct radeon_device *rdev)
2815 {
2816 	r600_audio_fini(rdev);
2817 	r600_cp_stop(rdev);
2818 	r600_dma_stop(rdev);
2819 	r600_irq_suspend(rdev);
2820 	radeon_wb_disable(rdev);
2821 	r600_pcie_gart_disable(rdev);
2822 
2823 	return 0;
2824 }
2825 
2826 /* Plan is to move initialization in that function and use
2827  * helper function so that radeon_device_init pretty much
2828  * do nothing more than calling asic specific function. This
2829  * should also allow to remove a bunch of callback function
2830  * like vram_info.
2831  */
2832 int r600_init(struct radeon_device *rdev)
2833 {
2834 	int r;
2835 
2836 	if (r600_debugfs_mc_info_init(rdev)) {
2837 		DRM_ERROR("Failed to register debugfs file for mc !\n");
2838 	}
2839 	/* Read BIOS */
2840 	if (!radeon_get_bios(rdev)) {
2841 		if (ASIC_IS_AVIVO(rdev))
2842 			return -EINVAL;
2843 	}
2844 	/* Must be an ATOMBIOS */
2845 	if (!rdev->is_atom_bios) {
2846 		dev_err(rdev->dev, "Expecting atombios for R600 GPU\n");
2847 		return -EINVAL;
2848 	}
2849 	r = radeon_atombios_init(rdev);
2850 	if (r)
2851 		return r;
2852 	/* Post card if necessary */
2853 	if (!radeon_card_posted(rdev)) {
2854 		if (!rdev->bios) {
2855 			dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
2856 			return -EINVAL;
2857 		}
2858 		DRM_INFO("GPU not posted. posting now...\n");
2859 		atom_asic_init(rdev->mode_info.atom_context);
2860 	}
2861 	/* Initialize scratch registers */
2862 	r600_scratch_init(rdev);
2863 	/* Initialize surface registers */
2864 	radeon_surface_init(rdev);
2865 	/* Initialize clocks */
2866 	radeon_get_clock_info(rdev->ddev);
2867 	/* Fence driver */
2868 	r = radeon_fence_driver_init(rdev);
2869 	if (r)
2870 		return r;
2871 	if (rdev->flags & RADEON_IS_AGP) {
2872 		r = radeon_agp_init(rdev);
2873 		if (r)
2874 			radeon_agp_disable(rdev);
2875 	}
2876 	r = r600_mc_init(rdev);
2877 	if (r)
2878 		return r;
2879 	/* Memory manager */
2880 	r = radeon_bo_init(rdev);
2881 	if (r)
2882 		return r;
2883 
2884 	r = radeon_irq_kms_init(rdev);
2885 	if (r)
2886 		return r;
2887 
2888 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ring_obj = NULL;
2889 	r600_ring_init(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX], 1024 * 1024);
2890 
2891 	rdev->ring[R600_RING_TYPE_DMA_INDEX].ring_obj = NULL;
2892 	r600_ring_init(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX], 64 * 1024);
2893 
2894 	rdev->ih.ring_obj = NULL;
2895 	r600_ih_ring_init(rdev, 64 * 1024);
2896 
2897 	r = r600_pcie_gart_init(rdev);
2898 	if (r)
2899 		return r;
2900 
2901 	rdev->accel_working = true;
2902 	r = r600_startup(rdev);
2903 	if (r) {
2904 		dev_err(rdev->dev, "disabling GPU acceleration\n");
2905 		r600_cp_fini(rdev);
2906 		r600_dma_fini(rdev);
2907 		r600_irq_fini(rdev);
2908 		radeon_wb_fini(rdev);
2909 		radeon_ib_pool_fini(rdev);
2910 		radeon_irq_kms_fini(rdev);
2911 		r600_pcie_gart_fini(rdev);
2912 		rdev->accel_working = false;
2913 	}
2914 
2915 	return 0;
2916 }
2917 
2918 void r600_fini(struct radeon_device *rdev)
2919 {
2920 	r600_audio_fini(rdev);
2921 	r600_blit_fini(rdev);
2922 	r600_cp_fini(rdev);
2923 	r600_dma_fini(rdev);
2924 	r600_irq_fini(rdev);
2925 	radeon_wb_fini(rdev);
2926 	radeon_ib_pool_fini(rdev);
2927 	radeon_irq_kms_fini(rdev);
2928 	r600_pcie_gart_fini(rdev);
2929 	r600_vram_scratch_fini(rdev);
2930 	radeon_agp_fini(rdev);
2931 	radeon_gem_fini(rdev);
2932 	radeon_fence_driver_fini(rdev);
2933 	radeon_bo_fini(rdev);
2934 	radeon_atombios_fini(rdev);
2935 	kfree(rdev->bios);
2936 	rdev->bios = NULL;
2937 }
2938 
2939 
2940 /*
2941  * CS stuff
2942  */
2943 void r600_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
2944 {
2945 	struct radeon_ring *ring = &rdev->ring[ib->ring];
2946 	u32 next_rptr;
2947 
2948 	if (ring->rptr_save_reg) {
2949 		next_rptr = ring->wptr + 3 + 4;
2950 		radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2951 		radeon_ring_write(ring, ((ring->rptr_save_reg -
2952 					 PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
2953 		radeon_ring_write(ring, next_rptr);
2954 	} else if (rdev->wb.enabled) {
2955 		next_rptr = ring->wptr + 5 + 4;
2956 		radeon_ring_write(ring, PACKET3(PACKET3_MEM_WRITE, 3));
2957 		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
2958 		radeon_ring_write(ring, (upper_32_bits(ring->next_rptr_gpu_addr) & 0xff) | (1 << 18));
2959 		radeon_ring_write(ring, next_rptr);
2960 		radeon_ring_write(ring, 0);
2961 	}
2962 
2963 	radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2964 	radeon_ring_write(ring,
2965 #ifdef __BIG_ENDIAN
2966 			  (2 << 0) |
2967 #endif
2968 			  (ib->gpu_addr & 0xFFFFFFFC));
2969 	radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFF);
2970 	radeon_ring_write(ring, ib->length_dw);
2971 }
2972 
2973 int r600_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
2974 {
2975 	struct radeon_ib ib;
2976 	uint32_t scratch;
2977 	uint32_t tmp = 0;
2978 	unsigned i;
2979 	int r;
2980 
2981 	r = radeon_scratch_get(rdev, &scratch);
2982 	if (r) {
2983 		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
2984 		return r;
2985 	}
2986 	WREG32(scratch, 0xCAFEDEAD);
2987 	r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
2988 	if (r) {
2989 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
2990 		goto free_scratch;
2991 	}
2992 	ib.ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1);
2993 	ib.ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
2994 	ib.ptr[2] = 0xDEADBEEF;
2995 	ib.length_dw = 3;
2996 	r = radeon_ib_schedule(rdev, &ib, NULL);
2997 	if (r) {
2998 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
2999 		goto free_ib;
3000 	}
3001 	r = radeon_fence_wait(ib.fence, false);
3002 	if (r) {
3003 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3004 		goto free_ib;
3005 	}
3006 	for (i = 0; i < rdev->usec_timeout; i++) {
3007 		tmp = RREG32(scratch);
3008 		if (tmp == 0xDEADBEEF)
3009 			break;
3010 		DRM_UDELAY(1);
3011 	}
3012 	if (i < rdev->usec_timeout) {
3013 		DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
3014 	} else {
3015 		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3016 			  scratch, tmp);
3017 		r = -EINVAL;
3018 	}
3019 free_ib:
3020 	radeon_ib_free(rdev, &ib);
3021 free_scratch:
3022 	radeon_scratch_free(rdev, scratch);
3023 	return r;
3024 }
3025 
3026 /**
3027  * r600_dma_ib_test - test an IB on the DMA engine
3028  *
3029  * @rdev: radeon_device pointer
3030  * @ring: radeon_ring structure holding ring information
3031  *
3032  * Test a simple IB in the DMA ring (r6xx-SI).
