xref: /openbmc/linux/drivers/gpu/drm/radeon/r100.c (revision 80ecbd24)
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/seq_file.h>
29 #include <linux/slab.h>
30 #include <drm/drmP.h>
31 #include <drm/radeon_drm.h>
32 #include "radeon_reg.h"
33 #include "radeon.h"
34 #include "radeon_asic.h"
35 #include "r100d.h"
36 #include "rs100d.h"
37 #include "rv200d.h"
38 #include "rv250d.h"
39 #include "atom.h"
40 
41 #include <linux/firmware.h>
42 #include <linux/module.h>
43 
44 #include "r100_reg_safe.h"
45 #include "rn50_reg_safe.h"
46 
47 /* Firmware Names */
48 #define FIRMWARE_R100		"radeon/R100_cp.bin"
49 #define FIRMWARE_R200		"radeon/R200_cp.bin"
50 #define FIRMWARE_R300		"radeon/R300_cp.bin"
51 #define FIRMWARE_R420		"radeon/R420_cp.bin"
52 #define FIRMWARE_RS690		"radeon/RS690_cp.bin"
53 #define FIRMWARE_RS600		"radeon/RS600_cp.bin"
54 #define FIRMWARE_R520		"radeon/R520_cp.bin"
55 
56 MODULE_FIRMWARE(FIRMWARE_R100);
57 MODULE_FIRMWARE(FIRMWARE_R200);
58 MODULE_FIRMWARE(FIRMWARE_R300);
59 MODULE_FIRMWARE(FIRMWARE_R420);
60 MODULE_FIRMWARE(FIRMWARE_RS690);
61 MODULE_FIRMWARE(FIRMWARE_RS600);
62 MODULE_FIRMWARE(FIRMWARE_R520);
63 
64 #include "r100_track.h"
65 
66 /* This files gather functions specifics to:
67  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
68  * and others in some cases.
69  */
70 
71 static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
72 {
73 	if (crtc == 0) {
74 		if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
75 			return true;
76 		else
77 			return false;
78 	} else {
79 		if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
80 			return true;
81 		else
82 			return false;
83 	}
84 }
85 
86 static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
87 {
88 	u32 vline1, vline2;
89 
90 	if (crtc == 0) {
91 		vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
92 		vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
93 	} else {
94 		vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
95 		vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
96 	}
97 	if (vline1 != vline2)
98 		return true;
99 	else
100 		return false;
101 }
102 
103 /**
104  * r100_wait_for_vblank - vblank wait asic callback.
105  *
106  * @rdev: radeon_device pointer
107  * @crtc: crtc to wait for vblank on
108  *
109  * Wait for vblank on the requested crtc (r1xx-r4xx).
110  */
111 void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
112 {
113 	unsigned i = 0;
114 
115 	if (crtc >= rdev->num_crtc)
116 		return;
117 
118 	if (crtc == 0) {
119 		if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
120 			return;
121 	} else {
122 		if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
123 			return;
124 	}
125 
126 	/* depending on when we hit vblank, we may be close to active; if so,
127 	 * wait for another frame.
128 	 */
129 	while (r100_is_in_vblank(rdev, crtc)) {
130 		if (i++ % 100 == 0) {
131 			if (!r100_is_counter_moving(rdev, crtc))
132 				break;
133 		}
134 	}
135 
136 	while (!r100_is_in_vblank(rdev, crtc)) {
137 		if (i++ % 100 == 0) {
138 			if (!r100_is_counter_moving(rdev, crtc))
139 				break;
140 		}
141 	}
142 }
143 
144 /**
145  * r100_pre_page_flip - pre-pageflip callback.
146  *
147  * @rdev: radeon_device pointer
148  * @crtc: crtc to prepare for pageflip on
149  *
150  * Pre-pageflip callback (r1xx-r4xx).
151  * Enables the pageflip irq (vblank irq).
152  */
153 void r100_pre_page_flip(struct radeon_device *rdev, int crtc)
154 {
155 	/* enable the pflip int */
156 	radeon_irq_kms_pflip_irq_get(rdev, crtc);
157 }
158 
159 /**
160  * r100_post_page_flip - pos-pageflip callback.
161  *
162  * @rdev: radeon_device pointer
163  * @crtc: crtc to cleanup pageflip on
164  *
165  * Post-pageflip callback (r1xx-r4xx).
166  * Disables the pageflip irq (vblank irq).
167  */
168 void r100_post_page_flip(struct radeon_device *rdev, int crtc)
169 {
170 	/* disable the pflip int */
171 	radeon_irq_kms_pflip_irq_put(rdev, crtc);
172 }
173 
174 /**
175  * r100_page_flip - pageflip callback.
176  *
177  * @rdev: radeon_device pointer
178  * @crtc_id: crtc to cleanup pageflip on
179  * @crtc_base: new address of the crtc (GPU MC address)
180  *
181  * Does the actual pageflip (r1xx-r4xx).
182  * During vblank we take the crtc lock and wait for the update_pending
183  * bit to go high, when it does, we release the lock, and allow the
184  * double buffered update to take place.
185  * Returns the current update pending status.
186  */
187 u32 r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
188 {
189 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
190 	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
191 	int i;
192 
193 	/* Lock the graphics update lock */
194 	/* update the scanout addresses */
195 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
196 
197 	/* Wait for update_pending to go high. */
198 	for (i = 0; i < rdev->usec_timeout; i++) {
199 		if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
200 			break;
201 		udelay(1);
202 	}
203 	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
204 
205 	/* Unlock the lock, so double-buffering can take place inside vblank */
206 	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
207 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
208 
209 	/* Return current update_pending status: */
210 	return RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET;
211 }
212 
213 /**
214  * r100_pm_get_dynpm_state - look up dynpm power state callback.
215  *
216  * @rdev: radeon_device pointer
217  *
218  * Look up the optimal power state based on the
219  * current state of the GPU (r1xx-r5xx).
220  * Used for dynpm only.
221  */
222 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
223 {
224 	int i;
225 	rdev->pm.dynpm_can_upclock = true;
226 	rdev->pm.dynpm_can_downclock = true;
227 
228 	switch (rdev->pm.dynpm_planned_action) {
229 	case DYNPM_ACTION_MINIMUM:
230 		rdev->pm.requested_power_state_index = 0;
231 		rdev->pm.dynpm_can_downclock = false;
232 		break;
233 	case DYNPM_ACTION_DOWNCLOCK:
234 		if (rdev->pm.current_power_state_index == 0) {
235 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
236 			rdev->pm.dynpm_can_downclock = false;
237 		} else {
238 			if (rdev->pm.active_crtc_count > 1) {
239 				for (i = 0; i < rdev->pm.num_power_states; i++) {
240 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
241 						continue;
242 					else if (i >= rdev->pm.current_power_state_index) {
243 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
244 						break;
245 					} else {
246 						rdev->pm.requested_power_state_index = i;
247 						break;
248 					}
249 				}
250 			} else
251 				rdev->pm.requested_power_state_index =
252 					rdev->pm.current_power_state_index - 1;
253 		}
254 		/* don't use the power state if crtcs are active and no display flag is set */
255 		if ((rdev->pm.active_crtc_count > 0) &&
256 		    (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
257 		     RADEON_PM_MODE_NO_DISPLAY)) {
258 			rdev->pm.requested_power_state_index++;
259 		}
260 		break;
261 	case DYNPM_ACTION_UPCLOCK:
262 		if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
263 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
264 			rdev->pm.dynpm_can_upclock = false;
265 		} else {
266 			if (rdev->pm.active_crtc_count > 1) {
267 				for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
268 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
269 						continue;
270 					else if (i <= rdev->pm.current_power_state_index) {
271 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
272 						break;
273 					} else {
274 						rdev->pm.requested_power_state_index = i;
275 						break;
276 					}
277 				}
278 			} else
279 				rdev->pm.requested_power_state_index =
280 					rdev->pm.current_power_state_index + 1;
281 		}
282 		break;
283 	case DYNPM_ACTION_DEFAULT:
284 		rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
285 		rdev->pm.dynpm_can_upclock = false;
286 		break;
287 	case DYNPM_ACTION_NONE:
288 	default:
289 		DRM_ERROR("Requested mode for not defined action\n");
290 		return;
291 	}
292 	/* only one clock mode per power state */
293 	rdev->pm.requested_clock_mode_index = 0;
294 
295 	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
296 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
297 		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
298 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
299 		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
300 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
301 		  pcie_lanes);
302 }
303 
304 /**
305  * r100_pm_init_profile - Initialize power profiles callback.
306  *
307  * @rdev: radeon_device pointer
308  *
309  * Initialize the power states used in profile mode
310  * (r1xx-r3xx).
311  * Used for profile mode only.
312  */
313 void r100_pm_init_profile(struct radeon_device *rdev)
314 {
315 	/* default */
316 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
317 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
318 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
319 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
320 	/* low sh */
321 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
322 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
323 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
324 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
325 	/* mid sh */
326 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
327 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
328 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
329 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
330 	/* high sh */
331 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
332 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
333 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
334 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
335 	/* low mh */
336 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
337 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
338 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
339 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
340 	/* mid mh */
341 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
342 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
343 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
344 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
345 	/* high mh */
346 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
347 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
348 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
349 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
350 }
351 
352 /**
353  * r100_pm_misc - set additional pm hw parameters callback.
354  *
355  * @rdev: radeon_device pointer
356  *
357  * Set non-clock parameters associated with a power state
358  * (voltage, pcie lanes, etc.) (r1xx-r4xx).
359  */
360 void r100_pm_misc(struct radeon_device *rdev)
361 {
362 	int requested_index = rdev->pm.requested_power_state_index;
363 	struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
364 	struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
365 	u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
366 
367 	if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
368 		if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
369 			tmp = RREG32(voltage->gpio.reg);
370 			if (voltage->active_high)
371 				tmp |= voltage->gpio.mask;
372 			else
373 				tmp &= ~(voltage->gpio.mask);
374 			WREG32(voltage->gpio.reg, tmp);
375 			if (voltage->delay)
376 				udelay(voltage->delay);
377 		} else {
378 			tmp = RREG32(voltage->gpio.reg);
379 			if (voltage->active_high)
380 				tmp &= ~voltage->gpio.mask;
381 			else
382 				tmp |= voltage->gpio.mask;
383 			WREG32(voltage->gpio.reg, tmp);
384 			if (voltage->delay)
385 				udelay(voltage->delay);
386 		}
387 	}
388 
389 	sclk_cntl = RREG32_PLL(SCLK_CNTL);
390 	sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
391 	sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
392 	sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
393 	sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
394 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
395 		sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
396 		if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
397 			sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
398 		else
399 			sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
400 		if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
401 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
402 		else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
403 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
404 	} else
405 		sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
406 
407 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
408 		sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
409 		if (voltage->delay) {
410 			sclk_more_cntl |= VOLTAGE_DROP_SYNC;
411 			switch (voltage->delay) {
412 			case 33:
413 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
414 				break;
415 			case 66:
416 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
417 				break;
418 			case 99:
419 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
420 				break;
421 			case 132:
422 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
423 				break;
424 			}
425 		} else
426 			sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
427 	} else
428 		sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
429 
430 	if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
431 		sclk_cntl &= ~FORCE_HDP;
432 	else
433 		sclk_cntl |= FORCE_HDP;
434 
435 	WREG32_PLL(SCLK_CNTL, sclk_cntl);
436 	WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
437 	WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
438 
439 	/* set pcie lanes */
440 	if ((rdev->flags & RADEON_IS_PCIE) &&
441 	    !(rdev->flags & RADEON_IS_IGP) &&
442 	    rdev->asic->pm.set_pcie_lanes &&
443 	    (ps->pcie_lanes !=
444 	     rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
445 		radeon_set_pcie_lanes(rdev,
446 				      ps->pcie_lanes);
447 		DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
448 	}
449 }
450 
451 /**
452  * r100_pm_prepare - pre-power state change callback.
453  *
454  * @rdev: radeon_device pointer
455  *
456  * Prepare for a power state change (r1xx-r4xx).
457  */
458 void r100_pm_prepare(struct radeon_device *rdev)
459 {
460 	struct drm_device *ddev = rdev->ddev;
461 	struct drm_crtc *crtc;
462 	struct radeon_crtc *radeon_crtc;
463 	u32 tmp;
464 
465 	/* disable any active CRTCs */
466 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
467 		radeon_crtc = to_radeon_crtc(crtc);
468 		if (radeon_crtc->enabled) {
469 			if (radeon_crtc->crtc_id) {
470 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
471 				tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
472 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
473 			} else {
474 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
475 				tmp |= RADEON_CRTC_DISP_REQ_EN_B;
476 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
477 			}
478 		}
479 	}
480 }
481 
482 /**
483  * r100_pm_finish - post-power state change callback.
484  *
485  * @rdev: radeon_device pointer
486  *
487  * Clean up after a power state change (r1xx-r4xx).
488  */
489 void r100_pm_finish(struct radeon_device *rdev)
490 {
491 	struct drm_device *ddev = rdev->ddev;
492 	struct drm_crtc *crtc;
493 	struct radeon_crtc *radeon_crtc;
494 	u32 tmp;
495 
496 	/* enable any active CRTCs */
497 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
498 		radeon_crtc = to_radeon_crtc(crtc);
499 		if (radeon_crtc->enabled) {
500 			if (radeon_crtc->crtc_id) {
501 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
502 				tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
503 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
504 			} else {
505 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
506 				tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
507 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
508 			}
509 		}
510 	}
511 }
512 
513 /**
514  * r100_gui_idle - gui idle callback.
515  *
516  * @rdev: radeon_device pointer
517  *
518  * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
519  * Returns true if idle, false if not.
520  */
521 bool r100_gui_idle(struct radeon_device *rdev)
522 {
523 	if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
524 		return false;
525 	else
526 		return true;
527 }
528 
529 /* hpd for digital panel detect/disconnect */
530 /**
531  * r100_hpd_sense - hpd sense callback.
532  *
533  * @rdev: radeon_device pointer
534  * @hpd: hpd (hotplug detect) pin
535  *
536  * Checks if a digital monitor is connected (r1xx-r4xx).
537  * Returns true if connected, false if not connected.
538  */
539 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
540 {
541 	bool connected = false;
542 
543 	switch (hpd) {
544 	case RADEON_HPD_1:
545 		if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
546 			connected = true;
547 		break;
548 	case RADEON_HPD_2:
549 		if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
550 			connected = true;
551 		break;
552 	default:
553 		break;
554 	}
555 	return connected;
556 }
557 
558 /**
559  * r100_hpd_set_polarity - hpd set polarity callback.
560  *
561  * @rdev: radeon_device pointer
562  * @hpd: hpd (hotplug detect) pin
563  *
564  * Set the polarity of the hpd pin (r1xx-r4xx).
565  */
566 void r100_hpd_set_polarity(struct radeon_device *rdev,
567 			   enum radeon_hpd_id hpd)
568 {
569 	u32 tmp;
570 	bool connected = r100_hpd_sense(rdev, hpd);
571 
572 	switch (hpd) {
573 	case RADEON_HPD_1:
574 		tmp = RREG32(RADEON_FP_GEN_CNTL);
575 		if (connected)
576 			tmp &= ~RADEON_FP_DETECT_INT_POL;
577 		else
578 			tmp |= RADEON_FP_DETECT_INT_POL;
579 		WREG32(RADEON_FP_GEN_CNTL, tmp);
580 		break;
581 	case RADEON_HPD_2:
582 		tmp = RREG32(RADEON_FP2_GEN_CNTL);
583 		if (connected)
584 			tmp &= ~RADEON_FP2_DETECT_INT_POL;
585 		else
586 			tmp |= RADEON_FP2_DETECT_INT_POL;
587 		WREG32(RADEON_FP2_GEN_CNTL, tmp);
588 		break;
589 	default:
590 		break;
591 	}
592 }
593 
594 /**
595  * r100_hpd_init - hpd setup callback.
596  *
597  * @rdev: radeon_device pointer
598  *
599  * Setup the hpd pins used by the card (r1xx-r4xx).
600  * Set the polarity, and enable the hpd interrupts.
