xref: /openbmc/linux/drivers/gpu/drm/radeon/r100.c (revision 94cdda6b)
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_page_flip - pageflip callback.
146  *
147  * @rdev: radeon_device pointer
148  * @crtc_id: crtc to cleanup pageflip on
149  * @crtc_base: new address of the crtc (GPU MC address)
150  *
151  * Does the actual pageflip (r1xx-r4xx).
152  * During vblank we take the crtc lock and wait for the update_pending
153  * bit to go high, when it does, we release the lock, and allow the
154  * double buffered update to take place.
155  */
156 void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
157 {
158 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
159 	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
160 	int i;
161 
162 	/* Lock the graphics update lock */
163 	/* update the scanout addresses */
164 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
165 
166 	/* Wait for update_pending to go high. */
167 	for (i = 0; i < rdev->usec_timeout; i++) {
168 		if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
169 			break;
170 		udelay(1);
171 	}
172 	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
173 
174 	/* Unlock the lock, so double-buffering can take place inside vblank */
175 	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
176 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
177 
178 }
179 
180 /**
181  * r100_page_flip_pending - check if page flip is still pending
182  *
183  * @rdev: radeon_device pointer
184  * @crtc_id: crtc to check
185  *
186  * Check if the last pagefilp is still pending (r1xx-r4xx).
187  * Returns the current update pending status.
188  */
189 bool r100_page_flip_pending(struct radeon_device *rdev, int crtc_id)
190 {
191 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
192 
193 	/* Return current update_pending status: */
194 	return !!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) &
195 		RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET);
196 }
197 
198 /**
199  * r100_pm_get_dynpm_state - look up dynpm power state callback.
200  *
201  * @rdev: radeon_device pointer
202  *
203  * Look up the optimal power state based on the
204  * current state of the GPU (r1xx-r5xx).
205  * Used for dynpm only.
206  */
207 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
208 {
209 	int i;
210 	rdev->pm.dynpm_can_upclock = true;
211 	rdev->pm.dynpm_can_downclock = true;
212 
213 	switch (rdev->pm.dynpm_planned_action) {
214 	case DYNPM_ACTION_MINIMUM:
215 		rdev->pm.requested_power_state_index = 0;
216 		rdev->pm.dynpm_can_downclock = false;
217 		break;
218 	case DYNPM_ACTION_DOWNCLOCK:
219 		if (rdev->pm.current_power_state_index == 0) {
220 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
221 			rdev->pm.dynpm_can_downclock = false;
222 		} else {
223 			if (rdev->pm.active_crtc_count > 1) {
224 				for (i = 0; i < rdev->pm.num_power_states; i++) {
225 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
226 						continue;
227 					else if (i >= rdev->pm.current_power_state_index) {
228 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
229 						break;
230 					} else {
231 						rdev->pm.requested_power_state_index = i;
232 						break;
233 					}
234 				}
235 			} else
236 				rdev->pm.requested_power_state_index =
237 					rdev->pm.current_power_state_index - 1;
238 		}
239 		/* don't use the power state if crtcs are active and no display flag is set */
240 		if ((rdev->pm.active_crtc_count > 0) &&
241 		    (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
242 		     RADEON_PM_MODE_NO_DISPLAY)) {
243 			rdev->pm.requested_power_state_index++;
244 		}
245 		break;
246 	case DYNPM_ACTION_UPCLOCK:
247 		if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
248 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
249 			rdev->pm.dynpm_can_upclock = false;
250 		} else {
251 			if (rdev->pm.active_crtc_count > 1) {
252 				for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
253 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
254 						continue;
255 					else if (i <= rdev->pm.current_power_state_index) {
256 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
257 						break;
258 					} else {
259 						rdev->pm.requested_power_state_index = i;
260 						break;
261 					}
262 				}
263 			} else
264 				rdev->pm.requested_power_state_index =
265 					rdev->pm.current_power_state_index + 1;
266 		}
267 		break;
268 	case DYNPM_ACTION_DEFAULT:
269 		rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
270 		rdev->pm.dynpm_can_upclock = false;
271 		break;
272 	case DYNPM_ACTION_NONE:
273 	default:
274 		DRM_ERROR("Requested mode for not defined action\n");
275 		return;
276 	}
277 	/* only one clock mode per power state */
278 	rdev->pm.requested_clock_mode_index = 0;
279 
280 	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
281 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
282 		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
283 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
284 		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
285 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
286 		  pcie_lanes);
287 }
288 
289 /**
290  * r100_pm_init_profile - Initialize power profiles callback.
291  *
292  * @rdev: radeon_device pointer
293  *
294  * Initialize the power states used in profile mode
295  * (r1xx-r3xx).
296  * Used for profile mode only.
297  */
298 void r100_pm_init_profile(struct radeon_device *rdev)
299 {
300 	/* default */
301 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
302 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
303 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
304 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
305 	/* low sh */
306 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
307 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
308 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
309 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
310 	/* mid sh */
311 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
312 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
313 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
314 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
315 	/* high sh */
316 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
317 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
318 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
319 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
320 	/* low mh */
321 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
322 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
323 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
324 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
325 	/* mid mh */
326 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
327 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
328 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
329 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
330 	/* high mh */
331 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
332 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
333 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
334 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
335 }
336 
337 /**
338  * r100_pm_misc - set additional pm hw parameters callback.
339  *
340  * @rdev: radeon_device pointer
341  *
342  * Set non-clock parameters associated with a power state
343  * (voltage, pcie lanes, etc.) (r1xx-r4xx).
344  */
345 void r100_pm_misc(struct radeon_device *rdev)
346 {
347 	int requested_index = rdev->pm.requested_power_state_index;
348 	struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
349 	struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
350 	u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
351 
352 	if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
353 		if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
354 			tmp = RREG32(voltage->gpio.reg);
355 			if (voltage->active_high)
356 				tmp |= voltage->gpio.mask;
357 			else
358 				tmp &= ~(voltage->gpio.mask);
359 			WREG32(voltage->gpio.reg, tmp);
360 			if (voltage->delay)
361 				udelay(voltage->delay);
362 		} else {
363 			tmp = RREG32(voltage->gpio.reg);
364 			if (voltage->active_high)
365 				tmp &= ~voltage->gpio.mask;
366 			else
367 				tmp |= voltage->gpio.mask;
368 			WREG32(voltage->gpio.reg, tmp);
369 			if (voltage->delay)
370 				udelay(voltage->delay);
371 		}
372 	}
373 
374 	sclk_cntl = RREG32_PLL(SCLK_CNTL);
375 	sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
376 	sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
377 	sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
378 	sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
379 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
380 		sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
381 		if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
382 			sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
383 		else
384 			sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
385 		if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
386 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
387 		else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
388 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
389 	} else
390 		sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
391 
392 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
393 		sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
394 		if (voltage->delay) {
395 			sclk_more_cntl |= VOLTAGE_DROP_SYNC;
396 			switch (voltage->delay) {
397 			case 33:
398 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
399 				break;
400 			case 66:
401 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
402 				break;
403 			case 99:
404 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
405 				break;
406 			case 132:
407 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
408 				break;
409 			}
410 		} else
411 			sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
412 	} else
413 		sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
414 
415 	if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
416 		sclk_cntl &= ~FORCE_HDP;
417 	else
418 		sclk_cntl |= FORCE_HDP;
419 
420 	WREG32_PLL(SCLK_CNTL, sclk_cntl);
421 	WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
422 	WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
423 
424 	/* set pcie lanes */
425 	if ((rdev->flags & RADEON_IS_PCIE) &&
426 	    !(rdev->flags & RADEON_IS_IGP) &&
427 	    rdev->asic->pm.set_pcie_lanes &&
428 	    (ps->pcie_lanes !=
429 	     rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
430 		radeon_set_pcie_lanes(rdev,
431 				      ps->pcie_lanes);
432 		DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
433 	}
434 }
435 
436 /**
437  * r100_pm_prepare - pre-power state change callback.
438  *
439  * @rdev: radeon_device pointer
440  *
441  * Prepare for a power state change (r1xx-r4xx).
442  */
443 void r100_pm_prepare(struct radeon_device *rdev)
444 {
445 	struct drm_device *ddev = rdev->ddev;
446 	struct drm_crtc *crtc;
447 	struct radeon_crtc *radeon_crtc;
448 	u32 tmp;
449 
450 	/* disable any active CRTCs */
451 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
452 		radeon_crtc = to_radeon_crtc(crtc);
453 		if (radeon_crtc->enabled) {
454 			if (radeon_crtc->crtc_id) {
455 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
456 				tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
457 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
458 			} else {
459 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
460 				tmp |= RADEON_CRTC_DISP_REQ_EN_B;
461 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
462 			}
463 		}
464 	}
465 }
466 
467 /**
468  * r100_pm_finish - post-power state change callback.
469  *
470  * @rdev: radeon_device pointer
471  *
472  * Clean up after a power state change (r1xx-r4xx).
473  */
474 void r100_pm_finish(struct radeon_device *rdev)
475 {
476 	struct drm_device *ddev = rdev->ddev;
477 	struct drm_crtc *crtc;
478 	struct radeon_crtc *radeon_crtc;
479 	u32 tmp;
480 
481 	/* enable any active CRTCs */
482 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
483 		radeon_crtc = to_radeon_crtc(crtc);
484 		if (radeon_crtc->enabled) {
485 			if (radeon_crtc->crtc_id) {
486 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
487 				tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
488 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
489 			} else {
490 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
491 				tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
492 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
493 			}
494 		}
495 	}
496 }
497 
498 /**
499  * r100_gui_idle - gui idle callback.
500  *
501  * @rdev: radeon_device pointer
502  *
503  * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
504  * Returns true if idle, false if not.
505  */
506 bool r100_gui_idle(struct radeon_device *rdev)
507 {
508 	if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
509 		return false;
510 	else
511 		return true;
512 }
513 
514 /* hpd for digital panel detect/disconnect */
515 /**
516  * r100_hpd_sense - hpd sense callback.
517  *
518  * @rdev: radeon_device pointer
519  * @hpd: hpd (hotplug detect) pin
520  *
521  * Checks if a digital monitor is connected (r1xx-r4xx).
522  * Returns true if connected, false if not connected.
523  */
524 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
525 {
526 	bool connected = false;
527 
528 	switch (hpd) {
529 	case RADEON_HPD_1:
530 		if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
531 			connected = true;
532 		break;
533 	case RADEON_HPD_2:
534 		if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
535 			connected = true;
536 		break;
537 	default:
538 		break;
539 	}
540 	return connected;
541 }
542 
543 /**
544  * r100_hpd_set_polarity - hpd set polarity callback.
545  *
546  * @rdev: radeon_device pointer
547  * @hpd: hpd (hotplug detect) pin
548  *
549  * Set the polarity of the hpd pin (r1xx-r4xx).
550  */
551 void r100_hpd_set_polarity(struct radeon_device *rdev,
552 			   enum radeon_hpd_id hpd)
553 {
554 	u32 tmp;
555 	bool connected = r100_hpd_sense(rdev, hpd);
556 
557 	switch (hpd) {
558 	case RADEON_HPD_1:
559 		tmp = RREG32(RADEON_FP_GEN_CNTL);
560 		if (connected)
561 			tmp &= ~RADEON_FP_DETECT_INT_POL;
562 		else
563 			tmp |= RADEON_FP_DETECT_INT_POL;
564 		WREG32(RADEON_FP_GEN_CNTL, tmp);
565 		break;
566 	case RADEON_HPD_2:
567 		tmp = RREG32(RADEON_FP2_GEN_CNTL);
568 		if (connected)
569 			tmp &= ~RADEON_FP2_DETECT_INT_POL;
570 		else
571 			tmp |= RADEON_FP2_DETECT_INT_POL;
572 		WREG32(RADEON_FP2_GEN_CNTL, tmp);
573 		break;
574 	default:
575 		break;
576 	}
577 }
578 
579 /**
580  * r100_hpd_init - hpd setup callback.
581  *
582  * @rdev: radeon_device pointer
583  *
584  * Setup the hpd pins used by the card (r1xx-r4xx).
585  * Set the polarity, and enable the hpd interrupts.
586  */
587 void r100_hpd_init(struct radeon_device *rdev)
588 {
589 	struct drm_device *dev = rdev->ddev;
590 	struct drm_connector *connector;
591 	unsigned enable = 0;
592 
593 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
594 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
595 		enable |= 1 << radeon_connector->hpd.hpd;
596 		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
597 	}
598 	radeon_irq_kms_enable_hpd(rdev, enable);
599 }
600 
601 /**
602  * r100_hpd_fini - hpd tear down callback.
603  *
604  * @rdev: radeon_device pointer
605  *
606  * Tear down the hpd pins used by the card (r1xx-r4xx).
607  * Disable the hpd interrupts.