3033  * Returns 0 on success, error on failure.
3034  */
3035 int r600_dma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3036 {
3037 	struct radeon_ib ib;
3038 	unsigned i;
3039 	int r;
3040 	void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
3041 	u32 tmp = 0;
3042 
3043 	if (!ptr) {
3044 		DRM_ERROR("invalid vram scratch pointer\n");
3045 		return -EINVAL;
3046 	}
3047 
3048 	tmp = 0xCAFEDEAD;
3049 	writel(tmp, ptr);
3050 
3051 	r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256);
3052 	if (r) {
3053 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3054 		return r;
3055 	}
3056 
3057 	ib.ptr[0] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1);
3058 	ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc;
3059 	ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xff;
3060 	ib.ptr[3] = 0xDEADBEEF;
3061 	ib.length_dw = 4;
3062 
3063 	r = radeon_ib_schedule(rdev, &ib, NULL);
3064 	if (r) {
3065 		radeon_ib_free(rdev, &ib);
3066 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3067 		return r;
3068 	}
3069 	r = radeon_fence_wait(ib.fence, false);
3070 	if (r) {
3071 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3072 		return r;
3073 	}
3074 	for (i = 0; i < rdev->usec_timeout; i++) {
3075 		tmp = readl(ptr);
3076 		if (tmp == 0xDEADBEEF)
3077 			break;
3078 		DRM_UDELAY(1);
3079 	}
3080 	if (i < rdev->usec_timeout) {
3081 		DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i);
3082 	} else {
3083 		DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp);
3084 		r = -EINVAL;
3085 	}
3086 	radeon_ib_free(rdev, &ib);
3087 	return r;
3088 }
3089 
3090 /**
3091  * r600_dma_ring_ib_execute - Schedule an IB on the DMA engine
3092  *
3093  * @rdev: radeon_device pointer
3094  * @ib: IB object to schedule
3095  *
3096  * Schedule an IB in the DMA ring (r6xx-r7xx).
3097  */
3098 void r600_dma_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3099 {
3100 	struct radeon_ring *ring = &rdev->ring[ib->ring];
3101 
3102 	if (rdev->wb.enabled) {
3103 		u32 next_rptr = ring->wptr + 4;
3104 		while ((next_rptr & 7) != 5)
3105 			next_rptr++;
3106 		next_rptr += 3;
3107 		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1));
3108 		radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
3109 		radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff);
3110 		radeon_ring_write(ring, next_rptr);
3111 	}
3112 
3113 	/* The indirect buffer packet must end on an 8 DW boundary in the DMA ring.
3114 	 * Pad as necessary with NOPs.
3115 	 */
3116 	while ((ring->wptr & 7) != 5)
3117 		radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0));
3118 	radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_INDIRECT_BUFFER, 0, 0, 0));
3119 	radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0));
3120 	radeon_ring_write(ring, (ib->length_dw << 16) | (upper_32_bits(ib->gpu_addr) & 0xFF));
3121 
3122 }
3123 
3124 /*
3125  * Interrupts
3126  *
3127  * Interrupts use a ring buffer on r6xx/r7xx hardware.  It works pretty
3128  * the same as the CP ring buffer, but in reverse.  Rather than the CPU
3129  * writing to the ring and the GPU consuming, the GPU writes to the ring
3130  * and host consumes.  As the host irq handler processes interrupts, it
3131  * increments the rptr.  When the rptr catches up with the wptr, all the
3132  * current interrupts have been processed.
3133  */
3134 
3135 void r600_ih_ring_init(struct radeon_device *rdev, unsigned ring_size)
3136 {
3137 	u32 rb_bufsz;
3138 
3139 	/* Align ring size */
3140 	rb_bufsz = drm_order(ring_size / 4);
3141 	ring_size = (1 << rb_bufsz) * 4;
3142 	rdev->ih.ring_size = ring_size;
3143 	rdev->ih.ptr_mask = rdev->ih.ring_size - 1;
3144 	rdev->ih.rptr = 0;
3145 }
3146 
3147 int r600_ih_ring_alloc(struct radeon_device *rdev)
3148 {
3149 	int r;
3150 
3151 	/* Allocate ring buffer */
3152 	if (rdev->ih.ring_obj == NULL) {
3153 		r = radeon_bo_create(rdev, rdev->ih.ring_size,
3154 				     PAGE_SIZE, true,
3155 				     RADEON_GEM_DOMAIN_GTT,
3156 				     NULL, &rdev->ih.ring_obj);
3157 		if (r) {
3158 			DRM_ERROR("radeon: failed to create ih ring buffer (%d).\n", r);
3159 			return r;
3160 		}
3161 		r = radeon_bo_reserve(rdev->ih.ring_obj, false);
3162 		if (unlikely(r != 0))
3163 			return r;
3164 		r = radeon_bo_pin(rdev->ih.ring_obj,
3165 				  RADEON_GEM_DOMAIN_GTT,
3166 				  &rdev->ih.gpu_addr);
3167 		if (r) {
3168 			radeon_bo_unreserve(rdev->ih.ring_obj);
3169 			DRM_ERROR("radeon: failed to pin ih ring buffer (%d).\n", r);
3170 			return r;
3171 		}
3172 		r = radeon_bo_kmap(rdev->ih.ring_obj,
3173 				   (void **)&rdev->ih.ring);
3174 		radeon_bo_unreserve(rdev->ih.ring_obj);
3175 		if (r) {
3176 			DRM_ERROR("radeon: failed to map ih ring buffer (%d).\n", r);
3177 			return r;
3178 		}
3179 	}
3180 	return 0;
3181 }
3182 
3183 void r600_ih_ring_fini(struct radeon_device *rdev)
3184 {
3185 	int r;
3186 	if (rdev->ih.ring_obj) {
3187 		r = radeon_bo_reserve(rdev->ih.ring_obj, false);
3188 		if (likely(r == 0)) {
3189 			radeon_bo_kunmap(rdev->ih.ring_obj);
3190 			radeon_bo_unpin(rdev->ih.ring_obj);
3191 			radeon_bo_unreserve(rdev->ih.ring_obj);
3192 		}
3193 		radeon_bo_unref(&rdev->ih.ring_obj);
3194 		rdev->ih.ring = NULL;
3195 		rdev->ih.ring_obj = NULL;
3196 	}
3197 }
3198 
3199 void r600_rlc_stop(struct radeon_device *rdev)
3200 {
3201 
3202 	if ((rdev->family >= CHIP_RV770) &&
3203 	    (rdev->family <= CHIP_RV740)) {
3204 		/* r7xx asics need to soft reset RLC before halting */
3205 		WREG32(SRBM_SOFT_RESET, SOFT_RESET_RLC);
3206 		RREG32(SRBM_SOFT_RESET);
3207 		mdelay(15);
3208 		WREG32(SRBM_SOFT_RESET, 0);
3209 		RREG32(SRBM_SOFT_RESET);
3210 	}
3211 
3212 	WREG32(RLC_CNTL, 0);
3213 }
3214 
3215 static void r600_rlc_start(struct radeon_device *rdev)
3216 {
3217 	WREG32(RLC_CNTL, RLC_ENABLE);
3218 }
3219 
3220 static int r600_rlc_init(struct radeon_device *rdev)
3221 {
3222 	u32 i;
3223 	const __be32 *fw_data;
3224 
3225 	if (!rdev->rlc_fw)
3226 		return -EINVAL;
3227 
3228 	r600_rlc_stop(rdev);
3229 
3230 	WREG32(RLC_HB_CNTL, 0);
3231 
3232 	if (rdev->family == CHIP_ARUBA) {
3233 		WREG32(TN_RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
3234 		WREG32(TN_RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8);