601  */
602 void r100_hpd_init(struct radeon_device *rdev)
603 {
604 	struct drm_device *dev = rdev->ddev;
605 	struct drm_connector *connector;
606 	unsigned enable = 0;
607 
608 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
609 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
610 		enable |= 1 << radeon_connector->hpd.hpd;
611 		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
612 	}
613 	radeon_irq_kms_enable_hpd(rdev, enable);
614 }
615 
616 /**
617  * r100_hpd_fini - hpd tear down callback.
618  *
619  * @rdev: radeon_device pointer
620  *
621  * Tear down the hpd pins used by the card (r1xx-r4xx).
622  * Disable the hpd interrupts.
623  */
624 void r100_hpd_fini(struct radeon_device *rdev)
625 {
626 	struct drm_device *dev = rdev->ddev;
627 	struct drm_connector *connector;
628 	unsigned disable = 0;
629 
630 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
631 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
632 		disable |= 1 << radeon_connector->hpd.hpd;
633 	}
634 	radeon_irq_kms_disable_hpd(rdev, disable);
635 }
636 
637 /*
638  * PCI GART
639  */
640 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
641 {
642 	/* TODO: can we do somethings here ? */
643 	/* It seems hw only cache one entry so we should discard this
644 	 * entry otherwise if first GPU GART read hit this entry it
645 	 * could end up in wrong address. */
646 }
647 
648 int r100_pci_gart_init(struct radeon_device *rdev)
649 {
650 	int r;
651 
652 	if (rdev->gart.ptr) {
653 		WARN(1, "R100 PCI GART already initialized\n");
654 		return 0;
655 	}
656 	/* Initialize common gart structure */
657 	r = radeon_gart_init(rdev);
658 	if (r)
659 		return r;
660 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
661 	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
662 	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
663 	return radeon_gart_table_ram_alloc(rdev);
664 }
665 
666 int r100_pci_gart_enable(struct radeon_device *rdev)
667 {
668 	uint32_t tmp;
669 
670 	radeon_gart_restore(rdev);
671 	/* discard memory request outside of configured range */
672 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
673 	WREG32(RADEON_AIC_CNTL, tmp);
674 	/* set address range for PCI address translate */
675 	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
676 	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
677 	/* set PCI GART page-table base address */
678 	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
679 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
680 	WREG32(RADEON_AIC_CNTL, tmp);
681 	r100_pci_gart_tlb_flush(rdev);
682 	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
683 		 (unsigned)(rdev->mc.gtt_size >> 20),
684 		 (unsigned long long)rdev->gart.table_addr);
685 	rdev->gart.ready = true;
686 	return 0;
687 }
688 
689 void r100_pci_gart_disable(struct radeon_device *rdev)
690 {
691 	uint32_t tmp;
692 
693 	/* discard memory request outside of configured range */
694 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
695 	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
696 	WREG32(RADEON_AIC_LO_ADDR, 0);
697 	WREG32(RADEON_AIC_HI_ADDR, 0);
698 }
699 
700 int r100_pci_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
701 {
702 	u32 *gtt = rdev->gart.ptr;
703 
704 	if (i < 0 || i > rdev->gart.num_gpu_pages) {
705 		return -EINVAL;
706 	}
707 	gtt[i] = cpu_to_le32(lower_32_bits(addr));
708 	return 0;
709 }
710 
711 void r100_pci_gart_fini(struct radeon_device *rdev)
712 {
713 	radeon_gart_fini(rdev);
714 	r100_pci_gart_disable(rdev);
715 	radeon_gart_table_ram_free(rdev);
716 }
717 
718 int r100_irq_set(struct radeon_device *rdev)
719 {
720 	uint32_t tmp = 0;
721 
722 	if (!rdev->irq.installed) {
723 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
724 		WREG32(R_000040_GEN_INT_CNTL, 0);
725 		return -EINVAL;
726 	}
727 	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
728 		tmp |= RADEON_SW_INT_ENABLE;
729 	}
730 	if (rdev->irq.crtc_vblank_int[0] ||
731 	    atomic_read(&rdev->irq.pflip[0])) {
732 		tmp |= RADEON_CRTC_VBLANK_MASK;
733 	}
734 	if (rdev->irq.crtc_vblank_int[1] ||
735 	    atomic_read(&rdev->irq.pflip[1])) {
736 		tmp |= RADEON_CRTC2_VBLANK_MASK;
737 	}
738 	if (rdev->irq.hpd[0]) {
739 		tmp |= RADEON_FP_DETECT_MASK;
740 	}
741 	if (rdev->irq.hpd[1]) {
742 		tmp |= RADEON_FP2_DETECT_MASK;
743 	}
744 	WREG32(RADEON_GEN_INT_CNTL, tmp);
745 	return 0;
746 }
747 
748 void r100_irq_disable(struct radeon_device *rdev)
749 {
750 	u32 tmp;
751 
752 	WREG32(R_000040_GEN_INT_CNTL, 0);
753 	/* Wait and acknowledge irq */
754 	mdelay(1);
755 	tmp = RREG32(R_000044_GEN_INT_STATUS);
756 	WREG32(R_000044_GEN_INT_STATUS, tmp);
757 }
758 
759 static uint32_t r100_irq_ack(struct radeon_device *rdev)
760 {
761 	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
762 	uint32_t irq_mask = RADEON_SW_INT_TEST |
763 		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
764 		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
765 
766 	if (irqs) {
767 		WREG32(RADEON_GEN_INT_STATUS, irqs);
768 	}
769 	return irqs & irq_mask;
770 }
771 
772 int r100_irq_process(struct radeon_device *rdev)
773 {
774 	uint32_t status, msi_rearm;
775 	bool queue_hotplug = false;
776 
777 	status = r100_irq_ack(rdev);
778 	if (!status) {
779 		return IRQ_NONE;
780 	}
781 	if (rdev->shutdown) {
782 		return IRQ_NONE;
783 	}
784 	while (status) {
785 		/* SW interrupt */
786 		if (status & RADEON_SW_INT_TEST) {
787 			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
788 		}
789 		/* Vertical blank interrupts */
790 		if (status & RADEON_CRTC_VBLANK_STAT) {
791 			if (rdev->irq.crtc_vblank_int[0]) {
792 				drm_handle_vblank(rdev->ddev, 0);
793 				rdev->pm.vblank_sync = true;
794 				wake_up(&rdev->irq.vblank_queue);
795 			}
796 			if (atomic_read(&rdev->irq.pflip[0]))
797 				radeon_crtc_handle_flip(rdev, 0);
798 		}
799 		if (status & RADEON_CRTC2_VBLANK_STAT) {
800 			if (rdev->irq.crtc_vblank_int[1]) {
801 				drm_handle_vblank(rdev->ddev, 1);
802 				rdev->pm.vblank_sync = true;
803 				wake_up(&rdev->irq.vblank_queue);
804 			}
805 			if (atomic_read(&rdev->irq.pflip[1]))
806 				radeon_crtc_handle_flip(rdev, 1);
807 		}
808 		if (status & RADEON_FP_DETECT_STAT) {
809 			queue_hotplug = true;
810 			DRM_DEBUG("HPD1\n");
811 		}
812 		if (status & RADEON_FP2_DETECT_STAT) {
813 			queue_hotplug = true;
814 			DRM_DEBUG("HPD2\n");
815 		}
816 		status = r100_irq_ack(rdev);
817 	}
818 	if (queue_hotplug)
819 		schedule_work(&rdev->hotplug_work);
820 	if (rdev->msi_enabled) {
821 		switch (rdev->family) {
822 		case CHIP_RS400:
823 		case CHIP_RS480:
824 			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
825 			WREG32(RADEON_AIC_CNTL, msi_rearm);
826 			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
827 			break;
828 		default:
829 			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
830 			break;
831 		}
832 	}
833 	return IRQ_HANDLED;
834 }
835 
836 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
837 {
838 	if (crtc == 0)
839 		return RREG32(RADEON_CRTC_CRNT_FRAME);
840 	else
841 		return RREG32(RADEON_CRTC2_CRNT_FRAME);
842 }
843 
844 /* Who ever call radeon_fence_emit should call ring_lock and ask
845  * for enough space (today caller are ib schedule and buffer move) */
846 void r100_fence_ring_emit(struct radeon_device *rdev,
847 			  struct radeon_fence *fence)
848 {
849 	struct radeon_ring *ring = &rdev->ring[fence->ring];
850 
851 	/* We have to make sure that caches are flushed before
852 	 * CPU might read something from VRAM. */
853 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
854 	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
855 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
856 	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
857 	/* Wait until IDLE & CLEAN */
858 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
859 	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
860 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
861 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
862 				RADEON_HDP_READ_BUFFER_INVALIDATE);
863 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
864 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
865 	/* Emit fence sequence & fire IRQ */
866 	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
867 	radeon_ring_write(ring, fence->seq);
868 	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
869 	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
870 }
871 
872 void r100_semaphore_ring_emit(struct radeon_device *rdev,
873 			      struct radeon_ring *ring,
874 			      struct radeon_semaphore *semaphore,
875 			      bool emit_wait)
876 {
877 	/* Unused on older asics, since we don't have semaphores or multiple rings */
878 	BUG();
879 }
880 
881 int r100_copy_blit(struct radeon_device *rdev,
882 		   uint64_t src_offset,
883 		   uint64_t dst_offset,
884 		   unsigned num_gpu_pages,
885 		   struct radeon_fence **fence)
886 {
887 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
888 	uint32_t cur_pages;
889 	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
890 	uint32_t pitch;
891 	uint32_t stride_pixels;
892 	unsigned ndw;
893 	int num_loops;
894 	int r = 0;
895 
896 	/* radeon limited to 16k stride */
897 	stride_bytes &= 0x3fff;
898 	/* radeon pitch is /64 */
899 	pitch = stride_bytes / 64;
900 	stride_pixels = stride_bytes / 4;
901 	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
902 
903 	/* Ask for enough room for blit + flush + fence */
904 	ndw = 64 + (10 * num_loops);
905 	r = radeon_ring_lock(rdev, ring, ndw);
906 	if (r) {
907 		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
908 		return -EINVAL;
909 	}
910 	while (num_gpu_pages > 0) {
911 		cur_pages = num_gpu_pages;
912 		if (cur_pages > 8191) {
913 			cur_pages = 8191;
914 		}
915 		num_gpu_pages -= cur_pages;
916 
917 		/* pages are in Y direction - height
918 		   page width in X direction - width */
919 		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
920 		radeon_ring_write(ring,
921 				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
922 				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
923 				  RADEON_GMC_SRC_CLIPPING |
924 				  RADEON_GMC_DST_CLIPPING |
925 				  RADEON_GMC_BRUSH_NONE |
926 				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
927 				  RADEON_GMC_SRC_DATATYPE_COLOR |
928 				  RADEON_ROP3_S |
929 				  RADEON_DP_SRC_SOURCE_MEMORY |
930 				  RADEON_GMC_CLR_CMP_CNTL_DIS |
931 				  RADEON_GMC_WR_MSK_DIS);
932 		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
933 		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
934 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
935 		radeon_ring_write(ring, 0);
936 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
937 		radeon_ring_write(ring, num_gpu_pages);
938 		radeon_ring_write(ring, num_gpu_pages);
939 		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
940 	}
941 	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
942 	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
943 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
944 	radeon_ring_write(ring,
945 			  RADEON_WAIT_2D_IDLECLEAN |
946 			  RADEON_WAIT_HOST_IDLECLEAN |
947 			  RADEON_WAIT_DMA_GUI_IDLE);
948 	if (fence) {
949 		r = radeon_fence_emit(rdev, fence, RADEON_RING_TYPE_GFX_INDEX);
950 	}
951 	radeon_ring_unlock_commit(rdev, ring);
952 	return r;
953 }
954 
955 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
956 {
957 	unsigned i;
958 	u32 tmp;
959 
960 	for (i = 0; i < rdev->usec_timeout; i++) {
961 		tmp = RREG32(R_000E40_RBBM_STATUS);
962 		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
963 			return 0;
964 		}
965 		udelay(1);
966 	}
967 	return -1;
968 }
969 
970 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
971 {
972 	int r;
973 
974 	r = radeon_ring_lock(rdev, ring, 2);
975 	if (r) {
976 		return;
977 	}
978 	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
979 	radeon_ring_write(ring,
980 			  RADEON_ISYNC_ANY2D_IDLE3D |
981 			  RADEON_ISYNC_ANY3D_IDLE2D |
982 			  RADEON_ISYNC_WAIT_IDLEGUI |
983 			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
984 	radeon_ring_unlock_commit(rdev, ring);
985 }
986 
987 
988 /* Load the microcode for the CP */
989 static int r100_cp_init_microcode(struct radeon_device *rdev)
990 {
991 	const char *fw_name = NULL;
992 	int err;
993 
994 	DRM_DEBUG_KMS("\n");
995 
996 	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
997 	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
998 	    (rdev->family == CHIP_RS200)) {
999 		DRM_INFO("Loading R100 Microcode\n");
1000 		fw_name = FIRMWARE_R100;
1001 	} else if ((rdev->family == CHIP_R200) ||
1002 		   (rdev->family == CHIP_RV250) ||
1003 		   (rdev->family == CHIP_RV280) ||
1004 		   (rdev->family == CHIP_RS300)) {
1005 		DRM_INFO("Loading R200 Microcode\n");
1006 		fw_name = FIRMWARE_R200;
1007 	} else if ((rdev->family == CHIP_R300) ||
1008 		   (rdev->family == CHIP_R350) ||
1009 		   (rdev->family == CHIP_RV350) ||
1010 		   (rdev->family == CHIP_RV380) ||
1011 		   (rdev->family == CHIP_RS400) ||
1012 		   (rdev->family == CHIP_RS480)) {
1013 		DRM_INFO("Loading R300 Microcode\n");
1014 		fw_name = FIRMWARE_R300;
1015 	} else if ((rdev->family == CHIP_R420) ||
1016 		   (rdev->family == CHIP_R423) ||
1017 		   (rdev->family == CHIP_RV410)) {
1018 		DRM_INFO("Loading R400 Microcode\n");
1019 		fw_name = FIRMWARE_R420;
1020 	} else if ((rdev->family == CHIP_RS690) ||
1021 		   (rdev->family == CHIP_RS740)) {
1022 		DRM_INFO("Loading RS690/RS740 Microcode\n");
1023 		fw_name = FIRMWARE_RS690;
1024 	} else if (rdev->family == CHIP_RS600) {
1025 		DRM_INFO("Loading RS600 Microcode\n");
1026 		fw_name = FIRMWARE_RS600;
1027 	} else if ((rdev->family == CHIP_RV515) ||
1028 		   (rdev->family == CHIP_R520) ||
1029 		   (rdev->family == CHIP_RV530) ||
1030 		   (rdev->family == CHIP_R580) ||
1031 		   (rdev->family == CHIP_RV560) ||
1032 		   (rdev->family == CHIP_RV570)) {
1033 		DRM_INFO("Loading R500 Microcode\n");
1034 		fw_name = FIRMWARE_R520;
1035 	}
1036 
1037 	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1038 	if (err) {
1039 		printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1040 		       fw_name);
1041 	} else if (rdev->me_fw->size % 8) {
1042 		printk(KERN_ERR
1043 		       "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1044 		       rdev->me_fw->size, fw_name);
1045 		err = -EINVAL;
1046 		release_firmware(rdev->me_fw);
1047 		rdev->me_fw = NULL;
1048 	}
1049 	return err;
1050 }
1051 
1052 static void r100_cp_load_microcode(struct radeon_device *rdev)
1053 {
1054 	const __be32 *fw_data;
1055 	int i, size;
1056 
1057 	if (r100_gui_wait_for_idle(rdev)) {
1058 		printk(KERN_WARNING "Failed to wait GUI idle while "
1059 		       "programming pipes. Bad things might happen.\n");
1060 	}
1061 
1062 	if (rdev->me_fw) {
1063 		size = rdev->me_fw->size / 4;
1064 		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1065 		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1066 		for (i = 0; i < size; i += 2) {
1067 			WREG32(RADEON_CP_ME_RAM_DATAH,
1068 			       be32_to_cpup(&fw_data[i]));
1069 			WREG32(RADEON_CP_ME_RAM_DATAL,
1070 			       be32_to_cpup(&fw_data[i + 1]));
1071 		}
1072 	}
1073 }
1074 
1075 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1076 {
1077 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1078 	unsigned rb_bufsz;
1079 	unsigned rb_blksz;
1080 	unsigned max_fetch;
1081 	unsigned pre_write_timer;
1082 	unsigned pre_write_limit;
1083 	unsigned indirect2_start;
1084 	unsigned indirect1_start;
1085 	uint32_t tmp;
1086 	int r;
1087 
1088 	if (r100_debugfs_cp_init(rdev)) {
1089 		DRM_ERROR("Failed to register debugfs file for CP !\n");
1090 	}
1091 	if (!rdev->me_fw) {
1092 		r = r100_cp_init_microcode(rdev);
1093 		if (r) {
1094 			DRM_ERROR("Failed to load firmware!\n");
1095 			return r;
1096 		}
1097 	}
1098 
1099 	/* Align ring size */
1100 	rb_bufsz = drm_order(ring_size / 8);
1101 	ring_size = (1 << (rb_bufsz + 1)) * 4;
1102 	r100_cp_load_microcode(rdev);
1103 	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1104 			     RADEON_CP_RB_RPTR, RADEON_CP_RB_WPTR,
1105 			     0, 0x7fffff, RADEON_CP_PACKET2);
1106 	if (r) {
1107 		return r;
1108 	}
1109 	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1110 	 * the rptr copy in system ram */
1111 	rb_blksz = 9;
1112 	/* cp will read 128bytes at a time (4 dwords) */
1113 	max_fetch = 1;
1114 	ring->align_mask = 16 - 1;
1115 	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1116 	pre_write_timer = 64;
1117 	/* Force CP_RB_WPTR write if written more than one time before the
1118 	 * delay expire
1119 	 */
1120 	pre_write_limit = 0;
1121 	/* Setup the cp cache like this (cache size is 96 dwords) :
1122 	 *	RING		0  to 15
1123 	 *	INDIRECT1	16 to 79
1124 	 *	INDIRECT2	80 to 95
1125 	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1126 	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1127 	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1128 	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1129 	 * so it gets the bigger cache.
