xref: /openbmc/linux/drivers/gpu/drm/radeon/r100.c (revision 77d84ff8)
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 bool 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 	return false;
880 }
881 
882 int r100_copy_blit(struct radeon_device *rdev,
883 		   uint64_t src_offset,
884 		   uint64_t dst_offset,
885 		   unsigned num_gpu_pages,
886 		   struct radeon_fence **fence)
887 {
888 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
889 	uint32_t cur_pages;
890 	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
891 	uint32_t pitch;
892 	uint32_t stride_pixels;
893 	unsigned ndw;
894 	int num_loops;
895 	int r = 0;
896 
897 	/* radeon limited to 16k stride */
898 	stride_bytes &= 0x3fff;
899 	/* radeon pitch is /64 */
900 	pitch = stride_bytes / 64;
901 	stride_pixels = stride_bytes / 4;
902 	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
903 
904 	/* Ask for enough room for blit + flush + fence */
905 	ndw = 64 + (10 * num_loops);
906 	r = radeon_ring_lock(rdev, ring, ndw);
907 	if (r) {
908 		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
909 		return -EINVAL;
910 	}
911 	while (num_gpu_pages > 0) {
912 		cur_pages = num_gpu_pages;
913 		if (cur_pages > 8191) {
914 			cur_pages = 8191;
915 		}
916 		num_gpu_pages -= cur_pages;
917 
918 		/* pages are in Y direction - height
919 		   page width in X direction - width */
920 		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
921 		radeon_ring_write(ring,
922 				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
923 				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
924 				  RADEON_GMC_SRC_CLIPPING |
925 				  RADEON_GMC_DST_CLIPPING |
926 				  RADEON_GMC_BRUSH_NONE |
927 				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
928 				  RADEON_GMC_SRC_DATATYPE_COLOR |
929 				  RADEON_ROP3_S |
930 				  RADEON_DP_SRC_SOURCE_MEMORY |
931 				  RADEON_GMC_CLR_CMP_CNTL_DIS |
932 				  RADEON_GMC_WR_MSK_DIS);
933 		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
934 		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
935 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
936 		radeon_ring_write(ring, 0);
937 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
938 		radeon_ring_write(ring, num_gpu_pages);
939 		radeon_ring_write(ring, num_gpu_pages);
940 		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
941 	}
942 	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
943 	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
944 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
945 	radeon_ring_write(ring,
946 			  RADEON_WAIT_2D_IDLECLEAN |
947 			  RADEON_WAIT_HOST_IDLECLEAN |
948 			  RADEON_WAIT_DMA_GUI_IDLE);
949 	if (fence) {
950 		r = radeon_fence_emit(rdev, fence, RADEON_RING_TYPE_GFX_INDEX);
951 	}
952 	radeon_ring_unlock_commit(rdev, ring);
953 	return r;
954 }
955 
956 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
957 {
958 	unsigned i;
959 	u32 tmp;
960 
961 	for (i = 0; i < rdev->usec_timeout; i++) {
962 		tmp = RREG32(R_000E40_RBBM_STATUS);
963 		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
964 			return 0;
965 		}
966 		udelay(1);
967 	}
968 	return -1;
969 }
970 
971 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
972 {
973 	int r;
974 
975 	r = radeon_ring_lock(rdev, ring, 2);
976 	if (r) {
977 		return;
978 	}
979 	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
980 	radeon_ring_write(ring,
981 			  RADEON_ISYNC_ANY2D_IDLE3D |
982 			  RADEON_ISYNC_ANY3D_IDLE2D |
983 			  RADEON_ISYNC_WAIT_IDLEGUI |
984 			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
985 	radeon_ring_unlock_commit(rdev, ring);
986 }
987 
988 
989 /* Load the microcode for the CP */
990 static int r100_cp_init_microcode(struct radeon_device *rdev)
991 {
992 	const char *fw_name = NULL;
993 	int err;
994 
995 	DRM_DEBUG_KMS("\n");
996 
997 	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
998 	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
999 	    (rdev->family == CHIP_RS200)) {
1000 		DRM_INFO("Loading R100 Microcode\n");
1001 		fw_name = FIRMWARE_R100;
1002 	} else if ((rdev->family == CHIP_R200) ||
1003 		   (rdev->family == CHIP_RV250) ||
1004 		   (rdev->family == CHIP_RV280) ||
1005 		   (rdev->family == CHIP_RS300)) {
1006 		DRM_INFO("Loading R200 Microcode\n");
1007 		fw_name = FIRMWARE_R200;
1008 	} else if ((rdev->family == CHIP_R300) ||
1009 		   (rdev->family == CHIP_R350) ||
1010 		   (rdev->family == CHIP_RV350) ||
1011 		   (rdev->family == CHIP_RV380) ||
1012 		   (rdev->family == CHIP_RS400) ||
1013 		   (rdev->family == CHIP_RS480)) {
1014 		DRM_INFO("Loading R300 Microcode\n");
1015 		fw_name = FIRMWARE_R300;
1016 	} else if ((rdev->family == CHIP_R420) ||
1017 		   (rdev->family == CHIP_R423) ||
1018 		   (rdev->family == CHIP_RV410)) {
1019 		DRM_INFO("Loading R400 Microcode\n");
1020 		fw_name = FIRMWARE_R420;
1021 	} else if ((rdev->family == CHIP_RS690) ||
1022 		   (rdev->family == CHIP_RS740)) {
1023 		DRM_INFO("Loading RS690/RS740 Microcode\n");
1024 		fw_name = FIRMWARE_RS690;
1025 	} else if (rdev->family == CHIP_RS600) {
1026 		DRM_INFO("Loading RS600 Microcode\n");
1027 		fw_name = FIRMWARE_RS600;
1028 	} else if ((rdev->family == CHIP_RV515) ||
1029 		   (rdev->family == CHIP_R520) ||
1030 		   (rdev->family == CHIP_RV530) ||
1031 		   (rdev->family == CHIP_R580) ||
1032 		   (rdev->family == CHIP_RV560) ||
1033 		   (rdev->family == CHIP_RV570)) {
1034 		DRM_INFO("Loading R500 Microcode\n");
1035 		fw_name = FIRMWARE_R520;
1036 	}
1037 
1038 	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1039 	if (err) {
1040 		printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1041 		       fw_name);
1042 	} else if (rdev->me_fw->size % 8) {
1043 		printk(KERN_ERR
1044 		       "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1045 		       rdev->me_fw->size, fw_name);
1046 		err = -EINVAL;
1047 		release_firmware(rdev->me_fw);
1048 		rdev->me_fw = NULL;
1049 	}
1050 	return err;
1051 }
1052 
1053 static void r100_cp_load_microcode(struct radeon_device *rdev)
1054 {
1055 	const __be32 *fw_data;
1056 	int i, size;
1057 
1058 	if (r100_gui_wait_for_idle(rdev)) {
1059 		printk(KERN_WARNING "Failed to wait GUI idle while "
1060 		       "programming pipes. Bad things might happen.\n");
1061 	}
1062 
1063 	if (rdev->me_fw) {
1064 		size = rdev->me_fw->size / 4;
1065 		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1066 		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1067 		for (i = 0; i < size; i += 2) {
1068 			WREG32(RADEON_CP_ME_RAM_DATAH,
1069 			       be32_to_cpup(&fw_data[i]));
1070 			WREG32(RADEON_CP_ME_RAM_DATAL,
1071 			       be32_to_cpup(&fw_data[i + 1]));
1072 		}
1073 	}
1074 }
1075 
1076 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1077 {
1078 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1079 	unsigned rb_bufsz;
1080 	unsigned rb_blksz;
1081 	unsigned max_fetch;
1082 	unsigned pre_write_timer;
1083 	unsigned pre_write_limit;
1084 	unsigned indirect2_start;
1085 	unsigned indirect1_start;
1086 	uint32_t tmp;
1087 	int r;
1088 
1089 	if (r100_debugfs_cp_init(rdev)) {
1090 		DRM_ERROR("Failed to register debugfs file for CP !\n");
1091 	}
1092 	if (!rdev->me_fw) {
1093 		r = r100_cp_init_microcode(rdev);
1094 		if (r) {
1095 			DRM_ERROR("Failed to load firmware!\n");
1096 			return r;
1097 		}
1098 	}
1099 
1100 	/* Align ring size */
1101 	rb_bufsz = order_base_2(ring_size / 8);
1102 	ring_size = (1 << (rb_bufsz + 1)) * 4;
1103 	r100_cp_load_microcode(rdev);
1104 	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1105 			     RADEON_CP_RB_RPTR, RADEON_CP_RB_WPTR,
1106 			     RADEON_CP_PACKET2);
1107 	if (r) {
1108 		return r;
1109 	}
1110 	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1111 	 * the rptr copy in system ram */
1112 	rb_blksz = 9;
1113 	/* cp will read 128bytes at a time (4 dwords) */
1114 	max_fetch = 1;
1115 	ring->align_mask = 16 - 1;
1116 	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1117 	pre_write_timer = 64;
1118 	/* Force CP_RB_WPTR write if written more than one time before the
1119 	 * delay expire
1120 	 */
1121 	pre_write_limit = 0;
1122 	/* Setup the cp cache like this (cache size is 96 dwords) :
1123 	 *	RING		0  to 15
1124 	 *	INDIRECT1	16 to 79
1125 	 *	INDIRECT2	80 to 95
1126 	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1127 	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1128 	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1129 	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1130 	 * so it gets the bigger cache.
1131 	 */
1132 	indirect2_start = 80;
1133 	indirect1_start = 16;
1134 	/* cp setup */
1135 	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1136 	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1137 	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1138 	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1139 #ifdef __BIG_ENDIAN
1140 	tmp |= RADEON_BUF_SWAP_32BIT;
1141 #endif
1142 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1143 
1144 	/* Set ring address */
1145 	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1146 	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1147 	/* Force read & write ptr to 0 */
1148 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1149 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1150 	ring->wptr = 0;
1151 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1152 
1153 	/* set the wb address whether it's enabled or not */
1154 	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1155 		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1156 	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1157 
1158 	if (rdev->wb.enabled)
1159 		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1160 	else {
1161 		tmp |= RADEON_RB_NO_UPDATE;
1162 		WREG32(R_000770_SCRATCH_UMSK, 0);
1163 	}
1164 
1165 	WREG32(RADEON_CP_RB_CNTL, tmp);
1166 	udelay(10);
1167 	ring->rptr = RREG32(RADEON_CP_RB_RPTR);
1168 	/* Set cp mode to bus mastering & enable cp*/
1169 	WREG32(RADEON_CP_CSQ_MODE,
1170 	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1171 	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1172 	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1173 	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1174 	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1175 
1176 	/* at this point everything should be setup correctly to enable master */
1177 	pci_set_master(rdev->pdev);
1178 
1179 	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1180 	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1181 	if (r) {
1182 		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1183 		return r;
1184 	}
1185 	ring->ready = true;
1186 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1187 
1188 	if (!ring->rptr_save_reg /* not resuming from suspend */
1189 	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1190 		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1191 		if (r) {
1192 			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1193 			ring->rptr_save_reg = 0;
1194 		}
1195 	}
1196 	return 0;
1197 }
1198 
1199 void r100_cp_fini(struct radeon_device *rdev)
1200 {
1201 	if (r100_cp_wait_for_idle(rdev)) {
1202 		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1203 	}
1204 	/* Disable ring */
1205 	r100_cp_disable(rdev);
1206 	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1207 	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1208 	DRM_INFO("radeon: cp finalized\n");
1209 }
1210 
1211 void r100_cp_disable(struct radeon_device *rdev)
1212 {
1213 	/* Disable ring */
1214 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1215 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1216 	WREG32(RADEON_CP_CSQ_MODE, 0);
1217 	WREG32(RADEON_CP_CSQ_CNTL, 0);
1218 	WREG32(R_000770_SCRATCH_UMSK, 0);
1219 	if (r100_gui_wait_for_idle(rdev)) {
1220 		printk(KERN_WARNING "Failed to wait GUI idle while "
1221 		       "programming pipes. Bad things might happen.\n");
1222 	}
1223 }
1224 
1225 /*
1226  * CS functions
1227  */
1228 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1229 			    struct radeon_cs_packet *pkt,
1230 			    unsigned idx,
1231 			    unsigned reg)
1232 {
1233 	int r;
1234 	u32 tile_flags = 0;
1235 	u32 tmp;
1236 	struct radeon_cs_reloc *reloc;
1237 	u32 value;
1238 
1239 	r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1240 	if (r) {
1241 		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1242 			  idx, reg);
1243 		radeon_cs_dump_packet(p, pkt);
1244 		return r;
1245 	}
1246 
1247 	value = radeon_get_ib_value(p, idx);
1248 	tmp = value & 0x003fffff;
1249 	tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
1250 
1251 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1252 		if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1253 			tile_flags |= RADEON_DST_TILE_MACRO;
1254 		if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
1255 			if (reg == RADEON_SRC_PITCH_OFFSET) {
1256 				DRM_ERROR("Cannot src blit from microtiled surface\n");
1257 				radeon_cs_dump_packet(p, pkt);
1258 				return -EINVAL;
1259 			}
1260 			tile_flags |= RADEON_DST_TILE_MICRO;
1261 		}
1262 
1263 		tmp |= tile_flags;
1264 		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1265 	} else
1266 		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1267 	return 0;
1268 }
1269 
1270 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1271 			     struct radeon_cs_packet *pkt,
1272 			     int idx)
1273 {
1274 	unsigned c, i;
1275 	struct radeon_cs_reloc *reloc;
1276 	struct r100_cs_track *track;
1277 	int r = 0;
1278 	volatile uint32_t *ib;
1279 	u32 idx_value;
1280 
1281 	ib = p->ib.ptr;
1282 	track = (struct r100_cs_track *)p->track;
1283 	c = radeon_get_ib_value(p, idx++) & 0x1F;
1284 	if (c > 16) {
1285 	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1286 		      pkt->opcode);
1287 	    radeon_cs_dump_packet(p, pkt);
1288 	    return -EINVAL;
1289 	}
1290 	track->num_arrays = c;
1291 	for (i = 0; i < (c - 1); i+=2, idx+=3) {
1292 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1293 		if (r) {
1294 			DRM_ERROR("No reloc for packet3 %d\n",
1295 				  pkt->opcode);
1296 			radeon_cs_dump_packet(p, pkt);
1297 			return r;
1298 		}
1299 		idx_value = radeon_get_ib_value(p, idx);
1300 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1301 
1302 		track->arrays[i + 0].esize = idx_value >> 8;
1303 		track->arrays[i + 0].robj = reloc->robj;
1304 		track->arrays[i + 0].esize &= 0x7F;
1305 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1306 		if (r) {
1307 			DRM_ERROR("No reloc for packet3 %d\n",
1308 				  pkt->opcode);
1309 			radeon_cs_dump_packet(p, pkt);
1310 			return r;
1311 		}
1312 		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->lobj.gpu_offset);
1313 		track->arrays[i + 1].robj = reloc->robj;
1314 		track->arrays[i + 1].esize = idx_value >> 24;
1315 		track->arrays[i + 1].esize &= 0x7F;
1316 	}
1317 	if (c & 1) {
1318 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1319 		if (r) {
1320 			DRM_ERROR("No reloc for packet3 %d\n",
1321 					  pkt->opcode);
1322 			radeon_cs_dump_packet(p, pkt);
1323 			return r;
1324 		}
1325 		idx_value = radeon_get_ib_value(p, idx);
1326 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1327 		track->arrays[i + 0].robj = reloc->robj;
1328 		track->arrays[i + 0].esize = idx_value >> 8;
1329 		track->arrays[i + 0].esize &= 0x7F;
1330 	}
1331 	return r;
1332 }
1333 
1334 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1335 			  struct radeon_cs_packet *pkt,
1336 			  const unsigned *auth, unsigned n,
1337 			  radeon_packet0_check_t check)
1338 {
1339 	unsigned reg;
1340 	unsigned i, j, m;
1341 	unsigned idx;
1342 	int r;
1343 
1344 	idx = pkt->idx + 1;
1345 	reg = pkt->reg;
1346 	/* Check that register fall into register range
1347 	 * determined by the number of entry (n) in the
1348 	 * safe register bitmap.