3235 	}
3236 	if (rdev->family <= CHIP_CAYMAN) {
3237 		WREG32(RLC_HB_BASE, 0);
3238 		WREG32(RLC_HB_RPTR, 0);
3239 		WREG32(RLC_HB_WPTR, 0);
3240 	}
3241 	if (rdev->family <= CHIP_CAICOS) {
3242 		WREG32(RLC_HB_WPTR_LSB_ADDR, 0);
3243 		WREG32(RLC_HB_WPTR_MSB_ADDR, 0);
3244 	}
3245 	WREG32(RLC_MC_CNTL, 0);
3246 	WREG32(RLC_UCODE_CNTL, 0);
3247 
3248 	fw_data = (const __be32 *)rdev->rlc_fw->data;
3249 	if (rdev->family >= CHIP_ARUBA) {
3250 		for (i = 0; i < ARUBA_RLC_UCODE_SIZE; i++) {
3251 			WREG32(RLC_UCODE_ADDR, i);
3252 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3253 		}
3254 	} else if (rdev->family >= CHIP_CAYMAN) {
3255 		for (i = 0; i < CAYMAN_RLC_UCODE_SIZE; i++) {
3256 			WREG32(RLC_UCODE_ADDR, i);
3257 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3258 		}
3259 	} else if (rdev->family >= CHIP_CEDAR) {
3260 		for (i = 0; i < EVERGREEN_RLC_UCODE_SIZE; i++) {
3261 			WREG32(RLC_UCODE_ADDR, i);
3262 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3263 		}
3264 	} else if (rdev->family >= CHIP_RV770) {
3265 		for (i = 0; i < R700_RLC_UCODE_SIZE; i++) {
3266 			WREG32(RLC_UCODE_ADDR, i);
3267 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3268 		}
3269 	} else {
3270 		for (i = 0; i < RLC_UCODE_SIZE; i++) {
3271 			WREG32(RLC_UCODE_ADDR, i);
3272 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3273 		}
3274 	}
3275 	WREG32(RLC_UCODE_ADDR, 0);
3276 
3277 	r600_rlc_start(rdev);
3278 
3279 	return 0;
3280 }
3281 
3282 static void r600_enable_interrupts(struct radeon_device *rdev)
3283 {
3284 	u32 ih_cntl = RREG32(IH_CNTL);
3285 	u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3286 
3287 	ih_cntl |= ENABLE_INTR;
3288 	ih_rb_cntl |= IH_RB_ENABLE;
3289 	WREG32(IH_CNTL, ih_cntl);
3290 	WREG32(IH_RB_CNTL, ih_rb_cntl);
3291 	rdev->ih.enabled = true;
3292 }
3293 
3294 void r600_disable_interrupts(struct radeon_device *rdev)
3295 {
3296 	u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3297 	u32 ih_cntl = RREG32(IH_CNTL);
3298 
3299 	ih_rb_cntl &= ~IH_RB_ENABLE;
3300 	ih_cntl &= ~ENABLE_INTR;
3301 	WREG32(IH_RB_CNTL, ih_rb_cntl);
3302 	WREG32(IH_CNTL, ih_cntl);
3303 	/* set rptr, wptr to 0 */
3304 	WREG32(IH_RB_RPTR, 0);
3305 	WREG32(IH_RB_WPTR, 0);
3306 	rdev->ih.enabled = false;
3307 	rdev->ih.rptr = 0;
3308 }
3309 
3310 static void r600_disable_interrupt_state(struct radeon_device *rdev)
3311 {
3312 	u32 tmp;
3313 
3314 	WREG32(CP_INT_CNTL, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
3315 	tmp = RREG32(DMA_CNTL) & ~TRAP_ENABLE;
3316 	WREG32(DMA_CNTL, tmp);
3317 	WREG32(GRBM_INT_CNTL, 0);
3318 	WREG32(DxMODE_INT_MASK, 0);
3319 	WREG32(D1GRPH_INTERRUPT_CONTROL, 0);
3320 	WREG32(D2GRPH_INTERRUPT_CONTROL, 0);
3321 	if (ASIC_IS_DCE3(rdev)) {
3322 		WREG32(DCE3_DACA_AUTODETECT_INT_CONTROL, 0);
3323 		WREG32(DCE3_DACB_AUTODETECT_INT_CONTROL, 0);
3324 		tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3325 		WREG32(DC_HPD1_INT_CONTROL, tmp);
3326 		tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3327 		WREG32(DC_HPD2_INT_CONTROL, tmp);
3328 		tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3329 		WREG32(DC_HPD3_INT_CONTROL, tmp);
3330 		tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3331 		WREG32(DC_HPD4_INT_CONTROL, tmp);
3332 		if (ASIC_IS_DCE32(rdev)) {
3333 			tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3334 			WREG32(DC_HPD5_INT_CONTROL, tmp);
3335 			tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3336 			WREG32(DC_HPD6_INT_CONTROL, tmp);
3337 			tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET0) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3338 			WREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET0, tmp);
3339 			tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET1) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3340 			WREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET1, tmp);
3341 		} else {
3342 			tmp = RREG32(HDMI0_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3343 			WREG32(HDMI0_AUDIO_PACKET_CONTROL, tmp);
3344 			tmp = RREG32(DCE3_HDMI1_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3345 			WREG32(DCE3_HDMI1_AUDIO_PACKET_CONTROL, tmp);
3346 		}
3347 	} else {
3348 		WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
3349 		WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
3350 		tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
3351 		WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
3352 		tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
3353 		WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
3354 		tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
3355 		WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
3356 		tmp = RREG32(HDMI0_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3357 		WREG32(HDMI0_AUDIO_PACKET_CONTROL, tmp);
3358 		tmp = RREG32(HDMI1_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3359 		WREG32(HDMI1_AUDIO_PACKET_CONTROL, tmp);
3360 	}
3361 }
3362 
3363 int r600_irq_init(struct radeon_device *rdev)
3364 {
3365 	int ret = 0;
3366 	int rb_bufsz;
3367 	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
3368 
3369 	/* allocate ring */
3370 	ret = r600_ih_ring_alloc(rdev);
3371 	if (ret)
3372 		return ret;
3373 
3374 	/* disable irqs */
3375 	r600_disable_interrupts(rdev);
3376 
3377 	/* init rlc */
3378 	ret = r600_rlc_init(rdev);
3379 	if (ret) {
3380 		r600_ih_ring_fini(rdev);
3381 		return ret;
3382 	}
3383 
3384 	/* setup interrupt control */
3385 	/* set dummy read address to ring address */
3386 	WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
3387 	interrupt_cntl = RREG32(INTERRUPT_CNTL);
3388 	/* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
3389 	 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
3390 	 */
3391 	interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
3392 	/* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
3393 	interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
3394 	WREG32(INTERRUPT_CNTL, interrupt_cntl);
3395 
3396 	WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
3397 	rb_bufsz = drm_order(rdev->ih.ring_size / 4);
3398 
3399 	ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
3400 		      IH_WPTR_OVERFLOW_CLEAR |