1130 	 */
1131 	indirect2_start = 80;
1132 	indirect1_start = 16;
1133 	/* cp setup */
1134 	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1135 	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1136 	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1137 	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1138 #ifdef __BIG_ENDIAN
1139 	tmp |= RADEON_BUF_SWAP_32BIT;
1140 #endif
1141 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1142 
1143 	/* Set ring address */
1144 	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1145 	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1146 	/* Force read & write ptr to 0 */
1147 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1148 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1149 	ring->wptr = 0;
1150 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1151 
1152 	/* set the wb address whether it's enabled or not */
1153 	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1154 		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1155 	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1156 
1157 	if (rdev->wb.enabled)
1158 		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1159 	else {
1160 		tmp |= RADEON_RB_NO_UPDATE;
1161 		WREG32(R_000770_SCRATCH_UMSK, 0);
1162 	}
1163 
1164 	WREG32(RADEON_CP_RB_CNTL, tmp);
1165 	udelay(10);
1166 	ring->rptr = RREG32(RADEON_CP_RB_RPTR);
1167 	/* Set cp mode to bus mastering & enable cp*/
1168 	WREG32(RADEON_CP_CSQ_MODE,
1169 	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1170 	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1171 	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1172 	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1173 	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1174 
1175 	/* at this point everything should be setup correctly to enable master */
1176 	pci_set_master(rdev->pdev);
1177 
1178 	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1179 	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1180 	if (r) {
1181 		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1182 		return r;
1183 	}
1184 	ring->ready = true;
1185 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1186 
1187 	if (!ring->rptr_save_reg /* not resuming from suspend */
1188 	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1189 		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1190 		if (r) {
1191 			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1192 			ring->rptr_save_reg = 0;
1193 		}
1194 	}
1195 	return 0;
1196 }
1197 
1198 void r100_cp_fini(struct radeon_device *rdev)
1199 {
1200 	if (r100_cp_wait_for_idle(rdev)) {
1201 		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1202 	}
1203 	/* Disable ring */
1204 	r100_cp_disable(rdev);
1205 	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1206 	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1207 	DRM_INFO("radeon: cp finalized\n");
1208 }
1209 
1210 void r100_cp_disable(struct radeon_device *rdev)
1211 {
1212 	/* Disable ring */
1213 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1214 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1215 	WREG32(RADEON_CP_CSQ_MODE, 0);
1216 	WREG32(RADEON_CP_CSQ_CNTL, 0);
1217 	WREG32(R_000770_SCRATCH_UMSK, 0);
1218 	if (r100_gui_wait_for_idle(rdev)) {
1219 		printk(KERN_WARNING "Failed to wait GUI idle while "
1220 		       "programming pipes. Bad things might happen.\n");
1221 	}
1222 }
1223 
1224 /*
1225  * CS functions
1226  */
1227 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1228 			    struct radeon_cs_packet *pkt,
1229 			    unsigned idx,
1230 			    unsigned reg)
1231 {
1232 	int r;
1233 	u32 tile_flags = 0;
1234 	u32 tmp;
1235 	struct radeon_cs_reloc *reloc;
1236 	u32 value;
1237 
1238 	r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1239 	if (r) {
1240 		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1241 			  idx, reg);
1242 		radeon_cs_dump_packet(p, pkt);
1243 		return r;
1244 	}
1245 
1246 	value = radeon_get_ib_value(p, idx);
1247 	tmp = value & 0x003fffff;
1248 	tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
1249 
1250 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1251 		if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1252 			tile_flags |= RADEON_DST_TILE_MACRO;
1253 		if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
1254 			if (reg == RADEON_SRC_PITCH_OFFSET) {
1255 				DRM_ERROR("Cannot src blit from microtiled surface\n");
1256 				radeon_cs_dump_packet(p, pkt);
1257 				return -EINVAL;
1258 			}
1259 			tile_flags |= RADEON_DST_TILE_MICRO;
1260 		}
1261 
1262 		tmp |= tile_flags;
1263 		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1264 	} else
1265 		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1266 	return 0;
1267 }
1268 
1269 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1270 			     struct radeon_cs_packet *pkt,
1271 			     int idx)
1272 {
1273 	unsigned c, i;
1274 	struct radeon_cs_reloc *reloc;
1275 	struct r100_cs_track *track;
1276 	int r = 0;
1277 	volatile uint32_t *ib;
1278 	u32 idx_value;
1279 
1280 	ib = p->ib.ptr;
1281 	track = (struct r100_cs_track *)p->track;
1282 	c = radeon_get_ib_value(p, idx++) & 0x1F;
1283 	if (c > 16) {
1284 	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1285 		      pkt->opcode);
1286 	    radeon_cs_dump_packet(p, pkt);
1287 	    return -EINVAL;
1288 	}
1289 	track->num_arrays = c;
1290 	for (i = 0; i < (c - 1); i+=2, idx+=3) {
1291 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1292 		if (r) {
1293 			DRM_ERROR("No reloc for packet3 %d\n",
1294 				  pkt->opcode);
1295 			radeon_cs_dump_packet(p, pkt);
1296 			return r;
1297 		}
1298 		idx_value = radeon_get_ib_value(p, idx);
1299 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1300 
1301 		track->arrays[i + 0].esize = idx_value >> 8;
1302 		track->arrays[i + 0].robj = reloc->robj;
1303 		track->arrays[i + 0].esize &= 0x7F;
1304 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1305 		if (r) {
1306 			DRM_ERROR("No reloc for packet3 %d\n",
1307 				  pkt->opcode);
1308 			radeon_cs_dump_packet(p, pkt);
1309 			return r;
1310 		}
1311 		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->lobj.gpu_offset);
1312 		track->arrays[i + 1].robj = reloc->robj;
1313 		track->arrays[i + 1].esize = idx_value >> 24;
1314 		track->arrays[i + 1].esize &= 0x7F;
1315 	}
1316 	if (c & 1) {
1317 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1318 		if (r) {
1319 			DRM_ERROR("No reloc for packet3 %d\n",
1320 					  pkt->opcode);
1321 			radeon_cs_dump_packet(p, pkt);
1322 			return r;
1323 		}
1324 		idx_value = radeon_get_ib_value(p, idx);
1325 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1326 		track->arrays[i + 0].robj = reloc->robj;
1327 		track->arrays[i + 0].esize = idx_value >> 8;
1328 		track->arrays[i + 0].esize &= 0x7F;
1329 	}
1330 	return r;
1331 }
1332 
1333 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1334 			  struct radeon_cs_packet *pkt,
1335 			  const unsigned *auth, unsigned n,
1336 			  radeon_packet0_check_t check)
1337 {
1338 	unsigned reg;
1339 	unsigned i, j, m;
1340 	unsigned idx;
1341 	int r;
1342 
1343 	idx = pkt->idx + 1;
1344 	reg = pkt->reg;
1345 	/* Check that register fall into register range
1346 	 * determined by the number of entry (n) in the
1347 	 * safe register bitmap.
1348 	 */
1349 	if (pkt->one_reg_wr) {
1350 		if ((reg >> 7) > n) {
1351 			return -EINVAL;
1352 		}
1353 	} else {
1354 		if (((reg + (pkt->count << 2)) >> 7) > n) {
1355 			return -EINVAL;
1356 		}
1357 	}
1358 	for (i = 0; i <= pkt->count; i++, idx++) {
1359 		j = (reg >> 7);
1360 		m = 1 << ((reg >> 2) & 31);
1361 		if (auth[j] & m) {
1362 			r = check(p, pkt, idx, reg);
1363 			if (r) {
1364 				return r;
1365 			}
1366 		}
1367 		if (pkt->one_reg_wr) {
1368 			if (!(auth[j] & m)) {
1369 				break;
1370 			}
1371 		} else {
1372 			reg += 4;
1373 		}
1374 	}
1375 	return 0;
1376 }
1377 
1378 /**
1379  * r100_cs_packet_next_vline() - parse userspace VLINE packet
1380  * @parser:		parser structure holding parsing context.
1381  *
1382  * Userspace sends a special sequence for VLINE waits.
1383  * PACKET0 - VLINE_START_END + value
1384  * PACKET0 - WAIT_UNTIL +_value
1385  * RELOC (P3) - crtc_id in reloc.
1386  *
1387  * This function parses this and relocates the VLINE START END
1388  * and WAIT UNTIL packets to the correct crtc.
1389  * It also detects a switched off crtc and nulls out the
1390  * wait in that case.
1391  */
1392 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1393 {
1394 	struct drm_mode_object *obj;
1395 	struct drm_crtc *crtc;
1396 	struct radeon_crtc *radeon_crtc;
1397 	struct radeon_cs_packet p3reloc, waitreloc;
1398 	int crtc_id;
1399 	int r;
1400 	uint32_t header, h_idx, reg;
1401 	volatile uint32_t *ib;
1402 
1403 	ib = p->ib.ptr;
1404 
1405 	/* parse the wait until */
1406 	r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1407 	if (r)
1408 		return r;
1409 
1410 	/* check its a wait until and only 1 count */
1411 	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1412 	    waitreloc.count != 0) {
1413 		DRM_ERROR("vline wait had illegal wait until segment\n");
1414 		return -EINVAL;
1415 	}
1416 
1417 	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1418 		DRM_ERROR("vline wait had illegal wait until\n");
1419 		return -EINVAL;
1420 	}
1421 
1422 	/* jump over the NOP */
1423 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1424 	if (r)
1425 		return r;
1426 
1427 	h_idx = p->idx - 2;
1428 	p->idx += waitreloc.count + 2;
1429 	p->idx += p3reloc.count + 2;
1430 
1431 	header = radeon_get_ib_value(p, h_idx);
1432 	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1433 	reg = R100_CP_PACKET0_GET_REG(header);
1434 	obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
1435 	if (!obj) {
1436 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1437 		return -EINVAL;
1438 	}
1439 	crtc = obj_to_crtc(obj);
1440 	radeon_crtc = to_radeon_crtc(crtc);
1441 	crtc_id = radeon_crtc->crtc_id;
1442 
1443 	if (!crtc->enabled) {
1444 		/* if the CRTC isn't enabled - we need to nop out the wait until */
1445 		ib[h_idx + 2] = PACKET2(0);
1446 		ib[h_idx + 3] = PACKET2(0);
1447 	} else if (crtc_id == 1) {
1448 		switch (reg) {
1449 		case AVIVO_D1MODE_VLINE_START_END:
1450 			header &= ~R300_CP_PACKET0_REG_MASK;
1451 			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1452 			break;
1453 		case RADEON_CRTC_GUI_TRIG_VLINE:
1454 			header &= ~R300_CP_PACKET0_REG_MASK;
1455 			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1456 			break;
1457 		default:
1458 			DRM_ERROR("unknown crtc reloc\n");
1459 			return -EINVAL;
1460 		}
1461 		ib[h_idx] = header;
1462 		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1463 	}
1464 
1465 	return 0;
1466 }
1467 
1468 static int r100_get_vtx_size(uint32_t vtx_fmt)
1469 {
1470 	int vtx_size;
1471 	vtx_size = 2;
1472 	/* ordered according to bits in spec */
1473 	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1474 		vtx_size++;
1475 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1476 		vtx_size += 3;
1477 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1478 		vtx_size++;
1479 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1480 		vtx_size++;
1481 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1482 		vtx_size += 3;
1483 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1484 		vtx_size++;
1485 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1486 		vtx_size++;
1487 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1488 		vtx_size += 2;
1489 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1490 		vtx_size += 2;
1491 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1492 		vtx_size++;
1493 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1494 		vtx_size += 2;
1495 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1496 		vtx_size++;
1497 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1498 		vtx_size += 2;
1499 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1500 		vtx_size++;
1501 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1502 		vtx_size++;
1503 	/* blend weight */
1504 	if (vtx_fmt & (0x7 << 15))
1505 		vtx_size += (vtx_fmt >> 15) & 0x7;
1506 	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1507 		vtx_size += 3;
1508 	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1509 		vtx_size += 2;
1510 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1511 		vtx_size++;
1512 	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1513 		vtx_size++;
1514 	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1515 		vtx_size++;
1516 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1517 		vtx_size++;
1518 	return vtx_size;
1519 }
1520 
1521 static int r100_packet0_check(struct radeon_cs_parser *p,
1522 			      struct radeon_cs_packet *pkt,
1523 			      unsigned idx, unsigned reg)
1524 {
1525 	struct radeon_cs_reloc *reloc;
1526 	struct r100_cs_track *track;
1527 	volatile uint32_t *ib;
1528 	uint32_t tmp;
1529 	int r;
1530 	int i, face;
1531 	u32 tile_flags = 0;
1532 	u32 idx_value;
1533 
1534 	ib = p->ib.ptr;
1535 	track = (struct r100_cs_track *)p->track;
1536 
1537 	idx_value = radeon_get_ib_value(p, idx);
1538 
1539 	switch (reg) {
1540 	case RADEON_CRTC_GUI_TRIG_VLINE:
1541 		r = r100_cs_packet_parse_vline(p);
1542 		if (r) {
1543 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1544 				  idx, reg);
1545 			radeon_cs_dump_packet(p, pkt);
1546 			return r;
1547 		}
1548 		break;
1549 		/* FIXME: only allow PACKET3 blit? easier to check for out of
1550 		 * range access */
1551 	case RADEON_DST_PITCH_OFFSET:
1552 	case RADEON_SRC_PITCH_OFFSET:
1553 		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1554 		if (r)
1555 			return r;
1556 		break;
1557 	case RADEON_RB3D_DEPTHOFFSET:
1558 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1559 		if (r) {
1560 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1561 				  idx, reg);
1562 			radeon_cs_dump_packet(p, pkt);
1563 			return r;
1564 		}
1565 		track->zb.robj = reloc->robj;
1566 		track->zb.offset = idx_value;
1567 		track->zb_dirty = true;
1568 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1569 		break;
1570 	case RADEON_RB3D_COLOROFFSET:
1571 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1572 		if (r) {
1573 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1574 				  idx, reg);
1575 			radeon_cs_dump_packet(p, pkt);
1576 			return r;
1577 		}
1578 		track->cb[0].robj = reloc->robj;
1579 		track->cb[0].offset = idx_value;
1580 		track->cb_dirty = true;
1581 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1582 		break;
1583 	case RADEON_PP_TXOFFSET_0:
1584 	case RADEON_PP_TXOFFSET_1:
1585 	case RADEON_PP_TXOFFSET_2:
1586 		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1587 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1588 		if (r) {
1589 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1590 				  idx, reg);
1591 			radeon_cs_dump_packet(p, pkt);
1592 			return r;
1593 		}
1594 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1595 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1596 				tile_flags |= RADEON_TXO_MACRO_TILE;
1597 			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1598 				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1599 
1600 			tmp = idx_value & ~(0x7 << 2);
1601 			tmp |= tile_flags;
1602 			ib[idx] = tmp + ((u32)reloc->lobj.gpu_offset);
1603 		} else
1604 			ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1605 		track->textures[i].robj = reloc->robj;
1606 		track->tex_dirty = true;
1607 		break;
1608 	case RADEON_PP_CUBIC_OFFSET_T0_0:
1609 	case RADEON_PP_CUBIC_OFFSET_T0_1:
1610 	case RADEON_PP_CUBIC_OFFSET_T0_2:
1611 	case RADEON_PP_CUBIC_OFFSET_T0_3:
1612 	case RADEON_PP_CUBIC_OFFSET_T0_4:
1613 		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1614 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1615 		if (r) {
1616 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1617 				  idx, reg);
1618 			radeon_cs_dump_packet(p, pkt);
1619 			return r;