1349 	 */
1350 	if (pkt->one_reg_wr) {
1351 		if ((reg >> 7) > n) {
1352 			return -EINVAL;
1353 		}
1354 	} else {
1355 		if (((reg + (pkt->count << 2)) >> 7) > n) {
1356 			return -EINVAL;
1357 		}
1358 	}
1359 	for (i = 0; i <= pkt->count; i++, idx++) {
1360 		j = (reg >> 7);
1361 		m = 1 << ((reg >> 2) & 31);
1362 		if (auth[j] & m) {
1363 			r = check(p, pkt, idx, reg);
1364 			if (r) {
1365 				return r;
1366 			}
1367 		}
1368 		if (pkt->one_reg_wr) {
1369 			if (!(auth[j] & m)) {
1370 				break;
1371 			}
1372 		} else {
1373 			reg += 4;
1374 		}
1375 	}
1376 	return 0;
1377 }
1378 
1379 /**
1380  * r100_cs_packet_next_vline() - parse userspace VLINE packet
1381  * @parser:		parser structure holding parsing context.
1382  *
1383  * Userspace sends a special sequence for VLINE waits.
1384  * PACKET0 - VLINE_START_END + value
1385  * PACKET0 - WAIT_UNTIL +_value
1386  * RELOC (P3) - crtc_id in reloc.
1387  *
1388  * This function parses this and relocates the VLINE START END
1389  * and WAIT UNTIL packets to the correct crtc.
1390  * It also detects a switched off crtc and nulls out the
1391  * wait in that case.
1392  */
1393 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1394 {
1395 	struct drm_mode_object *obj;
1396 	struct drm_crtc *crtc;
1397 	struct radeon_crtc *radeon_crtc;
1398 	struct radeon_cs_packet p3reloc, waitreloc;
1399 	int crtc_id;
1400 	int r;
1401 	uint32_t header, h_idx, reg;
1402 	volatile uint32_t *ib;
1403 
1404 	ib = p->ib.ptr;
1405 
1406 	/* parse the wait until */
1407 	r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1408 	if (r)
1409 		return r;
1410 
1411 	/* check its a wait until and only 1 count */
1412 	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1413 	    waitreloc.count != 0) {
1414 		DRM_ERROR("vline wait had illegal wait until segment\n");
1415 		return -EINVAL;
1416 	}
1417 
1418 	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1419 		DRM_ERROR("vline wait had illegal wait until\n");
1420 		return -EINVAL;
1421 	}
1422 
1423 	/* jump over the NOP */
1424 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1425 	if (r)
1426 		return r;
1427 
1428 	h_idx = p->idx - 2;
1429 	p->idx += waitreloc.count + 2;
1430 	p->idx += p3reloc.count + 2;
1431 
1432 	header = radeon_get_ib_value(p, h_idx);
1433 	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1434 	reg = R100_CP_PACKET0_GET_REG(header);
1435 	obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
1436 	if (!obj) {
1437 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1438 		return -ENOENT;
1439 	}
1440 	crtc = obj_to_crtc(obj);
1441 	radeon_crtc = to_radeon_crtc(crtc);
1442 	crtc_id = radeon_crtc->crtc_id;
1443 
1444 	if (!crtc->enabled) {
1445 		/* if the CRTC isn't enabled - we need to nop out the wait until */
1446 		ib[h_idx + 2] = PACKET2(0);
1447 		ib[h_idx + 3] = PACKET2(0);
1448 	} else if (crtc_id == 1) {
1449 		switch (reg) {
1450 		case AVIVO_D1MODE_VLINE_START_END:
1451 			header &= ~R300_CP_PACKET0_REG_MASK;
1452 			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1453 			break;
1454 		case RADEON_CRTC_GUI_TRIG_VLINE:
1455 			header &= ~R300_CP_PACKET0_REG_MASK;
1456 			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1457 			break;
1458 		default:
1459 			DRM_ERROR("unknown crtc reloc\n");
1460 			return -EINVAL;
1461 		}
1462 		ib[h_idx] = header;
1463 		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1464 	}
1465 
1466 	return 0;
1467 }
1468 
1469 static int r100_get_vtx_size(uint32_t vtx_fmt)
1470 {
1471 	int vtx_size;
1472 	vtx_size = 2;
1473 	/* ordered according to bits in spec */
1474 	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1475 		vtx_size++;
1476 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1477 		vtx_size += 3;
1478 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1479 		vtx_size++;
1480 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1481 		vtx_size++;
1482 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1483 		vtx_size += 3;
1484 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1485 		vtx_size++;
1486 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1487 		vtx_size++;
1488 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1489 		vtx_size += 2;
1490 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1491 		vtx_size += 2;
1492 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1493 		vtx_size++;
1494 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1495 		vtx_size += 2;
1496 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1497 		vtx_size++;
1498 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1499 		vtx_size += 2;
1500 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1501 		vtx_size++;
1502 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1503 		vtx_size++;
1504 	/* blend weight */
1505 	if (vtx_fmt & (0x7 << 15))
1506 		vtx_size += (vtx_fmt >> 15) & 0x7;
1507 	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1508 		vtx_size += 3;
1509 	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1510 		vtx_size += 2;
1511 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1512 		vtx_size++;
1513 	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1514 		vtx_size++;
1515 	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1516 		vtx_size++;
1517 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1518 		vtx_size++;
1519 	return vtx_size;
1520 }
1521 
1522 static int r100_packet0_check(struct radeon_cs_parser *p,
1523 			      struct radeon_cs_packet *pkt,
1524 			      unsigned idx, unsigned reg)
1525 {
1526 	struct radeon_cs_reloc *reloc;
1527 	struct r100_cs_track *track;
1528 	volatile uint32_t *ib;
1529 	uint32_t tmp;
1530 	int r;
1531 	int i, face;
1532 	u32 tile_flags = 0;
1533 	u32 idx_value;
1534 
1535 	ib = p->ib.ptr;
1536 	track = (struct r100_cs_track *)p->track;
1537 
1538 	idx_value = radeon_get_ib_value(p, idx);
1539 
1540 	switch (reg) {
1541 	case RADEON_CRTC_GUI_TRIG_VLINE:
1542 		r = r100_cs_packet_parse_vline(p);
1543 		if (r) {
1544 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1545 				  idx, reg);
1546 			radeon_cs_dump_packet(p, pkt);
1547 			return r;
1548 		}
1549 		break;
1550 		/* FIXME: only allow PACKET3 blit? easier to check for out of
1551 		 * range access */
1552 	case RADEON_DST_PITCH_OFFSET:
1553 	case RADEON_SRC_PITCH_OFFSET:
1554 		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1555 		if (r)
1556 			return r;
1557 		break;
1558 	case RADEON_RB3D_DEPTHOFFSET:
1559 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1560 		if (r) {
1561 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1562 				  idx, reg);
1563 			radeon_cs_dump_packet(p, pkt);
1564 			return r;
1565 		}
1566 		track->zb.robj = reloc->robj;
1567 		track->zb.offset = idx_value;
1568 		track->zb_dirty = true;
1569 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1570 		break;
1571 	case RADEON_RB3D_COLOROFFSET:
1572 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1573 		if (r) {
1574 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1575 				  idx, reg);
1576 			radeon_cs_dump_packet(p, pkt);
1577 			return r;
1578 		}
1579 		track->cb[0].robj = reloc->robj;
1580 		track->cb[0].offset = idx_value;
1581 		track->cb_dirty = true;
1582 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1583 		break;
1584 	case RADEON_PP_TXOFFSET_0:
1585 	case RADEON_PP_TXOFFSET_1:
1586 	case RADEON_PP_TXOFFSET_2:
1587 		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1588 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1589 		if (r) {
1590 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1591 				  idx, reg);
1592 			radeon_cs_dump_packet(p, pkt);
1593 			return r;
1594 		}
1595 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1596 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1597 				tile_flags |= RADEON_TXO_MACRO_TILE;
1598 			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1599 				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1600 
1601 			tmp = idx_value & ~(0x7 << 2);
1602 			tmp |= tile_flags;
1603 			ib[idx] = tmp + ((u32)reloc->lobj.gpu_offset);
1604 		} else
1605 			ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1606 		track->textures[i].robj = reloc->robj;
1607 		track->tex_dirty = true;
1608 		break;
1609 	case RADEON_PP_CUBIC_OFFSET_T0_0:
1610 	case RADEON_PP_CUBIC_OFFSET_T0_1:
1611 	case RADEON_PP_CUBIC_OFFSET_T0_2:
1612 	case RADEON_PP_CUBIC_OFFSET_T0_3:
1613 	case RADEON_PP_CUBIC_OFFSET_T0_4:
1614 		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1615 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1616 		if (r) {
1617 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1618 				  idx, reg);
1619 			radeon_cs_dump_packet(p, pkt);
1620 			return r;
1621 		}
1622 		track->textures[0].cube_info[i].offset = idx_value;
1623 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1624 		track->textures[0].cube_info[i].robj = reloc->robj;
1625 		track->tex_dirty = true;
1626 		break;
1627 	case RADEON_PP_CUBIC_OFFSET_T1_0:
1628 	case RADEON_PP_CUBIC_OFFSET_T1_1:
1629 	case RADEON_PP_CUBIC_OFFSET_T1_2:
1630 	case RADEON_PP_CUBIC_OFFSET_T1_3:
1631 	case RADEON_PP_CUBIC_OFFSET_T1_4:
1632 		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1633 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1634 		if (r) {
1635 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1636 				  idx, reg);
1637 			radeon_cs_dump_packet(p, pkt);
1638 			return r;
1639 		}
1640 		track->textures[1].cube_info[i].offset = idx_value;
1641 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1642 		track->textures[1].cube_info[i].robj = reloc->robj;
1643 		track->tex_dirty = true;
1644 		break;
1645 	case RADEON_PP_CUBIC_OFFSET_T2_0:
1646 	case RADEON_PP_CUBIC_OFFSET_T2_1:
1647 	case RADEON_PP_CUBIC_OFFSET_T2_2:
1648 	case RADEON_PP_CUBIC_OFFSET_T2_3:
1649 	case RADEON_PP_CUBIC_OFFSET_T2_4:
1650 		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1651 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1652 		if (r) {
1653 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1654 				  idx, reg);
1655 			radeon_cs_dump_packet(p, pkt);
1656 			return r;
1657 		}
1658 		track->textures[2].cube_info[i].offset = idx_value;
1659 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1660 		track->textures[2].cube_info[i].robj = reloc->robj;
1661 		track->tex_dirty = true;
1662 		break;
1663 	case RADEON_RE_WIDTH_HEIGHT:
1664 		track->maxy = ((idx_value >> 16) & 0x7FF);
1665 		track->cb_dirty = true;
1666 		track->zb_dirty = true;
1667 		break;
1668 	case RADEON_RB3D_COLORPITCH:
1669 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1670 		if (r) {
1671 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1672 				  idx, reg);
1673 			radeon_cs_dump_packet(p, pkt);
1674 			return r;
1675 		}
1676 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1677 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1678 				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1679 			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1680 				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1681 
1682 			tmp = idx_value & ~(0x7 << 16);
1683 			tmp |= tile_flags;
1684 			ib[idx] = tmp;
1685 		} else
1686 			ib[idx] = idx_value;
1687 
1688 		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1689 		track->cb_dirty = true;
1690 		break;
1691 	case RADEON_RB3D_DEPTHPITCH:
1692 		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1693 		track->zb_dirty = true;
1694 		break;
1695 	case RADEON_RB3D_CNTL:
1696 		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1697 		case 7:
1698 		case 8:
1699 		case 9:
1700 		case 11:
1701 		case 12:
1702 			track->cb[0].cpp = 1;
1703 			break;
1704 		case 3:
1705 		case 4:
1706 		case 15:
1707 			track->cb[0].cpp = 2;
1708 			break;
1709 		case 6:
1710 			track->cb[0].cpp = 4;
1711 			break;
1712 		default:
1713 			DRM_ERROR("Invalid color buffer format (%d) !\n",
1714 				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1715 			return -EINVAL;
1716 		}
1717 		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1718 		track->cb_dirty = true;
1719 		track->zb_dirty = true;
1720 		break;
1721 	case RADEON_RB3D_ZSTENCILCNTL:
1722 		switch (idx_value & 0xf) {