3401 		      (rb_bufsz << 1));
3402 
3403 	if (rdev->wb.enabled)
3404 		ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
3405 
3406 	/* set the writeback address whether it's enabled or not */
3407 	WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
3408 	WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
3409 
3410 	WREG32(IH_RB_CNTL, ih_rb_cntl);
3411 
3412 	/* set rptr, wptr to 0 */
3413 	WREG32(IH_RB_RPTR, 0);
3414 	WREG32(IH_RB_WPTR, 0);
3415 
3416 	/* Default settings for IH_CNTL (disabled at first) */
3417 	ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10);
3418 	/* RPTR_REARM only works if msi's are enabled */
3419 	if (rdev->msi_enabled)
3420 		ih_cntl |= RPTR_REARM;
3421 	WREG32(IH_CNTL, ih_cntl);
3422 
3423 	/* force the active interrupt state to all disabled */
3424 	if (rdev->family >= CHIP_CEDAR)
3425 		evergreen_disable_interrupt_state(rdev);
3426 	else
3427 		r600_disable_interrupt_state(rdev);
3428 
3429 	/* at this point everything should be setup correctly to enable master */
3430 	pci_set_master(rdev->pdev);
3431 
3432 	/* enable irqs */
3433 	r600_enable_interrupts(rdev);
3434 
3435 	return ret;
3436 }
3437 
3438 void r600_irq_suspend(struct radeon_device *rdev)
3439 {
3440 	r600_irq_disable(rdev);
3441 	r600_rlc_stop(rdev);
3442 }
3443 
3444 void r600_irq_fini(struct radeon_device *rdev)
3445 {
3446 	r600_irq_suspend(rdev);
3447 	r600_ih_ring_fini(rdev);
3448 }
3449 
3450 int r600_irq_set(struct radeon_device *rdev)
3451 {
3452 	u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
3453 	u32 mode_int = 0;
3454 	u32 hpd1, hpd2, hpd3, hpd4 = 0, hpd5 = 0, hpd6 = 0;
3455 	u32 grbm_int_cntl = 0;
3456 	u32 hdmi0, hdmi1;
3457 	u32 d1grph = 0, d2grph = 0;
3458 	u32 dma_cntl;
3459 
3460 	if (!rdev->irq.installed) {
3461 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
3462 		return -EINVAL;
3463 	}
3464 	/* don't enable anything if the ih is disabled */
3465 	if (!rdev->ih.enabled) {
3466 		r600_disable_interrupts(rdev);
3467 		/* force the active interrupt state to all disabled */
3468 		r600_disable_interrupt_state(rdev);
3469 		return 0;
3470 	}
3471 
3472 	if (ASIC_IS_DCE3(rdev)) {
3473 		hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
3474 		hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
3475 		hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
3476 		hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
3477 		if (ASIC_IS_DCE32(rdev)) {
3478 			hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
3479 			hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
3480 			hdmi0 = RREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET0) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
3481 			hdmi1 = RREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET1) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
3482 		} else {
3483 			hdmi0 = RREG32(HDMI0_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3484 			hdmi1 = RREG32(DCE3_HDMI1_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3485 		}
3486 	} else {
3487 		hpd1 = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & ~DC_HPDx_INT_EN;
3488 		hpd2 = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & ~DC_HPDx_INT_EN;
3489 		hpd3 = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & ~DC_HPDx_INT_EN;
3490 		hdmi0 = RREG32(HDMI0_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3491 		hdmi1 = RREG32(HDMI1_AUDIO_PACKET_CONTROL) & ~HDMI0_AZ_FORMAT_WTRIG_MASK;
3492 	}
3493 	dma_cntl = RREG32(DMA_CNTL) & ~TRAP_ENABLE;
3494 
3495 	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
3496 		DRM_DEBUG("r600_irq_set: sw int\n");
3497 		cp_int_cntl |= RB_INT_ENABLE;
3498 		cp_int_cntl |= TIME_STAMP_INT_ENABLE;
3499 	}
3500 
3501 	if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
3502 		DRM_DEBUG("r600_irq_set: sw int dma\n");
3503 		dma_cntl |= TRAP_ENABLE;
3504 	}
3505 
3506 	if (rdev->irq.crtc_vblank_int[0] ||
3507 	    atomic_read(&rdev->irq.pflip[0])) {
3508 		DRM_DEBUG("r600_irq_set: vblank 0\n");
3509 		mode_int |= D1MODE_VBLANK_INT_MASK;
3510 	}
3511 	if (rdev->irq.crtc_vblank_int[1] ||
3512 	    atomic_read(&rdev->irq.pflip[1])) {
3513 		DRM_DEBUG("r600_irq_set: vblank 1\n");
3514 		mode_int |= D2MODE_VBLANK_INT_MASK;
3515 	}
3516 	if (rdev->irq.hpd[0]) {
3517 		DRM_DEBUG("r600_irq_set: hpd 1\n");
3518 		hpd1 |= DC_HPDx_INT_EN;
3519 	}
3520 	if (rdev->irq.hpd[1]) {
3521 		DRM_DEBUG("r600_irq_set: hpd 2\n");
3522 		hpd2 |= DC_HPDx_INT_EN;
3523 	}
3524 	if (rdev->irq.hpd[2]) {
3525 		DRM_DEBUG("r600_irq_set: hpd 3\n");
3526 		hpd3 |= DC_HPDx_INT_EN;
3527 	}
3528 	if (rdev->irq.hpd[3]) {
3529 		DRM_DEBUG("r600_irq_set: hpd 4\n");
3530 		hpd4 |= DC_HPDx_INT_EN;
3531 	}
3532 	if (rdev->irq.hpd[4]) {
3533 		DRM_DEBUG("r600_irq_set: hpd 5\n");
3534 		hpd5 |= DC_HPDx_INT_EN;
3535 	}
3536 	if (rdev->irq.hpd[5]) {
3537 		DRM_DEBUG("r600_irq_set: hpd 6\n");
3538 		hpd6 |= DC_HPDx_INT_EN;
3539 	}
3540 	if (rdev->irq.afmt[0]) {
3541 		DRM_DEBUG("r600_irq_set: hdmi 0\n");
3542 		hdmi0 |= HDMI0_AZ_FORMAT_WTRIG_MASK;
3543 	}
3544 	if (rdev->irq.afmt[1]) {
3545 		DRM_DEBUG("r600_irq_set: hdmi 0\n");
3546 		hdmi1 |= HDMI0_AZ_FORMAT_WTRIG_MASK;
3547 	}
3548 
3549 	WREG32(CP_INT_CNTL, cp_int_cntl);
3550 	WREG32(DMA_CNTL, dma_cntl);
3551 	WREG32(DxMODE_INT_MASK, mode_int);
3552 	WREG32(D1GRPH_INTERRUPT_CONTROL, d1grph);
3553 	WREG32(D2GRPH_INTERRUPT_CONTROL, d2grph);
3554 	WREG32(GRBM_INT_CNTL, grbm_int_cntl);
3555 	if (ASIC_IS_DCE3(rdev)) {
3556 		WREG32(DC_HPD1_INT_CONTROL, hpd1);
3557 		WREG32(DC_HPD2_INT_CONTROL, hpd2);
3558 		WREG32(DC_HPD3_INT_CONTROL, hpd3);
3559 		WREG32(DC_HPD4_INT_CONTROL, hpd4);
3560 		if (ASIC_IS_DCE32(rdev)) {
3561 			WREG32(DC_HPD5_INT_CONTROL, hpd5);
3562 			WREG32(DC_HPD6_INT_CONTROL, hpd6);
3563 			WREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET0, hdmi0);
3564 			WREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET1, hdmi1);
3565 		} else {
3566 			WREG32(HDMI0_AUDIO_PACKET_CONTROL, hdmi0);
3567 			WREG32(DCE3_HDMI1_AUDIO_PACKET_CONTROL, hdmi1);
3568 		}
3569 	} else {
3570 		WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, hpd1);
3571 		WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, hpd2);
3572 		WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, hpd3);