1620 		}
1621 		track->textures[0].cube_info[i].offset = idx_value;
1622 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1623 		track->textures[0].cube_info[i].robj = reloc->robj;
1624 		track->tex_dirty = true;
1625 		break;
1626 	case RADEON_PP_CUBIC_OFFSET_T1_0:
1627 	case RADEON_PP_CUBIC_OFFSET_T1_1:
1628 	case RADEON_PP_CUBIC_OFFSET_T1_2:
1629 	case RADEON_PP_CUBIC_OFFSET_T1_3:
1630 	case RADEON_PP_CUBIC_OFFSET_T1_4:
1631 		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1632 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1633 		if (r) {
1634 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1635 				  idx, reg);
1636 			radeon_cs_dump_packet(p, pkt);
1637 			return r;
1638 		}
1639 		track->textures[1].cube_info[i].offset = idx_value;
1640 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1641 		track->textures[1].cube_info[i].robj = reloc->robj;
1642 		track->tex_dirty = true;
1643 		break;
1644 	case RADEON_PP_CUBIC_OFFSET_T2_0:
1645 	case RADEON_PP_CUBIC_OFFSET_T2_1:
1646 	case RADEON_PP_CUBIC_OFFSET_T2_2:
1647 	case RADEON_PP_CUBIC_OFFSET_T2_3:
1648 	case RADEON_PP_CUBIC_OFFSET_T2_4:
1649 		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1650 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1651 		if (r) {
1652 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1653 				  idx, reg);
1654 			radeon_cs_dump_packet(p, pkt);
1655 			return r;
1656 		}
1657 		track->textures[2].cube_info[i].offset = idx_value;
1658 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1659 		track->textures[2].cube_info[i].robj = reloc->robj;
1660 		track->tex_dirty = true;
1661 		break;
1662 	case RADEON_RE_WIDTH_HEIGHT:
1663 		track->maxy = ((idx_value >> 16) & 0x7FF);
1664 		track->cb_dirty = true;
1665 		track->zb_dirty = true;
1666 		break;
1667 	case RADEON_RB3D_COLORPITCH:
1668 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1669 		if (r) {
1670 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1671 				  idx, reg);
1672 			radeon_cs_dump_packet(p, pkt);
1673 			return r;
1674 		}
1675 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1676 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1677 				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1678 			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1679 				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1680 
1681 			tmp = idx_value & ~(0x7 << 16);
1682 			tmp |= tile_flags;
1683 			ib[idx] = tmp;
1684 		} else
1685 			ib[idx] = idx_value;
1686 
1687 		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1688 		track->cb_dirty = true;
1689 		break;
1690 	case RADEON_RB3D_DEPTHPITCH:
1691 		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1692 		track->zb_dirty = true;
1693 		break;
1694 	case RADEON_RB3D_CNTL:
1695 		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1696 		case 7:
1697 		case 8:
1698 		case 9:
1699 		case 11:
1700 		case 12:
1701 			track->cb[0].cpp = 1;
1702 			break;
1703 		case 3:
1704 		case 4:
1705 		case 15:
1706 			track->cb[0].cpp = 2;
1707 			break;
1708 		case 6:
1709 			track->cb[0].cpp = 4;
1710 			break;
1711 		default:
1712 			DRM_ERROR("Invalid color buffer format (%d) !\n",
1713 				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1714 			return -EINVAL;
1715 		}
1716 		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1717 		track->cb_dirty = true;
1718 		track->zb_dirty = true;
1719 		break;
1720 	case RADEON_RB3D_ZSTENCILCNTL:
1721 		switch (idx_value & 0xf) {
1722 		case 0:
1723 			track->zb.cpp = 2;
1724 			break;
1725 		case 2:
1726 		case 3:
1727 		case 4:
1728 		case 5:
1729 		case 9:
1730 		case 11:
1731 			track->zb.cpp = 4;
1732 			break;
1733 		default:
1734 			break;
1735 		}
1736 		track->zb_dirty = true;
1737 		break;
1738 	case RADEON_RB3D_ZPASS_ADDR:
1739 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1740 		if (r) {
1741 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1742 				  idx, reg);
1743 			radeon_cs_dump_packet(p, pkt);
1744 			return r;
1745 		}
1746 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1747 		break;
1748 	case RADEON_PP_CNTL:
1749 		{
1750 			uint32_t temp = idx_value >> 4;
1751 			for (i = 0; i < track->num_texture; i++)
1752 				track->textures[i].enabled = !!(temp & (1 << i));
1753 			track->tex_dirty = true;
1754 		}
1755 		break;
1756 	case RADEON_SE_VF_CNTL:
1757 		track->vap_vf_cntl = idx_value;
1758 		break;
1759 	case RADEON_SE_VTX_FMT:
1760 		track->vtx_size = r100_get_vtx_size(idx_value);
1761 		break;
1762 	case RADEON_PP_TEX_SIZE_0:
1763 	case RADEON_PP_TEX_SIZE_1:
1764 	case RADEON_PP_TEX_SIZE_2:
1765 		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1766 		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1767 		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1768 		track->tex_dirty = true;
1769 		break;
1770 	case RADEON_PP_TEX_PITCH_0:
1771 	case RADEON_PP_TEX_PITCH_1:
1772 	case RADEON_PP_TEX_PITCH_2:
1773 		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1774 		track->textures[i].pitch = idx_value + 32;
1775 		track->tex_dirty = true;
1776 		break;
1777 	case RADEON_PP_TXFILTER_0:
1778 	case RADEON_PP_TXFILTER_1:
1779 	case RADEON_PP_TXFILTER_2:
1780 		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1781 		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1782 						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1783 		tmp = (idx_value >> 23) & 0x7;
1784 		if (tmp == 2 || tmp == 6)
1785 			track->textures[i].roundup_w = false;
1786 		tmp = (idx_value >> 27) & 0x7;
1787 		if (tmp == 2 || tmp == 6)
1788 			track->textures[i].roundup_h = false;
1789 		track->tex_dirty = true;
1790 		break;
1791 	case RADEON_PP_TXFORMAT_0:
1792 	case RADEON_PP_TXFORMAT_1:
1793 	case RADEON_PP_TXFORMAT_2:
1794 		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1795 		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1796 			track->textures[i].use_pitch = 1;
1797 		} else {
1798 			track->textures[i].use_pitch = 0;
1799 			track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1800 			track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1801 		}
1802 		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1803 			track->textures[i].tex_coord_type = 2;
1804 		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1805 		case RADEON_TXFORMAT_I8:
1806 		case RADEON_TXFORMAT_RGB332:
1807 		case RADEON_TXFORMAT_Y8:
1808 			track->textures[i].cpp = 1;
1809 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1810 			break;
1811 		case RADEON_TXFORMAT_AI88:
1812 		case RADEON_TXFORMAT_ARGB1555:
1813 		case RADEON_TXFORMAT_RGB565:
1814 		case RADEON_TXFORMAT_ARGB4444:
1815 		case RADEON_TXFORMAT_VYUY422:
1816 		case RADEON_TXFORMAT_YVYU422:
1817 		case RADEON_TXFORMAT_SHADOW16:
1818 		case RADEON_TXFORMAT_LDUDV655:
1819 		case RADEON_TXFORMAT_DUDV88:
1820 			track->textures[i].cpp = 2;
1821 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1822 			break;
1823 		case RADEON_TXFORMAT_ARGB8888:
1824 		case RADEON_TXFORMAT_RGBA8888:
1825 		case RADEON_TXFORMAT_SHADOW32:
1826 		case RADEON_TXFORMAT_LDUDUV8888:
1827 			track->textures[i].cpp = 4;
1828 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1829 			break;
1830 		case RADEON_TXFORMAT_DXT1:
1831 			track->textures[i].cpp = 1;
1832 			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1833 			break;
1834 		case RADEON_TXFORMAT_DXT23:
1835 		case RADEON_TXFORMAT_DXT45:
1836 			track->textures[i].cpp = 1;
1837 			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1838 			break;
1839 		}
1840 		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1841 		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1842 		track->tex_dirty = true;
1843 		break;
1844 	case RADEON_PP_CUBIC_FACES_0:
1845 	case RADEON_PP_CUBIC_FACES_1:
1846 	case RADEON_PP_CUBIC_FACES_2:
1847 		tmp = idx_value;
1848 		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1849 		for (face = 0; face < 4; face++) {
1850 			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1851 			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1852 		}
1853 		track->tex_dirty = true;
1854 		break;
1855 	default:
1856 		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1857 		       reg, idx);
1858 		return -EINVAL;
1859 	}
1860 	return 0;
1861 }
1862 
1863 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1864 					 struct radeon_cs_packet *pkt,
1865 					 struct radeon_bo *robj)
1866 {
1867 	unsigned idx;
1868 	u32 value;
1869 	idx = pkt->idx + 1;
1870 	value = radeon_get_ib_value(p, idx + 2);
1871 	if ((value + 1) > radeon_bo_size(robj)) {
1872 		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1873 			  "(need %u have %lu) !\n",
1874 			  value + 1,
1875 			  radeon_bo_size(robj));
1876 		return -EINVAL;
1877 	}
1878 	return 0;
1879 }
1880 
1881 static int r100_packet3_check(struct radeon_cs_parser *p,
1882 			      struct radeon_cs_packet *pkt)
1883 {
1884 	struct radeon_cs_reloc *reloc;
1885 	struct r100_cs_track *track;
1886 	unsigned idx;
1887 	volatile uint32_t *ib;
1888 	int r;
1889 
1890 	ib = p->ib.ptr;
1891 	idx = pkt->idx + 1;
1892 	track = (struct r100_cs_track *)p->track;
1893 	switch (pkt->opcode) {
1894 	case PACKET3_3D_LOAD_VBPNTR:
1895 		r = r100_packet3_load_vbpntr(p, pkt, idx);
1896 		if (r)
1897 			return r;
1898 		break;
1899 	case PACKET3_INDX_BUFFER:
1900 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1901 		if (r) {
1902 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1903 			radeon_cs_dump_packet(p, pkt);
1904 			return r;
1905 		}
1906 		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->lobj.gpu_offset);
1907 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1908 		if (r) {
1909 			return r;
1910 		}
1911 		break;
1912 	case 0x23:
1913 		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1914 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1915 		if (r) {
1916 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1917 			radeon_cs_dump_packet(p, pkt);
1918 			return r;
1919 		}
1920 		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->lobj.gpu_offset);
1921 		track->num_arrays = 1;
1922 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1923 
1924 		track->arrays[0].robj = reloc->robj;
1925 		track->arrays[0].esize = track->vtx_size;
1926 
1927 		track->max_indx = radeon_get_ib_value(p, idx+1);
1928 
1929 		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1930 		track->immd_dwords = pkt->count - 1;
1931 		r = r100_cs_track_check(p->rdev, track);
1932 		if (r)
1933 			return r;
1934 		break;
1935 	case PACKET3_3D_DRAW_IMMD:
1936 		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1937 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1938 			return -EINVAL;
1939 		}
1940 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1941 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1942 		track->immd_dwords = pkt->count - 1;
1943 		r = r100_cs_track_check(p->rdev, track);
1944 		if (r)
1945 			return r;
1946 		break;
1947 		/* triggers drawing using in-packet vertex data */
1948 	case PACKET3_3D_DRAW_IMMD_2:
1949 		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1950 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1951 			return -EINVAL;
1952 		}
1953 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1954 		track->immd_dwords = pkt->count;
1955 		r = r100_cs_track_check(p->rdev, track);
1956 		if (r)
1957 			return r;
1958 		break;
1959 		/* triggers drawing using in-packet vertex data */
1960 	case PACKET3_3D_DRAW_VBUF_2:
1961 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1962 		r = r100_cs_track_check(p->rdev, track);
1963 		if (r)
1964 			return r;
1965 		break;
1966 		/* triggers drawing of vertex buffers setup elsewhere */
1967 	case PACKET3_3D_DRAW_INDX_2:
1968 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1969 		r = r100_cs_track_check(p->rdev, track);
1970 		if (r)
1971 			return r;
1972 		break;
1973 		/* triggers drawing using indices to vertex buffer */
1974 	case PACKET3_3D_DRAW_VBUF:
1975 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1976 		r = r100_cs_track_check(p->rdev, track);
1977 		if (r)
1978 			return r;
1979 		break;
1980 		/* triggers drawing of vertex buffers setup elsewhere */
1981 	case PACKET3_3D_DRAW_INDX:
1982 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1983 		r = r100_cs_track_check(p->rdev, track);
1984 		if (r)
1985 			return r;
1986 		break;
1987 		/* triggers drawing using indices to vertex buffer */
1988 	case PACKET3_3D_CLEAR_HIZ:
1989 	case PACKET3_3D_CLEAR_ZMASK:
1990 		if (p->rdev->hyperz_filp != p->filp)
1991 			return -EINVAL;
1992 		break;
1993 	case PACKET3_NOP:
1994 		break;
1995 	default:
1996 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1997 		return -EINVAL;
1998 	}
1999 	return 0;
2000 }
2001 
2002 int r100_cs_parse(struct radeon_cs_parser *p)
2003 {
2004 	struct radeon_cs_packet pkt;
2005 	struct r100_cs_track *track;
2006 	int r;
2007 
2008 	track = kzalloc(sizeof(*track), GFP_KERNEL);
2009 	if (!track)
2010 		return -ENOMEM;
2011 	r100_cs_track_clear(p->rdev, track);
2012 	p->track = track;
2013 	do {
2014 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2015 		if (r) {
2016 			return r;
2017 		}
2018 		p->idx += pkt.count + 2;
2019 		switch (pkt.type) {
2020 		case RADEON_PACKET_TYPE0:
2021 			if (p->rdev->family >= CHIP_R200)
2022 				r = r100_cs_parse_packet0(p, &pkt,
2023 					p->rdev->config.r100.reg_safe_bm,
2024 					p->rdev->config.r100.reg_safe_bm_size,
2025 					&r200_packet0_check);
2026 			else
2027 				r = r100_cs_parse_packet0(p, &pkt,
2028 					p->rdev->config.r100.reg_safe_bm,
2029 					p->rdev->config.r100.reg_safe_bm_size,
2030 					&r100_packet0_check);
2031 			break;
2032 		case RADEON_PACKET_TYPE2:
2033 			break;
2034 		case RADEON_PACKET_TYPE3:
2035 			r = r100_packet3_check(p, &pkt);
2036 			break;
2037 		default:
2038 			DRM_ERROR("Unknown packet type %d !\n",
2039 				  pkt.type);
2040 			return -EINVAL;
2041 		}
2042 		if (r)
2043 			return r;
2044 	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
2045 	return 0;
2046 }
2047 
2048 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2049 {
2050 	DRM_ERROR("pitch                      %d\n", t->pitch);
2051 	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2052 	DRM_ERROR("width                      %d\n", t->width);
2053 	DRM_ERROR("width_11                   %d\n", t->width_11);
2054 	DRM_ERROR("height                     %d\n", t->height);
2055 	DRM_ERROR("height_11                  %d\n", t->height_11);
2056 	DRM_ERROR("num levels                 %d\n", t->num_levels);
2057 	DRM_ERROR("depth                      %d\n", t->txdepth);
2058 	DRM_ERROR("bpp                        %d\n", t->cpp);
2059 	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2060 	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2061 	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2062 	DRM_ERROR("compress format            %d\n", t->compress_format);
2063 }
2064 
2065 static int r100_track_compress_size(int compress_format, int w, int h)
2066 {
2067 	int block_width, block_height, block_bytes;
2068 	int wblocks, hblocks;
2069 	int min_wblocks;
2070 	int sz;
2071 
2072 	block_width = 4;
2073 	block_height = 4;
2074 
2075 	switch (compress_format) {
2076 	case R100_TRACK_COMP_DXT1:
2077 		block_bytes = 8;
2078 		min_wblocks = 4;
2079 		break;
2080 	default:
2081 	case R100_TRACK_COMP_DXT35:
2082 		block_bytes = 16;
2083 		min_wblocks = 2;
2084 		break;
2085 	}
2086 
2087 	hblocks = (h + block_height - 1) / block_height;
2088 	wblocks = (w + block_width - 1) / block_width;
2089 	if (wblocks < min_wblocks)
2090 		wblocks = min_wblocks;
2091 	sz = wblocks * hblocks * block_bytes;
2092 	return sz;
2093 }
2094 
2095 static int r100_cs_track_cube(struct radeon_device *rdev,
2096 			      struct r100_cs_track *track, unsigned idx)
2097 {
2098 	unsigned face, w, h;
2099 	struct radeon_bo *cube_robj;
2100 	unsigned long size;
2101 	unsigned compress_format = track->textures[idx].compress_format;
2102 
2103 	for (face = 0; face < 5; face++) {