608  */
609 void r100_hpd_fini(struct radeon_device *rdev)
610 {
611 	struct drm_device *dev = rdev->ddev;
612 	struct drm_connector *connector;
613 	unsigned disable = 0;
614 
615 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
616 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
617 		disable |= 1 << radeon_connector->hpd.hpd;
618 	}
619 	radeon_irq_kms_disable_hpd(rdev, disable);
620 }
621 
622 /*
623  * PCI GART
624  */
625 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
626 {
627 	/* TODO: can we do somethings here ? */
628 	/* It seems hw only cache one entry so we should discard this
629 	 * entry otherwise if first GPU GART read hit this entry it
630 	 * could end up in wrong address. */
631 }
632 
633 int r100_pci_gart_init(struct radeon_device *rdev)
634 {
635 	int r;
636 
637 	if (rdev->gart.ptr) {
638 		WARN(1, "R100 PCI GART already initialized\n");
639 		return 0;
640 	}
641 	/* Initialize common gart structure */
642 	r = radeon_gart_init(rdev);
643 	if (r)
644 		return r;
645 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
646 	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
647 	rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
648 	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
649 	return radeon_gart_table_ram_alloc(rdev);
650 }
651 
652 int r100_pci_gart_enable(struct radeon_device *rdev)
653 {
654 	uint32_t tmp;
655 
656 	/* discard memory request outside of configured range */
657 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
658 	WREG32(RADEON_AIC_CNTL, tmp);
659 	/* set address range for PCI address translate */
660 	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
661 	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
662 	/* set PCI GART page-table base address */
663 	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
664 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
665 	WREG32(RADEON_AIC_CNTL, tmp);
666 	r100_pci_gart_tlb_flush(rdev);
667 	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
668 		 (unsigned)(rdev->mc.gtt_size >> 20),
669 		 (unsigned long long)rdev->gart.table_addr);
670 	rdev->gart.ready = true;
671 	return 0;
672 }
673 
674 void r100_pci_gart_disable(struct radeon_device *rdev)
675 {
676 	uint32_t tmp;
677 
678 	/* discard memory request outside of configured range */
679 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
680 	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
681 	WREG32(RADEON_AIC_LO_ADDR, 0);
682 	WREG32(RADEON_AIC_HI_ADDR, 0);
683 }
684 
685 uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
686 {
687 	return addr;
688 }
689 
690 void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
691 			    uint64_t entry)
692 {
693 	u32 *gtt = rdev->gart.ptr;
694 	gtt[i] = cpu_to_le32(lower_32_bits(entry));
695 }
696 
697 void r100_pci_gart_fini(struct radeon_device *rdev)
698 {
699 	radeon_gart_fini(rdev);
700 	r100_pci_gart_disable(rdev);
701 	radeon_gart_table_ram_free(rdev);
702 }
703 
704 int r100_irq_set(struct radeon_device *rdev)
705 {
706 	uint32_t tmp = 0;
707 
708 	if (!rdev->irq.installed) {
709 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
710 		WREG32(R_000040_GEN_INT_CNTL, 0);
711 		return -EINVAL;
712 	}
713 	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
714 		tmp |= RADEON_SW_INT_ENABLE;
715 	}
716 	if (rdev->irq.crtc_vblank_int[0] ||
717 	    atomic_read(&rdev->irq.pflip[0])) {
718 		tmp |= RADEON_CRTC_VBLANK_MASK;
719 	}
720 	if (rdev->irq.crtc_vblank_int[1] ||
721 	    atomic_read(&rdev->irq.pflip[1])) {
722 		tmp |= RADEON_CRTC2_VBLANK_MASK;
723 	}
724 	if (rdev->irq.hpd[0]) {
725 		tmp |= RADEON_FP_DETECT_MASK;
726 	}
727 	if (rdev->irq.hpd[1]) {
728 		tmp |= RADEON_FP2_DETECT_MASK;
729 	}
730 	WREG32(RADEON_GEN_INT_CNTL, tmp);
731 
732 	/* read back to post the write */
733 	RREG32(RADEON_GEN_INT_CNTL);
734 
735 	return 0;
736 }
737 
738 void r100_irq_disable(struct radeon_device *rdev)
739 {
740 	u32 tmp;
741 
742 	WREG32(R_000040_GEN_INT_CNTL, 0);
743 	/* Wait and acknowledge irq */
744 	mdelay(1);
745 	tmp = RREG32(R_000044_GEN_INT_STATUS);
746 	WREG32(R_000044_GEN_INT_STATUS, tmp);
747 }
748 
749 static uint32_t r100_irq_ack(struct radeon_device *rdev)
750 {
751 	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
752 	uint32_t irq_mask = RADEON_SW_INT_TEST |
753 		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
754 		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
755 
756 	if (irqs) {
757 		WREG32(RADEON_GEN_INT_STATUS, irqs);
758 	}
759 	return irqs & irq_mask;
760 }
761 
762 int r100_irq_process(struct radeon_device *rdev)
763 {
764 	uint32_t status, msi_rearm;
765 	bool queue_hotplug = false;
766 
767 	status = r100_irq_ack(rdev);
768 	if (!status) {
769 		return IRQ_NONE;
770 	}
771 	if (rdev->shutdown) {
772 		return IRQ_NONE;
773 	}
774 	while (status) {
775 		/* SW interrupt */
776 		if (status & RADEON_SW_INT_TEST) {
777 			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
778 		}
779 		/* Vertical blank interrupts */
780 		if (status & RADEON_CRTC_VBLANK_STAT) {
781 			if (rdev->irq.crtc_vblank_int[0]) {
782 				drm_handle_vblank(rdev->ddev, 0);
783 				rdev->pm.vblank_sync = true;
784 				wake_up(&rdev->irq.vblank_queue);
785 			}
786 			if (atomic_read(&rdev->irq.pflip[0]))
787 				radeon_crtc_handle_vblank(rdev, 0);
788 		}
789 		if (status & RADEON_CRTC2_VBLANK_STAT) {
790 			if (rdev->irq.crtc_vblank_int[1]) {
791 				drm_handle_vblank(rdev->ddev, 1);
792 				rdev->pm.vblank_sync = true;
793 				wake_up(&rdev->irq.vblank_queue);
794 			}
795 			if (atomic_read(&rdev->irq.pflip[1]))
796 				radeon_crtc_handle_vblank(rdev, 1);
797 		}
798 		if (status & RADEON_FP_DETECT_STAT) {
799 			queue_hotplug = true;
800 			DRM_DEBUG("HPD1\n");
801 		}
802 		if (status & RADEON_FP2_DETECT_STAT) {
803 			queue_hotplug = true;
804 			DRM_DEBUG("HPD2\n");
805 		}
806 		status = r100_irq_ack(rdev);
807 	}
808 	if (queue_hotplug)
809 		schedule_work(&rdev->hotplug_work);
810 	if (rdev->msi_enabled) {
811 		switch (rdev->family) {
812 		case CHIP_RS400:
813 		case CHIP_RS480:
814 			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
815 			WREG32(RADEON_AIC_CNTL, msi_rearm);
816 			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
817 			break;
818 		default:
819 			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
820 			break;
821 		}
822 	}
823 	return IRQ_HANDLED;
824 }
825 
826 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
827 {
828 	if (crtc == 0)
829 		return RREG32(RADEON_CRTC_CRNT_FRAME);
830 	else
831 		return RREG32(RADEON_CRTC2_CRNT_FRAME);
832 }
833 
834 /**
835  * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
836  * rdev: radeon device structure
837  * ring: ring buffer struct for emitting packets
838  */
839 static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
840 {
841 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
842 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
843 				RADEON_HDP_READ_BUFFER_INVALIDATE);
844 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
845 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
846 }
847 
848 /* Who ever call radeon_fence_emit should call ring_lock and ask
849  * for enough space (today caller are ib schedule and buffer move) */
850 void r100_fence_ring_emit(struct radeon_device *rdev,
851 			  struct radeon_fence *fence)
852 {
853 	struct radeon_ring *ring = &rdev->ring[fence->ring];
854 
855 	/* We have to make sure that caches are flushed before
856 	 * CPU might read something from VRAM. */
857 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
858 	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
859 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
860 	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
861 	/* Wait until IDLE & CLEAN */
862 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
863 	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
864 	r100_ring_hdp_flush(rdev, ring);
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 struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
883 				    uint64_t src_offset,
884 				    uint64_t dst_offset,
885 				    unsigned num_gpu_pages,
886 				    struct reservation_object *resv)
887 {
888 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
889 	struct radeon_fence *fence;
890 	uint32_t cur_pages;
891 	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
892 	uint32_t pitch;
893 	uint32_t stride_pixels;
894 	unsigned ndw;
895 	int num_loops;
896 	int r = 0;
897 
898 	/* radeon limited to 16k stride */
899 	stride_bytes &= 0x3fff;
900 	/* radeon pitch is /64 */
901 	pitch = stride_bytes / 64;
902 	stride_pixels = stride_bytes / 4;
903 	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
904 
905 	/* Ask for enough room for blit + flush + fence */
906 	ndw = 64 + (10 * num_loops);
907 	r = radeon_ring_lock(rdev, ring, ndw);
908 	if (r) {
909 		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
910 		return ERR_PTR(-EINVAL);
911 	}
912 	while (num_gpu_pages > 0) {
913 		cur_pages = num_gpu_pages;
914 		if (cur_pages > 8191) {
915 			cur_pages = 8191;
916 		}
917 		num_gpu_pages -= cur_pages;
918 
919 		/* pages are in Y direction - height
920 		   page width in X direction - width */
921 		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
922 		radeon_ring_write(ring,
923 				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
924 				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
925 				  RADEON_GMC_SRC_CLIPPING |
926 				  RADEON_GMC_DST_CLIPPING |
927 				  RADEON_GMC_BRUSH_NONE |
928 				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
929 				  RADEON_GMC_SRC_DATATYPE_COLOR |
930 				  RADEON_ROP3_S |
931 				  RADEON_DP_SRC_SOURCE_MEMORY |
932 				  RADEON_GMC_CLR_CMP_CNTL_DIS |
933 				  RADEON_GMC_WR_MSK_DIS);
934 		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
935 		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
936 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
937 		radeon_ring_write(ring, 0);
938 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
939 		radeon_ring_write(ring, num_gpu_pages);
940 		radeon_ring_write(ring, num_gpu_pages);
941 		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
942 	}
943 	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
944 	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
945 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
946 	radeon_ring_write(ring,
947 			  RADEON_WAIT_2D_IDLECLEAN |
948 			  RADEON_WAIT_HOST_IDLECLEAN |
949 			  RADEON_WAIT_DMA_GUI_IDLE);
950 	r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
951 	if (r) {
952 		radeon_ring_unlock_undo(rdev, ring);
953 		return ERR_PTR(r);
954 	}
955 	radeon_ring_unlock_commit(rdev, ring, false);
956 	return fence;
957 }
958 
959 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
960 {
961 	unsigned i;
962 	u32 tmp;
963 
964 	for (i = 0; i < rdev->usec_timeout; i++) {
965 		tmp = RREG32(R_000E40_RBBM_STATUS);
966 		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
967 			return 0;
968 		}
969 		udelay(1);
970 	}
971 	return -1;
972 }
973 
974 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
975 {
976 	int r;
977 
978 	r = radeon_ring_lock(rdev, ring, 2);
979 	if (r) {
980 		return;
981 	}
982 	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
983 	radeon_ring_write(ring,
984 			  RADEON_ISYNC_ANY2D_IDLE3D |
985 			  RADEON_ISYNC_ANY3D_IDLE2D |
986 			  RADEON_ISYNC_WAIT_IDLEGUI |
987 			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
988 	radeon_ring_unlock_commit(rdev, ring, false);
989 }
990 
991 
992 /* Load the microcode for the CP */
993 static int r100_cp_init_microcode(struct radeon_device *rdev)
994 {
995 	const char *fw_name = NULL;
996 	int err;
997 
998 	DRM_DEBUG_KMS("\n");
999 
1000 	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
1001 	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
1002 	    (rdev->family == CHIP_RS200)) {
1003 		DRM_INFO("Loading R100 Microcode\n");
1004 		fw_name = FIRMWARE_R100;
1005 	} else if ((rdev->family == CHIP_R200) ||
1006 		   (rdev->family == CHIP_RV250) ||
1007 		   (rdev->family == CHIP_RV280) ||
1008 		   (rdev->family == CHIP_RS300)) {
1009 		DRM_INFO("Loading R200 Microcode\n");
1010 		fw_name = FIRMWARE_R200;
1011 	} else if ((rdev->family == CHIP_R300) ||
1012 		   (rdev->family == CHIP_R350) ||
1013 		   (rdev->family == CHIP_RV350) ||
1014 		   (rdev->family == CHIP_RV380) ||
1015 		   (rdev->family == CHIP_RS400) ||
1016 		   (rdev->family == CHIP_RS480)) {
1017 		DRM_INFO("Loading R300 Microcode\n");
1018 		fw_name = FIRMWARE_R300;
1019 	} else if ((rdev->family == CHIP_R420) ||
1020 		   (rdev->family == CHIP_R423) ||
1021 		   (rdev->family == CHIP_RV410)) {
1022 		DRM_INFO("Loading R400 Microcode\n");