1723 		case 0:
1724 			track->zb.cpp = 2;
1725 			break;
1726 		case 2:
1727 		case 3:
1728 		case 4:
1729 		case 5:
1730 		case 9:
1731 		case 11:
1732 			track->zb.cpp = 4;
1733 			break;
1734 		default:
1735 			break;
1736 		}
1737 		track->zb_dirty = true;
1738 		break;
1739 	case RADEON_RB3D_ZPASS_ADDR:
1740 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1741 		if (r) {
1742 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1743 				  idx, reg);
1744 			radeon_cs_dump_packet(p, pkt);
1745 			return r;
1746 		}
1747 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1748 		break;
1749 	case RADEON_PP_CNTL:
1750 		{
1751 			uint32_t temp = idx_value >> 4;
1752 			for (i = 0; i < track->num_texture; i++)
1753 				track->textures[i].enabled = !!(temp & (1 << i));
1754 			track->tex_dirty = true;
1755 		}
1756 		break;
1757 	case RADEON_SE_VF_CNTL:
1758 		track->vap_vf_cntl = idx_value;
1759 		break;
1760 	case RADEON_SE_VTX_FMT:
1761 		track->vtx_size = r100_get_vtx_size(idx_value);
1762 		break;
1763 	case RADEON_PP_TEX_SIZE_0:
1764 	case RADEON_PP_TEX_SIZE_1:
1765 	case RADEON_PP_TEX_SIZE_2:
1766 		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1767 		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1768 		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1769 		track->tex_dirty = true;
1770 		break;
1771 	case RADEON_PP_TEX_PITCH_0:
1772 	case RADEON_PP_TEX_PITCH_1:
1773 	case RADEON_PP_TEX_PITCH_2:
1774 		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1775 		track->textures[i].pitch = idx_value + 32;
1776 		track->tex_dirty = true;
1777 		break;
1778 	case RADEON_PP_TXFILTER_0:
1779 	case RADEON_PP_TXFILTER_1:
1780 	case RADEON_PP_TXFILTER_2:
1781 		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1782 		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1783 						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1784 		tmp = (idx_value >> 23) & 0x7;
1785 		if (tmp == 2 || tmp == 6)
1786 			track->textures[i].roundup_w = false;
1787 		tmp = (idx_value >> 27) & 0x7;
1788 		if (tmp == 2 || tmp == 6)
1789 			track->textures[i].roundup_h = false;
1790 		track->tex_dirty = true;
1791 		break;
1792 	case RADEON_PP_TXFORMAT_0:
1793 	case RADEON_PP_TXFORMAT_1:
1794 	case RADEON_PP_TXFORMAT_2:
1795 		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1796 		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1797 			track->textures[i].use_pitch = 1;
1798 		} else {
1799 			track->textures[i].use_pitch = 0;
1800 			track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1801 			track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1802 		}
1803 		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1804 			track->textures[i].tex_coord_type = 2;
1805 		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1806 		case RADEON_TXFORMAT_I8:
1807 		case RADEON_TXFORMAT_RGB332:
1808 		case RADEON_TXFORMAT_Y8:
1809 			track->textures[i].cpp = 1;
1810 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1811 			break;
1812 		case RADEON_TXFORMAT_AI88:
1813 		case RADEON_TXFORMAT_ARGB1555:
1814 		case RADEON_TXFORMAT_RGB565:
1815 		case RADEON_TXFORMAT_ARGB4444:
1816 		case RADEON_TXFORMAT_VYUY422:
1817 		case RADEON_TXFORMAT_YVYU422:
1818 		case RADEON_TXFORMAT_SHADOW16:
1819 		case RADEON_TXFORMAT_LDUDV655:
1820 		case RADEON_TXFORMAT_DUDV88:
1821 			track->textures[i].cpp = 2;
1822 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1823 			break;
1824 		case RADEON_TXFORMAT_ARGB8888:
1825 		case RADEON_TXFORMAT_RGBA8888:
1826 		case RADEON_TXFORMAT_SHADOW32:
1827 		case RADEON_TXFORMAT_LDUDUV8888:
1828 			track->textures[i].cpp = 4;
1829 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1830 			break;
1831 		case RADEON_TXFORMAT_DXT1:
1832 			track->textures[i].cpp = 1;
1833 			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1834 			break;
1835 		case RADEON_TXFORMAT_DXT23:
1836 		case RADEON_TXFORMAT_DXT45:
1837 			track->textures[i].cpp = 1;
1838 			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1839 			break;
1840 		}
1841 		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1842 		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1843 		track->tex_dirty = true;
1844 		break;
1845 	case RADEON_PP_CUBIC_FACES_0:
1846 	case RADEON_PP_CUBIC_FACES_1:
1847 	case RADEON_PP_CUBIC_FACES_2:
1848 		tmp = idx_value;
1849 		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1850 		for (face = 0; face < 4; face++) {
1851 			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1852 			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1853 		}
1854 		track->tex_dirty = true;
1855 		break;
1856 	default:
1857 		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1858 		       reg, idx);
1859 		return -EINVAL;
1860 	}
1861 	return 0;
1862 }
1863 
1864 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1865 					 struct radeon_cs_packet *pkt,
1866 					 struct radeon_bo *robj)
1867 {
1868 	unsigned idx;
1869 	u32 value;
1870 	idx = pkt->idx + 1;
1871 	value = radeon_get_ib_value(p, idx + 2);
1872 	if ((value + 1) > radeon_bo_size(robj)) {
1873 		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1874 			  "(need %u have %lu) !\n",
1875 			  value + 1,
1876 			  radeon_bo_size(robj));
1877 		return -EINVAL;
1878 	}
1879 	return 0;
1880 }
1881 
1882 static int r100_packet3_check(struct radeon_cs_parser *p,
1883 			      struct radeon_cs_packet *pkt)
1884 {
1885 	struct radeon_cs_reloc *reloc;
1886 	struct r100_cs_track *track;
1887 	unsigned idx;
1888 	volatile uint32_t *ib;
1889 	int r;
1890 
1891 	ib = p->ib.ptr;
1892 	idx = pkt->idx + 1;
1893 	track = (struct r100_cs_track *)p->track;
1894 	switch (pkt->opcode) {
1895 	case PACKET3_3D_LOAD_VBPNTR:
1896 		r = r100_packet3_load_vbpntr(p, pkt, idx);
1897 		if (r)
1898 			return r;
1899 		break;
1900 	case PACKET3_INDX_BUFFER:
1901 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1902 		if (r) {
1903 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1904 			radeon_cs_dump_packet(p, pkt);
1905 			return r;
1906 		}
1907 		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->lobj.gpu_offset);
1908 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1909 		if (r) {
1910 			return r;
1911 		}
1912 		break;
1913 	case 0x23:
1914 		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1915 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1916 		if (r) {
1917 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1918 			radeon_cs_dump_packet(p, pkt);
1919 			return r;
1920 		}
1921 		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->lobj.gpu_offset);
1922 		track->num_arrays = 1;
1923 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1924 
1925 		track->arrays[0].robj = reloc->robj;
1926 		track->arrays[0].esize = track->vtx_size;
1927 
1928 		track->max_indx = radeon_get_ib_value(p, idx+1);
1929 
1930 		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1931 		track->immd_dwords = pkt->count - 1;
1932 		r = r100_cs_track_check(p->rdev, track);
1933 		if (r)
1934 			return r;
1935 		break;
1936 	case PACKET3_3D_DRAW_IMMD:
1937 		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1938 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1939 			return -EINVAL;
1940 		}
1941 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1942 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1943 		track->immd_dwords = pkt->count - 1;
1944 		r = r100_cs_track_check(p->rdev, track);
1945 		if (r)
1946 			return r;
1947 		break;
1948 		/* triggers drawing using in-packet vertex data */
1949 	case PACKET3_3D_DRAW_IMMD_2:
1950 		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1951 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1952 			return -EINVAL;
1953 		}
1954 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1955 		track->immd_dwords = pkt->count;
1956 		r = r100_cs_track_check(p->rdev, track);
1957 		if (r)
1958 			return r;
1959 		break;
1960 		/* triggers drawing using in-packet vertex data */
1961 	case PACKET3_3D_DRAW_VBUF_2:
1962 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1963 		r = r100_cs_track_check(p->rdev, track);
1964 		if (r)
1965 			return r;
1966 		break;
1967 		/* triggers drawing of vertex buffers setup elsewhere */
1968 	case PACKET3_3D_DRAW_INDX_2:
1969 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1970 		r = r100_cs_track_check(p->rdev, track);
1971 		if (r)
1972 			return r;
1973 		break;
1974 		/* triggers drawing using indices to vertex buffer */
1975 	case PACKET3_3D_DRAW_VBUF:
1976 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1977 		r = r100_cs_track_check(p->rdev, track);
1978 		if (r)
1979 			return r;
1980 		break;
1981 		/* triggers drawing of vertex buffers setup elsewhere */
1982 	case PACKET3_3D_DRAW_INDX:
1983 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1984 		r = r100_cs_track_check(p->rdev, track);
1985 		if (r)
1986 			return r;
1987 		break;
1988 		/* triggers drawing using indices to vertex buffer */
1989 	case PACKET3_3D_CLEAR_HIZ:
1990 	case PACKET3_3D_CLEAR_ZMASK:
1991 		if (p->rdev->hyperz_filp != p->filp)
1992 			return -EINVAL;
1993 		break;
1994 	case PACKET3_NOP:
1995 		break;
1996 	default:
1997 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1998 		return -EINVAL;
1999 	}
2000 	return 0;
2001 }
2002 
2003 int r100_cs_parse(struct radeon_cs_parser *p)
2004 {
2005 	struct radeon_cs_packet pkt;
2006 	struct r100_cs_track *track;
2007 	int r;
2008 
2009 	track = kzalloc(sizeof(*track), GFP_KERNEL);
2010 	if (!track)
2011 		return -ENOMEM;
2012 	r100_cs_track_clear(p->rdev, track);
2013 	p->track = track;
2014 	do {
2015 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2016 		if (r) {
2017 			return r;
2018 		}
2019 		p->idx += pkt.count + 2;
2020 		switch (pkt.type) {
2021 		case RADEON_PACKET_TYPE0:
2022 			if (p->rdev->family >= CHIP_R200)
2023 				r = r100_cs_parse_packet0(p, &pkt,
2024 					p->rdev->config.r100.reg_safe_bm,
2025 					p->rdev->config.r100.reg_safe_bm_size,
2026 					&r200_packet0_check);
2027 			else
2028 				r = r100_cs_parse_packet0(p, &pkt,
2029 					p->rdev->config.r100.reg_safe_bm,
2030 					p->rdev->config.r100.reg_safe_bm_size,
2031 					&r100_packet0_check);
2032 			break;
2033 		case RADEON_PACKET_TYPE2:
2034 			break;
2035 		case RADEON_PACKET_TYPE3:
2036 			r = r100_packet3_check(p, &pkt);
2037 			break;
2038 		default:
2039 			DRM_ERROR("Unknown packet type %d !\n",
2040 				  pkt.type);
2041 			return -EINVAL;
2042 		}
2043 		if (r)
2044 			return r;
2045 	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
2046 	return 0;
2047 }
2048 
2049 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2050 {
2051 	DRM_ERROR("pitch                      %d\n", t->pitch);
2052 	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2053 	DRM_ERROR("width                      %d\n", t->width);
2054 	DRM_ERROR("width_11                   %d\n", t->width_11);
2055 	DRM_ERROR("height                     %d\n", t->height);
2056 	DRM_ERROR("height_11                  %d\n", t->height_11);
2057 	DRM_ERROR("num levels                 %d\n", t->num_levels);
2058 	DRM_ERROR("depth                      %d\n", t->txdepth);
2059 	DRM_ERROR("bpp                        %d\n", t->cpp);
2060 	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2061 	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2062 	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2063 	DRM_ERROR("compress format            %d\n", t->compress_format);
2064 }
2065 
2066 static int r100_track_compress_size(int compress_format, int w, int h)
2067 {
2068 	int block_width, block_height, block_bytes;
2069 	int wblocks, hblocks;
2070 	int min_wblocks;
2071 	int sz;
2072 
2073 	block_width = 4;
2074 	block_height = 4;
2075 
2076 	switch (compress_format) {
2077 	case R100_TRACK_COMP_DXT1:
2078 		block_bytes = 8;
2079 		min_wblocks = 4;
2080 		break;
2081 	default:
2082 	case R100_TRACK_COMP_DXT35:
2083 		block_bytes = 16;
2084 		min_wblocks = 2;
2085 		break;
2086 	}
2087 
2088 	hblocks = (h + block_height - 1) / block_height;
2089 	wblocks = (w + block_width - 1) / block_width;
2090 	if (wblocks < min_wblocks)
2091 		wblocks = min_wblocks;
2092 	sz = wblocks * hblocks * block_bytes;