2104 		cube_robj = track->textures[idx].cube_info[face].robj;
2105 		w = track->textures[idx].cube_info[face].width;
2106 		h = track->textures[idx].cube_info[face].height;
2107 
2108 		if (compress_format) {
2109 			size = r100_track_compress_size(compress_format, w, h);
2110 		} else
2111 			size = w * h;
2112 		size *= track->textures[idx].cpp;
2113 
2114 		size += track->textures[idx].cube_info[face].offset;
2115 
2116 		if (size > radeon_bo_size(cube_robj)) {
2117 			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2118 				  size, radeon_bo_size(cube_robj));
2119 			r100_cs_track_texture_print(&track->textures[idx]);
2120 			return -1;
2121 		}
2122 	}
2123 	return 0;
2124 }
2125 
2126 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2127 				       struct r100_cs_track *track)
2128 {
2129 	struct radeon_bo *robj;
2130 	unsigned long size;
2131 	unsigned u, i, w, h, d;
2132 	int ret;
2133 
2134 	for (u = 0; u < track->num_texture; u++) {
2135 		if (!track->textures[u].enabled)
2136 			continue;
2137 		if (track->textures[u].lookup_disable)
2138 			continue;
2139 		robj = track->textures[u].robj;
2140 		if (robj == NULL) {
2141 			DRM_ERROR("No texture bound to unit %u\n", u);
2142 			return -EINVAL;
2143 		}
2144 		size = 0;
2145 		for (i = 0; i <= track->textures[u].num_levels; i++) {
2146 			if (track->textures[u].use_pitch) {
2147 				if (rdev->family < CHIP_R300)
2148 					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2149 				else
2150 					w = track->textures[u].pitch / (1 << i);
2151 			} else {
2152 				w = track->textures[u].width;
2153 				if (rdev->family >= CHIP_RV515)
2154 					w |= track->textures[u].width_11;
2155 				w = w / (1 << i);
2156 				if (track->textures[u].roundup_w)
2157 					w = roundup_pow_of_two(w);
2158 			}
2159 			h = track->textures[u].height;
2160 			if (rdev->family >= CHIP_RV515)
2161 				h |= track->textures[u].height_11;
2162 			h = h / (1 << i);
2163 			if (track->textures[u].roundup_h)
2164 				h = roundup_pow_of_two(h);
2165 			if (track->textures[u].tex_coord_type == 1) {
2166 				d = (1 << track->textures[u].txdepth) / (1 << i);
2167 				if (!d)
2168 					d = 1;
2169 			} else {
2170 				d = 1;
2171 			}
2172 			if (track->textures[u].compress_format) {
2173 
2174 				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2175 				/* compressed textures are block based */
2176 			} else
2177 				size += w * h * d;
2178 		}
2179 		size *= track->textures[u].cpp;
2180 
2181 		switch (track->textures[u].tex_coord_type) {
2182 		case 0:
2183 		case 1:
2184 			break;
2185 		case 2:
2186 			if (track->separate_cube) {
2187 				ret = r100_cs_track_cube(rdev, track, u);
2188 				if (ret)
2189 					return ret;
2190 			} else
2191 				size *= 6;
2192 			break;
2193 		default:
2194 			DRM_ERROR("Invalid texture coordinate type %u for unit "
2195 				  "%u\n", track->textures[u].tex_coord_type, u);
2196 			return -EINVAL;
2197 		}
2198 		if (size > radeon_bo_size(robj)) {
2199 			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2200 				  "%lu\n", u, size, radeon_bo_size(robj));
2201 			r100_cs_track_texture_print(&track->textures[u]);
2202 			return -EINVAL;
2203 		}
2204 	}
2205 	return 0;
2206 }
2207 
2208 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2209 {
2210 	unsigned i;
2211 	unsigned long size;
2212 	unsigned prim_walk;
2213 	unsigned nverts;
2214 	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2215 
2216 	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2217 	    !track->blend_read_enable)
2218 		num_cb = 0;
2219 
2220 	for (i = 0; i < num_cb; i++) {
2221 		if (track->cb[i].robj == NULL) {
2222 			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2223 			return -EINVAL;
2224 		}
2225 		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2226 		size += track->cb[i].offset;
2227 		if (size > radeon_bo_size(track->cb[i].robj)) {
2228 			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2229 				  "(need %lu have %lu) !\n", i, size,
2230 				  radeon_bo_size(track->cb[i].robj));
2231 			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2232 				  i, track->cb[i].pitch, track->cb[i].cpp,
2233 				  track->cb[i].offset, track->maxy);
2234 			return -EINVAL;
2235 		}
2236 	}
2237 	track->cb_dirty = false;
2238 
2239 	if (track->zb_dirty && track->z_enabled) {
2240 		if (track->zb.robj == NULL) {
2241 			DRM_ERROR("[drm] No buffer for z buffer !\n");
2242 			return -EINVAL;
2243 		}
2244 		size = track->zb.pitch * track->zb.cpp * track->maxy;
2245 		size += track->zb.offset;
2246 		if (size > radeon_bo_size(track->zb.robj)) {
2247 			DRM_ERROR("[drm] Buffer too small for z buffer "
2248 				  "(need %lu have %lu) !\n", size,
2249 				  radeon_bo_size(track->zb.robj));
2250 			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2251 				  track->zb.pitch, track->zb.cpp,
2252 				  track->zb.offset, track->maxy);
2253 			return -EINVAL;
2254 		}
2255 	}
2256 	track->zb_dirty = false;
2257 
2258 	if (track->aa_dirty && track->aaresolve) {
2259 		if (track->aa.robj == NULL) {
2260 			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2261 			return -EINVAL;
2262 		}
2263 		/* I believe the format comes from colorbuffer0. */
2264 		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2265 		size += track->aa.offset;
2266 		if (size > radeon_bo_size(track->aa.robj)) {
2267 			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2268 				  "(need %lu have %lu) !\n", i, size,
2269 				  radeon_bo_size(track->aa.robj));
2270 			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2271 				  i, track->aa.pitch, track->cb[0].cpp,
2272 				  track->aa.offset, track->maxy);
2273 			return -EINVAL;
2274 		}
2275 	}
2276 	track->aa_dirty = false;
2277 
2278 	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2279 	if (track->vap_vf_cntl & (1 << 14)) {
2280 		nverts = track->vap_alt_nverts;
2281 	} else {
2282 		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2283 	}
2284 	switch (prim_walk) {
2285 	case 1:
2286 		for (i = 0; i < track->num_arrays; i++) {
2287 			size = track->arrays[i].esize * track->max_indx * 4;
2288 			if (track->arrays[i].robj == NULL) {
2289 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2290 					  "bound\n", prim_walk, i);
2291 				return -EINVAL;
2292 			}
2293 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2294 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2295 					"need %lu dwords have %lu dwords\n",
2296 					prim_walk, i, size >> 2,
2297 					radeon_bo_size(track->arrays[i].robj)
2298 					>> 2);
2299 				DRM_ERROR("Max indices %u\n", track->max_indx);
2300 				return -EINVAL;
2301 			}
2302 		}
2303 		break;
2304 	case 2:
2305 		for (i = 0; i < track->num_arrays; i++) {
2306 			size = track->arrays[i].esize * (nverts - 1) * 4;
2307 			if (track->arrays[i].robj == NULL) {
2308 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2309 					  "bound\n", prim_walk, i);
2310 				return -EINVAL;
2311 			}
2312 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2313 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2314 					"need %lu dwords have %lu dwords\n",
2315 					prim_walk, i, size >> 2,
2316 					radeon_bo_size(track->arrays[i].robj)
2317 					>> 2);
2318 				return -EINVAL;
2319 			}
2320 		}
2321 		break;
2322 	case 3:
2323 		size = track->vtx_size * nverts;
2324 		if (size != track->immd_dwords) {
2325 			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2326 				  track->immd_dwords, size);
2327 			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2328 				  nverts, track->vtx_size);
2329 			return -EINVAL;
2330 		}
2331 		break;
2332 	default:
2333 		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2334 			  prim_walk);
2335 		return -EINVAL;
2336 	}
2337 
2338 	if (track->tex_dirty) {
2339 		track->tex_dirty = false;
2340 		return r100_cs_track_texture_check(rdev, track);
2341 	}
2342 	return 0;
2343 }
2344 
2345 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2346 {
2347 	unsigned i, face;
2348 
2349 	track->cb_dirty = true;
2350 	track->zb_dirty = true;
2351 	track->tex_dirty = true;
2352 	track->aa_dirty = true;
2353 
2354 	if (rdev->family < CHIP_R300) {
2355 		track->num_cb = 1;
2356 		if (rdev->family <= CHIP_RS200)
2357 			track->num_texture = 3;
2358 		else
2359 			track->num_texture = 6;
2360 		track->maxy = 2048;
2361 		track->separate_cube = 1;
2362 	} else {
2363 		track->num_cb = 4;
2364 		track->num_texture = 16;
2365 		track->maxy = 4096;
2366 		track->separate_cube = 0;
2367 		track->aaresolve = false;
2368 		track->aa.robj = NULL;
2369 	}
2370 
2371 	for (i = 0; i < track->num_cb; i++) {
2372 		track->cb[i].robj = NULL;
2373 		track->cb[i].pitch = 8192;
2374 		track->cb[i].cpp = 16;
2375 		track->cb[i].offset = 0;
2376 	}
2377 	track->z_enabled = true;
2378 	track->zb.robj = NULL;
2379 	track->zb.pitch = 8192;
2380 	track->zb.cpp = 4;
2381 	track->zb.offset = 0;
2382 	track->vtx_size = 0x7F;
2383 	track->immd_dwords = 0xFFFFFFFFUL;
2384 	track->num_arrays = 11;
2385 	track->max_indx = 0x00FFFFFFUL;
2386 	for (i = 0; i < track->num_arrays; i++) {
2387 		track->arrays[i].robj = NULL;
2388 		track->arrays[i].esize = 0x7F;
2389 	}
2390 	for (i = 0; i < track->num_texture; i++) {
2391 		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2392 		track->textures[i].pitch = 16536;
2393 		track->textures[i].width = 16536;
2394 		track->textures[i].height = 16536;
2395 		track->textures[i].width_11 = 1 << 11;
2396 		track->textures[i].height_11 = 1 << 11;
2397 		track->textures[i].num_levels = 12;
2398 		if (rdev->family <= CHIP_RS200) {
2399 			track->textures[i].tex_coord_type = 0;
2400 			track->textures[i].txdepth = 0;
2401 		} else {
2402 			track->textures[i].txdepth = 16;
2403 			track->textures[i].tex_coord_type = 1;
2404 		}
2405 		track->textures[i].cpp = 64;
2406 		track->textures[i].robj = NULL;
2407 		/* CS IB emission code makes sure texture unit are disabled */
2408 		track->textures[i].enabled = false;
2409 		track->textures[i].lookup_disable = false;
2410 		track->textures[i].roundup_w = true;
2411 		track->textures[i].roundup_h = true;
2412 		if (track->separate_cube)
2413 			for (face = 0; face < 5; face++) {
2414 				track->textures[i].cube_info[face].robj = NULL;
2415 				track->textures[i].cube_info[face].width = 16536;
2416 				track->textures[i].cube_info[face].height = 16536;
2417 				track->textures[i].cube_info[face].offset = 0;
2418 			}
2419 	}
2420 }
2421 
2422 /*
2423  * Global GPU functions
2424  */
2425 static void r100_errata(struct radeon_device *rdev)
2426 {
2427 	rdev->pll_errata = 0;
2428 
2429 	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2430 		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2431 	}
2432 
2433 	if (rdev->family == CHIP_RV100 ||
2434 	    rdev->family == CHIP_RS100 ||
2435 	    rdev->family == CHIP_RS200) {
2436 		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2437 	}
2438 }
2439 
2440 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2441 {
2442 	unsigned i;
2443 	uint32_t tmp;
2444 
2445 	for (i = 0; i < rdev->usec_timeout; i++) {
2446 		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2447 		if (tmp >= n) {
2448 			return 0;
2449 		}
2450 		DRM_UDELAY(1);
2451 	}
2452 	return -1;
2453 }
2454 
2455 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2456 {
2457 	unsigned i;
2458 	uint32_t tmp;
2459 
2460 	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2461 		printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
2462 		       " Bad things might happen.\n");
2463 	}
2464 	for (i = 0; i < rdev->usec_timeout; i++) {
2465 		tmp = RREG32(RADEON_RBBM_STATUS);
2466 		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2467 			return 0;
2468 		}
2469 		DRM_UDELAY(1);
2470 	}
2471 	return -1;
2472 }
2473 
2474 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2475 {
2476 	unsigned i;
2477 	uint32_t tmp;
2478 
2479 	for (i = 0; i < rdev->usec_timeout; i++) {
2480 		/* read MC_STATUS */
2481 		tmp = RREG32(RADEON_MC_STATUS);
2482 		if (tmp & RADEON_MC_IDLE) {
2483 			return 0;
2484 		}
2485 		DRM_UDELAY(1);
2486 	}
2487 	return -1;
2488 }
2489 
2490 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2491 {
2492 	u32 rbbm_status;
2493 
2494 	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2495 	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2496 		radeon_ring_lockup_update(ring);
2497 		return false;
2498 	}
2499 	/* force CP activities */
2500 	radeon_ring_force_activity(rdev, ring);
2501 	return radeon_ring_test_lockup(rdev, ring);
2502 }
2503 
2504 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2505 void r100_enable_bm(struct radeon_device *rdev)
2506 {
2507 	uint32_t tmp;
2508 	/* Enable bus mastering */
2509 	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2510 	WREG32(RADEON_BUS_CNTL, tmp);
2511 }
2512 
2513 void r100_bm_disable(struct radeon_device *rdev)
2514 {
2515 	u32 tmp;
2516 
2517 	/* disable bus mastering */
2518 	tmp = RREG32(R_000030_BUS_CNTL);
2519 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2520 	mdelay(1);
2521 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2522 	mdelay(1);
2523 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2524 	tmp = RREG32(RADEON_BUS_CNTL);
2525 	mdelay(1);
2526 	pci_clear_master(rdev->pdev);
2527 	mdelay(1);
2528 }
2529 
2530 int r100_asic_reset(struct radeon_device *rdev)
2531 {
2532 	struct r100_mc_save save;
2533 	u32 status, tmp;
2534 	int ret = 0;
2535 
2536 	status = RREG32(R_000E40_RBBM_STATUS);
2537 	if (!G_000E40_GUI_ACTIVE(status)) {
2538 		return 0;
2539 	}
2540 	r100_mc_stop(rdev, &save);
2541 	status = RREG32(R_000E40_RBBM_STATUS);
2542 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2543 	/* stop CP */
2544 	WREG32(RADEON_CP_CSQ_CNTL, 0);
2545 	tmp = RREG32(RADEON_CP_RB_CNTL);
2546 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2547 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2548 	WREG32(RADEON_CP_RB_WPTR, 0);
2549 	WREG32(RADEON_CP_RB_CNTL, tmp);
2550 	/* save PCI state */
2551 	pci_save_state(rdev->pdev);
2552 	/* disable bus mastering */
2553 	r100_bm_disable(rdev);
2554 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2555 					S_0000F0_SOFT_RESET_RE(1) |
2556 					S_0000F0_SOFT_RESET_PP(1) |
2557 					S_0000F0_SOFT_RESET_RB(1));
2558 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2559 	mdelay(500);
2560 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2561 	mdelay(1);
2562 	status = RREG32(R_000E40_RBBM_STATUS);
2563 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2564 	/* reset CP */
2565 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2566 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2567 	mdelay(500);
2568 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2569 	mdelay(1);
2570 	status = RREG32(R_000E40_RBBM_STATUS);
2571 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2572 	/* restore PCI & busmastering */
2573 	pci_restore_state(rdev->pdev);
2574 	r100_enable_bm(rdev);
2575 	/* Check if GPU is idle */
2576 	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2577 		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2578 		dev_err(rdev->dev, "failed to reset GPU\n");
2579 		ret = -1;
2580 	} else
2581 		dev_info(rdev->dev, "GPU reset succeed\n");
2582 	r100_mc_resume(rdev, &save);
2583 	return ret;
2584 }
2585 
2586 void r100_set_common_regs(struct radeon_device *rdev)
2587 {
2588 	struct drm_device *dev = rdev->ddev;
2589 	bool force_dac2 = false;
2590 	u32 tmp;
2591 
2592 	/* set these so they don't interfere with anything */
2593 	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2594 	WREG32(RADEON_SUBPIC_CNTL, 0);
2595 	WREG32(RADEON_VIPH_CONTROL, 0);
2596 	WREG32(RADEON_I2C_CNTL_1, 0);
2597 	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2598 	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2599 	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2600 
2601 	/* always set up dac2 on rn50 and some rv100 as lots
2602 	 * of servers seem to wire it up to a VGA port but
2603 	 * don't report it in the bios connector
2604 	 * table.