1023 		fw_name = FIRMWARE_R420;
1024 	} else if ((rdev->family == CHIP_RS690) ||
1025 		   (rdev->family == CHIP_RS740)) {
1026 		DRM_INFO("Loading RS690/RS740 Microcode\n");
1027 		fw_name = FIRMWARE_RS690;
1028 	} else if (rdev->family == CHIP_RS600) {
1029 		DRM_INFO("Loading RS600 Microcode\n");
1030 		fw_name = FIRMWARE_RS600;
1031 	} else if ((rdev->family == CHIP_RV515) ||
1032 		   (rdev->family == CHIP_R520) ||
1033 		   (rdev->family == CHIP_RV530) ||
1034 		   (rdev->family == CHIP_R580) ||
1035 		   (rdev->family == CHIP_RV560) ||
1036 		   (rdev->family == CHIP_RV570)) {
1037 		DRM_INFO("Loading R500 Microcode\n");
1038 		fw_name = FIRMWARE_R520;
1039 	}
1040 
1041 	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1042 	if (err) {
1043 		printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1044 		       fw_name);
1045 	} else if (rdev->me_fw->size % 8) {
1046 		printk(KERN_ERR
1047 		       "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1048 		       rdev->me_fw->size, fw_name);
1049 		err = -EINVAL;
1050 		release_firmware(rdev->me_fw);
1051 		rdev->me_fw = NULL;
1052 	}
1053 	return err;
1054 }
1055 
1056 u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1057 		      struct radeon_ring *ring)
1058 {
1059 	u32 rptr;
1060 
1061 	if (rdev->wb.enabled)
1062 		rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1063 	else
1064 		rptr = RREG32(RADEON_CP_RB_RPTR);
1065 
1066 	return rptr;
1067 }
1068 
1069 u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1070 		      struct radeon_ring *ring)
1071 {
1072 	u32 wptr;
1073 
1074 	wptr = RREG32(RADEON_CP_RB_WPTR);
1075 
1076 	return wptr;
1077 }
1078 
1079 void r100_gfx_set_wptr(struct radeon_device *rdev,
1080 		       struct radeon_ring *ring)
1081 {
1082 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1083 	(void)RREG32(RADEON_CP_RB_WPTR);
1084 }
1085 
1086 static void r100_cp_load_microcode(struct radeon_device *rdev)
1087 {
1088 	const __be32 *fw_data;
1089 	int i, size;
1090 
1091 	if (r100_gui_wait_for_idle(rdev)) {
1092 		printk(KERN_WARNING "Failed to wait GUI idle while "
1093 		       "programming pipes. Bad things might happen.\n");
1094 	}
1095 
1096 	if (rdev->me_fw) {
1097 		size = rdev->me_fw->size / 4;
1098 		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1099 		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1100 		for (i = 0; i < size; i += 2) {
1101 			WREG32(RADEON_CP_ME_RAM_DATAH,
1102 			       be32_to_cpup(&fw_data[i]));
1103 			WREG32(RADEON_CP_ME_RAM_DATAL,
1104 			       be32_to_cpup(&fw_data[i + 1]));
1105 		}
1106 	}
1107 }
1108 
1109 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1110 {
1111 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1112 	unsigned rb_bufsz;
1113 	unsigned rb_blksz;
1114 	unsigned max_fetch;
1115 	unsigned pre_write_timer;
1116 	unsigned pre_write_limit;
1117 	unsigned indirect2_start;
1118 	unsigned indirect1_start;
1119 	uint32_t tmp;
1120 	int r;
1121 
1122 	if (r100_debugfs_cp_init(rdev)) {
1123 		DRM_ERROR("Failed to register debugfs file for CP !\n");
1124 	}
1125 	if (!rdev->me_fw) {
1126 		r = r100_cp_init_microcode(rdev);
1127 		if (r) {
1128 			DRM_ERROR("Failed to load firmware!\n");
1129 			return r;
1130 		}
1131 	}
1132 
1133 	/* Align ring size */
1134 	rb_bufsz = order_base_2(ring_size / 8);
1135 	ring_size = (1 << (rb_bufsz + 1)) * 4;
1136 	r100_cp_load_microcode(rdev);
1137 	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1138 			     RADEON_CP_PACKET2);
1139 	if (r) {
1140 		return r;
1141 	}
1142 	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1143 	 * the rptr copy in system ram */
1144 	rb_blksz = 9;
1145 	/* cp will read 128bytes at a time (4 dwords) */
1146 	max_fetch = 1;
1147 	ring->align_mask = 16 - 1;
1148 	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1149 	pre_write_timer = 64;
1150 	/* Force CP_RB_WPTR write if written more than one time before the
1151 	 * delay expire
1152 	 */
1153 	pre_write_limit = 0;
1154 	/* Setup the cp cache like this (cache size is 96 dwords) :
1155 	 *	RING		0  to 15
1156 	 *	INDIRECT1	16 to 79
1157 	 *	INDIRECT2	80 to 95
1158 	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1159 	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1160 	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1161 	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1162 	 * so it gets the bigger cache.
1163 	 */
1164 	indirect2_start = 80;
1165 	indirect1_start = 16;
1166 	/* cp setup */
1167 	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1168 	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1169 	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1170 	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1171 #ifdef __BIG_ENDIAN
1172 	tmp |= RADEON_BUF_SWAP_32BIT;
1173 #endif
1174 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1175 
1176 	/* Set ring address */
1177 	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1178 	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1179 	/* Force read & write ptr to 0 */
1180 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1181 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1182 	ring->wptr = 0;
1183 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1184 
1185 	/* set the wb address whether it's enabled or not */
1186 	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1187 		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1188 	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1189 
1190 	if (rdev->wb.enabled)
1191 		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1192 	else {
1193 		tmp |= RADEON_RB_NO_UPDATE;
1194 		WREG32(R_000770_SCRATCH_UMSK, 0);
1195 	}
1196 
1197 	WREG32(RADEON_CP_RB_CNTL, tmp);
1198 	udelay(10);
1199 	/* Set cp mode to bus mastering & enable cp*/
1200 	WREG32(RADEON_CP_CSQ_MODE,
1201 	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1202 	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1203 	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1204 	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1205 	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1206 
1207 	/* at this point everything should be setup correctly to enable master */
1208 	pci_set_master(rdev->pdev);
1209 
1210 	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1211 	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1212 	if (r) {
1213 		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1214 		return r;
1215 	}
1216 	ring->ready = true;
1217 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1218 
1219 	if (!ring->rptr_save_reg /* not resuming from suspend */
1220 	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1221 		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1222 		if (r) {
1223 			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1224 			ring->rptr_save_reg = 0;
1225 		}
1226 	}
1227 	return 0;
1228 }
1229 
1230 void r100_cp_fini(struct radeon_device *rdev)
1231 {
1232 	if (r100_cp_wait_for_idle(rdev)) {
1233 		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1234 	}
1235 	/* Disable ring */
1236 	r100_cp_disable(rdev);
1237 	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1238 	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1239 	DRM_INFO("radeon: cp finalized\n");
1240 }
1241 
1242 void r100_cp_disable(struct radeon_device *rdev)
1243 {
1244 	/* Disable ring */
1245 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1246 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1247 	WREG32(RADEON_CP_CSQ_MODE, 0);
1248 	WREG32(RADEON_CP_CSQ_CNTL, 0);
1249 	WREG32(R_000770_SCRATCH_UMSK, 0);
1250 	if (r100_gui_wait_for_idle(rdev)) {
1251 		printk(KERN_WARNING "Failed to wait GUI idle while "
1252 		       "programming pipes. Bad things might happen.\n");
1253 	}
1254 }
1255 
1256 /*
1257  * CS functions
1258  */
1259 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1260 			    struct radeon_cs_packet *pkt,
1261 			    unsigned idx,
1262 			    unsigned reg)
1263 {
1264 	int r;
1265 	u32 tile_flags = 0;
1266 	u32 tmp;
1267 	struct radeon_bo_list *reloc;
1268 	u32 value;
1269 
1270 	r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1271 	if (r) {
1272 		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1273 			  idx, reg);
1274 		radeon_cs_dump_packet(p, pkt);
1275 		return r;
1276 	}
1277 
1278 	value = radeon_get_ib_value(p, idx);
1279 	tmp = value & 0x003fffff;
1280 	tmp += (((u32)reloc->gpu_offset) >> 10);
1281 
1282 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1283 		if (reloc->tiling_flags & RADEON_TILING_MACRO)
1284 			tile_flags |= RADEON_DST_TILE_MACRO;
1285 		if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1286 			if (reg == RADEON_SRC_PITCH_OFFSET) {
1287 				DRM_ERROR("Cannot src blit from microtiled surface\n");
1288 				radeon_cs_dump_packet(p, pkt);
1289 				return -EINVAL;
1290 			}
1291 			tile_flags |= RADEON_DST_TILE_MICRO;
1292 		}
1293 
1294 		tmp |= tile_flags;
1295 		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1296 	} else
1297 		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1298 	return 0;
1299 }
1300 
1301 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1302 			     struct radeon_cs_packet *pkt,
1303 			     int idx)
1304 {
1305 	unsigned c, i;
1306 	struct radeon_bo_list *reloc;
1307 	struct r100_cs_track *track;
1308 	int r = 0;
1309 	volatile uint32_t *ib;
1310 	u32 idx_value;
1311 
1312 	ib = p->ib.ptr;
1313 	track = (struct r100_cs_track *)p->track;
1314 	c = radeon_get_ib_value(p, idx++) & 0x1F;
1315 	if (c > 16) {
1316 	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1317 		      pkt->opcode);
1318 	    radeon_cs_dump_packet(p, pkt);
1319 	    return -EINVAL;
1320 	}
1321 	track->num_arrays = c;
1322 	for (i = 0; i < (c - 1); i+=2, idx+=3) {
1323 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1324 		if (r) {
1325 			DRM_ERROR("No reloc for packet3 %d\n",
1326 				  pkt->opcode);
1327 			radeon_cs_dump_packet(p, pkt);
1328 			return r;
1329 		}
1330 		idx_value = radeon_get_ib_value(p, idx);
1331 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1332 
1333 		track->arrays[i + 0].esize = idx_value >> 8;
1334 		track->arrays[i + 0].robj = reloc->robj;
1335 		track->arrays[i + 0].esize &= 0x7F;
1336 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1337 		if (r) {
1338 			DRM_ERROR("No reloc for packet3 %d\n",
1339 				  pkt->opcode);
1340 			radeon_cs_dump_packet(p, pkt);
1341 			return r;
1342 		}
1343 		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
1344 		track->arrays[i + 1].robj = reloc->robj;
1345 		track->arrays[i + 1].esize = idx_value >> 24;
1346 		track->arrays[i + 1].esize &= 0x7F;
1347 	}
1348 	if (c & 1) {
1349 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1350 		if (r) {
1351 			DRM_ERROR("No reloc for packet3 %d\n",
1352 					  pkt->opcode);
1353 			radeon_cs_dump_packet(p, pkt);
1354 			return r;
1355 		}
1356 		idx_value = radeon_get_ib_value(p, idx);
1357 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1358 		track->arrays[i + 0].robj = reloc->robj;
1359 		track->arrays[i + 0].esize = idx_value >> 8;
1360 		track->arrays[i + 0].esize &= 0x7F;
1361 	}
1362 	return r;
1363 }
1364 
1365 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1366 			  struct radeon_cs_packet *pkt,
1367 			  const unsigned *auth, unsigned n,
1368 			  radeon_packet0_check_t check)
1369 {
1370 	unsigned reg;
1371 	unsigned i, j, m;
1372 	unsigned idx;
1373 	int r;
1374 
1375 	idx = pkt->idx + 1;
1376 	reg = pkt->reg;
1377 	/* Check that register fall into register range
1378 	 * determined by the number of entry (n) in the
1379 	 * safe register bitmap.
1380 	 */
1381 	if (pkt->one_reg_wr) {
1382 		if ((reg >> 7) > n) {
1383 			return -EINVAL;
1384 		}
1385 	} else {
1386 		if (((reg + (pkt->count << 2)) >> 7) > n) {
1387 			return -EINVAL;
1388 		}
1389 	}
1390 	for (i = 0; i <= pkt->count; i++, idx++) {
1391 		j = (reg >> 7);
1392 		m = 1 << ((reg >> 2) & 31);
1393 		if (auth[j] & m) {
1394 			r = check(p, pkt, idx, reg);
1395 			if (r) {
1396 				return r;
1397 			}
1398 		}
1399 		if (pkt->one_reg_wr) {
1400 			if (!(auth[j] & m)) {
1401 				break;
1402 			}
1403 		} else {
1404 			reg += 4;
1405 		}
1406 	}
1407 	return 0;
1408 }
1409 
1410 /**
1411  * r100_cs_packet_next_vline() - parse userspace VLINE packet
1412  * @parser:		parser structure holding parsing context.
1413  *
1414  * Userspace sends a special sequence for VLINE waits.
1415  * PACKET0 - VLINE_START_END + value
1416  * PACKET0 - WAIT_UNTIL +_value
1417  * RELOC (P3) - crtc_id in reloc.
1418  *
1419  * This function parses this and relocates the VLINE START END
1420  * and WAIT UNTIL packets to the correct crtc.
1421  * It also detects a switched off crtc and nulls out the
1422  * wait in that case.