2093 	return sz;
2094 }
2095 
2096 static int r100_cs_track_cube(struct radeon_device *rdev,
2097 			      struct r100_cs_track *track, unsigned idx)
2098 {
2099 	unsigned face, w, h;
2100 	struct radeon_bo *cube_robj;
2101 	unsigned long size;
2102 	unsigned compress_format = track->textures[idx].compress_format;
2103 
2104 	for (face = 0; face < 5; face++) {
2105 		cube_robj = track->textures[idx].cube_info[face].robj;
2106 		w = track->textures[idx].cube_info[face].width;
2107 		h = track->textures[idx].cube_info[face].height;
2108 
2109 		if (compress_format) {
2110 			size = r100_track_compress_size(compress_format, w, h);
2111 		} else
2112 			size = w * h;
2113 		size *= track->textures[idx].cpp;
2114 
2115 		size += track->textures[idx].cube_info[face].offset;
2116 
2117 		if (size > radeon_bo_size(cube_robj)) {
2118 			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2119 				  size, radeon_bo_size(cube_robj));
2120 			r100_cs_track_texture_print(&track->textures[idx]);
2121 			return -1;
2122 		}
2123 	}
2124 	return 0;
2125 }
2126 
2127 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2128 				       struct r100_cs_track *track)
2129 {
2130 	struct radeon_bo *robj;
2131 	unsigned long size;
2132 	unsigned u, i, w, h, d;
2133 	int ret;
2134 
2135 	for (u = 0; u < track->num_texture; u++) {
2136 		if (!track->textures[u].enabled)
2137 			continue;
2138 		if (track->textures[u].lookup_disable)
2139 			continue;
2140 		robj = track->textures[u].robj;
2141 		if (robj == NULL) {
2142 			DRM_ERROR("No texture bound to unit %u\n", u);
2143 			return -EINVAL;
2144 		}
2145 		size = 0;
2146 		for (i = 0; i <= track->textures[u].num_levels; i++) {
2147 			if (track->textures[u].use_pitch) {
2148 				if (rdev->family < CHIP_R300)
2149 					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2150 				else
2151 					w = track->textures[u].pitch / (1 << i);
2152 			} else {
2153 				w = track->textures[u].width;
2154 				if (rdev->family >= CHIP_RV515)
2155 					w |= track->textures[u].width_11;
2156 				w = w / (1 << i);
2157 				if (track->textures[u].roundup_w)
2158 					w = roundup_pow_of_two(w);
2159 			}
2160 			h = track->textures[u].height;
2161 			if (rdev->family >= CHIP_RV515)
2162 				h |= track->textures[u].height_11;
2163 			h = h / (1 << i);
2164 			if (track->textures[u].roundup_h)
2165 				h = roundup_pow_of_two(h);
2166 			if (track->textures[u].tex_coord_type == 1) {
2167 				d = (1 << track->textures[u].txdepth) / (1 << i);
2168 				if (!d)
2169 					d = 1;
2170 			} else {
2171 				d = 1;
2172 			}
2173 			if (track->textures[u].compress_format) {
2174 
2175 				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2176 				/* compressed textures are block based */
2177 			} else
2178 				size += w * h * d;
2179 		}
2180 		size *= track->textures[u].cpp;
2181 
2182 		switch (track->textures[u].tex_coord_type) {
2183 		case 0:
2184 		case 1:
2185 			break;
2186 		case 2:
2187 			if (track->separate_cube) {
2188 				ret = r100_cs_track_cube(rdev, track, u);
2189 				if (ret)
2190 					return ret;
2191 			} else
2192 				size *= 6;
2193 			break;
2194 		default:
2195 			DRM_ERROR("Invalid texture coordinate type %u for unit "
2196 				  "%u\n", track->textures[u].tex_coord_type, u);
2197 			return -EINVAL;
2198 		}
2199 		if (size > radeon_bo_size(robj)) {
2200 			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2201 				  "%lu\n", u, size, radeon_bo_size(robj));
2202 			r100_cs_track_texture_print(&track->textures[u]);
2203 			return -EINVAL;
2204 		}
2205 	}
2206 	return 0;
2207 }
2208 
2209 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2210 {
2211 	unsigned i;
2212 	unsigned long size;
2213 	unsigned prim_walk;
2214 	unsigned nverts;
2215 	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2216 
2217 	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2218 	    !track->blend_read_enable)
2219 		num_cb = 0;
2220 
2221 	for (i = 0; i < num_cb; i++) {
2222 		if (track->cb[i].robj == NULL) {
2223 			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2224 			return -EINVAL;
2225 		}
2226 		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2227 		size += track->cb[i].offset;
2228 		if (size > radeon_bo_size(track->cb[i].robj)) {
2229 			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2230 				  "(need %lu have %lu) !\n", i, size,
2231 				  radeon_bo_size(track->cb[i].robj));
2232 			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2233 				  i, track->cb[i].pitch, track->cb[i].cpp,
2234 				  track->cb[i].offset, track->maxy);
2235 			return -EINVAL;
2236 		}
2237 	}
2238 	track->cb_dirty = false;
2239 
2240 	if (track->zb_dirty && track->z_enabled) {
2241 		if (track->zb.robj == NULL) {
2242 			DRM_ERROR("[drm] No buffer for z buffer !\n");
2243 			return -EINVAL;
2244 		}
2245 		size = track->zb.pitch * track->zb.cpp * track->maxy;
2246 		size += track->zb.offset;
2247 		if (size > radeon_bo_size(track->zb.robj)) {
2248 			DRM_ERROR("[drm] Buffer too small for z buffer "
2249 				  "(need %lu have %lu) !\n", size,
2250 				  radeon_bo_size(track->zb.robj));
2251 			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2252 				  track->zb.pitch, track->zb.cpp,
2253 				  track->zb.offset, track->maxy);
2254 			return -EINVAL;
2255 		}
2256 	}
2257 	track->zb_dirty = false;
2258 
2259 	if (track->aa_dirty && track->aaresolve) {
2260 		if (track->aa.robj == NULL) {
2261 			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2262 			return -EINVAL;
2263 		}
2264 		/* I believe the format comes from colorbuffer0. */
2265 		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2266 		size += track->aa.offset;
2267 		if (size > radeon_bo_size(track->aa.robj)) {
2268 			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2269 				  "(need %lu have %lu) !\n", i, size,
2270 				  radeon_bo_size(track->aa.robj));
2271 			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2272 				  i, track->aa.pitch, track->cb[0].cpp,
2273 				  track->aa.offset, track->maxy);
2274 			return -EINVAL;
2275 		}
2276 	}
2277 	track->aa_dirty = false;
2278 
2279 	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2280 	if (track->vap_vf_cntl & (1 << 14)) {
2281 		nverts = track->vap_alt_nverts;
2282 	} else {
2283 		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2284 	}
2285 	switch (prim_walk) {
2286 	case 1:
2287 		for (i = 0; i < track->num_arrays; i++) {
2288 			size = track->arrays[i].esize * track->max_indx * 4;
2289 			if (track->arrays[i].robj == NULL) {
2290 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2291 					  "bound\n", prim_walk, i);
2292 				return -EINVAL;
2293 			}
2294 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2295 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2296 					"need %lu dwords have %lu dwords\n",
2297 					prim_walk, i, size >> 2,
2298 					radeon_bo_size(track->arrays[i].robj)
2299 					>> 2);
2300 				DRM_ERROR("Max indices %u\n", track->max_indx);
2301 				return -EINVAL;
2302 			}
2303 		}
2304 		break;
2305 	case 2:
2306 		for (i = 0; i < track->num_arrays; i++) {
2307 			size = track->arrays[i].esize * (nverts - 1) * 4;
2308 			if (track->arrays[i].robj == NULL) {
2309 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2310 					  "bound\n", prim_walk, i);
2311 				return -EINVAL;
2312 			}
2313 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2314 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2315 					"need %lu dwords have %lu dwords\n",
2316 					prim_walk, i, size >> 2,
2317 					radeon_bo_size(track->arrays[i].robj)
2318 					>> 2);
2319 				return -EINVAL;
2320 			}
2321 		}
2322 		break;
2323 	case 3:
2324 		size = track->vtx_size * nverts;
2325 		if (size != track->immd_dwords) {
2326 			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2327 				  track->immd_dwords, size);
2328 			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2329 				  nverts, track->vtx_size);
2330 			return -EINVAL;
2331 		}
2332 		break;
2333 	default:
2334 		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2335 			  prim_walk);
2336 		return -EINVAL;
2337 	}
2338 
2339 	if (track->tex_dirty) {
2340 		track->tex_dirty = false;
2341 		return r100_cs_track_texture_check(rdev, track);
2342 	}
2343 	return 0;
2344 }
2345 
2346 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2347 {
2348 	unsigned i, face;
2349 
2350 	track->cb_dirty = true;
2351 	track->zb_dirty = true;
2352 	track->tex_dirty = true;
2353 	track->aa_dirty = true;
2354 
2355 	if (rdev->family < CHIP_R300) {
2356 		track->num_cb = 1;
2357 		if (rdev->family <= CHIP_RS200)
2358 			track->num_texture = 3;
2359 		else
2360 			track->num_texture = 6;
2361 		track->maxy = 2048;
2362 		track->separate_cube = 1;
2363 	} else {
2364 		track->num_cb = 4;
2365 		track->num_texture = 16;
2366 		track->maxy = 4096;
2367 		track->separate_cube = 0;
2368 		track->aaresolve = false;
2369 		track->aa.robj = NULL;
2370 	}
2371 
2372 	for (i = 0; i < track->num_cb; i++) {
2373 		track->cb[i].robj = NULL;
2374 		track->cb[i].pitch = 8192;
2375 		track->cb[i].cpp = 16;
2376 		track->cb[i].offset = 0;
2377 	}
2378 	track->z_enabled = true;
2379 	track->zb.robj = NULL;
2380 	track->zb.pitch = 8192;
2381 	track->zb.cpp = 4;
2382 	track->zb.offset = 0;
2383 	track->vtx_size = 0x7F;
2384 	track->immd_dwords = 0xFFFFFFFFUL;
2385 	track->num_arrays = 11;
2386 	track->max_indx = 0x00FFFFFFUL;
2387 	for (i = 0; i < track->num_arrays; i++) {
2388 		track->arrays[i].robj = NULL;
2389 		track->arrays[i].esize = 0x7F;
2390 	}
2391 	for (i = 0; i < track->num_texture; i++) {
2392 		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2393 		track->textures[i].pitch = 16536;
2394 		track->textures[i].width = 16536;
2395 		track->textures[i].height = 16536;
2396 		track->textures[i].width_11 = 1 << 11;
2397 		track->textures[i].height_11 = 1 << 11;
2398 		track->textures[i].num_levels = 12;
2399 		if (rdev->family <= CHIP_RS200) {
2400 			track->textures[i].tex_coord_type = 0;
2401 			track->textures[i].txdepth = 0;
2402 		} else {
2403 			track->textures[i].txdepth = 16;
2404 			track->textures[i].tex_coord_type = 1;
2405 		}
2406 		track->textures[i].cpp = 64;
2407 		track->textures[i].robj = NULL;
2408 		/* CS IB emission code makes sure texture unit are disabled */
2409 		track->textures[i].enabled = false;
2410 		track->textures[i].lookup_disable = false;
2411 		track->textures[i].roundup_w = true;
2412 		track->textures[i].roundup_h = true;
2413 		if (track->separate_cube)
2414 			for (face = 0; face < 5; face++) {
2415 				track->textures[i].cube_info[face].robj = NULL;
2416 				track->textures[i].cube_info[face].width = 16536;
2417 				track->textures[i].cube_info[face].height = 16536;
2418 				track->textures[i].cube_info[face].offset = 0;
2419 			}
2420 	}
2421 }
2422 
2423 /*
2424  * Global GPU functions
2425  */
2426 static void r100_errata(struct radeon_device *rdev)
2427 {
2428 	rdev->pll_errata = 0;
2429 
2430 	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2431 		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2432 	}
2433 
2434 	if (rdev->family == CHIP_RV100 ||
2435 	    rdev->family == CHIP_RS100 ||
2436 	    rdev->family == CHIP_RS200) {
2437 		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2438 	}
2439 }
2440 
2441 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2442 {
2443 	unsigned i;
2444 	uint32_t tmp;
2445 
2446 	for (i = 0; i < rdev->usec_timeout; i++) {
2447 		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2448 		if (tmp >= n) {
2449 			return 0;
2450 		}
2451 		DRM_UDELAY(1);
2452 	}
2453 	return -1;
2454 }
2455 
2456 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2457 {
2458 	unsigned i;
2459 	uint32_t tmp;
2460 
2461 	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2462 		printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
2463 		       " Bad things might happen.\n");
2464 	}
2465 	for (i = 0; i < rdev->usec_timeout; i++) {
2466 		tmp = RREG32(RADEON_RBBM_STATUS);
2467 		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2468 			return 0;
2469 		}
2470 		DRM_UDELAY(1);
2471 	}
2472 	return -1;
2473 }
2474 
2475 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2476 {