2605 	 */
2606 	switch (dev->pdev->device) {
2607 		/* RN50 */
2608 	case 0x515e:
2609 	case 0x5969:
2610 		force_dac2 = true;
2611 		break;
2612 		/* RV100*/
2613 	case 0x5159:
2614 	case 0x515a:
2615 		/* DELL triple head servers */
2616 		if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2617 		    ((dev->pdev->subsystem_device == 0x016c) ||
2618 		     (dev->pdev->subsystem_device == 0x016d) ||
2619 		     (dev->pdev->subsystem_device == 0x016e) ||
2620 		     (dev->pdev->subsystem_device == 0x016f) ||
2621 		     (dev->pdev->subsystem_device == 0x0170) ||
2622 		     (dev->pdev->subsystem_device == 0x017d) ||
2623 		     (dev->pdev->subsystem_device == 0x017e) ||
2624 		     (dev->pdev->subsystem_device == 0x0183) ||
2625 		     (dev->pdev->subsystem_device == 0x018a) ||
2626 		     (dev->pdev->subsystem_device == 0x019a)))
2627 			force_dac2 = true;
2628 		break;
2629 	}
2630 
2631 	if (force_dac2) {
2632 		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2633 		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2634 		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2635 
2636 		/* For CRT on DAC2, don't turn it on if BIOS didn't
2637 		   enable it, even it's detected.
2638 		*/
2639 
2640 		/* force it to crtc0 */
2641 		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2642 		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2643 		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2644 
2645 		/* set up the TV DAC */
2646 		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2647 				 RADEON_TV_DAC_STD_MASK |
2648 				 RADEON_TV_DAC_RDACPD |
2649 				 RADEON_TV_DAC_GDACPD |
2650 				 RADEON_TV_DAC_BDACPD |
2651 				 RADEON_TV_DAC_BGADJ_MASK |
2652 				 RADEON_TV_DAC_DACADJ_MASK);
2653 		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2654 				RADEON_TV_DAC_NHOLD |
2655 				RADEON_TV_DAC_STD_PS2 |
2656 				(0x58 << 16));
2657 
2658 		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2659 		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2660 		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2661 	}
2662 
2663 	/* switch PM block to ACPI mode */
2664 	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2665 	tmp &= ~RADEON_PM_MODE_SEL;
2666 	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2667 
2668 }
2669 
2670 /*
2671  * VRAM info
2672  */
2673 static void r100_vram_get_type(struct radeon_device *rdev)
2674 {
2675 	uint32_t tmp;
2676 
2677 	rdev->mc.vram_is_ddr = false;
2678 	if (rdev->flags & RADEON_IS_IGP)
2679 		rdev->mc.vram_is_ddr = true;
2680 	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2681 		rdev->mc.vram_is_ddr = true;
2682 	if ((rdev->family == CHIP_RV100) ||
2683 	    (rdev->family == CHIP_RS100) ||
2684 	    (rdev->family == CHIP_RS200)) {
2685 		tmp = RREG32(RADEON_MEM_CNTL);
2686 		if (tmp & RV100_HALF_MODE) {
2687 			rdev->mc.vram_width = 32;
2688 		} else {
2689 			rdev->mc.vram_width = 64;
2690 		}
2691 		if (rdev->flags & RADEON_SINGLE_CRTC) {
2692 			rdev->mc.vram_width /= 4;
2693 			rdev->mc.vram_is_ddr = true;
2694 		}
2695 	} else if (rdev->family <= CHIP_RV280) {
2696 		tmp = RREG32(RADEON_MEM_CNTL);
2697 		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2698 			rdev->mc.vram_width = 128;
2699 		} else {
2700 			rdev->mc.vram_width = 64;
2701 		}
2702 	} else {
2703 		/* newer IGPs */
2704 		rdev->mc.vram_width = 128;
2705 	}
2706 }
2707 
2708 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2709 {
2710 	u32 aper_size;
2711 	u8 byte;
2712 
2713 	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2714 
2715 	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2716 	 * that is has the 2nd generation multifunction PCI interface
2717 	 */
2718 	if (rdev->family == CHIP_RV280 ||
2719 	    rdev->family >= CHIP_RV350) {
2720 		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2721 		       ~RADEON_HDP_APER_CNTL);
2722 		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2723 		return aper_size * 2;
2724 	}
2725 
2726 	/* Older cards have all sorts of funny issues to deal with. First
2727 	 * check if it's a multifunction card by reading the PCI config
2728 	 * header type... Limit those to one aperture size
2729 	 */
2730 	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2731 	if (byte & 0x80) {
2732 		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2733 		DRM_INFO("Limiting VRAM to one aperture\n");
2734 		return aper_size;
2735 	}
2736 
2737 	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2738 	 * have set it up. We don't write this as it's broken on some ASICs but
2739 	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2740 	 */
2741 	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2742 		return aper_size * 2;
2743 	return aper_size;
2744 }
2745 
2746 void r100_vram_init_sizes(struct radeon_device *rdev)
2747 {
2748 	u64 config_aper_size;
2749 
2750 	/* work out accessible VRAM */
2751 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2752 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2753 	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2754 	/* FIXME we don't use the second aperture yet when we could use it */
2755 	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2756 		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2757 	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2758 	if (rdev->flags & RADEON_IS_IGP) {
2759 		uint32_t tom;
2760 		/* read NB_TOM to get the amount of ram stolen for the GPU */
2761 		tom = RREG32(RADEON_NB_TOM);
2762 		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2763 		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2764 		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2765 	} else {
2766 		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2767 		/* Some production boards of m6 will report 0
2768 		 * if it's 8 MB
2769 		 */
2770 		if (rdev->mc.real_vram_size == 0) {
2771 			rdev->mc.real_vram_size = 8192 * 1024;
2772 			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2773 		}
2774 		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2775 		 * Novell bug 204882 + along with lots of ubuntu ones
2776 		 */
2777 		if (rdev->mc.aper_size > config_aper_size)
2778 			config_aper_size = rdev->mc.aper_size;
2779 
2780 		if (config_aper_size > rdev->mc.real_vram_size)
2781 			rdev->mc.mc_vram_size = config_aper_size;
2782 		else
2783 			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2784 	}
2785 }
2786 
2787 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2788 {
2789 	uint32_t temp;
2790 
2791 	temp = RREG32(RADEON_CONFIG_CNTL);
2792 	if (state == false) {
2793 		temp &= ~RADEON_CFG_VGA_RAM_EN;
2794 		temp |= RADEON_CFG_VGA_IO_DIS;
2795 	} else {
2796 		temp &= ~RADEON_CFG_VGA_IO_DIS;
2797 	}
2798 	WREG32(RADEON_CONFIG_CNTL, temp);
2799 }
2800 
2801 static void r100_mc_init(struct radeon_device *rdev)
2802 {
2803 	u64 base;
2804 
2805 	r100_vram_get_type(rdev);
2806 	r100_vram_init_sizes(rdev);
2807 	base = rdev->mc.aper_base;
2808 	if (rdev->flags & RADEON_IS_IGP)
2809 		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2810 	radeon_vram_location(rdev, &rdev->mc, base);
2811 	rdev->mc.gtt_base_align = 0;
2812 	if (!(rdev->flags & RADEON_IS_AGP))
2813 		radeon_gtt_location(rdev, &rdev->mc);
2814 	radeon_update_bandwidth_info(rdev);
2815 }
2816 
2817 
2818 /*
2819  * Indirect registers accessor
2820  */
2821 void r100_pll_errata_after_index(struct radeon_device *rdev)
2822 {
2823 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2824 		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2825 		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2826 	}
2827 }
2828 
2829 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2830 {
2831 	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2832 	 * or the chip could hang on a subsequent access
2833 	 */
2834 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2835 		mdelay(5);
2836 	}
2837 
2838 	/* This function is required to workaround a hardware bug in some (all?)
2839 	 * revisions of the R300.  This workaround should be called after every
2840 	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2841 	 * may not be correct.
2842 	 */
2843 	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2844 		uint32_t save, tmp;
2845 
2846 		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2847 		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2848 		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2849 		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2850 		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2851 	}
2852 }
2853 
2854 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2855 {
2856 	uint32_t data;
2857 
2858 	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2859 	r100_pll_errata_after_index(rdev);
2860 	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2861 	r100_pll_errata_after_data(rdev);
2862 	return data;
2863 }
2864 
2865 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2866 {
2867 	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2868 	r100_pll_errata_after_index(rdev);
2869 	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2870 	r100_pll_errata_after_data(rdev);
2871 }
2872 
2873 static void r100_set_safe_registers(struct radeon_device *rdev)
2874 {
2875 	if (ASIC_IS_RN50(rdev)) {
2876 		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2877 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2878 	} else if (rdev->family < CHIP_R200) {
2879 		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2880 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2881 	} else {
2882 		r200_set_safe_registers(rdev);
2883 	}
2884 }
2885 
2886 /*
2887  * Debugfs info
2888  */
2889 #if defined(CONFIG_DEBUG_FS)
2890 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2891 {
2892 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2893 	struct drm_device *dev = node->minor->dev;
2894 	struct radeon_device *rdev = dev->dev_private;
2895 	uint32_t reg, value;
2896 	unsigned i;
2897 
2898 	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2899 	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2900 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2901 	for (i = 0; i < 64; i++) {
2902 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2903 		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2904 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2905 		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2906 		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2907 	}
2908 	return 0;
2909 }
2910 
2911 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2912 {
2913 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2914 	struct drm_device *dev = node->minor->dev;
2915 	struct radeon_device *rdev = dev->dev_private;
2916 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2917 	uint32_t rdp, wdp;
2918 	unsigned count, i, j;
2919 
2920 	radeon_ring_free_size(rdev, ring);
2921 	rdp = RREG32(RADEON_CP_RB_RPTR);
2922 	wdp = RREG32(RADEON_CP_RB_WPTR);
2923 	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2924 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2925 	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2926 	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2927 	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2928 	seq_printf(m, "%u dwords in ring\n", count);
2929 	for (j = 0; j <= count; j++) {
2930 		i = (rdp + j) & ring->ptr_mask;
2931 		seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2932 	}
2933 	return 0;
2934 }
2935 
2936 
2937 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2938 {
2939 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2940 	struct drm_device *dev = node->minor->dev;
2941 	struct radeon_device *rdev = dev->dev_private;
2942 	uint32_t csq_stat, csq2_stat, tmp;
2943 	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2944 	unsigned i;
2945 
2946 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2947 	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2948 	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2949 	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2950 	r_rptr = (csq_stat >> 0) & 0x3ff;
2951 	r_wptr = (csq_stat >> 10) & 0x3ff;
2952 	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2953 	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2954 	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2955 	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2956 	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2957 	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2958 	seq_printf(m, "Ring rptr %u\n", r_rptr);
2959 	seq_printf(m, "Ring wptr %u\n", r_wptr);
2960 	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2961 	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2962 	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2963 	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2964 	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
2965 	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
2966 	seq_printf(m, "Ring fifo:\n");
2967 	for (i = 0; i < 256; i++) {
2968 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2969 		tmp = RREG32(RADEON_CP_CSQ_DATA);
2970 		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
2971 	}
2972 	seq_printf(m, "Indirect1 fifo:\n");
2973 	for (i = 256; i <= 512; i++) {
2974 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2975 		tmp = RREG32(RADEON_CP_CSQ_DATA);
2976 		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
2977 	}
2978 	seq_printf(m, "Indirect2 fifo:\n");
2979 	for (i = 640; i < ib1_wptr; i++) {
2980 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2981 		tmp = RREG32(RADEON_CP_CSQ_DATA);
2982 		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
2983 	}
2984 	return 0;
2985 }
2986 
2987 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
2988 {
2989 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2990 	struct drm_device *dev = node->minor->dev;
2991 	struct radeon_device *rdev = dev->dev_private;
2992 	uint32_t tmp;
2993 
2994 	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
2995 	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
2996 	tmp = RREG32(RADEON_MC_FB_LOCATION);
2997 	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
2998 	tmp = RREG32(RADEON_BUS_CNTL);
2999 	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3000 	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3001 	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3002 	tmp = RREG32(RADEON_AGP_BASE);
3003 	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3004 	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3005 	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3006 	tmp = RREG32(0x01D0);
3007 	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3008 	tmp = RREG32(RADEON_AIC_LO_ADDR);
3009 	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3010 	tmp = RREG32(RADEON_AIC_HI_ADDR);
3011 	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3012 	tmp = RREG32(0x01E4);
3013 	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3014 	return 0;
3015 }
3016 
3017 static struct drm_info_list r100_debugfs_rbbm_list[] = {
3018 	{"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3019 };
3020 
3021 static struct drm_info_list r100_debugfs_cp_list[] = {
3022 	{"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3023 	{"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3024 };
3025 
3026 static struct drm_info_list r100_debugfs_mc_info_list[] = {
3027 	{"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3028 };
3029 #endif
3030 
3031 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3032 {
3033 #if defined(CONFIG_DEBUG_FS)
3034 	return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3035 #else
3036 	return 0;
3037 #endif
3038 }
3039 
3040 int r100_debugfs_cp_init(struct radeon_device *rdev)
3041 {
3042 #if defined(CONFIG_DEBUG_FS)
3043 	return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3044 #else
3045 	return 0;
3046 #endif