1423  */
1424 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1425 {
1426 	struct drm_crtc *crtc;
1427 	struct radeon_crtc *radeon_crtc;
1428 	struct radeon_cs_packet p3reloc, waitreloc;
1429 	int crtc_id;
1430 	int r;
1431 	uint32_t header, h_idx, reg;
1432 	volatile uint32_t *ib;
1433 
1434 	ib = p->ib.ptr;
1435 
1436 	/* parse the wait until */
1437 	r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1438 	if (r)
1439 		return r;
1440 
1441 	/* check its a wait until and only 1 count */
1442 	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1443 	    waitreloc.count != 0) {
1444 		DRM_ERROR("vline wait had illegal wait until segment\n");
1445 		return -EINVAL;
1446 	}
1447 
1448 	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1449 		DRM_ERROR("vline wait had illegal wait until\n");
1450 		return -EINVAL;
1451 	}
1452 
1453 	/* jump over the NOP */
1454 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1455 	if (r)
1456 		return r;
1457 
1458 	h_idx = p->idx - 2;
1459 	p->idx += waitreloc.count + 2;
1460 	p->idx += p3reloc.count + 2;
1461 
1462 	header = radeon_get_ib_value(p, h_idx);
1463 	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1464 	reg = R100_CP_PACKET0_GET_REG(header);
1465 	crtc = drm_crtc_find(p->rdev->ddev, crtc_id);
1466 	if (!crtc) {
1467 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1468 		return -ENOENT;
1469 	}
1470 	radeon_crtc = to_radeon_crtc(crtc);
1471 	crtc_id = radeon_crtc->crtc_id;
1472 
1473 	if (!crtc->enabled) {
1474 		/* if the CRTC isn't enabled - we need to nop out the wait until */
1475 		ib[h_idx + 2] = PACKET2(0);
1476 		ib[h_idx + 3] = PACKET2(0);
1477 	} else if (crtc_id == 1) {
1478 		switch (reg) {
1479 		case AVIVO_D1MODE_VLINE_START_END:
1480 			header &= ~R300_CP_PACKET0_REG_MASK;
1481 			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1482 			break;
1483 		case RADEON_CRTC_GUI_TRIG_VLINE:
1484 			header &= ~R300_CP_PACKET0_REG_MASK;
1485 			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1486 			break;
1487 		default:
1488 			DRM_ERROR("unknown crtc reloc\n");
1489 			return -EINVAL;
1490 		}
1491 		ib[h_idx] = header;
1492 		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1493 	}
1494 
1495 	return 0;
1496 }
1497 
1498 static int r100_get_vtx_size(uint32_t vtx_fmt)
1499 {
1500 	int vtx_size;
1501 	vtx_size = 2;
1502 	/* ordered according to bits in spec */
1503 	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1504 		vtx_size++;
1505 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1506 		vtx_size += 3;
1507 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1508 		vtx_size++;
1509 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1510 		vtx_size++;
1511 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1512 		vtx_size += 3;
1513 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1514 		vtx_size++;
1515 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1516 		vtx_size++;
1517 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1518 		vtx_size += 2;
1519 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1520 		vtx_size += 2;
1521 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1522 		vtx_size++;
1523 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1524 		vtx_size += 2;
1525 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1526 		vtx_size++;
1527 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1528 		vtx_size += 2;
1529 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1530 		vtx_size++;
1531 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1532 		vtx_size++;
1533 	/* blend weight */
1534 	if (vtx_fmt & (0x7 << 15))
1535 		vtx_size += (vtx_fmt >> 15) & 0x7;
1536 	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1537 		vtx_size += 3;
1538 	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1539 		vtx_size += 2;
1540 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1541 		vtx_size++;
1542 	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1543 		vtx_size++;
1544 	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1545 		vtx_size++;
1546 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1547 		vtx_size++;
1548 	return vtx_size;
1549 }
1550 
1551 static int r100_packet0_check(struct radeon_cs_parser *p,
1552 			      struct radeon_cs_packet *pkt,
1553 			      unsigned idx, unsigned reg)
1554 {
1555 	struct radeon_bo_list *reloc;
1556 	struct r100_cs_track *track;
1557 	volatile uint32_t *ib;
1558 	uint32_t tmp;
1559 	int r;
1560 	int i, face;
1561 	u32 tile_flags = 0;
1562 	u32 idx_value;
1563 
1564 	ib = p->ib.ptr;
1565 	track = (struct r100_cs_track *)p->track;
1566 
1567 	idx_value = radeon_get_ib_value(p, idx);
1568 
1569 	switch (reg) {
1570 	case RADEON_CRTC_GUI_TRIG_VLINE:
1571 		r = r100_cs_packet_parse_vline(p);
1572 		if (r) {
1573 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1574 				  idx, reg);
1575 			radeon_cs_dump_packet(p, pkt);
1576 			return r;
1577 		}
1578 		break;
1579 		/* FIXME: only allow PACKET3 blit? easier to check for out of
1580 		 * range access */
1581 	case RADEON_DST_PITCH_OFFSET:
1582 	case RADEON_SRC_PITCH_OFFSET:
1583 		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1584 		if (r)
1585 			return r;
1586 		break;
1587 	case RADEON_RB3D_DEPTHOFFSET:
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 		track->zb.robj = reloc->robj;
1596 		track->zb.offset = idx_value;
1597 		track->zb_dirty = true;
1598 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1599 		break;
1600 	case RADEON_RB3D_COLOROFFSET:
1601 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1602 		if (r) {
1603 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1604 				  idx, reg);
1605 			radeon_cs_dump_packet(p, pkt);
1606 			return r;
1607 		}
1608 		track->cb[0].robj = reloc->robj;
1609 		track->cb[0].offset = idx_value;
1610 		track->cb_dirty = true;
1611 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1612 		break;
1613 	case RADEON_PP_TXOFFSET_0:
1614 	case RADEON_PP_TXOFFSET_1:
1615 	case RADEON_PP_TXOFFSET_2:
1616 		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1617 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1618 		if (r) {
1619 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1620 				  idx, reg);
1621 			radeon_cs_dump_packet(p, pkt);
1622 			return r;
1623 		}
1624 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1625 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1626 				tile_flags |= RADEON_TXO_MACRO_TILE;
1627 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1628 				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1629 
1630 			tmp = idx_value & ~(0x7 << 2);
1631 			tmp |= tile_flags;
1632 			ib[idx] = tmp + ((u32)reloc->gpu_offset);
1633 		} else
1634 			ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1635 		track->textures[i].robj = reloc->robj;
1636 		track->tex_dirty = true;
1637 		break;
1638 	case RADEON_PP_CUBIC_OFFSET_T0_0:
1639 	case RADEON_PP_CUBIC_OFFSET_T0_1:
1640 	case RADEON_PP_CUBIC_OFFSET_T0_2:
1641 	case RADEON_PP_CUBIC_OFFSET_T0_3:
1642 	case RADEON_PP_CUBIC_OFFSET_T0_4:
1643 		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1644 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1645 		if (r) {
1646 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1647 				  idx, reg);
1648 			radeon_cs_dump_packet(p, pkt);
1649 			return r;
1650 		}
1651 		track->textures[0].cube_info[i].offset = idx_value;
1652 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1653 		track->textures[0].cube_info[i].robj = reloc->robj;
1654 		track->tex_dirty = true;
1655 		break;
1656 	case RADEON_PP_CUBIC_OFFSET_T1_0:
1657 	case RADEON_PP_CUBIC_OFFSET_T1_1:
1658 	case RADEON_PP_CUBIC_OFFSET_T1_2:
1659 	case RADEON_PP_CUBIC_OFFSET_T1_3:
1660 	case RADEON_PP_CUBIC_OFFSET_T1_4:
1661 		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1662 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1663 		if (r) {
1664 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1665 				  idx, reg);
1666 			radeon_cs_dump_packet(p, pkt);
1667 			return r;
1668 		}
1669 		track->textures[1].cube_info[i].offset = idx_value;
1670 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1671 		track->textures[1].cube_info[i].robj = reloc->robj;
1672 		track->tex_dirty = true;
1673 		break;
1674 	case RADEON_PP_CUBIC_OFFSET_T2_0:
1675 	case RADEON_PP_CUBIC_OFFSET_T2_1:
1676 	case RADEON_PP_CUBIC_OFFSET_T2_2:
1677 	case RADEON_PP_CUBIC_OFFSET_T2_3:
1678 	case RADEON_PP_CUBIC_OFFSET_T2_4:
1679 		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1680 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1681 		if (r) {
1682 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1683 				  idx, reg);
1684 			radeon_cs_dump_packet(p, pkt);
1685 			return r;
1686 		}
1687 		track->textures[2].cube_info[i].offset = idx_value;
1688 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1689 		track->textures[2].cube_info[i].robj = reloc->robj;
1690 		track->tex_dirty = true;
1691 		break;
1692 	case RADEON_RE_WIDTH_HEIGHT:
1693 		track->maxy = ((idx_value >> 16) & 0x7FF);
1694 		track->cb_dirty = true;
1695 		track->zb_dirty = true;
1696 		break;
1697 	case RADEON_RB3D_COLORPITCH:
1698 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1699 		if (r) {
1700 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1701 				  idx, reg);
1702 			radeon_cs_dump_packet(p, pkt);
1703 			return r;
1704 		}
1705 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1706 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1707 				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1708 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1709 				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1710 
1711 			tmp = idx_value & ~(0x7 << 16);
1712 			tmp |= tile_flags;
1713 			ib[idx] = tmp;
1714 		} else
1715 			ib[idx] = idx_value;
1716 
1717 		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1718 		track->cb_dirty = true;
1719 		break;
1720 	case RADEON_RB3D_DEPTHPITCH:
1721 		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1722 		track->zb_dirty = true;
1723 		break;
1724 	case RADEON_RB3D_CNTL:
1725 		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1726 		case 7:
1727 		case 8:
1728 		case 9:
1729 		case 11:
1730 		case 12:
1731 			track->cb[0].cpp = 1;
1732 			break;
1733 		case 3:
1734 		case 4:
1735 		case 15:
1736 			track->cb[0].cpp = 2;
1737 			break;
1738 		case 6:
1739 			track->cb[0].cpp = 4;
1740 			break;
1741 		default:
1742 			DRM_ERROR("Invalid color buffer format (%d) !\n",
1743 				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1744 			return -EINVAL;
1745 		}
1746 		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1747 		track->cb_dirty = true;
1748 		track->zb_dirty = true;
1749 		break;
1750 	case RADEON_RB3D_ZSTENCILCNTL:
1751 		switch (idx_value & 0xf) {
1752 		case 0:
1753 			track->zb.cpp = 2;
1754 			break;
1755 		case 2:
1756 		case 3:
1757 		case 4:
1758 		case 5:
1759 		case 9:
1760 		case 11:
1761 			track->zb.cpp = 4;
1762 			break;
1763 		default:
1764 			break;
1765 		}
1766 		track->zb_dirty = true;
1767 		break;
1768 	case RADEON_RB3D_ZPASS_ADDR:
1769 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1770 		if (r) {
1771 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1772 				  idx, reg);
1773 			radeon_cs_dump_packet(p, pkt);
1774 			return r;
1775 		}
1776 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1777 		break;
1778 	case RADEON_PP_CNTL:
1779 		{
1780 			uint32_t temp = idx_value >> 4;
1781 			for (i = 0; i < track->num_texture; i++)
1782 				track->textures[i].enabled = !!(temp & (1 << i));
1783 			track->tex_dirty = true;
1784 		}
1785 		break;
1786 	case RADEON_SE_VF_CNTL:
1787 		track->vap_vf_cntl = idx_value;
1788 		break;
1789 	case RADEON_SE_VTX_FMT:
1790 		track->vtx_size = r100_get_vtx_size(idx_value);
1791 		break;
1792 	case RADEON_PP_TEX_SIZE_0:
1793 	case RADEON_PP_TEX_SIZE_1:
1794 	case RADEON_PP_TEX_SIZE_2:
1795 		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1796 		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1797 		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1798 		track->tex_dirty = true;
1799 		break;
1800 	case RADEON_PP_TEX_PITCH_0:
1801 	case RADEON_PP_TEX_PITCH_1:
1802 	case RADEON_PP_TEX_PITCH_2:
1803 		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1804 		track->textures[i].pitch = idx_value + 32;
1805 		track->tex_dirty = true;
1806 		break;
1807 	case RADEON_PP_TXFILTER_0:
1808 	case RADEON_PP_TXFILTER_1:
1809 	case RADEON_PP_TXFILTER_2:
1810 		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1811 		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1812 						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1813 		tmp = (idx_value >> 23) & 0x7;
1814 		if (tmp == 2 || tmp == 6)
1815 			track->textures[i].roundup_w = false;
1816 		tmp = (idx_value >> 27) & 0x7;
1817 		if (tmp == 2 || tmp == 6)
1818 			track->textures[i].roundup_h = false;
1819 		track->tex_dirty = true;
1820 		break;
1821 	case RADEON_PP_TXFORMAT_0:
1822 	case RADEON_PP_TXFORMAT_1:
1823 	case RADEON_PP_TXFORMAT_2:
1824 		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1825 		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1826 			track->textures[i].use_pitch = 1;
1827 		} else {
1828 			track->textures[i].use_pitch = 0;
1829 			track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1830 			track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1831 		}
1832 		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1833 			track->textures[i].tex_coord_type = 2;
1834 		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1835 		case RADEON_TXFORMAT_I8:
1836 		case RADEON_TXFORMAT_RGB332:
1837 		case RADEON_TXFORMAT_Y8:
1838 			track->textures[i].cpp = 1;
1839 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1840 			break;
1841 		case RADEON_TXFORMAT_AI88:
1842 		case RADEON_TXFORMAT_ARGB1555:
1843 		case RADEON_TXFORMAT_RGB565:
1844 		case RADEON_TXFORMAT_ARGB4444:
1845 		case RADEON_TXFORMAT_VYUY422:
1846 		case RADEON_TXFORMAT_YVYU422:
1847 		case RADEON_TXFORMAT_SHADOW16:
1848 		case RADEON_TXFORMAT_LDUDV655:
1849 		case RADEON_TXFORMAT_DUDV88:
1850 			track->textures[i].cpp = 2;
1851 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1852 			break;
1853 		case RADEON_TXFORMAT_ARGB8888:
1854 		case RADEON_TXFORMAT_RGBA8888:
1855 		case RADEON_TXFORMAT_SHADOW32:
1856 		case RADEON_TXFORMAT_LDUDUV8888:
1857 			track->textures[i].cpp = 4;
1858 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1859 			break;
1860 		case RADEON_TXFORMAT_DXT1:
1861 			track->textures[i].cpp = 1;
1862 			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1863 			break;
1864 		case RADEON_TXFORMAT_DXT23:
1865 		case RADEON_TXFORMAT_DXT45:
1866 			track->textures[i].cpp = 1;
1867 			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1868 			break;
1869 		}
1870 		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1871 		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1872 		track->tex_dirty = true;
1873 		break;
1874 	case RADEON_PP_CUBIC_FACES_0:
1875 	case RADEON_PP_CUBIC_FACES_1:
1876 	case RADEON_PP_CUBIC_FACES_2:
1877 		tmp = idx_value;
1878 		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1879 		for (face = 0; face < 4; face++) {
1880 			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1881 			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1882 		}
1883 		track->tex_dirty = true;
1884 		break;
1885 	default:
1886 		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1887 		       reg, idx);
1888 		return -EINVAL;
1889 	}
1890 	return 0;
1891 }
1892 
1893 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1894 					 struct radeon_cs_packet *pkt,
1895 					 struct radeon_bo *robj)
1896 {
1897 	unsigned idx;
1898 	u32 value;
1899 	idx = pkt->idx + 1;
1900 	value = radeon_get_ib_value(p, idx + 2);
1901 	if ((value + 1) > radeon_bo_size(robj)) {
1902 		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1903 			  "(need %u have %lu) !\n",
1904 			  value + 1,
1905 			  radeon_bo_size(robj));
1906 		return -EINVAL;
1907 	}
1908 	return 0;
1909 }
1910 
1911 static int r100_packet3_check(struct radeon_cs_parser *p,
1912 			      struct radeon_cs_packet *pkt)
1913 {
1914 	struct radeon_bo_list *reloc;
1915 	struct r100_cs_track *track;
1916 	unsigned idx;
1917 	volatile uint32_t *ib;
1918 	int r;
1919 
1920 	ib = p->ib.ptr;
1921 	idx = pkt->idx + 1;
1922 	track = (struct r100_cs_track *)p->track;
1923 	switch (pkt->opcode) {
1924 	case PACKET3_3D_LOAD_VBPNTR:
1925 		r = r100_packet3_load_vbpntr(p, pkt, idx);
1926 		if (r)
1927 			return r;
1928 		break;
1929 	case PACKET3_INDX_BUFFER:
1930 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1931 		if (r) {
1932 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1933 			radeon_cs_dump_packet(p, pkt);
1934 			return r;
1935 		}
1936 		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1937 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1938 		if (r) {
1939 			return r;
1940 		}
1941 		break;
1942 	case 0x23:
1943 		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1944 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1945 		if (r) {
1946 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1947 			radeon_cs_dump_packet(p, pkt);
1948 			return r;
1949 		}
1950 		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1951 		track->num_arrays = 1;
1952 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1953 
1954 		track->arrays[0].robj = reloc->robj;
1955 		track->arrays[0].esize = track->vtx_size;
1956 
1957 		track->max_indx = radeon_get_ib_value(p, idx+1);
1958 
1959 		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1960 		track->immd_dwords = pkt->count - 1;
1961 		r = r100_cs_track_check(p->rdev, track);
1962 		if (r)
1963 			return r;
1964 		break;
1965 	case PACKET3_3D_DRAW_IMMD:
1966 		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1967 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1968 			return -EINVAL;
1969 		}
1970 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1971 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1972 		track->immd_dwords = pkt->count - 1;
1973 		r = r100_cs_track_check(p->rdev, track);
1974 		if (r)
1975 			return r;
1976 		break;
1977 		/* triggers drawing using in-packet vertex data */
1978 	case PACKET3_3D_DRAW_IMMD_2:
1979 		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1980 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1981 			return -EINVAL;
1982 		}
1983 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1984 		track->immd_dwords = pkt->count;
1985 		r = r100_cs_track_check(p->rdev, track);
1986 		if (r)
1987 			return r;
1988 		break;
1989 		/* triggers drawing using in-packet vertex data */
1990 	case PACKET3_3D_DRAW_VBUF_2:
1991 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1992 		r = r100_cs_track_check(p->rdev, track);
1993 		if (r)
1994 			return r;
1995 		break;
1996 		/* triggers drawing of vertex buffers setup elsewhere */
1997 	case PACKET3_3D_DRAW_INDX_2:
1998 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1999 		r = r100_cs_track_check(p->rdev, track);
2000 		if (r)
2001 			return r;
2002 		break;
2003 		/* triggers drawing using indices to vertex buffer */
2004 	case PACKET3_3D_DRAW_VBUF:
2005 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2006 		r = r100_cs_track_check(p->rdev, track);
2007 		if (r)
2008 			return r;
2009 		break;
2010 		/* triggers drawing of vertex buffers setup elsewhere */
2011 	case PACKET3_3D_DRAW_INDX:
2012 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2013 		r = r100_cs_track_check(p->rdev, track);
2014 		if (r)
2015 			return r;
2016 		break;
2017 		/* triggers drawing using indices to vertex buffer */
2018 	case PACKET3_3D_CLEAR_HIZ:
2019 	case PACKET3_3D_CLEAR_ZMASK:
2020 		if (p->rdev->hyperz_filp != p->filp)
2021 			return -EINVAL;
2022 		break;
2023 	case PACKET3_NOP:
2024 		break;
2025 	default:
2026 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2027 		return -EINVAL;
2028 	}
2029 	return 0;
2030 }
2031 
2032 int r100_cs_parse(struct radeon_cs_parser *p)
2033 {
2034 	struct radeon_cs_packet pkt;
2035 	struct r100_cs_track *track;
2036 	int r;
2037 
2038 	track = kzalloc(sizeof(*track), GFP_KERNEL);
2039 	if (!track)
2040 		return -ENOMEM;
2041 	r100_cs_track_clear(p->rdev, track);
2042 	p->track = track;
2043 	do {
2044 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2045 		if (r) {
2046 			return r;
2047 		}
2048 		p->idx += pkt.count + 2;
2049 		switch (pkt.type) {
2050 		case RADEON_PACKET_TYPE0:
2051 			if (p->rdev->family >= CHIP_R200)
2052 				r = r100_cs_parse_packet0(p, &pkt,
2053 					p->rdev->config.r100.reg_safe_bm,
2054 					p->rdev->config.r100.reg_safe_bm_size,
2055 					&r200_packet0_check);
2056 			else
2057 				r = r100_cs_parse_packet0(p, &pkt,
2058 					p->rdev->config.r100.reg_safe_bm,
2059 					p->rdev->config.r100.reg_safe_bm_size,
2060 					&r100_packet0_check);
2061 			break;
2062 		case RADEON_PACKET_TYPE2:
2063 			break;
2064 		case RADEON_PACKET_TYPE3:
2065 			r = r100_packet3_check(p, &pkt);
2066 			break;
2067 		default:
2068 			DRM_ERROR("Unknown packet type %d !\n",
2069 				  pkt.type);
2070 			return -EINVAL;
2071 		}
2072 		if (r)
2073 			return r;
2074 	} while (p->idx < p->chunk_ib->length_dw);
2075 	return 0;
2076 }
2077 
2078 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2079 {
2080 	DRM_ERROR("pitch                      %d\n", t->pitch);
2081 	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2082 	DRM_ERROR("width                      %d\n", t->width);
2083 	DRM_ERROR("width_11                   %d\n", t->width_11);
2084 	DRM_ERROR("height                     %d\n", t->height);
2085 	DRM_ERROR("height_11                  %d\n", t->height_11);
2086 	DRM_ERROR("num levels                 %d\n", t->num_levels);
2087 	DRM_ERROR("depth                      %d\n", t->txdepth);
2088 	DRM_ERROR("bpp                        %d\n", t->cpp);
2089 	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2090 	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2091 	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2092 	DRM_ERROR("compress format            %d\n", t->compress_format);
2093 }
2094 
2095 static int r100_track_compress_size(int compress_format, int w, int h)
2096 {
2097 	int block_width, block_height, block_bytes;
2098 	int wblocks, hblocks;
2099 	int min_wblocks;
2100 	int sz;
2101 
2102 	block_width = 4;
2103 	block_height = 4;
2104 
2105 	switch (compress_format) {
2106 	case R100_TRACK_COMP_DXT1:
2107 		block_bytes = 8;
2108 		min_wblocks = 4;
2109 		break;
2110 	default:
2111 	case R100_TRACK_COMP_DXT35:
2112 		block_bytes = 16;
2113 		min_wblocks = 2;
2114 		break;
2115 	}
2116 
2117 	hblocks = (h + block_height - 1) / block_height;
2118 	wblocks = (w + block_width - 1) / block_width;
2119 	if (wblocks < min_wblocks)
2120 		wblocks = min_wblocks;
2121 	sz = wblocks * hblocks * block_bytes;
2122 	return sz;
2123 }
2124 
2125 static int r100_cs_track_cube(struct radeon_device *rdev,
2126 			      struct r100_cs_track *track, unsigned idx)
2127 {
2128 	unsigned face, w, h;
2129 	struct radeon_bo *cube_robj;
2130 	unsigned long size;
2131 	unsigned compress_format = track->textures[idx].compress_format;
2132 
2133 	for (face = 0; face < 5; face++) {
2134 		cube_robj = track->textures[idx].cube_info[face].robj;
2135 		w = track->textures[idx].cube_info[face].width;
2136 		h = track->textures[idx].cube_info[face].height;
2137 
2138 		if (compress_format) {
2139 			size = r100_track_compress_size(compress_format, w, h);
2140 		} else
2141 			size = w * h;
2142 		size *= track->textures[idx].cpp;
2143 
2144 		size += track->textures[idx].cube_info[face].offset;
2145 
2146 		if (size > radeon_bo_size(cube_robj)) {
2147 			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2148 				  size, radeon_bo_size(cube_robj));
2149 			r100_cs_track_texture_print(&track->textures[idx]);
2150 			return -1;
2151 		}
2152 	}
2153 	return 0;
2154 }
2155 
2156 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2157 				       struct r100_cs_track *track)
2158 {
2159 	struct radeon_bo *robj;
2160 	unsigned long size;
2161 	unsigned u, i, w, h, d;
2162 	int ret;
2163 
2164 	for (u = 0; u < track->num_texture; u++) {
2165 		if (!track->textures[u].enabled)
2166 			continue;
2167 		if (track->textures[u].lookup_disable)
2168 			continue;
2169 		robj = track->textures[u].robj;
2170 		if (robj == NULL) {
2171 			DRM_ERROR("No texture bound to unit %u\n", u);
2172 			return -EINVAL;
2173 		}
2174 		size = 0;
2175 		for (i = 0; i <= track->textures[u].num_levels; i++) {
2176 			if (track->textures[u].use_pitch) {
2177 				if (rdev->family < CHIP_R300)
2178 					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2179 				else
2180 					w = track->textures[u].pitch / (1 << i);
2181 			} else {
2182 				w = track->textures[u].width;
2183 				if (rdev->family >= CHIP_RV515)
2184 					w |= track->textures[u].width_11;
2185 				w = w / (1 << i);
2186 				if (track->textures[u].roundup_w)
2187 					w = roundup_pow_of_two(w);
2188 			}
2189 			h = track->textures[u].height;
2190 			if (rdev->family >= CHIP_RV515)
2191 				h |= track->textures[u].height_11;
2192 			h = h / (1 << i);
2193 			if (track->textures[u].roundup_h)
2194 				h = roundup_pow_of_two(h);
2195 			if (track->textures[u].tex_coord_type == 1) {
2196 				d = (1 << track->textures[u].txdepth) / (1 << i);
2197 				if (!d)
2198 					d = 1;
2199 			} else {
2200 				d = 1;
2201 			}
2202 			if (track->textures[u].compress_format) {
2203 
2204 				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2205 				/* compressed textures are block based */
2206 			} else
2207 				size += w * h * d;
2208 		}
2209 		size *= track->textures[u].cpp;
2210 
2211 		switch (track->textures[u].tex_coord_type) {
2212 		case 0:
2213 		case 1:
2214 			break;
2215 		case 2:
2216 			if (track->separate_cube) {
2217 				ret = r100_cs_track_cube(rdev, track, u);
2218 				if (ret)
2219 					return ret;
2220 			} else
2221 				size *= 6;
2222 			break;
2223 		default:
2224 			DRM_ERROR("Invalid texture coordinate type %u for unit "
2225 				  "%u\n", track->textures[u].tex_coord_type, u);