2477 	unsigned i;
2478 	uint32_t tmp;
2479 
2480 	for (i = 0; i < rdev->usec_timeout; i++) {
2481 		/* read MC_STATUS */
2482 		tmp = RREG32(RADEON_MC_STATUS);
2483 		if (tmp & RADEON_MC_IDLE) {
2484 			return 0;
2485 		}
2486 		DRM_UDELAY(1);
2487 	}
2488 	return -1;
2489 }
2490 
2491 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2492 {
2493 	u32 rbbm_status;
2494 
2495 	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2496 	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2497 		radeon_ring_lockup_update(ring);
2498 		return false;
2499 	}
2500 	/* force CP activities */
2501 	radeon_ring_force_activity(rdev, ring);
2502 	return radeon_ring_test_lockup(rdev, ring);
2503 }
2504 
2505 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2506 void r100_enable_bm(struct radeon_device *rdev)
2507 {
2508 	uint32_t tmp;
2509 	/* Enable bus mastering */
2510 	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2511 	WREG32(RADEON_BUS_CNTL, tmp);
2512 }
2513 
2514 void r100_bm_disable(struct radeon_device *rdev)
2515 {
2516 	u32 tmp;
2517 
2518 	/* disable bus mastering */
2519 	tmp = RREG32(R_000030_BUS_CNTL);
2520 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2521 	mdelay(1);
2522 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2523 	mdelay(1);
2524 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2525 	tmp = RREG32(RADEON_BUS_CNTL);
2526 	mdelay(1);
2527 	pci_clear_master(rdev->pdev);
2528 	mdelay(1);
2529 }
2530 
2531 int r100_asic_reset(struct radeon_device *rdev)
2532 {
2533 	struct r100_mc_save save;
2534 	u32 status, tmp;
2535 	int ret = 0;
2536 
2537 	status = RREG32(R_000E40_RBBM_STATUS);
2538 	if (!G_000E40_GUI_ACTIVE(status)) {
2539 		return 0;
2540 	}
2541 	r100_mc_stop(rdev, &save);
2542 	status = RREG32(R_000E40_RBBM_STATUS);
2543 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2544 	/* stop CP */
2545 	WREG32(RADEON_CP_CSQ_CNTL, 0);
2546 	tmp = RREG32(RADEON_CP_RB_CNTL);
2547 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2548 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2549 	WREG32(RADEON_CP_RB_WPTR, 0);
2550 	WREG32(RADEON_CP_RB_CNTL, tmp);
2551 	/* save PCI state */
2552 	pci_save_state(rdev->pdev);
2553 	/* disable bus mastering */
2554 	r100_bm_disable(rdev);
2555 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2556 					S_0000F0_SOFT_RESET_RE(1) |
2557 					S_0000F0_SOFT_RESET_PP(1) |
2558 					S_0000F0_SOFT_RESET_RB(1));
2559 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2560 	mdelay(500);
2561 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2562 	mdelay(1);
2563 	status = RREG32(R_000E40_RBBM_STATUS);
2564 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2565 	/* reset CP */
2566 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2567 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2568 	mdelay(500);
2569 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2570 	mdelay(1);
2571 	status = RREG32(R_000E40_RBBM_STATUS);
2572 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2573 	/* restore PCI & busmastering */
2574 	pci_restore_state(rdev->pdev);
2575 	r100_enable_bm(rdev);
2576 	/* Check if GPU is idle */
2577 	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2578 		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2579 		dev_err(rdev->dev, "failed to reset GPU\n");
2580 		ret = -1;
2581 	} else
2582 		dev_info(rdev->dev, "GPU reset succeed\n");
2583 	r100_mc_resume(rdev, &save);
2584 	return ret;
2585 }
2586 
2587 void r100_set_common_regs(struct radeon_device *rdev)
2588 {
2589 	struct drm_device *dev = rdev->ddev;
2590 	bool force_dac2 = false;
2591 	u32 tmp;
2592 
2593 	/* set these so they don't interfere with anything */
2594 	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2595 	WREG32(RADEON_SUBPIC_CNTL, 0);
2596 	WREG32(RADEON_VIPH_CONTROL, 0);
2597 	WREG32(RADEON_I2C_CNTL_1, 0);
2598 	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2599 	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2600 	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2601 
2602 	/* always set up dac2 on rn50 and some rv100 as lots
2603 	 * of servers seem to wire it up to a VGA port but
2604 	 * don't report it in the bios connector
2605 	 * table.
2606 	 */
2607 	switch (dev->pdev->device) {
2608 		/* RN50 */
2609 	case 0x515e:
2610 	case 0x5969:
2611 		force_dac2 = true;
2612 		break;
2613 		/* RV100*/
2614 	case 0x5159:
2615 	case 0x515a:
2616 		/* DELL triple head servers */
2617 		if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2618 		    ((dev->pdev->subsystem_device == 0x016c) ||
2619 		     (dev->pdev->subsystem_device == 0x016d) ||
2620 		     (dev->pdev->subsystem_device == 0x016e) ||
2621 		     (dev->pdev->subsystem_device == 0x016f) ||
2622 		     (dev->pdev->subsystem_device == 0x0170) ||
2623 		     (dev->pdev->subsystem_device == 0x017d) ||
2624 		     (dev->pdev->subsystem_device == 0x017e) ||
2625 		     (dev->pdev->subsystem_device == 0x0183) ||
2626 		     (dev->pdev->subsystem_device == 0x018a) ||
2627 		     (dev->pdev->subsystem_device == 0x019a)))
2628 			force_dac2 = true;
2629 		break;
2630 	}
2631 
2632 	if (force_dac2) {
2633 		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2634 		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2635 		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2636 
2637 		/* For CRT on DAC2, don't turn it on if BIOS didn't
2638 		   enable it, even it's detected.
2639 		*/
2640 
2641 		/* force it to crtc0 */
2642 		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2643 		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2644 		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2645 
2646 		/* set up the TV DAC */
2647 		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2648 				 RADEON_TV_DAC_STD_MASK |
2649 				 RADEON_TV_DAC_RDACPD |
2650 				 RADEON_TV_DAC_GDACPD |
2651 				 RADEON_TV_DAC_BDACPD |
2652 				 RADEON_TV_DAC_BGADJ_MASK |
2653 				 RADEON_TV_DAC_DACADJ_MASK);
2654 		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2655 				RADEON_TV_DAC_NHOLD |
2656 				RADEON_TV_DAC_STD_PS2 |
2657 				(0x58 << 16));
2658 
2659 		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2660 		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2661 		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2662 	}
2663 
2664 	/* switch PM block to ACPI mode */
2665 	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2666 	tmp &= ~RADEON_PM_MODE_SEL;
2667 	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2668 
2669 }
2670 
2671 /*
2672  * VRAM info
2673  */
2674 static void r100_vram_get_type(struct radeon_device *rdev)
2675 {
2676 	uint32_t tmp;
2677 
2678 	rdev->mc.vram_is_ddr = false;
2679 	if (rdev->flags & RADEON_IS_IGP)
2680 		rdev->mc.vram_is_ddr = true;
2681 	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2682 		rdev->mc.vram_is_ddr = true;
2683 	if ((rdev->family == CHIP_RV100) ||
2684 	    (rdev->family == CHIP_RS100) ||
2685 	    (rdev->family == CHIP_RS200)) {
2686 		tmp = RREG32(RADEON_MEM_CNTL);
2687 		if (tmp & RV100_HALF_MODE) {
2688 			rdev->mc.vram_width = 32;
2689 		} else {
2690 			rdev->mc.vram_width = 64;
2691 		}
2692 		if (rdev->flags & RADEON_SINGLE_CRTC) {
2693 			rdev->mc.vram_width /= 4;
2694 			rdev->mc.vram_is_ddr = true;
2695 		}
2696 	} else if (rdev->family <= CHIP_RV280) {
2697 		tmp = RREG32(RADEON_MEM_CNTL);
2698 		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2699 			rdev->mc.vram_width = 128;
2700 		} else {
2701 			rdev->mc.vram_width = 64;
2702 		}
2703 	} else {
2704 		/* newer IGPs */
2705 		rdev->mc.vram_width = 128;
2706 	}
2707 }
2708 
2709 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2710 {
2711 	u32 aper_size;
2712 	u8 byte;
2713 
2714 	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2715 
2716 	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2717 	 * that is has the 2nd generation multifunction PCI interface
2718 	 */
2719 	if (rdev->family == CHIP_RV280 ||
2720 	    rdev->family >= CHIP_RV350) {
2721 		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2722 		       ~RADEON_HDP_APER_CNTL);
2723 		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2724 		return aper_size * 2;
2725 	}
2726 
2727 	/* Older cards have all sorts of funny issues to deal with. First
2728 	 * check if it's a multifunction card by reading the PCI config
2729 	 * header type... Limit those to one aperture size
2730 	 */
2731 	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2732 	if (byte & 0x80) {
2733 		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2734 		DRM_INFO("Limiting VRAM to one aperture\n");
2735 		return aper_size;
2736 	}
2737 
2738 	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2739 	 * have set it up. We don't write this as it's broken on some ASICs but
2740 	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2741 	 */
2742 	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2743 		return aper_size * 2;
2744 	return aper_size;
2745 }
2746 
2747 void r100_vram_init_sizes(struct radeon_device *rdev)
2748 {
2749 	u64 config_aper_size;
2750 
2751 	/* work out accessible VRAM */
2752 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2753 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2754 	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2755 	/* FIXME we don't use the second aperture yet when we could use it */
2756 	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2757 		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2758 	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2759 	if (rdev->flags & RADEON_IS_IGP) {
2760 		uint32_t tom;
2761 		/* read NB_TOM to get the amount of ram stolen for the GPU */
2762 		tom = RREG32(RADEON_NB_TOM);
2763 		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2764 		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2765 		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2766 	} else {
2767 		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2768 		/* Some production boards of m6 will report 0
2769 		 * if it's 8 MB
2770 		 */
2771 		if (rdev->mc.real_vram_size == 0) {
2772 			rdev->mc.real_vram_size = 8192 * 1024;
2773 			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2774 		}
2775 		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2776 		 * Novell bug 204882 + along with lots of ubuntu ones
2777 		 */
2778 		if (rdev->mc.aper_size > config_aper_size)
2779 			config_aper_size = rdev->mc.aper_size;
2780 
2781 		if (config_aper_size > rdev->mc.real_vram_size)
2782 			rdev->mc.mc_vram_size = config_aper_size;
2783 		else
2784 			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2785 	}
2786 }
2787 
2788 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2789 {
2790 	uint32_t temp;
2791 
2792 	temp = RREG32(RADEON_CONFIG_CNTL);
2793 	if (state == false) {
2794 		temp &= ~RADEON_CFG_VGA_RAM_EN;
2795 		temp |= RADEON_CFG_VGA_IO_DIS;
2796 	} else {
2797 		temp &= ~RADEON_CFG_VGA_IO_DIS;
2798 	}
2799 	WREG32(RADEON_CONFIG_CNTL, temp);
2800 }
2801 
2802 static void r100_mc_init(struct radeon_device *rdev)
2803 {
2804 	u64 base;
2805 
2806 	r100_vram_get_type(rdev);
2807 	r100_vram_init_sizes(rdev);
2808 	base = rdev->mc.aper_base;
2809 	if (rdev->flags & RADEON_IS_IGP)
2810 		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2811 	radeon_vram_location(rdev, &rdev->mc, base);
2812 	rdev->mc.gtt_base_align = 0;
2813 	if (!(rdev->flags & RADEON_IS_AGP))
2814 		radeon_gtt_location(rdev, &rdev->mc);
2815 	radeon_update_bandwidth_info(rdev);
2816 }
2817 
2818 
2819 /*
2820  * Indirect registers accessor
2821  */
2822 void r100_pll_errata_after_index(struct radeon_device *rdev)
2823 {
2824 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2825 		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2826 		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2827 	}
2828 }
2829 
2830 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2831 {
2832 	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2833 	 * or the chip could hang on a subsequent access
2834 	 */
2835 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2836 		mdelay(5);
2837 	}
2838 
2839 	/* This function is required to workaround a hardware bug in some (all?)