3047 }
3048 
3049 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3050 {
3051 #if defined(CONFIG_DEBUG_FS)
3052 	return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3053 #else
3054 	return 0;
3055 #endif
3056 }
3057 
3058 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3059 			 uint32_t tiling_flags, uint32_t pitch,
3060 			 uint32_t offset, uint32_t obj_size)
3061 {
3062 	int surf_index = reg * 16;
3063 	int flags = 0;
3064 
3065 	if (rdev->family <= CHIP_RS200) {
3066 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3067 				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3068 			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3069 		if (tiling_flags & RADEON_TILING_MACRO)
3070 			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3071 		/* setting pitch to 0 disables tiling */
3072 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3073 				== 0)
3074 			pitch = 0;
3075 	} else if (rdev->family <= CHIP_RV280) {
3076 		if (tiling_flags & (RADEON_TILING_MACRO))
3077 			flags |= R200_SURF_TILE_COLOR_MACRO;
3078 		if (tiling_flags & RADEON_TILING_MICRO)
3079 			flags |= R200_SURF_TILE_COLOR_MICRO;
3080 	} else {
3081 		if (tiling_flags & RADEON_TILING_MACRO)
3082 			flags |= R300_SURF_TILE_MACRO;
3083 		if (tiling_flags & RADEON_TILING_MICRO)
3084 			flags |= R300_SURF_TILE_MICRO;
3085 	}
3086 
3087 	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3088 		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3089 	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3090 		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3091 
3092 	/* r100/r200 divide by 16 */
3093 	if (rdev->family < CHIP_R300)
3094 		flags |= pitch / 16;
3095 	else
3096 		flags |= pitch / 8;
3097 
3098 
3099 	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3100 	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3101 	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3102 	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3103 	return 0;
3104 }
3105 
3106 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3107 {
3108 	int surf_index = reg * 16;
3109 	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3110 }
3111 
3112 void r100_bandwidth_update(struct radeon_device *rdev)
3113 {
3114 	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3115 	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3116 	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
3117 	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3118 	fixed20_12 memtcas_ff[8] = {
3119 		dfixed_init(1),
3120 		dfixed_init(2),
3121 		dfixed_init(3),
3122 		dfixed_init(0),
3123 		dfixed_init_half(1),
3124 		dfixed_init_half(2),
3125 		dfixed_init(0),
3126 	};
3127 	fixed20_12 memtcas_rs480_ff[8] = {
3128 		dfixed_init(0),
3129 		dfixed_init(1),
3130 		dfixed_init(2),
3131 		dfixed_init(3),
3132 		dfixed_init(0),
3133 		dfixed_init_half(1),
3134 		dfixed_init_half(2),
3135 		dfixed_init_half(3),
3136 	};
3137 	fixed20_12 memtcas2_ff[8] = {
3138 		dfixed_init(0),
3139 		dfixed_init(1),
3140 		dfixed_init(2),
3141 		dfixed_init(3),
3142 		dfixed_init(4),
3143 		dfixed_init(5),
3144 		dfixed_init(6),
3145 		dfixed_init(7),
3146 	};
3147 	fixed20_12 memtrbs[8] = {
3148 		dfixed_init(1),
3149 		dfixed_init_half(1),
3150 		dfixed_init(2),
3151 		dfixed_init_half(2),
3152 		dfixed_init(3),
3153 		dfixed_init_half(3),
3154 		dfixed_init(4),
3155 		dfixed_init_half(4)
3156 	};
3157 	fixed20_12 memtrbs_r4xx[8] = {
3158 		dfixed_init(4),
3159 		dfixed_init(5),
3160 		dfixed_init(6),
3161 		dfixed_init(7),
3162 		dfixed_init(8),
3163 		dfixed_init(9),
3164 		dfixed_init(10),
3165 		dfixed_init(11)
3166 	};
3167 	fixed20_12 min_mem_eff;
3168 	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3169 	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3170 	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
3171 		disp_drain_rate2, read_return_rate;
3172 	fixed20_12 time_disp1_drop_priority;
3173 	int c;
3174 	int cur_size = 16;       /* in octawords */
3175 	int critical_point = 0, critical_point2;
3176 /* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3177 	int stop_req, max_stop_req;
3178 	struct drm_display_mode *mode1 = NULL;
3179 	struct drm_display_mode *mode2 = NULL;
3180 	uint32_t pixel_bytes1 = 0;
3181 	uint32_t pixel_bytes2 = 0;
3182 
3183 	radeon_update_display_priority(rdev);
3184 
3185 	if (rdev->mode_info.crtcs[0]->base.enabled) {
3186 		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3187 		pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
3188 	}
3189 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3190 		if (rdev->mode_info.crtcs[1]->base.enabled) {
3191 			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3192 			pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
3193 		}
3194 	}
3195 
3196 	min_mem_eff.full = dfixed_const_8(0);
3197 	/* get modes */
3198 	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3199 		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3200 		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3201 		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3202 		/* check crtc enables */
3203 		if (mode2)
3204 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3205 		if (mode1)
3206 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3207 		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3208 	}
3209 
3210 	/*
3211 	 * determine is there is enough bw for current mode
3212 	 */
3213 	sclk_ff = rdev->pm.sclk;
3214 	mclk_ff = rdev->pm.mclk;
3215 
3216 	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3217 	temp_ff.full = dfixed_const(temp);
3218 	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3219 
3220 	pix_clk.full = 0;
3221 	pix_clk2.full = 0;
3222 	peak_disp_bw.full = 0;
3223 	if (mode1) {
3224 		temp_ff.full = dfixed_const(1000);
3225 		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3226 		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3227 		temp_ff.full = dfixed_const(pixel_bytes1);
3228 		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3229 	}
3230 	if (mode2) {
3231 		temp_ff.full = dfixed_const(1000);
3232 		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3233 		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3234 		temp_ff.full = dfixed_const(pixel_bytes2);
3235 		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3236 	}
3237 
3238 	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3239 	if (peak_disp_bw.full >= mem_bw.full) {
3240 		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3241 			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3242 	}
3243 
3244 	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3245 	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3246 	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3247 		mem_trcd = ((temp >> 2) & 0x3) + 1;
3248 		mem_trp  = ((temp & 0x3)) + 1;
3249 		mem_tras = ((temp & 0x70) >> 4) + 1;
3250 	} else if (rdev->family == CHIP_R300 ||
3251 		   rdev->family == CHIP_R350) { /* r300, r350 */
3252 		mem_trcd = (temp & 0x7) + 1;
3253 		mem_trp = ((temp >> 8) & 0x7) + 1;
3254 		mem_tras = ((temp >> 11) & 0xf) + 4;
3255 	} else if (rdev->family == CHIP_RV350 ||
3256 		   rdev->family <= CHIP_RV380) {
3257 		/* rv3x0 */
3258 		mem_trcd = (temp & 0x7) + 3;
3259 		mem_trp = ((temp >> 8) & 0x7) + 3;
3260 		mem_tras = ((temp >> 11) & 0xf) + 6;
3261 	} else if (rdev->family == CHIP_R420 ||
3262 		   rdev->family == CHIP_R423 ||
3263 		   rdev->family == CHIP_RV410) {
3264 		/* r4xx */
3265 		mem_trcd = (temp & 0xf) + 3;
3266 		if (mem_trcd > 15)
3267 			mem_trcd = 15;
3268 		mem_trp = ((temp >> 8) & 0xf) + 3;
3269 		if (mem_trp > 15)
3270 			mem_trp = 15;
3271 		mem_tras = ((temp >> 12) & 0x1f) + 6;
3272 		if (mem_tras > 31)
3273 			mem_tras = 31;
3274 	} else { /* RV200, R200 */
3275 		mem_trcd = (temp & 0x7) + 1;
3276 		mem_trp = ((temp >> 8) & 0x7) + 1;
3277 		mem_tras = ((temp >> 12) & 0xf) + 4;
3278 	}
3279 	/* convert to FF */
3280 	trcd_ff.full = dfixed_const(mem_trcd);
3281 	trp_ff.full = dfixed_const(mem_trp);
3282 	tras_ff.full = dfixed_const(mem_tras);
3283 
3284 	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3285 	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3286 	data = (temp & (7 << 20)) >> 20;
3287 	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3288 		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3289 			tcas_ff = memtcas_rs480_ff[data];
3290 		else
3291 			tcas_ff = memtcas_ff[data];
3292 	} else
3293 		tcas_ff = memtcas2_ff[data];
3294 
3295 	if (rdev->family == CHIP_RS400 ||
3296 	    rdev->family == CHIP_RS480) {
3297 		/* extra cas latency stored in bits 23-25 0-4 clocks */
3298 		data = (temp >> 23) & 0x7;
3299 		if (data < 5)
3300 			tcas_ff.full += dfixed_const(data);
3301 	}
3302 
3303 	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3304 		/* on the R300, Tcas is included in Trbs.
3305 		 */
3306 		temp = RREG32(RADEON_MEM_CNTL);
3307 		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3308 		if (data == 1) {
3309 			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3310 				temp = RREG32(R300_MC_IND_INDEX);
3311 				temp &= ~R300_MC_IND_ADDR_MASK;
3312 				temp |= R300_MC_READ_CNTL_CD_mcind;
3313 				WREG32(R300_MC_IND_INDEX, temp);
3314 				temp = RREG32(R300_MC_IND_DATA);
3315 				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3316 			} else {
3317 				temp = RREG32(R300_MC_READ_CNTL_AB);
3318 				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3319 			}
3320 		} else {
3321 			temp = RREG32(R300_MC_READ_CNTL_AB);
3322 			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3323 		}
3324 		if (rdev->family == CHIP_RV410 ||
3325 		    rdev->family == CHIP_R420 ||
3326 		    rdev->family == CHIP_R423)
3327 			trbs_ff = memtrbs_r4xx[data];
3328 		else
3329 			trbs_ff = memtrbs[data];
3330 		tcas_ff.full += trbs_ff.full;
3331 	}
3332 
3333 	sclk_eff_ff.full = sclk_ff.full;
3334 
3335 	if (rdev->flags & RADEON_IS_AGP) {
3336 		fixed20_12 agpmode_ff;
3337 		agpmode_ff.full = dfixed_const(radeon_agpmode);
3338 		temp_ff.full = dfixed_const_666(16);
3339 		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3340 	}
3341 	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3342 
3343 	if (ASIC_IS_R300(rdev)) {
3344 		sclk_delay_ff.full = dfixed_const(250);
3345 	} else {
3346 		if ((rdev->family == CHIP_RV100) ||
3347 		    rdev->flags & RADEON_IS_IGP) {
3348 			if (rdev->mc.vram_is_ddr)
3349 				sclk_delay_ff.full = dfixed_const(41);
3350 			else
3351 				sclk_delay_ff.full = dfixed_const(33);
3352 		} else {
3353 			if (rdev->mc.vram_width == 128)
3354 				sclk_delay_ff.full = dfixed_const(57);
3355 			else
3356 				sclk_delay_ff.full = dfixed_const(41);
3357 		}
3358 	}
3359 
3360 	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3361 
3362 	if (rdev->mc.vram_is_ddr) {
3363 		if (rdev->mc.vram_width == 32) {
3364 			k1.full = dfixed_const(40);
3365 			c  = 3;
3366 		} else {
3367 			k1.full = dfixed_const(20);
3368 			c  = 1;
3369 		}
3370 	} else {
3371 		k1.full = dfixed_const(40);
3372 		c  = 3;
3373 	}
3374 
3375 	temp_ff.full = dfixed_const(2);
3376 	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3377 	temp_ff.full = dfixed_const(c);
3378 	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3379 	temp_ff.full = dfixed_const(4);
3380 	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3381 	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3382 	mc_latency_mclk.full += k1.full;
3383 
3384 	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3385 	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3386 
3387 	/*
3388 	  HW cursor time assuming worst case of full size colour cursor.
3389 	*/
3390 	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3391 	temp_ff.full += trcd_ff.full;
3392 	if (temp_ff.full < tras_ff.full)
3393 		temp_ff.full = tras_ff.full;
3394 	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3395 
3396 	temp_ff.full = dfixed_const(cur_size);
3397 	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3398 	/*
3399 	  Find the total latency for the display data.
3400 	*/
3401 	disp_latency_overhead.full = dfixed_const(8);
3402 	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3403 	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3404 	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3405 
3406 	if (mc_latency_mclk.full > mc_latency_sclk.full)
3407 		disp_latency.full = mc_latency_mclk.full;
3408 	else
3409 		disp_latency.full = mc_latency_sclk.full;
3410 
3411 	/* setup Max GRPH_STOP_REQ default value */
3412 	if (ASIC_IS_RV100(rdev))
3413 		max_stop_req = 0x5c;
3414 	else
3415 		max_stop_req = 0x7c;
3416 
3417 	if (mode1) {
3418 		/*  CRTC1
3419 		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3420 		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3421 		*/
3422 		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3423 
3424 		if (stop_req > max_stop_req)
3425 			stop_req = max_stop_req;
3426 
3427 		/*
3428 		  Find the drain rate of the display buffer.
3429 		*/
3430 		temp_ff.full = dfixed_const((16/pixel_bytes1));
3431 		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3432 
3433 		/*
3434 		  Find the critical point of the display buffer.
3435 		*/
3436 		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3437 		crit_point_ff.full += dfixed_const_half(0);
3438 
3439 		critical_point = dfixed_trunc(crit_point_ff);
3440 
3441 		if (rdev->disp_priority == 2) {
3442 			critical_point = 0;
3443 		}
3444 
3445 		/*
3446 		  The critical point should never be above max_stop_req-4.  Setting
3447 		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3448 		*/
3449 		if (max_stop_req - critical_point < 4)
3450 			critical_point = 0;
3451 
3452 		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3453 			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3454 			critical_point = 0x10;
3455 		}
3456 
3457 		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3458 		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3459 		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3460 		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3461 		if ((rdev->family == CHIP_R350) &&
3462 		    (stop_req > 0x15)) {
3463 			stop_req -= 0x10;
3464 		}
3465 		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3466 		temp |= RADEON_GRPH_BUFFER_SIZE;
3467 		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3468 			  RADEON_GRPH_CRITICAL_AT_SOF |
3469 			  RADEON_GRPH_STOP_CNTL);
3470 		/*
3471 		  Write the result into the register.
3472 		*/
3473 		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3474 						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3475 
3476 #if 0
3477 		if ((rdev->family == CHIP_RS400) ||
3478 		    (rdev->family == CHIP_RS480)) {
3479 			/* attempt to program RS400 disp regs correctly ??? */
3480 			temp = RREG32(RS400_DISP1_REG_CNTL);
3481 			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3482 				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3483 			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3484 						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3485 						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3486 			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3487 			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3488 				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3489 			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3490 						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3491 						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3492 		}
3493 #endif
3494 
3495 		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3496 			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3497 			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3498 	}
3499 
3500 	if (mode2) {
3501 		u32 grph2_cntl;
3502 		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3503 
3504 		if (stop_req > max_stop_req)
3505 			stop_req = max_stop_req;