3573 		WREG32(HDMI0_AUDIO_PACKET_CONTROL, hdmi0);
3574 		WREG32(HDMI1_AUDIO_PACKET_CONTROL, hdmi1);
3575 	}
3576 
3577 	return 0;
3578 }
3579 
3580 static void r600_irq_ack(struct radeon_device *rdev)
3581 {
3582 	u32 tmp;
3583 
3584 	if (ASIC_IS_DCE3(rdev)) {
3585 		rdev->irq.stat_regs.r600.disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
3586 		rdev->irq.stat_regs.r600.disp_int_cont = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE);
3587 		rdev->irq.stat_regs.r600.disp_int_cont2 = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE2);
3588 		if (ASIC_IS_DCE32(rdev)) {
3589 			rdev->irq.stat_regs.r600.hdmi0_status = RREG32(AFMT_STATUS + DCE3_HDMI_OFFSET0);
3590 			rdev->irq.stat_regs.r600.hdmi1_status = RREG32(AFMT_STATUS + DCE3_HDMI_OFFSET1);
3591 		} else {
3592 			rdev->irq.stat_regs.r600.hdmi0_status = RREG32(HDMI0_STATUS);
3593 			rdev->irq.stat_regs.r600.hdmi1_status = RREG32(DCE3_HDMI1_STATUS);
3594 		}
3595 	} else {
3596 		rdev->irq.stat_regs.r600.disp_int = RREG32(DISP_INTERRUPT_STATUS);
3597 		rdev->irq.stat_regs.r600.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
3598 		rdev->irq.stat_regs.r600.disp_int_cont2 = 0;
3599 		rdev->irq.stat_regs.r600.hdmi0_status = RREG32(HDMI0_STATUS);
3600 		rdev->irq.stat_regs.r600.hdmi1_status = RREG32(HDMI1_STATUS);
3601 	}
3602 	rdev->irq.stat_regs.r600.d1grph_int = RREG32(D1GRPH_INTERRUPT_STATUS);
3603 	rdev->irq.stat_regs.r600.d2grph_int = RREG32(D2GRPH_INTERRUPT_STATUS);
3604 
3605 	if (rdev->irq.stat_regs.r600.d1grph_int & DxGRPH_PFLIP_INT_OCCURRED)
3606 		WREG32(D1GRPH_INTERRUPT_STATUS, DxGRPH_PFLIP_INT_CLEAR);
3607 	if (rdev->irq.stat_regs.r600.d2grph_int & DxGRPH_PFLIP_INT_OCCURRED)
3608 		WREG32(D2GRPH_INTERRUPT_STATUS, DxGRPH_PFLIP_INT_CLEAR);
3609 	if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VBLANK_INTERRUPT)
3610 		WREG32(D1MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
3611 	if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VLINE_INTERRUPT)
3612 		WREG32(D1MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
3613 	if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VBLANK_INTERRUPT)
3614 		WREG32(D2MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
3615 	if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VLINE_INTERRUPT)
3616 		WREG32(D2MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
3617 	if (rdev->irq.stat_regs.r600.disp_int & DC_HPD1_INTERRUPT) {
3618 		if (ASIC_IS_DCE3(rdev)) {
3619 			tmp = RREG32(DC_HPD1_INT_CONTROL);
3620 			tmp |= DC_HPDx_INT_ACK;
3621 			WREG32(DC_HPD1_INT_CONTROL, tmp);
3622 		} else {
3623 			tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL);
3624 			tmp |= DC_HPDx_INT_ACK;
3625 			WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
3626 		}
3627 	}
3628 	if (rdev->irq.stat_regs.r600.disp_int & DC_HPD2_INTERRUPT) {
3629 		if (ASIC_IS_DCE3(rdev)) {
3630 			tmp = RREG32(DC_HPD2_INT_CONTROL);
3631 			tmp |= DC_HPDx_INT_ACK;
3632 			WREG32(DC_HPD2_INT_CONTROL, tmp);
3633 		} else {
3634 			tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL);
3635 			tmp |= DC_HPDx_INT_ACK;
3636 			WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
3637 		}
3638 	}
3639 	if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD3_INTERRUPT) {
3640 		if (ASIC_IS_DCE3(rdev)) {
3641 			tmp = RREG32(DC_HPD3_INT_CONTROL);
3642 			tmp |= DC_HPDx_INT_ACK;
3643 			WREG32(DC_HPD3_INT_CONTROL, tmp);
3644 		} else {
3645 			tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL);
3646 			tmp |= DC_HPDx_INT_ACK;
3647 			WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
3648 		}
3649 	}
3650 	if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD4_INTERRUPT) {
3651 		tmp = RREG32(DC_HPD4_INT_CONTROL);
3652 		tmp |= DC_HPDx_INT_ACK;
3653 		WREG32(DC_HPD4_INT_CONTROL, tmp);
3654 	}
3655 	if (ASIC_IS_DCE32(rdev)) {
3656 		if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD5_INTERRUPT) {
3657 			tmp = RREG32(DC_HPD5_INT_CONTROL);
3658 			tmp |= DC_HPDx_INT_ACK;
3659 			WREG32(DC_HPD5_INT_CONTROL, tmp);
3660 		}
3661 		if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD6_INTERRUPT) {
3662 			tmp = RREG32(DC_HPD5_INT_CONTROL);
3663 			tmp |= DC_HPDx_INT_ACK;
3664 			WREG32(DC_HPD6_INT_CONTROL, tmp);
3665 		}
3666 		if (rdev->irq.stat_regs.r600.hdmi0_status & AFMT_AZ_FORMAT_WTRIG) {
3667 			tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET0);
3668 			tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
3669 			WREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET0, tmp);
3670 		}
3671 		if (rdev->irq.stat_regs.r600.hdmi1_status & AFMT_AZ_FORMAT_WTRIG) {
3672 			tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET1);
3673 			tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
3674 			WREG32(AFMT_AUDIO_PACKET_CONTROL + DCE3_HDMI_OFFSET1, tmp);
3675 		}
3676 	} else {
3677 		if (rdev->irq.stat_regs.r600.hdmi0_status & HDMI0_AZ_FORMAT_WTRIG) {
3678 			tmp = RREG32(HDMI0_AUDIO_PACKET_CONTROL);
3679 			tmp |= HDMI0_AZ_FORMAT_WTRIG_ACK;
3680 			WREG32(HDMI0_AUDIO_PACKET_CONTROL, tmp);
3681 		}
3682 		if (rdev->irq.stat_regs.r600.hdmi1_status & HDMI0_AZ_FORMAT_WTRIG) {
3683 			if (ASIC_IS_DCE3(rdev)) {
3684 				tmp = RREG32(DCE3_HDMI1_AUDIO_PACKET_CONTROL);
3685 				tmp |= HDMI0_AZ_FORMAT_WTRIG_ACK;
3686 				WREG32(DCE3_HDMI1_AUDIO_PACKET_CONTROL, tmp);
3687 			} else {
3688 				tmp = RREG32(HDMI1_AUDIO_PACKET_CONTROL);
3689 				tmp |= HDMI0_AZ_FORMAT_WTRIG_ACK;
3690 				WREG32(HDMI1_AUDIO_PACKET_CONTROL, tmp);
3691 			}
3692 		}
3693 	}
3694 }
3695 
3696 void r600_irq_disable(struct radeon_device *rdev)
3697 {
3698 	r600_disable_interrupts(rdev);
3699 	/* Wait and acknowledge irq */
3700 	mdelay(1);
3701 	r600_irq_ack(rdev);
3702 	r600_disable_interrupt_state(rdev);
3703 }
3704 
3705 static u32 r600_get_ih_wptr(struct radeon_device *rdev)
3706 {
3707 	u32 wptr, tmp;
3708 
3709 	if (rdev->wb.enabled)
3710 		wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
3711 	else
3712 		wptr = RREG32(IH_RB_WPTR);
3713 
3714 	if (wptr & RB_OVERFLOW) {
3715 		/* When a ring buffer overflow happen start parsing interrupt
3716 		 * from the last not overwritten vector (wptr + 16). Hopefully
3717 		 * this should allow us to catchup.