2840 	 * revisions of the R300.  This workaround should be called after every
2841 	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2842 	 * may not be correct.
2843 	 */
2844 	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2845 		uint32_t save, tmp;
2846 
2847 		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2848 		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2849 		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2850 		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2851 		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2852 	}
2853 }
2854 
2855 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2856 {
2857 	unsigned long flags;
2858 	uint32_t data;
2859 
2860 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2861 	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2862 	r100_pll_errata_after_index(rdev);
2863 	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2864 	r100_pll_errata_after_data(rdev);
2865 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2866 	return data;
2867 }
2868 
2869 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2870 {
2871 	unsigned long flags;
2872 
2873 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2874 	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2875 	r100_pll_errata_after_index(rdev);
2876 	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2877 	r100_pll_errata_after_data(rdev);
2878 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2879 }
2880 
2881 static void r100_set_safe_registers(struct radeon_device *rdev)
2882 {
2883 	if (ASIC_IS_RN50(rdev)) {
2884 		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2885 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2886 	} else if (rdev->family < CHIP_R200) {
2887 		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2888 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2889 	} else {
2890 		r200_set_safe_registers(rdev);
2891 	}
2892 }
2893 
2894 /*
2895  * Debugfs info
2896  */
2897 #if defined(CONFIG_DEBUG_FS)
2898 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2899 {
2900 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2901 	struct drm_device *dev = node->minor->dev;
2902 	struct radeon_device *rdev = dev->dev_private;
2903 	uint32_t reg, value;
2904 	unsigned i;
2905 
2906 	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2907 	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2908 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2909 	for (i = 0; i < 64; i++) {
2910 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2911 		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2912 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2913 		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2914 		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2915 	}
2916 	return 0;
2917 }
2918 
2919 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2920 {
2921 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2922 	struct drm_device *dev = node->minor->dev;
2923 	struct radeon_device *rdev = dev->dev_private;
2924 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2925 	uint32_t rdp, wdp;
2926 	unsigned count, i, j;
2927 
2928 	radeon_ring_free_size(rdev, ring);
2929 	rdp = RREG32(RADEON_CP_RB_RPTR);
2930 	wdp = RREG32(RADEON_CP_RB_WPTR);
2931 	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2932 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2933 	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2934 	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2935 	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2936 	seq_printf(m, "%u dwords in ring\n", count);
2937 	if (ring->ready) {
2938 		for (j = 0; j <= count; j++) {
2939 			i = (rdp + j) & ring->ptr_mask;
2940 			seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2941 		}
2942 	}
2943 	return 0;
2944 }
2945 
2946 
2947 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2948 {
2949 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2950 	struct drm_device *dev = node->minor->dev;
2951 	struct radeon_device *rdev = dev->dev_private;
2952 	uint32_t csq_stat, csq2_stat, tmp;
2953 	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2954 	unsigned i;
2955 
2956 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2957 	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2958 	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2959 	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2960 	r_rptr = (csq_stat >> 0) & 0x3ff;
2961 	r_wptr = (csq_stat >> 10) & 0x3ff;
2962 	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2963 	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2964 	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2965 	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2966 	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2967 	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2968 	seq_printf(m, "Ring rptr %u\n", r_rptr);
2969 	seq_printf(m, "Ring wptr %u\n", r_wptr);
2970 	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2971 	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2972 	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2973 	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2974 	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
2975 	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
2976 	seq_printf(m, "Ring fifo:\n");
2977 	for (i = 0; i < 256; i++) {
2978 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2979 		tmp = RREG32(RADEON_CP_CSQ_DATA);
2980 		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
2981 	}
2982 	seq_printf(m, "Indirect1 fifo:\n");
2983 	for (i = 256; i <= 512; i++) {
2984 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2985 		tmp = RREG32(RADEON_CP_CSQ_DATA);
2986 		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
2987 	}
2988 	seq_printf(m, "Indirect2 fifo:\n");
2989 	for (i = 640; i < ib1_wptr; i++) {
2990 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2991 		tmp = RREG32(RADEON_CP_CSQ_DATA);
2992 		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
2993 	}
2994 	return 0;
2995 }
2996 
2997 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
2998 {
2999 	struct drm_info_node *node = (struct drm_info_node *) m->private;
3000 	struct drm_device *dev = node->minor->dev;
3001 	struct radeon_device *rdev = dev->dev_private;
3002 	uint32_t tmp;
3003 
3004 	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3005 	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3006 	tmp = RREG32(RADEON_MC_FB_LOCATION);
3007 	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3008 	tmp = RREG32(RADEON_BUS_CNTL);
3009 	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3010 	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3011 	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3012 	tmp = RREG32(RADEON_AGP_BASE);
3013 	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3014 	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3015 	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3016 	tmp = RREG32(0x01D0);
3017 	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3018 	tmp = RREG32(RADEON_AIC_LO_ADDR);
3019 	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3020 	tmp = RREG32(RADEON_AIC_HI_ADDR);
3021 	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3022 	tmp = RREG32(0x01E4);
3023 	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3024 	return 0;
3025 }
3026 
3027 static struct drm_info_list r100_debugfs_rbbm_list[] = {
3028 	{"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3029 };
3030 
3031 static struct drm_info_list r100_debugfs_cp_list[] = {
3032 	{"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3033 	{"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3034 };
3035 
3036 static struct drm_info_list r100_debugfs_mc_info_list[] = {
3037 	{"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3038 };
3039 #endif
3040 
3041 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3042 {
3043 #if defined(CONFIG_DEBUG_FS)
3044 	return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3045 #else
3046 	return 0;
3047 #endif
3048 }
3049 
3050 int r100_debugfs_cp_init(struct radeon_device *rdev)
3051 {
3052 #if defined(CONFIG_DEBUG_FS)
3053 	return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3054 #else
3055 	return 0;
3056 #endif
3057 }
3058 
3059 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3060 {
3061 #if defined(CONFIG_DEBUG_FS)
3062 	return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3063 #else
3064 	return 0;
3065 #endif
3066 }
3067 
3068 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3069 			 uint32_t tiling_flags, uint32_t pitch,
3070 			 uint32_t offset, uint32_t obj_size)
3071 {
3072 	int surf_index = reg * 16;
3073 	int flags = 0;
3074 
3075 	if (rdev->family <= CHIP_RS200) {
3076 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3077 				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3078 			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3079 		if (tiling_flags & RADEON_TILING_MACRO)
3080 			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3081 		/* setting pitch to 0 disables tiling */
3082 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3083 				== 0)
3084 			pitch = 0;
3085 	} else if (rdev->family <= CHIP_RV280) {
3086 		if (tiling_flags & (RADEON_TILING_MACRO))
3087 			flags |= R200_SURF_TILE_COLOR_MACRO;
3088 		if (tiling_flags & RADEON_TILING_MICRO)
3089 			flags |= R200_SURF_TILE_COLOR_MICRO;
3090 	} else {
3091 		if (tiling_flags & RADEON_TILING_MACRO)
3092 			flags |= R300_SURF_TILE_MACRO;
3093 		if (tiling_flags & RADEON_TILING_MICRO)
3094 			flags |= R300_SURF_TILE_MICRO;
3095 	}
3096 
3097 	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3098 		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3099 	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3100 		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3101 
3102 	/* r100/r200 divide by 16 */
3103 	if (rdev->family < CHIP_R300)
3104 		flags |= pitch / 16;
3105 	else
3106 		flags |= pitch / 8;
3107 
3108 
3109 	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3110 	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3111 	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3112 	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3113 	return 0;
3114 }
3115 
3116 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3117 {
3118 	int surf_index = reg * 16;
3119 	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3120 }
3121 
3122 void r100_bandwidth_update(struct radeon_device *rdev)
3123 {
3124 	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3125 	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3126 	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
3127 	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3128 	fixed20_12 memtcas_ff[8] = {
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(0),
3136 	};
3137 	fixed20_12 memtcas_rs480_ff[8] = {
3138 		dfixed_init(0),
3139 		dfixed_init(1),
3140 		dfixed_init(2),
3141 		dfixed_init(3),
3142 		dfixed_init(0),
3143 		dfixed_init_half(1),
3144 		dfixed_init_half(2),
3145 		dfixed_init_half(3),
3146 	};
3147 	fixed20_12 memtcas2_ff[8] = {
3148 		dfixed_init(0),
3149 		dfixed_init(1),
3150 		dfixed_init(2),
3151 		dfixed_init(3),
3152 		dfixed_init(4),
3153 		dfixed_init(5),
3154 		dfixed_init(6),
3155 		dfixed_init(7),
3156 	};
3157 	fixed20_12 memtrbs[8] = {
3158 		dfixed_init(1),
3159 		dfixed_init_half(1),
3160 		dfixed_init(2),
3161 		dfixed_init_half(2),
3162 		dfixed_init(3),
3163 		dfixed_init_half(3),
3164 		dfixed_init(4),
3165 		dfixed_init_half(4)
3166 	};
3167 	fixed20_12 memtrbs_r4xx[8] = {
3168 		dfixed_init(4),
3169 		dfixed_init(5),
3170 		dfixed_init(6),
3171 		dfixed_init(7),
3172 		dfixed_init(8),
3173 		dfixed_init(9),
3174 		dfixed_init(10),
3175 		dfixed_init(11)
3176 	};
3177 	fixed20_12 min_mem_eff;
3178 	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3179 	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3180 	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
3181 		disp_drain_rate2, read_return_rate;
3182 	fixed20_12 time_disp1_drop_priority;
3183 	int c;
3184 	int cur_size = 16;       /* in octawords */
3185 	int critical_point = 0, critical_point2;
3186 /* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3187 	int stop_req, max_stop_req;
3188 	struct drm_display_mode *mode1 = NULL;
3189 	struct drm_display_mode *mode2 = NULL;
3190 	uint32_t pixel_bytes1 = 0;
3191 	uint32_t pixel_bytes2 = 0;
3192 
3193 	radeon_update_display_priority(rdev);
3194 
3195 	if (rdev->mode_info.crtcs[0]->base.enabled) {
3196 		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3197 		pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
3198 	}
3199 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3200 		if (rdev->mode_info.crtcs[1]->base.enabled) {
3201 			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3202 			pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
3203 		}
3204 	}
3205 
3206 	min_mem_eff.full = dfixed_const_8(0);
3207 	/* get modes */
3208 	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3209 		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3210 		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3211 		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3212 		/* check crtc enables */
3213 		if (mode2)
3214 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3215 		if (mode1)
3216 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3217 		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3218 	}
3219 
3220 	/*
3221 	 * determine is there is enough bw for current mode
3222 	 */
3223 	sclk_ff = rdev->pm.sclk;
3224 	mclk_ff = rdev->pm.mclk;
3225 
3226 	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3227 	temp_ff.full = dfixed_const(temp);
3228 	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3229 
3230 	pix_clk.full = 0;
3231 	pix_clk2.full = 0;
3232 	peak_disp_bw.full = 0;
3233 	if (mode1) {
3234 		temp_ff.full = dfixed_const(1000);
3235 		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3236 		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3237 		temp_ff.full = dfixed_const(pixel_bytes1);
3238 		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3239 	}
3240 	if (mode2) {
3241 		temp_ff.full = dfixed_const(1000);
3242 		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3243 		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3244 		temp_ff.full = dfixed_const(pixel_bytes2);
3245 		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3246 	}
3247 
3248 	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3249 	if (peak_disp_bw.full >= mem_bw.full) {
3250 		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3251 			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3252 	}
3253 
3254 	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3255 	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3256 	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3257 		mem_trcd = ((temp >> 2) & 0x3) + 1;
3258 		mem_trp  = ((temp & 0x3)) + 1;
3259 		mem_tras = ((temp & 0x70) >> 4) + 1;
3260 	} else if (rdev->family == CHIP_R300 ||
3261 		   rdev->family == CHIP_R350) { /* r300, r350 */
3262 		mem_trcd = (temp & 0x7) + 1;
3263 		mem_trp = ((temp >> 8) & 0x7) + 1;
3264 		mem_tras = ((temp >> 11) & 0xf) + 4;
3265 	} else if (rdev->family == CHIP_RV350 ||
3266 		   rdev->family <= CHIP_RV380) {
3267 		/* rv3x0 */
3268 		mem_trcd = (temp & 0x7) + 3;
3269 		mem_trp = ((temp >> 8) & 0x7) + 3;
3270 		mem_tras = ((temp >> 11) & 0xf) + 6;
3271 	} else if (rdev->family == CHIP_R420 ||
3272 		   rdev->family == CHIP_R423 ||
3273 		   rdev->family == CHIP_RV410) {
3274 		/* r4xx */
3275 		mem_trcd = (temp & 0xf) + 3;
3276 		if (mem_trcd > 15)
3277 			mem_trcd = 15;
3278 		mem_trp = ((temp >> 8) & 0xf) + 3;
3279 		if (mem_trp > 15)
3280 			mem_trp = 15;
3281 		mem_tras = ((temp >> 12) & 0x1f) + 6;
3282 		if (mem_tras > 31)
3283 			mem_tras = 31;
3284 	} else { /* RV200, R200 */
3285 		mem_trcd = (temp & 0x7) + 1;
3286 		mem_trp = ((temp >> 8) & 0x7) + 1;
3287 		mem_tras = ((temp >> 12) & 0xf) + 4;
3288 	}
3289 	/* convert to FF */
3290 	trcd_ff.full = dfixed_const(mem_trcd);
3291 	trp_ff.full = dfixed_const(mem_trp);
3292 	tras_ff.full = dfixed_const(mem_tras);
3293 
3294 	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3295 	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3296 	data = (temp & (7 << 20)) >> 20;
3297 	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3298 		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3299 			tcas_ff = memtcas_rs480_ff[data];
3300 		else
3301 			tcas_ff = memtcas_ff[data];
3302 	} else
3303 		tcas_ff = memtcas2_ff[data];
3304 
3305 	if (rdev->family == CHIP_RS400 ||
3306 	    rdev->family == CHIP_RS480) {
3307 		/* extra cas latency stored in bits 23-25 0-4 clocks */
3308 		data = (temp >> 23) & 0x7;
3309 		if (data < 5)
3310 			tcas_ff.full += dfixed_const(data);
3311 	}
3312 
3313 	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3314 		/* on the R300, Tcas is included in Trbs.