2226 			return -EINVAL;
2227 		}
2228 		if (size > radeon_bo_size(robj)) {
2229 			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2230 				  "%lu\n", u, size, radeon_bo_size(robj));
2231 			r100_cs_track_texture_print(&track->textures[u]);
2232 			return -EINVAL;
2233 		}
2234 	}
2235 	return 0;
2236 }
2237 
2238 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2239 {
2240 	unsigned i;
2241 	unsigned long size;
2242 	unsigned prim_walk;
2243 	unsigned nverts;
2244 	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2245 
2246 	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2247 	    !track->blend_read_enable)
2248 		num_cb = 0;
2249 
2250 	for (i = 0; i < num_cb; i++) {
2251 		if (track->cb[i].robj == NULL) {
2252 			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2253 			return -EINVAL;
2254 		}
2255 		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2256 		size += track->cb[i].offset;
2257 		if (size > radeon_bo_size(track->cb[i].robj)) {
2258 			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2259 				  "(need %lu have %lu) !\n", i, size,
2260 				  radeon_bo_size(track->cb[i].robj));
2261 			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2262 				  i, track->cb[i].pitch, track->cb[i].cpp,
2263 				  track->cb[i].offset, track->maxy);
2264 			return -EINVAL;
2265 		}
2266 	}
2267 	track->cb_dirty = false;
2268 
2269 	if (track->zb_dirty && track->z_enabled) {
2270 		if (track->zb.robj == NULL) {
2271 			DRM_ERROR("[drm] No buffer for z buffer !\n");
2272 			return -EINVAL;
2273 		}
2274 		size = track->zb.pitch * track->zb.cpp * track->maxy;
2275 		size += track->zb.offset;
2276 		if (size > radeon_bo_size(track->zb.robj)) {
2277 			DRM_ERROR("[drm] Buffer too small for z buffer "
2278 				  "(need %lu have %lu) !\n", size,
2279 				  radeon_bo_size(track->zb.robj));
2280 			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2281 				  track->zb.pitch, track->zb.cpp,
2282 				  track->zb.offset, track->maxy);
2283 			return -EINVAL;
2284 		}
2285 	}
2286 	track->zb_dirty = false;
2287 
2288 	if (track->aa_dirty && track->aaresolve) {
2289 		if (track->aa.robj == NULL) {
2290 			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2291 			return -EINVAL;
2292 		}
2293 		/* I believe the format comes from colorbuffer0. */
2294 		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2295 		size += track->aa.offset;
2296 		if (size > radeon_bo_size(track->aa.robj)) {
2297 			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2298 				  "(need %lu have %lu) !\n", i, size,
2299 				  radeon_bo_size(track->aa.robj));
2300 			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2301 				  i, track->aa.pitch, track->cb[0].cpp,
2302 				  track->aa.offset, track->maxy);
2303 			return -EINVAL;
2304 		}
2305 	}
2306 	track->aa_dirty = false;
2307 
2308 	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2309 	if (track->vap_vf_cntl & (1 << 14)) {
2310 		nverts = track->vap_alt_nverts;
2311 	} else {
2312 		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2313 	}
2314 	switch (prim_walk) {
2315 	case 1:
2316 		for (i = 0; i < track->num_arrays; i++) {
2317 			size = track->arrays[i].esize * track->max_indx * 4;
2318 			if (track->arrays[i].robj == NULL) {
2319 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2320 					  "bound\n", prim_walk, i);
2321 				return -EINVAL;
2322 			}
2323 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2324 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2325 					"need %lu dwords have %lu dwords\n",
2326 					prim_walk, i, size >> 2,
2327 					radeon_bo_size(track->arrays[i].robj)
2328 					>> 2);
2329 				DRM_ERROR("Max indices %u\n", track->max_indx);
2330 				return -EINVAL;
2331 			}
2332 		}
2333 		break;
2334 	case 2:
2335 		for (i = 0; i < track->num_arrays; i++) {
2336 			size = track->arrays[i].esize * (nverts - 1) * 4;
2337 			if (track->arrays[i].robj == NULL) {
2338 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2339 					  "bound\n", prim_walk, i);
2340 				return -EINVAL;
2341 			}
2342 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2343 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2344 					"need %lu dwords have %lu dwords\n",
2345 					prim_walk, i, size >> 2,
2346 					radeon_bo_size(track->arrays[i].robj)
2347 					>> 2);
2348 				return -EINVAL;
2349 			}
2350 		}
2351 		break;
2352 	case 3:
2353 		size = track->vtx_size * nverts;
2354 		if (size != track->immd_dwords) {
2355 			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2356 				  track->immd_dwords, size);
2357 			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2358 				  nverts, track->vtx_size);
2359 			return -EINVAL;
2360 		}
2361 		break;
2362 	default:
2363 		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2364 			  prim_walk);
2365 		return -EINVAL;
2366 	}
2367 
2368 	if (track->tex_dirty) {
2369 		track->tex_dirty = false;
2370 		return r100_cs_track_texture_check(rdev, track);
2371 	}
2372 	return 0;
2373 }
2374 
2375 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2376 {
2377 	unsigned i, face;
2378 
2379 	track->cb_dirty = true;
2380 	track->zb_dirty = true;
2381 	track->tex_dirty = true;
2382 	track->aa_dirty = true;
2383 
2384 	if (rdev->family < CHIP_R300) {
2385 		track->num_cb = 1;
2386 		if (rdev->family <= CHIP_RS200)
2387 			track->num_texture = 3;
2388 		else
2389 			track->num_texture = 6;
2390 		track->maxy = 2048;
2391 		track->separate_cube = 1;
2392 	} else {
2393 		track->num_cb = 4;
2394 		track->num_texture = 16;
2395 		track->maxy = 4096;
2396 		track->separate_cube = 0;
2397 		track->aaresolve = false;
2398 		track->aa.robj = NULL;
2399 	}
2400 
2401 	for (i = 0; i < track->num_cb; i++) {
2402 		track->cb[i].robj = NULL;
2403 		track->cb[i].pitch = 8192;
2404 		track->cb[i].cpp = 16;
2405 		track->cb[i].offset = 0;
2406 	}
2407 	track->z_enabled = true;
2408 	track->zb.robj = NULL;
2409 	track->zb.pitch = 8192;
2410 	track->zb.cpp = 4;
2411 	track->zb.offset = 0;
2412 	track->vtx_size = 0x7F;
2413 	track->immd_dwords = 0xFFFFFFFFUL;
2414 	track->num_arrays = 11;
2415 	track->max_indx = 0x00FFFFFFUL;
2416 	for (i = 0; i < track->num_arrays; i++) {
2417 		track->arrays[i].robj = NULL;
2418 		track->arrays[i].esize = 0x7F;
2419 	}
2420 	for (i = 0; i < track->num_texture; i++) {
2421 		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2422 		track->textures[i].pitch = 16536;
2423 		track->textures[i].width = 16536;
2424 		track->textures[i].height = 16536;
2425 		track->textures[i].width_11 = 1 << 11;
2426 		track->textures[i].height_11 = 1 << 11;
2427 		track->textures[i].num_levels = 12;
2428 		if (rdev->family <= CHIP_RS200) {
2429 			track->textures[i].tex_coord_type = 0;
2430 			track->textures[i].txdepth = 0;
2431 		} else {
2432 			track->textures[i].txdepth = 16;
2433 			track->textures[i].tex_coord_type = 1;
2434 		}
2435 		track->textures[i].cpp = 64;
2436 		track->textures[i].robj = NULL;
2437 		/* CS IB emission code makes sure texture unit are disabled */
2438 		track->textures[i].enabled = false;
2439 		track->textures[i].lookup_disable = false;
2440 		track->textures[i].roundup_w = true;
2441 		track->textures[i].roundup_h = true;
2442 		if (track->separate_cube)
2443 			for (face = 0; face < 5; face++) {
2444 				track->textures[i].cube_info[face].robj = NULL;
2445 				track->textures[i].cube_info[face].width = 16536;
2446 				track->textures[i].cube_info[face].height = 16536;
2447 				track->textures[i].cube_info[face].offset = 0;
2448 			}
2449 	}
2450 }
2451 
2452 /*
2453  * Global GPU functions
2454  */
2455 static void r100_errata(struct radeon_device *rdev)
2456 {
2457 	rdev->pll_errata = 0;
2458 
2459 	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2460 		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2461 	}
2462 
2463 	if (rdev->family == CHIP_RV100 ||
2464 	    rdev->family == CHIP_RS100 ||
2465 	    rdev->family == CHIP_RS200) {
2466 		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2467 	}
2468 }
2469 
2470 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2471 {
2472 	unsigned i;
2473 	uint32_t tmp;
2474 
2475 	for (i = 0; i < rdev->usec_timeout; i++) {
2476 		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2477 		if (tmp >= n) {
2478 			return 0;
2479 		}
2480 		DRM_UDELAY(1);
2481 	}
2482 	return -1;
2483 }
2484 
2485 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2486 {
2487 	unsigned i;
2488 	uint32_t tmp;
2489 
2490 	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2491 		printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
2492 		       " Bad things might happen.\n");
2493 	}
2494 	for (i = 0; i < rdev->usec_timeout; i++) {
2495 		tmp = RREG32(RADEON_RBBM_STATUS);
2496 		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2497 			return 0;
2498 		}
2499 		DRM_UDELAY(1);
2500 	}
2501 	return -1;
2502 }
2503 
2504 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2505 {
2506 	unsigned i;
2507 	uint32_t tmp;
2508 
2509 	for (i = 0; i < rdev->usec_timeout; i++) {
2510 		/* read MC_STATUS */
2511 		tmp = RREG32(RADEON_MC_STATUS);
2512 		if (tmp & RADEON_MC_IDLE) {
2513 			return 0;
2514 		}
2515 		DRM_UDELAY(1);
2516 	}
2517 	return -1;
2518 }
2519 
2520 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2521 {
2522 	u32 rbbm_status;
2523 
2524 	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2525 	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2526 		radeon_ring_lockup_update(rdev, ring);
2527 		return false;
2528 	}
2529 	return radeon_ring_test_lockup(rdev, ring);
2530 }
2531 
2532 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2533 void r100_enable_bm(struct radeon_device *rdev)
2534 {
2535 	uint32_t tmp;
2536 	/* Enable bus mastering */
2537 	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2538 	WREG32(RADEON_BUS_CNTL, tmp);
2539 }
2540 
2541 void r100_bm_disable(struct radeon_device *rdev)
2542 {
2543 	u32 tmp;
2544 
2545 	/* disable bus mastering */
2546 	tmp = RREG32(R_000030_BUS_CNTL);
2547 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2548 	mdelay(1);
2549 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2550 	mdelay(1);
2551 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2552 	tmp = RREG32(RADEON_BUS_CNTL);
2553 	mdelay(1);
2554 	pci_clear_master(rdev->pdev);
2555 	mdelay(1);
2556 }
2557 
2558 int r100_asic_reset(struct radeon_device *rdev)
2559 {
2560 	struct r100_mc_save save;
2561 	u32 status, tmp;
2562 	int ret = 0;
2563 
2564 	status = RREG32(R_000E40_RBBM_STATUS);
2565 	if (!G_000E40_GUI_ACTIVE(status)) {
2566 		return 0;
2567 	}
2568 	r100_mc_stop(rdev, &save);
2569 	status = RREG32(R_000E40_RBBM_STATUS);
2570 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2571 	/* stop CP */
2572 	WREG32(RADEON_CP_CSQ_CNTL, 0);
2573 	tmp = RREG32(RADEON_CP_RB_CNTL);
2574 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2575 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2576 	WREG32(RADEON_CP_RB_WPTR, 0);
2577 	WREG32(RADEON_CP_RB_CNTL, tmp);
2578 	/* save PCI state */
2579 	pci_save_state(rdev->pdev);
2580 	/* disable bus mastering */
2581 	r100_bm_disable(rdev);
2582 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2583 					S_0000F0_SOFT_RESET_RE(1) |
2584 					S_0000F0_SOFT_RESET_PP(1) |
2585 					S_0000F0_SOFT_RESET_RB(1));
2586 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2587 	mdelay(500);
2588 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2589 	mdelay(1);
2590 	status = RREG32(R_000E40_RBBM_STATUS);
2591 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2592 	/* reset CP */
2593 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2594 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2595 	mdelay(500);
2596 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2597 	mdelay(1);
2598 	status = RREG32(R_000E40_RBBM_STATUS);
2599 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2600 	/* restore PCI & busmastering */
2601 	pci_restore_state(rdev->pdev);
2602 	r100_enable_bm(rdev);
2603 	/* Check if GPU is idle */
2604 	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2605 		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2606 		dev_err(rdev->dev, "failed to reset GPU\n");
2607 		ret = -1;
2608 	} else
2609 		dev_info(rdev->dev, "GPU reset succeed\n");
2610 	r100_mc_resume(rdev, &save);
2611 	return ret;
2612 }
2613 
2614 void r100_set_common_regs(struct radeon_device *rdev)
2615 {
2616 	struct drm_device *dev = rdev->ddev;
2617 	bool force_dac2 = false;
2618 	u32 tmp;
2619 
2620 	/* set these so they don't interfere with anything */
2621 	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2622 	WREG32(RADEON_SUBPIC_CNTL, 0);
2623 	WREG32(RADEON_VIPH_CONTROL, 0);
2624 	WREG32(RADEON_I2C_CNTL_1, 0);
2625 	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2626 	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2627 	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2628 
2629 	/* always set up dac2 on rn50 and some rv100 as lots
2630 	 * of servers seem to wire it up to a VGA port but
2631 	 * don't report it in the bios connector
2632 	 * table.