3718 		 */
3719 		dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
3720 			wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
3721 		rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
3722 		tmp = RREG32(IH_RB_CNTL);
3723 		tmp |= IH_WPTR_OVERFLOW_CLEAR;
3724 		WREG32(IH_RB_CNTL, tmp);
3725 	}
3726 	return (wptr & rdev->ih.ptr_mask);
3727 }
3728 
3729 /*        r600 IV Ring
3730  * Each IV ring entry is 128 bits:
3731  * [7:0]    - interrupt source id
3732  * [31:8]   - reserved
3733  * [59:32]  - interrupt source data
3734  * [127:60]  - reserved
3735  *
3736  * The basic interrupt vector entries
3737  * are decoded as follows:
3738  * src_id  src_data  description
3739  *      1         0  D1 Vblank
3740  *      1         1  D1 Vline
3741  *      5         0  D2 Vblank
3742  *      5         1  D2 Vline
3743  *     19         0  FP Hot plug detection A
3744  *     19         1  FP Hot plug detection B
3745  *     19         2  DAC A auto-detection
3746  *     19         3  DAC B auto-detection
3747  *     21         4  HDMI block A
3748  *     21         5  HDMI block B
3749  *    176         -  CP_INT RB
3750  *    177         -  CP_INT IB1
3751  *    178         -  CP_INT IB2
3752  *    181         -  EOP Interrupt
3753  *    233         -  GUI Idle
3754  *
3755  * Note, these are based on r600 and may need to be
3756  * adjusted or added to on newer asics
3757  */
3758 
3759 int r600_irq_process(struct radeon_device *rdev)
3760 {
3761 	u32 wptr;
3762 	u32 rptr;
3763 	u32 src_id, src_data;
3764 	u32 ring_index;
3765 	bool queue_hotplug = false;
3766 	bool queue_hdmi = false;
3767 
3768 	if (!rdev->ih.enabled || rdev->shutdown)
3769 		return IRQ_NONE;
3770 
3771 	/* No MSIs, need a dummy read to flush PCI DMAs */
3772 	if (!rdev->msi_enabled)
3773 		RREG32(IH_RB_WPTR);
3774 
3775 	wptr = r600_get_ih_wptr(rdev);
3776 
3777 restart_ih:
3778 	/* is somebody else already processing irqs? */
3779 	if (atomic_xchg(&rdev->ih.lock, 1))
3780 		return IRQ_NONE;
3781 
3782 	rptr = rdev->ih.rptr;
3783 	DRM_DEBUG("r600_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
3784 
3785 	/* Order reading of wptr vs. reading of IH ring data */
3786 	rmb();
3787 
3788 	/* display interrupts */
3789 	r600_irq_ack(rdev);
3790 
3791 	while (rptr != wptr) {
3792 		/* wptr/rptr are in bytes! */
3793 		ring_index = rptr / 4;
3794 		src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
3795 		src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
3796 
3797 		switch (src_id) {
3798 		case 1: /* D1 vblank/vline */
3799 			switch (src_data) {
3800 			case 0: /* D1 vblank */
3801 				if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VBLANK_INTERRUPT) {
3802 					if (rdev->irq.crtc_vblank_int[0]) {
3803 						drm_handle_vblank(rdev->ddev, 0);
3804 						rdev->pm.vblank_sync = true;
3805 						wake_up(&rdev->irq.vblank_queue);
3806 					}
3807 					if (atomic_read(&rdev->irq.pflip[0]))
3808 						radeon_crtc_handle_flip(rdev, 0);
3809 					rdev->irq.stat_regs.r600.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
3810 					DRM_DEBUG("IH: D1 vblank\n");
3811 				}
3812 				break;
3813 			case 1: /* D1 vline */
3814 				if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VLINE_INTERRUPT) {
3815 					rdev->irq.stat_regs.r600.disp_int &= ~LB_D1_VLINE_INTERRUPT;
3816 					DRM_DEBUG("IH: D1 vline\n");
3817 				}
3818 				break;
3819 			default:
3820 				DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3821 				break;
3822 			}
3823 			break;
3824 		case 5: /* D2 vblank/vline */
3825 			switch (src_data) {
3826 			case 0: /* D2 vblank */
3827 				if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VBLANK_INTERRUPT) {
3828 					if (rdev->irq.crtc_vblank_int[1]) {
3829 						drm_handle_vblank(rdev->ddev, 1);
3830 						rdev->pm.vblank_sync = true;
3831 						wake_up(&rdev->irq.vblank_queue);
3832 					}
3833 					if (atomic_read(&rdev->irq.pflip[1]))
3834 						radeon_crtc_handle_flip(rdev, 1);
3835 					rdev->irq.stat_regs.r600.disp_int &= ~LB_D2_VBLANK_INTERRUPT;
3836 					DRM_DEBUG("IH: D2 vblank\n");
3837 				}
3838 				break;
3839 			case 1: /* D1 vline */
3840 				if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VLINE_INTERRUPT) {
3841 					rdev->irq.stat_regs.r600.disp_int &= ~LB_D2_VLINE_INTERRUPT;
3842 					DRM_DEBUG("IH: D2 vline\n");
3843 				}
3844 				break;
3845 			default:
3846 				DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3847 				break;
3848 			}
3849 			break;
3850 		case 19: /* HPD/DAC hotplug */
3851 			switch (src_data) {
3852 			case 0:
3853 				if (rdev->irq.stat_regs.r600.disp_int & DC_HPD1_INTERRUPT) {
3854 					rdev->irq.stat_regs.r600.disp_int &= ~DC_HPD1_INTERRUPT;
3855 					queue_hotplug = true;
3856 					DRM_DEBUG("IH: HPD1\n");
3857 				}
3858 				break;
3859 			case 1:
3860 				if (rdev->irq.stat_regs.r600.disp_int & DC_HPD2_INTERRUPT) {
3861 					rdev->irq.stat_regs.r600.disp_int &= ~DC_HPD2_INTERRUPT;
3862 					queue_hotplug = true;
3863 					DRM_DEBUG("IH: HPD2\n");
3864 				}
3865 				break;
3866 			case 4:
3867 				if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD3_INTERRUPT) {
3868 					rdev->irq.stat_regs.r600.disp_int_cont &= ~DC_HPD3_INTERRUPT;
3869 					queue_hotplug = true;
3870 					DRM_DEBUG("IH: HPD3\n");
3871 				}
3872 				break;
3873 			case 5:
3874 				if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD4_INTERRUPT) {
3875 					rdev->irq.stat_regs.r600.disp_int_cont &= ~DC_HPD4_INTERRUPT;
3876 					queue_hotplug = true;
3877 					DRM_DEBUG("IH: HPD4\n");
3878 				}
3879 				break;
3880 			case 10:
3881 				if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD5_INTERRUPT) {
3882 					rdev->irq.stat_regs.r600.disp_int_cont2 &= ~DC_HPD5_INTERRUPT;
3883 					queue_hotplug = true;
3884 					DRM_DEBUG("IH: HPD5\n");
3885 				}
3886 				break;
3887 			case 12:
3888 				if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD6_INTERRUPT) {
3889 					rdev->irq.stat_regs.r600.disp_int_cont2 &= ~DC_HPD6_INTERRUPT;
3890 					queue_hotplug = true;