3315 		 */
3316 		temp = RREG32(RADEON_MEM_CNTL);
3317 		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3318 		if (data == 1) {
3319 			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3320 				temp = RREG32(R300_MC_IND_INDEX);
3321 				temp &= ~R300_MC_IND_ADDR_MASK;
3322 				temp |= R300_MC_READ_CNTL_CD_mcind;
3323 				WREG32(R300_MC_IND_INDEX, temp);
3324 				temp = RREG32(R300_MC_IND_DATA);
3325 				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3326 			} else {
3327 				temp = RREG32(R300_MC_READ_CNTL_AB);
3328 				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3329 			}
3330 		} else {
3331 			temp = RREG32(R300_MC_READ_CNTL_AB);
3332 			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3333 		}
3334 		if (rdev->family == CHIP_RV410 ||
3335 		    rdev->family == CHIP_R420 ||
3336 		    rdev->family == CHIP_R423)
3337 			trbs_ff = memtrbs_r4xx[data];
3338 		else
3339 			trbs_ff = memtrbs[data];
3340 		tcas_ff.full += trbs_ff.full;
3341 	}
3342 
3343 	sclk_eff_ff.full = sclk_ff.full;
3344 
3345 	if (rdev->flags & RADEON_IS_AGP) {
3346 		fixed20_12 agpmode_ff;
3347 		agpmode_ff.full = dfixed_const(radeon_agpmode);
3348 		temp_ff.full = dfixed_const_666(16);
3349 		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3350 	}
3351 	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3352 
3353 	if (ASIC_IS_R300(rdev)) {
3354 		sclk_delay_ff.full = dfixed_const(250);
3355 	} else {
3356 		if ((rdev->family == CHIP_RV100) ||
3357 		    rdev->flags & RADEON_IS_IGP) {
3358 			if (rdev->mc.vram_is_ddr)
3359 				sclk_delay_ff.full = dfixed_const(41);
3360 			else
3361 				sclk_delay_ff.full = dfixed_const(33);
3362 		} else {
3363 			if (rdev->mc.vram_width == 128)
3364 				sclk_delay_ff.full = dfixed_const(57);
3365 			else
3366 				sclk_delay_ff.full = dfixed_const(41);
3367 		}
3368 	}
3369 
3370 	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3371 
3372 	if (rdev->mc.vram_is_ddr) {
3373 		if (rdev->mc.vram_width == 32) {
3374 			k1.full = dfixed_const(40);
3375 			c  = 3;
3376 		} else {
3377 			k1.full = dfixed_const(20);
3378 			c  = 1;
3379 		}
3380 	} else {
3381 		k1.full = dfixed_const(40);
3382 		c  = 3;
3383 	}
3384 
3385 	temp_ff.full = dfixed_const(2);
3386 	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3387 	temp_ff.full = dfixed_const(c);
3388 	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3389 	temp_ff.full = dfixed_const(4);
3390 	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3391 	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3392 	mc_latency_mclk.full += k1.full;
3393 
3394 	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3395 	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3396 
3397 	/*
3398 	  HW cursor time assuming worst case of full size colour cursor.
3399 	*/
3400 	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3401 	temp_ff.full += trcd_ff.full;
3402 	if (temp_ff.full < tras_ff.full)
3403 		temp_ff.full = tras_ff.full;
3404 	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3405 
3406 	temp_ff.full = dfixed_const(cur_size);
3407 	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3408 	/*
3409 	  Find the total latency for the display data.
3410 	*/
3411 	disp_latency_overhead.full = dfixed_const(8);
3412 	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3413 	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3414 	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3415 
3416 	if (mc_latency_mclk.full > mc_latency_sclk.full)
3417 		disp_latency.full = mc_latency_mclk.full;
3418 	else
3419 		disp_latency.full = mc_latency_sclk.full;
3420 
3421 	/* setup Max GRPH_STOP_REQ default value */
3422 	if (ASIC_IS_RV100(rdev))
3423 		max_stop_req = 0x5c;
3424 	else
3425 		max_stop_req = 0x7c;
3426 
3427 	if (mode1) {
3428 		/*  CRTC1
3429 		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3430 		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3431 		*/
3432 		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3433 
3434 		if (stop_req > max_stop_req)
3435 			stop_req = max_stop_req;
3436 
3437 		/*
3438 		  Find the drain rate of the display buffer.
3439 		*/
3440 		temp_ff.full = dfixed_const((16/pixel_bytes1));
3441 		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3442 
3443 		/*
3444 		  Find the critical point of the display buffer.
3445 		*/
3446 		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3447 		crit_point_ff.full += dfixed_const_half(0);
3448 
3449 		critical_point = dfixed_trunc(crit_point_ff);
3450 
3451 		if (rdev->disp_priority == 2) {
3452 			critical_point = 0;
3453 		}
3454 
3455 		/*
3456 		  The critical point should never be above max_stop_req-4.  Setting
3457 		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3458 		*/
3459 		if (max_stop_req - critical_point < 4)
3460 			critical_point = 0;
3461 
3462 		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3463 			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3464 			critical_point = 0x10;
3465 		}
3466 
3467 		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3468 		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3469 		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3470 		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3471 		if ((rdev->family == CHIP_R350) &&
3472 		    (stop_req > 0x15)) {
3473 			stop_req -= 0x10;
3474 		}
3475 		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3476 		temp |= RADEON_GRPH_BUFFER_SIZE;
3477 		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3478 			  RADEON_GRPH_CRITICAL_AT_SOF |
3479 			  RADEON_GRPH_STOP_CNTL);
3480 		/*
3481 		  Write the result into the register.
3482 		*/
3483 		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3484 						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3485 
3486 #if 0
3487 		if ((rdev->family == CHIP_RS400) ||
3488 		    (rdev->family == CHIP_RS480)) {
3489 			/* attempt to program RS400 disp regs correctly ??? */
3490 			temp = RREG32(RS400_DISP1_REG_CNTL);
3491 			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3492 				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3493 			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3494 						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3495 						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3496 			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3497 			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3498 				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3499 			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3500 						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3501 						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3502 		}
3503 #endif
3504 
3505 		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3506 			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3507 			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3508 	}
3509 
3510 	if (mode2) {
3511 		u32 grph2_cntl;
3512 		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3513 
3514 		if (stop_req > max_stop_req)
3515 			stop_req = max_stop_req;
3516 
3517 		/*
3518 		  Find the drain rate of the display buffer.
3519 		*/
3520 		temp_ff.full = dfixed_const((16/pixel_bytes2));
3521 		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3522 
3523 		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3524 		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3525 		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3526 		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3527 		if ((rdev->family == CHIP_R350) &&
3528 		    (stop_req > 0x15)) {
3529 			stop_req -= 0x10;
3530 		}
3531 		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3532 		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3533 		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3534 			  RADEON_GRPH_CRITICAL_AT_SOF |
3535 			  RADEON_GRPH_STOP_CNTL);
3536 
3537 		if ((rdev->family == CHIP_RS100) ||
3538 		    (rdev->family == CHIP_RS200))
3539 			critical_point2 = 0;
3540 		else {
3541 			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3542 			temp_ff.full = dfixed_const(temp);
3543 			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3544 			if (sclk_ff.full < temp_ff.full)
3545 				temp_ff.full = sclk_ff.full;
3546 
3547 			read_return_rate.full = temp_ff.full;
3548 
3549 			if (mode1) {
3550 				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3551 				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3552 			} else {
3553 				time_disp1_drop_priority.full = 0;
3554 			}
3555 			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3556 			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3557 			crit_point_ff.full += dfixed_const_half(0);
3558 
3559 			critical_point2 = dfixed_trunc(crit_point_ff);
3560 
3561 			if (rdev->disp_priority == 2) {
3562 				critical_point2 = 0;
3563 			}
3564 
3565 			if (max_stop_req - critical_point2 < 4)
3566 				critical_point2 = 0;
3567 
3568 		}
3569 
3570 		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3571 			/* some R300 cards have problem with this set to 0 */
3572 			critical_point2 = 0x10;
3573 		}
3574 
3575 		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3576 						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3577 
3578 		if ((rdev->family == CHIP_RS400) ||
3579 		    (rdev->family == CHIP_RS480)) {
3580 #if 0
3581 			/* attempt to program RS400 disp2 regs correctly ??? */
3582 			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3583 			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3584 				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3585 			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3586 						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3587 						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3588 			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3589 			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3590 				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3591 			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3592 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3593 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3594 #endif
3595 			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3596 			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3597 			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3598 			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3599 		}
3600 
3601 		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3602 			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3603 	}
3604 }
3605 
3606 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3607 {
3608 	uint32_t scratch;
3609 	uint32_t tmp = 0;
3610 	unsigned i;
3611 	int r;
3612 
3613 	r = radeon_scratch_get(rdev, &scratch);
3614 	if (r) {
3615 		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3616 		return r;
3617 	}
3618 	WREG32(scratch, 0xCAFEDEAD);
3619 	r = radeon_ring_lock(rdev, ring, 2);
3620 	if (r) {
3621 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3622 		radeon_scratch_free(rdev, scratch);
3623 		return r;
3624 	}
3625 	radeon_ring_write(ring, PACKET0(scratch, 0));
3626 	radeon_ring_write(ring, 0xDEADBEEF);
3627 	radeon_ring_unlock_commit(rdev, ring);
3628 	for (i = 0; i < rdev->usec_timeout; i++) {
3629 		tmp = RREG32(scratch);
3630 		if (tmp == 0xDEADBEEF) {
3631 			break;
3632 		}
3633 		DRM_UDELAY(1);
3634 	}
3635 	if (i < rdev->usec_timeout) {
3636 		DRM_INFO("ring test succeeded in %d usecs\n", i);
3637 	} else {
3638 		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3639 			  scratch, tmp);
3640 		r = -EINVAL;
3641 	}
3642 	radeon_scratch_free(rdev, scratch);
3643 	return r;
3644 }
3645 
3646 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3647 {
3648 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3649 
3650 	if (ring->rptr_save_reg) {
3651 		u32 next_rptr = ring->wptr + 2 + 3;
3652 		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3653 		radeon_ring_write(ring, next_rptr);
3654 	}
3655 
3656 	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3657 	radeon_ring_write(ring, ib->gpu_addr);
3658 	radeon_ring_write(ring, ib->length_dw);
3659 }
3660 
3661 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3662 {
3663 	struct radeon_ib ib;
3664 	uint32_t scratch;
3665 	uint32_t tmp = 0;
3666 	unsigned i;
3667 	int r;
3668 
3669 	r = radeon_scratch_get(rdev, &scratch);
3670 	if (r) {
3671 		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3672 		return r;
3673 	}
3674 	WREG32(scratch, 0xCAFEDEAD);
3675 	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3676 	if (r) {
3677 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3678 		goto free_scratch;
3679 	}
3680 	ib.ptr[0] = PACKET0(scratch, 0);
3681 	ib.ptr[1] = 0xDEADBEEF;
3682 	ib.ptr[2] = PACKET2(0);
3683 	ib.ptr[3] = PACKET2(0);
3684 	ib.ptr[4] = PACKET2(0);
3685 	ib.ptr[5] = PACKET2(0);
3686 	ib.ptr[6] = PACKET2(0);
3687 	ib.ptr[7] = PACKET2(0);
3688 	ib.length_dw = 8;
3689 	r = radeon_ib_schedule(rdev, &ib, NULL);
3690 	if (r) {
3691 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3692 		goto free_ib;