2633 	 */
2634 	switch (dev->pdev->device) {
2635 		/* RN50 */
2636 	case 0x515e:
2637 	case 0x5969:
2638 		force_dac2 = true;
2639 		break;
2640 		/* RV100*/
2641 	case 0x5159:
2642 	case 0x515a:
2643 		/* DELL triple head servers */
2644 		if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2645 		    ((dev->pdev->subsystem_device == 0x016c) ||
2646 		     (dev->pdev->subsystem_device == 0x016d) ||
2647 		     (dev->pdev->subsystem_device == 0x016e) ||
2648 		     (dev->pdev->subsystem_device == 0x016f) ||
2649 		     (dev->pdev->subsystem_device == 0x0170) ||
2650 		     (dev->pdev->subsystem_device == 0x017d) ||
2651 		     (dev->pdev->subsystem_device == 0x017e) ||
2652 		     (dev->pdev->subsystem_device == 0x0183) ||
2653 		     (dev->pdev->subsystem_device == 0x018a) ||
2654 		     (dev->pdev->subsystem_device == 0x019a)))
2655 			force_dac2 = true;
2656 		break;
2657 	}
2658 
2659 	if (force_dac2) {
2660 		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2661 		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2662 		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2663 
2664 		/* For CRT on DAC2, don't turn it on if BIOS didn't
2665 		   enable it, even it's detected.
2666 		*/
2667 
2668 		/* force it to crtc0 */
2669 		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2670 		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2671 		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2672 
2673 		/* set up the TV DAC */
2674 		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2675 				 RADEON_TV_DAC_STD_MASK |
2676 				 RADEON_TV_DAC_RDACPD |
2677 				 RADEON_TV_DAC_GDACPD |
2678 				 RADEON_TV_DAC_BDACPD |
2679 				 RADEON_TV_DAC_BGADJ_MASK |
2680 				 RADEON_TV_DAC_DACADJ_MASK);
2681 		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2682 				RADEON_TV_DAC_NHOLD |
2683 				RADEON_TV_DAC_STD_PS2 |
2684 				(0x58 << 16));
2685 
2686 		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2687 		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2688 		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2689 	}
2690 
2691 	/* switch PM block to ACPI mode */
2692 	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2693 	tmp &= ~RADEON_PM_MODE_SEL;
2694 	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2695 
2696 }
2697 
2698 /*
2699  * VRAM info
2700  */
2701 static void r100_vram_get_type(struct radeon_device *rdev)
2702 {
2703 	uint32_t tmp;
2704 
2705 	rdev->mc.vram_is_ddr = false;
2706 	if (rdev->flags & RADEON_IS_IGP)
2707 		rdev->mc.vram_is_ddr = true;
2708 	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2709 		rdev->mc.vram_is_ddr = true;
2710 	if ((rdev->family == CHIP_RV100) ||
2711 	    (rdev->family == CHIP_RS100) ||
2712 	    (rdev->family == CHIP_RS200)) {
2713 		tmp = RREG32(RADEON_MEM_CNTL);
2714 		if (tmp & RV100_HALF_MODE) {
2715 			rdev->mc.vram_width = 32;
2716 		} else {
2717 			rdev->mc.vram_width = 64;
2718 		}
2719 		if (rdev->flags & RADEON_SINGLE_CRTC) {
2720 			rdev->mc.vram_width /= 4;
2721 			rdev->mc.vram_is_ddr = true;
2722 		}
2723 	} else if (rdev->family <= CHIP_RV280) {
2724 		tmp = RREG32(RADEON_MEM_CNTL);
2725 		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2726 			rdev->mc.vram_width = 128;
2727 		} else {
2728 			rdev->mc.vram_width = 64;
2729 		}
2730 	} else {
2731 		/* newer IGPs */
2732 		rdev->mc.vram_width = 128;
2733 	}
2734 }
2735 
2736 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2737 {
2738 	u32 aper_size;
2739 	u8 byte;
2740 
2741 	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2742 
2743 	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2744 	 * that is has the 2nd generation multifunction PCI interface
2745 	 */
2746 	if (rdev->family == CHIP_RV280 ||
2747 	    rdev->family >= CHIP_RV350) {
2748 		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2749 		       ~RADEON_HDP_APER_CNTL);
2750 		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2751 		return aper_size * 2;
2752 	}
2753 
2754 	/* Older cards have all sorts of funny issues to deal with. First
2755 	 * check if it's a multifunction card by reading the PCI config
2756 	 * header type... Limit those to one aperture size
2757 	 */
2758 	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2759 	if (byte & 0x80) {
2760 		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2761 		DRM_INFO("Limiting VRAM to one aperture\n");
2762 		return aper_size;
2763 	}
2764 
2765 	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2766 	 * have set it up. We don't write this as it's broken on some ASICs but
2767 	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2768 	 */
2769 	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2770 		return aper_size * 2;
2771 	return aper_size;
2772 }
2773 
2774 void r100_vram_init_sizes(struct radeon_device *rdev)
2775 {
2776 	u64 config_aper_size;
2777 
2778 	/* work out accessible VRAM */
2779 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2780 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2781 	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2782 	/* FIXME we don't use the second aperture yet when we could use it */
2783 	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2784 		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2785 	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2786 	if (rdev->flags & RADEON_IS_IGP) {
2787 		uint32_t tom;
2788 		/* read NB_TOM to get the amount of ram stolen for the GPU */
2789 		tom = RREG32(RADEON_NB_TOM);
2790 		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2791 		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2792 		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2793 	} else {
2794 		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2795 		/* Some production boards of m6 will report 0
2796 		 * if it's 8 MB
2797 		 */
2798 		if (rdev->mc.real_vram_size == 0) {
2799 			rdev->mc.real_vram_size = 8192 * 1024;
2800 			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2801 		}
2802 		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2803 		 * Novell bug 204882 + along with lots of ubuntu ones
2804 		 */
2805 		if (rdev->mc.aper_size > config_aper_size)
2806 			config_aper_size = rdev->mc.aper_size;
2807 
2808 		if (config_aper_size > rdev->mc.real_vram_size)
2809 			rdev->mc.mc_vram_size = config_aper_size;
2810 		else
2811 			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2812 	}
2813 }
2814 
2815 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2816 {
2817 	uint32_t temp;
2818 
2819 	temp = RREG32(RADEON_CONFIG_CNTL);
2820 	if (state == false) {
2821 		temp &= ~RADEON_CFG_VGA_RAM_EN;
2822 		temp |= RADEON_CFG_VGA_IO_DIS;
2823 	} else {
2824 		temp &= ~RADEON_CFG_VGA_IO_DIS;
2825 	}
2826 	WREG32(RADEON_CONFIG_CNTL, temp);
2827 }
2828 
2829 static void r100_mc_init(struct radeon_device *rdev)
2830 {
2831 	u64 base;
2832 
2833 	r100_vram_get_type(rdev);
2834 	r100_vram_init_sizes(rdev);
2835 	base = rdev->mc.aper_base;
2836 	if (rdev->flags & RADEON_IS_IGP)
2837 		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2838 	radeon_vram_location(rdev, &rdev->mc, base);
2839 	rdev->mc.gtt_base_align = 0;
2840 	if (!(rdev->flags & RADEON_IS_AGP))
2841 		radeon_gtt_location(rdev, &rdev->mc);
2842 	radeon_update_bandwidth_info(rdev);
2843 }
2844 
2845 
2846 /*
2847  * Indirect registers accessor
2848  */
2849 void r100_pll_errata_after_index(struct radeon_device *rdev)
2850 {
2851 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2852 		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2853 		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2854 	}
2855 }
2856 
2857 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2858 {
2859 	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2860 	 * or the chip could hang on a subsequent access
2861 	 */
2862 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2863 		mdelay(5);
2864 	}
2865 
2866 	/* This function is required to workaround a hardware bug in some (all?)
2867 	 * revisions of the R300.  This workaround should be called after every
2868 	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2869 	 * may not be correct.
2870 	 */
2871 	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2872 		uint32_t save, tmp;
2873 
2874 		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2875 		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2876 		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2877 		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2878 		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2879 	}
2880 }
2881 
2882 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2883 {
2884 	unsigned long flags;
2885 	uint32_t data;
2886 
2887 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2888 	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2889 	r100_pll_errata_after_index(rdev);
2890 	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2891 	r100_pll_errata_after_data(rdev);
2892 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2893 	return data;
2894 }
2895 
2896 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2897 {
2898 	unsigned long flags;
2899 
2900 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2901 	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2902 	r100_pll_errata_after_index(rdev);
2903 	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2904 	r100_pll_errata_after_data(rdev);
2905 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2906 }
2907 
2908 static void r100_set_safe_registers(struct radeon_device *rdev)
2909 {
2910 	if (ASIC_IS_RN50(rdev)) {
2911 		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2912 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2913 	} else if (rdev->family < CHIP_R200) {
2914 		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2915 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2916 	} else {
2917 		r200_set_safe_registers(rdev);
2918 	}
2919 }
2920 
2921 /*
2922  * Debugfs info
2923  */
2924 #if defined(CONFIG_DEBUG_FS)
2925 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2926 {
2927 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2928 	struct drm_device *dev = node->minor->dev;
2929 	struct radeon_device *rdev = dev->dev_private;
2930 	uint32_t reg, value;
2931 	unsigned i;
2932 
2933 	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2934 	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2935 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2936 	for (i = 0; i < 64; i++) {
2937 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2938 		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2939 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2940 		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2941 		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2942 	}
2943 	return 0;
2944 }
2945 
2946 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2947 {
2948 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2949 	struct drm_device *dev = node->minor->dev;
2950 	struct radeon_device *rdev = dev->dev_private;
2951 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2952 	uint32_t rdp, wdp;
2953 	unsigned count, i, j;
2954 
2955 	radeon_ring_free_size(rdev, ring);
2956 	rdp = RREG32(RADEON_CP_RB_RPTR);
2957 	wdp = RREG32(RADEON_CP_RB_WPTR);
2958 	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2959 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2960 	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2961 	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2962 	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2963 	seq_printf(m, "%u dwords in ring\n", count);
2964 	if (ring->ready) {
2965 		for (j = 0; j <= count; j++) {
2966 			i = (rdp + j) & ring->ptr_mask;
2967 			seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2968 		}
2969 	}
2970 	return 0;
2971 }
2972 
2973 
2974 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2975 {
2976 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2977 	struct drm_device *dev = node->minor->dev;
2978 	struct radeon_device *rdev = dev->dev_private;
2979 	uint32_t csq_stat, csq2_stat, tmp;
2980 	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2981 	unsigned i;
2982 
2983 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2984 	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2985 	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2986 	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2987 	r_rptr = (csq_stat >> 0) & 0x3ff;
2988 	r_wptr = (csq_stat >> 10) & 0x3ff;
2989 	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2990 	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2991 	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2992 	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2993 	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2994 	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2995 	seq_printf(m, "Ring rptr %u\n", r_rptr);
2996 	seq_printf(m, "Ring wptr %u\n", r_wptr);
2997 	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2998 	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2999 	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
3000 	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
3001 	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
3002 	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
3003 	seq_printf(m, "Ring fifo:\n");
3004 	for (i = 0; i < 256; i++) {
3005 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3006 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3007 		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3008 	}
3009 	seq_printf(m, "Indirect1 fifo:\n");
3010 	for (i = 256; i <= 512; i++) {
3011 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3012 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3013 		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3014 	}
3015 	seq_printf(m, "Indirect2 fifo:\n");
3016 	for (i = 640; i < ib1_wptr; i++) {
3017 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3018 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3019 		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3020 	}
3021 	return 0;
3022 }
3023 
3024 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
3025 {
3026 	struct drm_info_node *node = (struct drm_info_node *) m->private;
3027 	struct drm_device *dev = node->minor->dev;
3028 	struct radeon_device *rdev = dev->dev_private;
3029 	uint32_t tmp;
3030 
3031 	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3032 	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3033 	tmp = RREG32(RADEON_MC_FB_LOCATION);
3034 	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3035 	tmp = RREG32(RADEON_BUS_CNTL);
3036 	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3037 	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3038 	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3039 	tmp = RREG32(RADEON_AGP_BASE);
3040 	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3041 	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3042 	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3043 	tmp = RREG32(0x01D0);