3506 
3507 		/*
3508 		  Find the drain rate of the display buffer.
3509 		*/
3510 		temp_ff.full = dfixed_const((16/pixel_bytes2));
3511 		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3512 
3513 		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3514 		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3515 		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3516 		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3517 		if ((rdev->family == CHIP_R350) &&
3518 		    (stop_req > 0x15)) {
3519 			stop_req -= 0x10;
3520 		}
3521 		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3522 		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3523 		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3524 			  RADEON_GRPH_CRITICAL_AT_SOF |
3525 			  RADEON_GRPH_STOP_CNTL);
3526 
3527 		if ((rdev->family == CHIP_RS100) ||
3528 		    (rdev->family == CHIP_RS200))
3529 			critical_point2 = 0;
3530 		else {
3531 			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3532 			temp_ff.full = dfixed_const(temp);
3533 			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3534 			if (sclk_ff.full < temp_ff.full)
3535 				temp_ff.full = sclk_ff.full;
3536 
3537 			read_return_rate.full = temp_ff.full;
3538 
3539 			if (mode1) {
3540 				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3541 				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3542 			} else {
3543 				time_disp1_drop_priority.full = 0;
3544 			}
3545 			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3546 			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3547 			crit_point_ff.full += dfixed_const_half(0);
3548 
3549 			critical_point2 = dfixed_trunc(crit_point_ff);
3550 
3551 			if (rdev->disp_priority == 2) {
3552 				critical_point2 = 0;
3553 			}
3554 
3555 			if (max_stop_req - critical_point2 < 4)
3556 				critical_point2 = 0;
3557 
3558 		}
3559 
3560 		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3561 			/* some R300 cards have problem with this set to 0 */
3562 			critical_point2 = 0x10;
3563 		}
3564 
3565 		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3566 						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3567 
3568 		if ((rdev->family == CHIP_RS400) ||
3569 		    (rdev->family == CHIP_RS480)) {
3570 #if 0
3571 			/* attempt to program RS400 disp2 regs correctly ??? */
3572 			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3573 			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3574 				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3575 			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3576 						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3577 						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3578 			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3579 			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3580 				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3581 			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3582 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3583 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3584 #endif
3585 			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3586 			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3587 			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3588 			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3589 		}
3590 
3591 		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3592 			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3593 	}
3594 }
3595 
3596 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3597 {
3598 	uint32_t scratch;
3599 	uint32_t tmp = 0;
3600 	unsigned i;
3601 	int r;
3602 
3603 	r = radeon_scratch_get(rdev, &scratch);
3604 	if (r) {
3605 		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3606 		return r;
3607 	}
3608 	WREG32(scratch, 0xCAFEDEAD);
3609 	r = radeon_ring_lock(rdev, ring, 2);
3610 	if (r) {
3611 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3612 		radeon_scratch_free(rdev, scratch);
3613 		return r;
3614 	}
3615 	radeon_ring_write(ring, PACKET0(scratch, 0));
3616 	radeon_ring_write(ring, 0xDEADBEEF);
3617 	radeon_ring_unlock_commit(rdev, ring);
3618 	for (i = 0; i < rdev->usec_timeout; i++) {
3619 		tmp = RREG32(scratch);
3620 		if (tmp == 0xDEADBEEF) {
3621 			break;
3622 		}
3623 		DRM_UDELAY(1);
3624 	}
3625 	if (i < rdev->usec_timeout) {
3626 		DRM_INFO("ring test succeeded in %d usecs\n", i);
3627 	} else {
3628 		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3629 			  scratch, tmp);
3630 		r = -EINVAL;
3631 	}
3632 	radeon_scratch_free(rdev, scratch);
3633 	return r;
3634 }
3635 
3636 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3637 {
3638 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3639 
3640 	if (ring->rptr_save_reg) {
3641 		u32 next_rptr = ring->wptr + 2 + 3;
3642 		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3643 		radeon_ring_write(ring, next_rptr);
3644 	}
3645 
3646 	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3647 	radeon_ring_write(ring, ib->gpu_addr);
3648 	radeon_ring_write(ring, ib->length_dw);
3649 }
3650 
3651 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3652 {
3653 	struct radeon_ib ib;
3654 	uint32_t scratch;
3655 	uint32_t tmp = 0;
3656 	unsigned i;
3657 	int r;
3658 
3659 	r = radeon_scratch_get(rdev, &scratch);
3660 	if (r) {
3661 		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3662 		return r;
3663 	}
3664 	WREG32(scratch, 0xCAFEDEAD);
3665 	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3666 	if (r) {
3667 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3668 		goto free_scratch;
3669 	}
3670 	ib.ptr[0] = PACKET0(scratch, 0);
3671 	ib.ptr[1] = 0xDEADBEEF;
3672 	ib.ptr[2] = PACKET2(0);
3673 	ib.ptr[3] = PACKET2(0);
3674 	ib.ptr[4] = PACKET2(0);
3675 	ib.ptr[5] = PACKET2(0);
3676 	ib.ptr[6] = PACKET2(0);
3677 	ib.ptr[7] = PACKET2(0);
3678 	ib.length_dw = 8;
3679 	r = radeon_ib_schedule(rdev, &ib, NULL);
3680 	if (r) {
3681 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3682 		goto free_ib;
3683 	}
3684 	r = radeon_fence_wait(ib.fence, false);
3685 	if (r) {
3686 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3687 		goto free_ib;
3688 	}
3689 	for (i = 0; i < rdev->usec_timeout; i++) {
3690 		tmp = RREG32(scratch);
3691 		if (tmp == 0xDEADBEEF) {
3692 			break;
3693 		}
3694 		DRM_UDELAY(1);
3695 	}
3696 	if (i < rdev->usec_timeout) {
3697 		DRM_INFO("ib test succeeded in %u usecs\n", i);
3698 	} else {
3699 		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3700 			  scratch, tmp);
3701 		r = -EINVAL;
3702 	}
3703 free_ib:
3704 	radeon_ib_free(rdev, &ib);
3705 free_scratch:
3706 	radeon_scratch_free(rdev, scratch);
3707 	return r;
3708 }
3709 
3710 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3711 {
3712 	/* Shutdown CP we shouldn't need to do that but better be safe than
3713 	 * sorry
3714 	 */
3715 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3716 	WREG32(R_000740_CP_CSQ_CNTL, 0);
3717 
3718 	/* Save few CRTC registers */
3719 	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3720 	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3721 	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3722 	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3723 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3724 		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3725 		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3726 	}
3727 
3728 	/* Disable VGA aperture access */
3729 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3730 	/* Disable cursor, overlay, crtc */
3731 	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3732 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3733 					S_000054_CRTC_DISPLAY_DIS(1));
3734 	WREG32(R_000050_CRTC_GEN_CNTL,
3735 			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3736 			S_000050_CRTC_DISP_REQ_EN_B(1));
3737 	WREG32(R_000420_OV0_SCALE_CNTL,
3738 		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3739 	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3740 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3741 		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3742 						S_000360_CUR2_LOCK(1));
3743 		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3744 			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3745 			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3746 			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3747 		WREG32(R_000360_CUR2_OFFSET,
3748 			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3749 	}
3750 }
3751 
3752 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3753 {
3754 	/* Update base address for crtc */
3755 	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3756 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3757 		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3758 	}
3759 	/* Restore CRTC registers */
3760 	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3761 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3762 	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3763 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3764 		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3765 	}
3766 }
3767 
3768 void r100_vga_render_disable(struct radeon_device *rdev)
3769 {
3770 	u32 tmp;
3771 
3772 	tmp = RREG8(R_0003C2_GENMO_WT);
3773 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3774 }
3775 
3776 static void r100_debugfs(struct radeon_device *rdev)
3777 {
3778 	int r;
3779 
3780 	r = r100_debugfs_mc_info_init(rdev);
3781 	if (r)
3782 		dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3783 }
3784 
3785 static void r100_mc_program(struct radeon_device *rdev)
3786 {
3787 	struct r100_mc_save save;
3788 
3789 	/* Stops all mc clients */
3790 	r100_mc_stop(rdev, &save);
3791 	if (rdev->flags & RADEON_IS_AGP) {
3792 		WREG32(R_00014C_MC_AGP_LOCATION,
3793 			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3794 			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3795 		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3796 		if (rdev->family > CHIP_RV200)
3797 			WREG32(R_00015C_AGP_BASE_2,
3798 				upper_32_bits(rdev->mc.agp_base) & 0xff);
3799 	} else {
3800 		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3801 		WREG32(R_000170_AGP_BASE, 0);
3802 		if (rdev->family > CHIP_RV200)
3803 			WREG32(R_00015C_AGP_BASE_2, 0);
3804 	}
3805 	/* Wait for mc idle */
3806 	if (r100_mc_wait_for_idle(rdev))
3807 		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3808 	/* Program MC, should be a 32bits limited address space */
3809 	WREG32(R_000148_MC_FB_LOCATION,
3810 		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3811 		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3812 	r100_mc_resume(rdev, &save);
3813 }
3814 
3815 static void r100_clock_startup(struct radeon_device *rdev)
3816 {
3817 	u32 tmp;
3818 
3819 	if (radeon_dynclks != -1 && radeon_dynclks)
3820 		radeon_legacy_set_clock_gating(rdev, 1);
3821 	/* We need to force on some of the block */
3822 	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3823 	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3824 	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3825 		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3826 	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3827 }
3828 
3829 static int r100_startup(struct radeon_device *rdev)
3830 {
3831 	int r;
3832 
3833 	/* set common regs */
3834 	r100_set_common_regs(rdev);
3835 	/* program mc */
3836 	r100_mc_program(rdev);
3837 	/* Resume clock */
3838 	r100_clock_startup(rdev);
3839 	/* Initialize GART (initialize after TTM so we can allocate
3840 	 * memory through TTM but finalize after TTM) */
3841 	r100_enable_bm(rdev);
3842 	if (rdev->flags & RADEON_IS_PCI) {
3843 		r = r100_pci_gart_enable(rdev);
3844 		if (r)
3845 			return r;
3846 	}
3847 
3848 	/* allocate wb buffer */
3849 	r = radeon_wb_init(rdev);
3850 	if (r)
3851 		return r;
3852 
3853 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3854 	if (r) {
3855 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3856 		return r;
3857 	}
3858 
3859 	/* Enable IRQ */
3860 	if (!rdev->irq.installed) {
3861 		r = radeon_irq_kms_init(rdev);
3862 		if (r)
3863 			return r;
3864 	}
3865 
3866 	r100_irq_set(rdev);
3867 	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3868 	/* 1M ring buffer */
3869 	r = r100_cp_init(rdev, 1024 * 1024);
3870 	if (r) {
3871 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3872 		return r;
3873 	}
3874 
3875 	r = radeon_ib_pool_init(rdev);
3876 	if (r) {
3877 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3878 		return r;
3879 	}
3880 
3881 	return 0;
3882 }
3883 
3884 int r100_resume(struct radeon_device *rdev)
3885 {
3886 	int r;
3887 
3888 	/* Make sur GART are not working */
3889 	if (rdev->flags & RADEON_IS_PCI)
3890 		r100_pci_gart_disable(rdev);
3891 	/* Resume clock before doing reset */
3892 	r100_clock_startup(rdev);
3893 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3894 	if (radeon_asic_reset(rdev)) {
3895 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3896 			RREG32(R_000E40_RBBM_STATUS),
3897 			RREG32(R_0007C0_CP_STAT));
3898 	}
3899 	/* post */
3900 	radeon_combios_asic_init(rdev->ddev);
3901 	/* Resume clock after posting */
3902 	r100_clock_startup(rdev);
3903 	/* Initialize surface registers */
3904 	radeon_surface_init(rdev);
3905 
3906 	rdev->accel_working = true;
3907 	r = r100_startup(rdev);
3908 	if (r) {
3909 		rdev->accel_working = false;
3910 	}
3911 	return r;
3912 }
3913 
3914 int r100_suspend(struct radeon_device *rdev)
3915 {
3916 	r100_cp_disable(rdev);
3917 	radeon_wb_disable(rdev);
3918 	r100_irq_disable(rdev);
3919 	if (rdev->flags & RADEON_IS_PCI)
3920 		r100_pci_gart_disable(rdev);
3921 	return 0;
3922 }
3923 
3924 void r100_fini(struct radeon_device *rdev)
3925 {
3926 	r100_cp_fini(rdev);
3927 	radeon_wb_fini(rdev);
3928 	radeon_ib_pool_fini(rdev);
3929 	radeon_gem_fini(rdev);
3930 	if (rdev->flags & RADEON_IS_PCI)
3931 		r100_pci_gart_fini(rdev);
3932 	radeon_agp_fini(rdev);
3933 	radeon_irq_kms_fini(rdev);
3934 	radeon_fence_driver_fini(rdev);
3935 	radeon_bo_fini(rdev);
3936 	radeon_atombios_fini(rdev);
3937 	kfree(rdev->bios);
3938 	rdev->bios = NULL;
3939 }
3940 
3941 /*
3942  * Due to how kexec works, it can leave the hw fully initialised when it
3943  * boots the new kernel. However doing our init sequence with the CP and
3944  * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3945  * do some quick sanity checks and restore sane values to avoid this
3946  * problem.
3947  */
3948 void r100_restore_sanity(struct radeon_device *rdev)
3949 {
3950 	u32 tmp;
3951 
3952 	tmp = RREG32(RADEON_CP_CSQ_CNTL);
3953 	if (tmp) {
3954 		WREG32(RADEON_CP_CSQ_CNTL, 0);
3955 	}
3956 	tmp = RREG32(RADEON_CP_RB_CNTL);
3957 	if (tmp) {
3958 		WREG32(RADEON_CP_RB_CNTL, 0);
3959 	}
3960 	tmp = RREG32(RADEON_SCRATCH_UMSK);
3961 	if (tmp) {
3962 		WREG32(RADEON_SCRATCH_UMSK, 0);
3963 	}
3964 }
3965 
3966 int r100_init(struct radeon_device *rdev)
3967 {
3968 	int r;
3969 
3970 	/* Register debugfs file specific to this group of asics */
3971 	r100_debugfs(rdev);
3972 	/* Disable VGA */
3973 	r100_vga_render_disable(rdev);
3974 	/* Initialize scratch registers */
3975 	radeon_scratch_init(rdev);
3976 	/* Initialize surface registers */
3977 	radeon_surface_init(rdev);
3978 	/* sanity check some register to avoid hangs like after kexec */
3979 	r100_restore_sanity(rdev);
3980 	/* TODO: disable VGA need to use VGA request */
3981 	/* BIOS*/
3982 	if (!radeon_get_bios(rdev)) {
3983 		if (ASIC_IS_AVIVO(rdev))
3984 			return -EINVAL;
3985 	}
3986 	if (rdev->is_atom_bios) {
3987 		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
3988 		return -EINVAL;
3989 	} else {
3990 		r = radeon_combios_init(rdev);
3991 		if (r)
3992 			return r;
3993 	}
3994 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3995 	if (radeon_asic_reset(rdev)) {
3996 		dev_warn(rdev->dev,
3997 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3998 			RREG32(R_000E40_RBBM_STATUS),
3999 			RREG32(R_0007C0_CP_STAT));
4000 	}
4001 	/* check if cards are posted or not */
4002 	if (radeon_boot_test_post_card(rdev) == false)
4003 		return -EINVAL;
4004 	/* Set asic errata */
4005 	r100_errata(rdev);
4006 	/* Initialize clocks */
4007 	radeon_get_clock_info(rdev->ddev);
4008 	/* initialize AGP */
4009 	if (rdev->flags & RADEON_IS_AGP) {
4010 		r = radeon_agp_init(rdev);
4011 		if (r) {
4012 			radeon_agp_disable(rdev);
4013 		}
4014 	}
4015 	/* initialize VRAM */
4016 	r100_mc_init(rdev);
4017 	/* Fence driver */
4018 	r = radeon_fence_driver_init(rdev);
4019 	if (r)
4020 		return r;
4021 	/* Memory manager */
4022 	r = radeon_bo_init(rdev);
4023 	if (r)
4024 		return r;
4025 	if (rdev->flags & RADEON_IS_PCI) {
4026 		r = r100_pci_gart_init(rdev);
4027 		if (r)
4028 			return r;
4029 	}
4030 	r100_set_safe_registers(rdev);
4031 
4032 	rdev->accel_working = true;
4033 	r = r100_startup(rdev);
4034 	if (r) {
4035 		/* Somethings want wront with the accel init stop accel */
4036 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4037 		r100_cp_fini(rdev);
4038 		radeon_wb_fini(rdev);
4039 		radeon_ib_pool_fini(rdev);
4040 		radeon_irq_kms_fini(rdev);
4041 		if (rdev->flags & RADEON_IS_PCI)
4042 			r100_pci_gart_fini(rdev);
4043 		rdev->accel_working = false;
4044 	}
4045 	return 0;
4046 }
4047 
4048 uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg,
4049 		      bool always_indirect)
4050 {
4051 	if (reg < rdev->rmmio_size && !always_indirect)
4052 		return readl(((void __iomem *)rdev->rmmio) + reg);
4053 	else {
4054 		unsigned long flags;
4055 		uint32_t ret;
4056 
4057 		spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4058 		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4059 		ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4060 		spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4061 
4062 		return ret;
4063 	}
4064 }
4065 
4066 void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v,
4067 		  bool always_indirect)
4068 {
4069 	if (reg < rdev->rmmio_size && !always_indirect)
4070 		writel(v, ((void __iomem *)rdev->rmmio) + reg);
4071 	else {
4072 		unsigned long flags;
4073 
4074 		spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4075 		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4076 		writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4077 		spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4078 	}
4079 }
4080 
4081 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4082 {
4083 	if (reg < rdev->rio_mem_size)
4084 		return ioread32(rdev->rio_mem + reg);
4085 	else {
4086 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4087 		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4088 	}
4089 }
4090 
4091 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4092 {
4093 	if (reg < rdev->rio_mem_size)
4094 		iowrite32(v, rdev->rio_mem + reg);
4095 	else {
4096 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4097 		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4098 	}
4099 }
4100