3693 	}
3694 	r = radeon_fence_wait(ib.fence, false);
3695 	if (r) {
3696 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3697 		goto free_ib;
3698 	}
3699 	for (i = 0; i < rdev->usec_timeout; i++) {
3700 		tmp = RREG32(scratch);
3701 		if (tmp == 0xDEADBEEF) {
3702 			break;
3703 		}
3704 		DRM_UDELAY(1);
3705 	}
3706 	if (i < rdev->usec_timeout) {
3707 		DRM_INFO("ib test succeeded in %u usecs\n", i);
3708 	} else {
3709 		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3710 			  scratch, tmp);
3711 		r = -EINVAL;
3712 	}
3713 free_ib:
3714 	radeon_ib_free(rdev, &ib);
3715 free_scratch:
3716 	radeon_scratch_free(rdev, scratch);
3717 	return r;
3718 }
3719 
3720 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3721 {
3722 	/* Shutdown CP we shouldn't need to do that but better be safe than
3723 	 * sorry
3724 	 */
3725 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3726 	WREG32(R_000740_CP_CSQ_CNTL, 0);
3727 
3728 	/* Save few CRTC registers */
3729 	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3730 	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3731 	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3732 	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3733 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3734 		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3735 		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3736 	}
3737 
3738 	/* Disable VGA aperture access */
3739 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3740 	/* Disable cursor, overlay, crtc */
3741 	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3742 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3743 					S_000054_CRTC_DISPLAY_DIS(1));
3744 	WREG32(R_000050_CRTC_GEN_CNTL,
3745 			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3746 			S_000050_CRTC_DISP_REQ_EN_B(1));
3747 	WREG32(R_000420_OV0_SCALE_CNTL,
3748 		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3749 	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3750 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3751 		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3752 						S_000360_CUR2_LOCK(1));
3753 		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3754 			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3755 			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3756 			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3757 		WREG32(R_000360_CUR2_OFFSET,
3758 			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3759 	}
3760 }
3761 
3762 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3763 {
3764 	/* Update base address for crtc */
3765 	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3766 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3767 		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3768 	}
3769 	/* Restore CRTC registers */
3770 	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3771 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3772 	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3773 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3774 		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3775 	}
3776 }
3777 
3778 void r100_vga_render_disable(struct radeon_device *rdev)
3779 {
3780 	u32 tmp;
3781 
3782 	tmp = RREG8(R_0003C2_GENMO_WT);
3783 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3784 }
3785 
3786 static void r100_debugfs(struct radeon_device *rdev)
3787 {
3788 	int r;
3789 
3790 	r = r100_debugfs_mc_info_init(rdev);
3791 	if (r)
3792 		dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3793 }
3794 
3795 static void r100_mc_program(struct radeon_device *rdev)
3796 {
3797 	struct r100_mc_save save;
3798 
3799 	/* Stops all mc clients */
3800 	r100_mc_stop(rdev, &save);
3801 	if (rdev->flags & RADEON_IS_AGP) {
3802 		WREG32(R_00014C_MC_AGP_LOCATION,
3803 			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3804 			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3805 		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3806 		if (rdev->family > CHIP_RV200)
3807 			WREG32(R_00015C_AGP_BASE_2,
3808 				upper_32_bits(rdev->mc.agp_base) & 0xff);
3809 	} else {
3810 		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3811 		WREG32(R_000170_AGP_BASE, 0);
3812 		if (rdev->family > CHIP_RV200)
3813 			WREG32(R_00015C_AGP_BASE_2, 0);
3814 	}
3815 	/* Wait for mc idle */
3816 	if (r100_mc_wait_for_idle(rdev))
3817 		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3818 	/* Program MC, should be a 32bits limited address space */
3819 	WREG32(R_000148_MC_FB_LOCATION,
3820 		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3821 		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3822 	r100_mc_resume(rdev, &save);
3823 }
3824 
3825 static void r100_clock_startup(struct radeon_device *rdev)
3826 {
3827 	u32 tmp;
3828 
3829 	if (radeon_dynclks != -1 && radeon_dynclks)
3830 		radeon_legacy_set_clock_gating(rdev, 1);
3831 	/* We need to force on some of the block */
3832 	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3833 	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3834 	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3835 		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3836 	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3837 }
3838 
3839 static int r100_startup(struct radeon_device *rdev)
3840 {
3841 	int r;
3842 
3843 	/* set common regs */
3844 	r100_set_common_regs(rdev);
3845 	/* program mc */
3846 	r100_mc_program(rdev);
3847 	/* Resume clock */
3848 	r100_clock_startup(rdev);
3849 	/* Initialize GART (initialize after TTM so we can allocate
3850 	 * memory through TTM but finalize after TTM) */
3851 	r100_enable_bm(rdev);
3852 	if (rdev->flags & RADEON_IS_PCI) {
3853 		r = r100_pci_gart_enable(rdev);
3854 		if (r)
3855 			return r;
3856 	}
3857 
3858 	/* allocate wb buffer */
3859 	r = radeon_wb_init(rdev);
3860 	if (r)
3861 		return r;
3862 
3863 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3864 	if (r) {
3865 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3866 		return r;
3867 	}
3868 
3869 	/* Enable IRQ */
3870 	if (!rdev->irq.installed) {
3871 		r = radeon_irq_kms_init(rdev);
3872 		if (r)
3873 			return r;
3874 	}
3875 
3876 	r100_irq_set(rdev);
3877 	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3878 	/* 1M ring buffer */
3879 	r = r100_cp_init(rdev, 1024 * 1024);
3880 	if (r) {
3881 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3882 		return r;
3883 	}
3884 
3885 	r = radeon_ib_pool_init(rdev);
3886 	if (r) {
3887 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3888 		return r;
3889 	}
3890 
3891 	return 0;
3892 }
3893 
3894 int r100_resume(struct radeon_device *rdev)
3895 {
3896 	int r;
3897 
3898 	/* Make sur GART are not working */
3899 	if (rdev->flags & RADEON_IS_PCI)
3900 		r100_pci_gart_disable(rdev);
3901 	/* Resume clock before doing reset */
3902 	r100_clock_startup(rdev);
3903 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3904 	if (radeon_asic_reset(rdev)) {
3905 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3906 			RREG32(R_000E40_RBBM_STATUS),
3907 			RREG32(R_0007C0_CP_STAT));
3908 	}
3909 	/* post */
3910 	radeon_combios_asic_init(rdev->ddev);
3911 	/* Resume clock after posting */
3912 	r100_clock_startup(rdev);
3913 	/* Initialize surface registers */
3914 	radeon_surface_init(rdev);
3915 
3916 	rdev->accel_working = true;
3917 	r = r100_startup(rdev);
3918 	if (r) {
3919 		rdev->accel_working = false;
3920 	}
3921 	return r;
3922 }
3923 
3924 int r100_suspend(struct radeon_device *rdev)
3925 {
3926 	r100_cp_disable(rdev);
3927 	radeon_wb_disable(rdev);
3928 	r100_irq_disable(rdev);
3929 	if (rdev->flags & RADEON_IS_PCI)
3930 		r100_pci_gart_disable(rdev);
3931 	return 0;
3932 }
3933 
3934 void r100_fini(struct radeon_device *rdev)
3935 {
3936 	r100_cp_fini(rdev);
3937 	radeon_wb_fini(rdev);
3938 	radeon_ib_pool_fini(rdev);
3939 	radeon_gem_fini(rdev);
3940 	if (rdev->flags & RADEON_IS_PCI)
3941 		r100_pci_gart_fini(rdev);
3942 	radeon_agp_fini(rdev);
3943 	radeon_irq_kms_fini(rdev);
3944 	radeon_fence_driver_fini(rdev);
3945 	radeon_bo_fini(rdev);
3946 	radeon_atombios_fini(rdev);
3947 	kfree(rdev->bios);
3948 	rdev->bios = NULL;
3949 }
3950 
3951 /*
3952  * Due to how kexec works, it can leave the hw fully initialised when it
3953  * boots the new kernel. However doing our init sequence with the CP and
3954  * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3955  * do some quick sanity checks and restore sane values to avoid this
3956  * problem.
3957  */
3958 void r100_restore_sanity(struct radeon_device *rdev)
3959 {
3960 	u32 tmp;
3961 
3962 	tmp = RREG32(RADEON_CP_CSQ_CNTL);
3963 	if (tmp) {
3964 		WREG32(RADEON_CP_CSQ_CNTL, 0);
3965 	}
3966 	tmp = RREG32(RADEON_CP_RB_CNTL);
3967 	if (tmp) {
3968 		WREG32(RADEON_CP_RB_CNTL, 0);
3969 	}
3970 	tmp = RREG32(RADEON_SCRATCH_UMSK);
3971 	if (tmp) {
3972 		WREG32(RADEON_SCRATCH_UMSK, 0);
3973 	}
3974 }
3975 
3976 int r100_init(struct radeon_device *rdev)
3977 {
3978 	int r;
3979 
3980 	/* Register debugfs file specific to this group of asics */
3981 	r100_debugfs(rdev);
3982 	/* Disable VGA */
3983 	r100_vga_render_disable(rdev);
3984 	/* Initialize scratch registers */
3985 	radeon_scratch_init(rdev);
3986 	/* Initialize surface registers */
3987 	radeon_surface_init(rdev);
3988 	/* sanity check some register to avoid hangs like after kexec */
3989 	r100_restore_sanity(rdev);
3990 	/* TODO: disable VGA need to use VGA request */
3991 	/* BIOS*/
3992 	if (!radeon_get_bios(rdev)) {
3993 		if (ASIC_IS_AVIVO(rdev))
3994 			return -EINVAL;
3995 	}
3996 	if (rdev->is_atom_bios) {
3997 		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
3998 		return -EINVAL;
3999 	} else {
4000 		r = radeon_combios_init(rdev);
4001 		if (r)
4002 			return r;
4003 	}
4004 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4005 	if (radeon_asic_reset(rdev)) {
4006 		dev_warn(rdev->dev,
4007 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4008 			RREG32(R_000E40_RBBM_STATUS),
4009 			RREG32(R_0007C0_CP_STAT));
4010 	}
4011 	/* check if cards are posted or not */
4012 	if (radeon_boot_test_post_card(rdev) == false)
4013 		return -EINVAL;
4014 	/* Set asic errata */
4015 	r100_errata(rdev);
4016 	/* Initialize clocks */
4017 	radeon_get_clock_info(rdev->ddev);
4018 	/* initialize AGP */
4019 	if (rdev->flags & RADEON_IS_AGP) {
4020 		r = radeon_agp_init(rdev);
4021 		if (r) {
4022 			radeon_agp_disable(rdev);
4023 		}
4024 	}
4025 	/* initialize VRAM */
4026 	r100_mc_init(rdev);
4027 	/* Fence driver */
4028 	r = radeon_fence_driver_init(rdev);
4029 	if (r)
4030 		return r;
4031 	/* Memory manager */
4032 	r = radeon_bo_init(rdev);
4033 	if (r)
4034 		return r;
4035 	if (rdev->flags & RADEON_IS_PCI) {
4036 		r = r100_pci_gart_init(rdev);
4037 		if (r)
4038 			return r;
4039 	}
4040 	r100_set_safe_registers(rdev);
4041 
4042 	rdev->accel_working = true;
4043 	r = r100_startup(rdev);
4044 	if (r) {
4045 		/* Somethings want wront with the accel init stop accel */
4046 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4047 		r100_cp_fini(rdev);
4048 		radeon_wb_fini(rdev);
4049 		radeon_ib_pool_fini(rdev);
4050 		radeon_irq_kms_fini(rdev);
4051 		if (rdev->flags & RADEON_IS_PCI)
4052 			r100_pci_gart_fini(rdev);
4053 		rdev->accel_working = false;
4054 	}
4055 	return 0;
4056 }
4057 
4058 uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg,
4059 		      bool always_indirect)
4060 {
4061 	if (reg < rdev->rmmio_size && !always_indirect)
4062 		return readl(((void __iomem *)rdev->rmmio) + reg);
4063 	else {
4064 		unsigned long flags;
4065 		uint32_t ret;
4066 
4067 		spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4068 		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4069 		ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4070 		spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4071 
4072 		return ret;
4073 	}
4074 }
4075 
4076 void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v,
4077 		  bool always_indirect)
4078 {
4079 	if (reg < rdev->rmmio_size && !always_indirect)
4080 		writel(v, ((void __iomem *)rdev->rmmio) + reg);
4081 	else {
4082 		unsigned long flags;
4083 
4084 		spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4085 		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4086 		writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4087 		spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4088 	}
4089 }
4090 
4091 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4092 {
4093 	if (reg < rdev->rio_mem_size)
4094 		return ioread32(rdev->rio_mem + reg);
4095 	else {
4096 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4097 		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4098 	}
4099 }
4100 
4101 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4102 {
4103 	if (reg < rdev->rio_mem_size)
4104 		iowrite32(v, rdev->rio_mem + reg);
4105 	else {
4106 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4107 		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4108 	}
4109 }
4110