3891 					DRM_DEBUG("IH: HPD6\n");
3892 				}
3893 				break;
3894 			default:
3895 				DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3896 				break;
3897 			}
3898 			break;
3899 		case 21: /* hdmi */
3900 			switch (src_data) {
3901 			case 4:
3902 				if (rdev->irq.stat_regs.r600.hdmi0_status & HDMI0_AZ_FORMAT_WTRIG) {
3903 					rdev->irq.stat_regs.r600.hdmi0_status &= ~HDMI0_AZ_FORMAT_WTRIG;
3904 					queue_hdmi = true;
3905 					DRM_DEBUG("IH: HDMI0\n");
3906 				}
3907 				break;
3908 			case 5:
3909 				if (rdev->irq.stat_regs.r600.hdmi1_status & HDMI0_AZ_FORMAT_WTRIG) {
3910 					rdev->irq.stat_regs.r600.hdmi1_status &= ~HDMI0_AZ_FORMAT_WTRIG;
3911 					queue_hdmi = true;
3912 					DRM_DEBUG("IH: HDMI1\n");
3913 				}
3914 				break;
3915 			default:
3916 				DRM_ERROR("Unhandled interrupt: %d %d\n", src_id, src_data);
3917 				break;
3918 			}
3919 			break;
3920 		case 176: /* CP_INT in ring buffer */
3921 		case 177: /* CP_INT in IB1 */
3922 		case 178: /* CP_INT in IB2 */
3923 			DRM_DEBUG("IH: CP int: 0x%08x\n", src_data);
3924 			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
3925 			break;
3926 		case 181: /* CP EOP event */
3927 			DRM_DEBUG("IH: CP EOP\n");
3928 			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
3929 			break;
3930 		case 224: /* DMA trap event */
3931 			DRM_DEBUG("IH: DMA trap\n");
3932 			radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
3933 			break;
3934 		case 233: /* GUI IDLE */
3935 			DRM_DEBUG("IH: GUI idle\n");
3936 			break;
3937 		default:
3938 			DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3939 			break;
3940 		}
3941 
3942 		/* wptr/rptr are in bytes! */
3943 		rptr += 16;
3944 		rptr &= rdev->ih.ptr_mask;
3945 	}
3946 	if (queue_hotplug)
3947 		schedule_work(&rdev->hotplug_work);
3948 	if (queue_hdmi)
3949 		schedule_work(&rdev->audio_work);
3950 	rdev->ih.rptr = rptr;
3951 	WREG32(IH_RB_RPTR, rdev->ih.rptr);
3952 	atomic_set(&rdev->ih.lock, 0);
3953 
3954 	/* make sure wptr hasn't changed while processing */
3955 	wptr = r600_get_ih_wptr(rdev);
3956 	if (wptr != rptr)
3957 		goto restart_ih;
3958 
3959 	return IRQ_HANDLED;
3960 }
3961 
3962 /*
3963  * Debugfs info
3964  */
3965 #if defined(CONFIG_DEBUG_FS)
3966 
3967 static int r600_debugfs_mc_info(struct seq_file *m, void *data)
3968 {
3969 	struct drm_info_node *node = (struct drm_info_node *) m->private;
3970 	struct drm_device *dev = node->minor->dev;
3971 	struct radeon_device *rdev = dev->dev_private;
3972 
3973 	DREG32_SYS(m, rdev, R_000E50_SRBM_STATUS);
3974 	DREG32_SYS(m, rdev, VM_L2_STATUS);
3975 	return 0;
3976 }
3977 
3978 static struct drm_info_list r600_mc_info_list[] = {
3979 	{"r600_mc_info", r600_debugfs_mc_info, 0, NULL},
3980 };
3981 #endif
3982 
3983 int r600_debugfs_mc_info_init(struct radeon_device *rdev)
3984 {
3985 #if defined(CONFIG_DEBUG_FS)
3986 	return radeon_debugfs_add_files(rdev, r600_mc_info_list, ARRAY_SIZE(r600_mc_info_list));
3987 #else
3988 	return 0;
3989 #endif
3990 }
3991 
3992 /**
3993  * r600_ioctl_wait_idle - flush host path cache on wait idle ioctl
3994  * rdev: radeon device structure
3995  * bo: buffer object struct which userspace is waiting for idle
3996  *
3997  * Some R6XX/R7XX doesn't seems to take into account HDP flush performed
3998  * through ring buffer, this leads to corruption in rendering, see
3999  * http://bugzilla.kernel.org/show_bug.cgi?id=15186 to avoid this we
4000  * directly perform HDP flush by writing register through MMIO.
4001  */
4002 void r600_ioctl_wait_idle(struct radeon_device *rdev, struct radeon_bo *bo)
4003 {
4004 	/* r7xx hw bug.  write to HDP_DEBUG1 followed by fb read
4005 	 * rather than write to HDP_REG_COHERENCY_FLUSH_CNTL.
4006 	 * This seems to cause problems on some AGP cards. Just use the old
4007 	 * method for them.
4008 	 */
4009 	if ((rdev->family >= CHIP_RV770) && (rdev->family <= CHIP_RV740) &&
4010 	    rdev->vram_scratch.ptr && !(rdev->flags & RADEON_IS_AGP)) {
4011 		void __iomem *ptr = (void *)rdev->vram_scratch.ptr;
4012 		u32 tmp;
4013 
4014 		WREG32(HDP_DEBUG1, 0);
4015 		tmp = readl((void __iomem *)ptr);
4016 	} else
4017 		WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
4018 }
4019 
4020 void r600_set_pcie_lanes(struct radeon_device *rdev, int lanes)
4021 {
4022 	u32 link_width_cntl, mask, target_reg;
4023 
4024 	if (rdev->flags & RADEON_IS_IGP)
4025 		return;
4026 
4027 	if (!(rdev->flags & RADEON_IS_PCIE))
4028 		return;
4029 
4030 	/* x2 cards have a special sequence */
4031 	if (ASIC_IS_X2(rdev))
4032 		return;
4033 
4034 	/* FIXME wait for idle */
4035 
4036 	switch (lanes) {
4037 	case 0:
4038 		mask = RADEON_PCIE_LC_LINK_WIDTH_X0;
4039 		break;
4040 	case 1:
4041 		mask = RADEON_PCIE_LC_LINK_WIDTH_X1;
4042 		break;
4043 	case 2:
4044 		mask = RADEON_PCIE_LC_LINK_WIDTH_X2;
4045 		break;
4046 	case 4:
4047 		mask = RADEON_PCIE_LC_LINK_WIDTH_X4;
4048 		break;
4049 	case 8:
4050 		mask = RADEON_PCIE_LC_LINK_WIDTH_X8;
4051 		break;
4052 	case 12:
4053 		mask = RADEON_PCIE_LC_LINK_WIDTH_X12;
4054 		break;
4055 	case 16:
4056 	default:
4057 		mask = RADEON_PCIE_LC_LINK_WIDTH_X16;
4058 		break;
4059 	}
4060 
4061 	link_width_cntl = RREG32_PCIE_P(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
4062 
4063 	if ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) ==
4064 	    (mask << RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT))
4065 		return;
4066 
4067 	if (link_width_cntl & R600_PCIE_LC_UPCONFIGURE_DIS)
4068 		return;
4069 
4070 	link_width_cntl &= ~(RADEON_PCIE_LC_LINK_WIDTH_MASK |
4071 			     RADEON_PCIE_LC_RECONFIG_NOW |
4072 			     R600_PCIE_LC_RENEGOTIATE_EN |
4073 			     R600_PCIE_LC_RECONFIG_ARC_MISSING_ESCAPE);
4074 	link_width_cntl |= mask;
4075 
4076 	WREG32_PCIE_P(RADEON_PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
4077 
4078         /* some northbridges can renegotiate the link rather than requiring
4079          * a complete re-config.
4080          * e.g., AMD 780/790 northbridges (pci ids: 0x5956, 0x5957, 0x5958, etc.)
4081          */
4082         if (link_width_cntl & R600_PCIE_LC_RENEGOTIATION_SUPPORT)
4083 		link_width_cntl |= R600_PCIE_LC_RENEGOTIATE_EN | R600_PCIE_LC_UPCONFIGURE_SUPPORT;
4084         else
4085 		link_width_cntl |= R600_PCIE_LC_RECONFIG_ARC_MISSING_ESCAPE;
4086 
4087 	WREG32_PCIE_P(RADEON_PCIE_LC_LINK_WIDTH_CNTL, (link_width_cntl |
4088 						       RADEON_PCIE_LC_RECONFIG_NOW));
4089 
4090         if (rdev->family >= CHIP_RV770)
4091 		target_reg = R700_TARGET_AND_CURRENT_PROFILE_INDEX;
4092         else
4093 		target_reg = R600_TARGET_AND_CURRENT_PROFILE_INDEX;
4094 
4095         /* wait for lane set to complete */
4096         link_width_cntl = RREG32(target_reg);
4097         while (link_width_cntl == 0xffffffff)
4098 		link_width_cntl = RREG32(target_reg);
4099 
4100 }
4101 
4102 int r600_get_pcie_lanes(struct radeon_device *rdev)
4103 {
4104 	u32 link_width_cntl;
4105 
4106 	if (rdev->flags & RADEON_IS_IGP)
4107 		return 0;
4108 
4109 	if (!(rdev->flags & RADEON_IS_PCIE))
4110 		return 0;
4111 
4112 	/* x2 cards have a special sequence */
4113 	if (ASIC_IS_X2(rdev))
4114 		return 0;
4115 
4116 	/* FIXME wait for idle */
4117 
4118 	link_width_cntl = RREG32_PCIE_P(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
4119 
4120 	switch ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) >> RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT) {
4121 	case RADEON_PCIE_LC_LINK_WIDTH_X0:
4122 		return 0;
4123 	case RADEON_PCIE_LC_LINK_WIDTH_X1:
4124 		return 1;
4125 	case RADEON_PCIE_LC_LINK_WIDTH_X2:
4126 		return 2;
4127 	case RADEON_PCIE_LC_LINK_WIDTH_X4:
4128 		return 4;
4129 	case RADEON_PCIE_LC_LINK_WIDTH_X8:
4130 		return 8;
4131 	case RADEON_PCIE_LC_LINK_WIDTH_X16:
4132 	default:
4133 		return 16;
4134 	}
4135 }
4136 
4137 static void r600_pcie_gen2_enable(struct radeon_device *rdev)
4138 {
4139 	u32 link_width_cntl, lanes, speed_cntl, training_cntl, tmp;
4140 	u16 link_cntl2;
4141 	u32 mask;
4142 	int ret;
4143 
4144 	if (radeon_pcie_gen2 == 0)
4145 		return;
4146 
4147 	if (rdev->flags & RADEON_IS_IGP)
4148 		return;
4149 
4150 	if (!(rdev->flags & RADEON_IS_PCIE))
4151 		return;
4152 
4153 	/* x2 cards have a special sequence */
4154 	if (ASIC_IS_X2(rdev))
4155 		return;
4156 
4157 	/* only RV6xx+ chips are supported */
4158 	if (rdev->family <= CHIP_R600)
4159 		return;
4160 
4161 	ret = drm_pcie_get_speed_cap_mask(rdev->ddev, &mask);
4162 	if (ret != 0)
4163 		return;
4164 
4165 	if (!(mask & DRM_PCIE_SPEED_50))
4166 		return;
4167 
4168 	speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
4169 	if (speed_cntl & LC_CURRENT_DATA_RATE) {
4170 		DRM_INFO("PCIE gen 2 link speeds already enabled\n");
4171 		return;
4172 	}
4173 
4174 	DRM_INFO("enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0\n");
4175 
4176 	/* 55 nm r6xx asics */
4177 	if ((rdev->family == CHIP_RV670) ||
4178 	    (rdev->family == CHIP_RV620) ||
4179 	    (rdev->family == CHIP_RV635)) {
4180 		/* advertise upconfig capability */
4181 		link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL);
4182 		link_width_cntl &= ~LC_UPCONFIGURE_DIS;
4183 		WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
4184 		link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL);
4185 		if (link_width_cntl & LC_RENEGOTIATION_SUPPORT) {
4186 			lanes = (link_width_cntl & LC_LINK_WIDTH_RD_MASK) >> LC_LINK_WIDTH_RD_SHIFT;
4187 			link_width_cntl &= ~(LC_LINK_WIDTH_MASK |
4188 					     LC_RECONFIG_ARC_MISSING_ESCAPE);
4189 			link_width_cntl |= lanes | LC_RECONFIG_NOW | LC_RENEGOTIATE_EN;
4190 			WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
4191 		} else {
4192 			link_width_cntl |= LC_UPCONFIGURE_DIS;
4193 			WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
4194 		}
4195 	}
4196 
4197 	speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
4198 	if ((speed_cntl & LC_OTHER_SIDE_EVER_SENT_GEN2) &&
4199 	    (speed_cntl & LC_OTHER_SIDE_SUPPORTS_GEN2)) {
4200 
4201 		/* 55 nm r6xx asics */
4202 		if ((rdev->family == CHIP_RV670) ||
4203 		    (rdev->family == CHIP_RV620) ||
4204 		    (rdev->family == CHIP_RV635)) {
4205 			WREG32(MM_CFGREGS_CNTL, 0x8);
4206 			link_cntl2 = RREG32(0x4088);
4207 			WREG32(MM_CFGREGS_CNTL, 0);
4208 			/* not supported yet */
4209 			if (link_cntl2 & SELECTABLE_DEEMPHASIS)
4210 				return;
4211 		}
4212 
4213 		speed_cntl &= ~LC_SPEED_CHANGE_ATTEMPTS_ALLOWED_MASK;
4214 		speed_cntl |= (0x3 << LC_SPEED_CHANGE_ATTEMPTS_ALLOWED_SHIFT);
4215 		speed_cntl &= ~LC_VOLTAGE_TIMER_SEL_MASK;
4216 		speed_cntl &= ~LC_FORCE_DIS_HW_SPEED_CHANGE;
4217 		speed_cntl |= LC_FORCE_EN_HW_SPEED_CHANGE;
4218 		WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl);
4219 
4220 		tmp = RREG32(0x541c);
4221 		WREG32(0x541c, tmp | 0x8);
4222 		WREG32(MM_CFGREGS_CNTL, MM_WR_TO_CFG_EN);
4223 		link_cntl2 = RREG16(0x4088);
4224 		link_cntl2 &= ~TARGET_LINK_SPEED_MASK;
4225 		link_cntl2 |= 0x2;
4226 		WREG16(0x4088, link_cntl2);
4227 		WREG32(MM_CFGREGS_CNTL, 0);
4228 
4229 		if ((rdev->family == CHIP_RV670) ||
4230 		    (rdev->family == CHIP_RV620) ||
4231 		    (rdev->family == CHIP_RV635)) {
4232 			training_cntl = RREG32_PCIE_P(PCIE_LC_TRAINING_CNTL);
4233 			training_cntl &= ~LC_POINT_7_PLUS_EN;
4234 			WREG32_PCIE_P(PCIE_LC_TRAINING_CNTL, training_cntl);
4235 		} else {
4236 			speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
4237 			speed_cntl &= ~LC_TARGET_LINK_SPEED_OVERRIDE_EN;
4238 			WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl);
4239 		}
4240 
4241 		speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
4242 		speed_cntl |= LC_GEN2_EN_STRAP;
4243 		WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl);
4244 
4245 	} else {
4246 		link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL);
4247 		/* XXX: only disable it if gen1 bridge vendor == 0x111d or 0x1106 */
4248 		if (1)
4249 			link_width_cntl |= LC_UPCONFIGURE_DIS;
4250 		else
4251 			link_width_cntl &= ~LC_UPCONFIGURE_DIS;
4252 		WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
4253 	}
4254 }
4255 
4256 /**
4257  * r600_get_gpu_clock - return GPU clock counter snapshot
4258  *
4259  * @rdev: radeon_device pointer
4260  *
4261  * Fetches a GPU clock counter snapshot (R6xx-cayman).
4262  * Returns the 64 bit clock counter snapshot.
4263  */
4264 uint64_t r600_get_gpu_clock(struct radeon_device *rdev)
4265 {
4266 	uint64_t clock;
4267 
4268 	mutex_lock(&rdev->gpu_clock_mutex);
4269 	WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4270 	clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) |
4271 	        ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4272 	mutex_unlock(&rdev->gpu_clock_mutex);
4273 	return clock;
4274 }
4275