3044 	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3045 	tmp = RREG32(RADEON_AIC_LO_ADDR);
3046 	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3047 	tmp = RREG32(RADEON_AIC_HI_ADDR);
3048 	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3049 	tmp = RREG32(0x01E4);
3050 	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3051 	return 0;
3052 }
3053 
3054 static struct drm_info_list r100_debugfs_rbbm_list[] = {
3055 	{"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3056 };
3057 
3058 static struct drm_info_list r100_debugfs_cp_list[] = {
3059 	{"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3060 	{"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3061 };
3062 
3063 static struct drm_info_list r100_debugfs_mc_info_list[] = {
3064 	{"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3065 };
3066 #endif
3067 
3068 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3069 {
3070 #if defined(CONFIG_DEBUG_FS)
3071 	return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3072 #else
3073 	return 0;
3074 #endif
3075 }
3076 
3077 int r100_debugfs_cp_init(struct radeon_device *rdev)
3078 {
3079 #if defined(CONFIG_DEBUG_FS)
3080 	return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3081 #else
3082 	return 0;
3083 #endif
3084 }
3085 
3086 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3087 {
3088 #if defined(CONFIG_DEBUG_FS)
3089 	return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3090 #else
3091 	return 0;
3092 #endif
3093 }
3094 
3095 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3096 			 uint32_t tiling_flags, uint32_t pitch,
3097 			 uint32_t offset, uint32_t obj_size)
3098 {
3099 	int surf_index = reg * 16;
3100 	int flags = 0;
3101 
3102 	if (rdev->family <= CHIP_RS200) {
3103 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3104 				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3105 			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3106 		if (tiling_flags & RADEON_TILING_MACRO)
3107 			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3108 		/* setting pitch to 0 disables tiling */
3109 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3110 				== 0)
3111 			pitch = 0;
3112 	} else if (rdev->family <= CHIP_RV280) {
3113 		if (tiling_flags & (RADEON_TILING_MACRO))
3114 			flags |= R200_SURF_TILE_COLOR_MACRO;
3115 		if (tiling_flags & RADEON_TILING_MICRO)
3116 			flags |= R200_SURF_TILE_COLOR_MICRO;
3117 	} else {
3118 		if (tiling_flags & RADEON_TILING_MACRO)
3119 			flags |= R300_SURF_TILE_MACRO;
3120 		if (tiling_flags & RADEON_TILING_MICRO)
3121 			flags |= R300_SURF_TILE_MICRO;
3122 	}
3123 
3124 	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3125 		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3126 	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3127 		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3128 
3129 	/* r100/r200 divide by 16 */
3130 	if (rdev->family < CHIP_R300)
3131 		flags |= pitch / 16;
3132 	else
3133 		flags |= pitch / 8;
3134 
3135 
3136 	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3137 	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3138 	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3139 	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3140 	return 0;
3141 }
3142 
3143 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3144 {
3145 	int surf_index = reg * 16;
3146 	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3147 }
3148 
3149 void r100_bandwidth_update(struct radeon_device *rdev)
3150 {
3151 	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3152 	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3153 	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
3154 	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3155 	fixed20_12 memtcas_ff[8] = {
3156 		dfixed_init(1),
3157 		dfixed_init(2),
3158 		dfixed_init(3),
3159 		dfixed_init(0),
3160 		dfixed_init_half(1),
3161 		dfixed_init_half(2),
3162 		dfixed_init(0),
3163 	};
3164 	fixed20_12 memtcas_rs480_ff[8] = {
3165 		dfixed_init(0),
3166 		dfixed_init(1),
3167 		dfixed_init(2),
3168 		dfixed_init(3),
3169 		dfixed_init(0),
3170 		dfixed_init_half(1),
3171 		dfixed_init_half(2),
3172 		dfixed_init_half(3),
3173 	};
3174 	fixed20_12 memtcas2_ff[8] = {
3175 		dfixed_init(0),
3176 		dfixed_init(1),
3177 		dfixed_init(2),
3178 		dfixed_init(3),
3179 		dfixed_init(4),
3180 		dfixed_init(5),
3181 		dfixed_init(6),
3182 		dfixed_init(7),
3183 	};
3184 	fixed20_12 memtrbs[8] = {
3185 		dfixed_init(1),
3186 		dfixed_init_half(1),
3187 		dfixed_init(2),
3188 		dfixed_init_half(2),
3189 		dfixed_init(3),
3190 		dfixed_init_half(3),
3191 		dfixed_init(4),
3192 		dfixed_init_half(4)
3193 	};
3194 	fixed20_12 memtrbs_r4xx[8] = {
3195 		dfixed_init(4),
3196 		dfixed_init(5),
3197 		dfixed_init(6),
3198 		dfixed_init(7),
3199 		dfixed_init(8),
3200 		dfixed_init(9),
3201 		dfixed_init(10),
3202 		dfixed_init(11)
3203 	};
3204 	fixed20_12 min_mem_eff;
3205 	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3206 	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3207 	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
3208 		disp_drain_rate2, read_return_rate;
3209 	fixed20_12 time_disp1_drop_priority;
3210 	int c;
3211 	int cur_size = 16;       /* in octawords */
3212 	int critical_point = 0, critical_point2;
3213 /* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3214 	int stop_req, max_stop_req;
3215 	struct drm_display_mode *mode1 = NULL;
3216 	struct drm_display_mode *mode2 = NULL;
3217 	uint32_t pixel_bytes1 = 0;
3218 	uint32_t pixel_bytes2 = 0;
3219 
3220 	if (!rdev->mode_info.mode_config_initialized)
3221 		return;
3222 
3223 	radeon_update_display_priority(rdev);
3224 
3225 	if (rdev->mode_info.crtcs[0]->base.enabled) {
3226 		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3227 		pixel_bytes1 = rdev->mode_info.crtcs[0]->base.primary->fb->bits_per_pixel / 8;
3228 	}
3229 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3230 		if (rdev->mode_info.crtcs[1]->base.enabled) {
3231 			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3232 			pixel_bytes2 = rdev->mode_info.crtcs[1]->base.primary->fb->bits_per_pixel / 8;
3233 		}
3234 	}
3235 
3236 	min_mem_eff.full = dfixed_const_8(0);
3237 	/* get modes */
3238 	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3239 		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3240 		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3241 		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3242 		/* check crtc enables */
3243 		if (mode2)
3244 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3245 		if (mode1)
3246 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3247 		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3248 	}
3249 
3250 	/*
3251 	 * determine is there is enough bw for current mode
3252 	 */
3253 	sclk_ff = rdev->pm.sclk;
3254 	mclk_ff = rdev->pm.mclk;
3255 
3256 	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3257 	temp_ff.full = dfixed_const(temp);
3258 	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3259 
3260 	pix_clk.full = 0;
3261 	pix_clk2.full = 0;
3262 	peak_disp_bw.full = 0;
3263 	if (mode1) {
3264 		temp_ff.full = dfixed_const(1000);
3265 		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3266 		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3267 		temp_ff.full = dfixed_const(pixel_bytes1);
3268 		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3269 	}
3270 	if (mode2) {
3271 		temp_ff.full = dfixed_const(1000);
3272 		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3273 		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3274 		temp_ff.full = dfixed_const(pixel_bytes2);
3275 		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3276 	}
3277 
3278 	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3279 	if (peak_disp_bw.full >= mem_bw.full) {
3280 		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3281 			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3282 	}
3283 
3284 	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3285 	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3286 	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3287 		mem_trcd = ((temp >> 2) & 0x3) + 1;
3288 		mem_trp  = ((temp & 0x3)) + 1;
3289 		mem_tras = ((temp & 0x70) >> 4) + 1;
3290 	} else if (rdev->family == CHIP_R300 ||
3291 		   rdev->family == CHIP_R350) { /* r300, r350 */
3292 		mem_trcd = (temp & 0x7) + 1;
3293 		mem_trp = ((temp >> 8) & 0x7) + 1;
3294 		mem_tras = ((temp >> 11) & 0xf) + 4;
3295 	} else if (rdev->family == CHIP_RV350 ||
3296 		   rdev->family <= CHIP_RV380) {
3297 		/* rv3x0 */
3298 		mem_trcd = (temp & 0x7) + 3;
3299 		mem_trp = ((temp >> 8) & 0x7) + 3;
3300 		mem_tras = ((temp >> 11) & 0xf) + 6;
3301 	} else if (rdev->family == CHIP_R420 ||
3302 		   rdev->family == CHIP_R423 ||
3303 		   rdev->family == CHIP_RV410) {
3304 		/* r4xx */
3305 		mem_trcd = (temp & 0xf) + 3;
3306 		if (mem_trcd > 15)
3307 			mem_trcd = 15;
3308 		mem_trp = ((temp >> 8) & 0xf) + 3;
3309 		if (mem_trp > 15)
3310 			mem_trp = 15;
3311 		mem_tras = ((temp >> 12) & 0x1f) + 6;
3312 		if (mem_tras > 31)
3313 			mem_tras = 31;
3314 	} else { /* RV200, R200 */
3315 		mem_trcd = (temp & 0x7) + 1;
3316 		mem_trp = ((temp >> 8) & 0x7) + 1;
3317 		mem_tras = ((temp >> 12) & 0xf) + 4;
3318 	}
3319 	/* convert to FF */
3320 	trcd_ff.full = dfixed_const(mem_trcd);
3321 	trp_ff.full = dfixed_const(mem_trp);
3322 	tras_ff.full = dfixed_const(mem_tras);
3323 
3324 	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3325 	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3326 	data = (temp & (7 << 20)) >> 20;
3327 	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3328 		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3329 			tcas_ff = memtcas_rs480_ff[data];
3330 		else
3331 			tcas_ff = memtcas_ff[data];
3332 	} else
3333 		tcas_ff = memtcas2_ff[data];
3334 
3335 	if (rdev->family == CHIP_RS400 ||
3336 	    rdev->family == CHIP_RS480) {
3337 		/* extra cas latency stored in bits 23-25 0-4 clocks */
3338 		data = (temp >> 23) & 0x7;
3339 		if (data < 5)
3340 			tcas_ff.full += dfixed_const(data);
3341 	}
3342 
3343 	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3344 		/* on the R300, Tcas is included in Trbs.
3345 		 */
3346 		temp = RREG32(RADEON_MEM_CNTL);
3347 		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3348 		if (data == 1) {
3349 			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3350 				temp = RREG32(R300_MC_IND_INDEX);
3351 				temp &= ~R300_MC_IND_ADDR_MASK;
3352 				temp |= R300_MC_READ_CNTL_CD_mcind;
3353 				WREG32(R300_MC_IND_INDEX, temp);
3354 				temp = RREG32(R300_MC_IND_DATA);
3355 				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3356 			} else {
3357 				temp = RREG32(R300_MC_READ_CNTL_AB);
3358 				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3359 			}
3360 		} else {
3361 			temp = RREG32(R300_MC_READ_CNTL_AB);
3362 			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3363 		}
3364 		if (rdev->family == CHIP_RV410 ||
3365 		    rdev->family == CHIP_R420 ||
3366 		    rdev->family == CHIP_R423)
3367 			trbs_ff = memtrbs_r4xx[data];
3368 		else
3369 			trbs_ff = memtrbs[data];
3370 		tcas_ff.full += trbs_ff.full;
3371 	}
3372 
3373 	sclk_eff_ff.full = sclk_ff.full;
3374 
3375 	if (rdev->flags & RADEON_IS_AGP) {
3376 		fixed20_12 agpmode_ff;
3377 		agpmode_ff.full = dfixed_const(radeon_agpmode);
3378 		temp_ff.full = dfixed_const_666(16);
3379 		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3380 	}
3381 	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3382 
3383 	if (ASIC_IS_R300(rdev)) {
3384 		sclk_delay_ff.full = dfixed_const(250);
3385 	} else {
3386 		if ((rdev->family == CHIP_RV100) ||
3387 		    rdev->flags & RADEON_IS_IGP) {
3388 			if (rdev->mc.vram_is_ddr)
3389 				sclk_delay_ff.full = dfixed_const(41);
3390 			else
3391 				sclk_delay_ff.full = dfixed_const(33);
3392 		} else {
3393 			if (rdev->mc.vram_width == 128)
3394 				sclk_delay_ff.full = dfixed_const(57);
3395 			else
3396 				sclk_delay_ff.full = dfixed_const(41);
3397 		}
3398 	}
3399 
3400 	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3401 
3402 	if (rdev->mc.vram_is_ddr) {
3403 		if (rdev->mc.vram_width == 32) {
3404 			k1.full = dfixed_const(40);
3405 			c  = 3;
3406 		} else {
3407 			k1.full = dfixed_const(20);
3408 			c  = 1;
3409 		}
3410 	} else {
3411 		k1.full = dfixed_const(40);
3412 		c  = 3;
3413 	}
3414 
3415 	temp_ff.full = dfixed_const(2);
3416 	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3417 	temp_ff.full = dfixed_const(c);
3418 	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3419 	temp_ff.full = dfixed_const(4);
3420 	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3421 	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3422 	mc_latency_mclk.full += k1.full;
3423 
3424 	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3425 	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3426 
3427 	/*
3428 	  HW cursor time assuming worst case of full size colour cursor.
3429 	*/
3430 	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3431 	temp_ff.full += trcd_ff.full;
3432 	if (temp_ff.full < tras_ff.full)
3433 		temp_ff.full = tras_ff.full;
3434 	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3435 
3436 	temp_ff.full = dfixed_const(cur_size);
3437 	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3438 	/*
3439 	  Find the total latency for the display data.
3440 	*/
3441 	disp_latency_overhead.full = dfixed_const(8);
3442 	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3443 	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3444 	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3445 
3446 	if (mc_latency_mclk.full > mc_latency_sclk.full)
3447 		disp_latency.full = mc_latency_mclk.full;
3448 	else
3449 		disp_latency.full = mc_latency_sclk.full;
3450 
3451 	/* setup Max GRPH_STOP_REQ default value */
3452 	if (ASIC_IS_RV100(rdev))
3453 		max_stop_req = 0x5c;
3454 	else
3455 		max_stop_req = 0x7c;
3456 
3457 	if (mode1) {
3458 		/*  CRTC1
3459 		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3460 		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3461 		*/
3462 		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3463 
3464 		if (stop_req > max_stop_req)
3465 			stop_req = max_stop_req;
3466 
3467 		/*
3468 		  Find the drain rate of the display buffer.
3469 		*/
3470 		temp_ff.full = dfixed_const((16/pixel_bytes1));
3471 		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3472 
3473 		/*
3474 		  Find the critical point of the display buffer.
3475 		*/
3476 		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3477 		crit_point_ff.full += dfixed_const_half(0);
3478 
3479 		critical_point = dfixed_trunc(crit_point_ff);
3480 
3481 		if (rdev->disp_priority == 2) {
3482 			critical_point = 0;
3483 		}
3484 
3485 		/*
3486 		  The critical point should never be above max_stop_req-4.  Setting
3487 		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3488 		*/
3489 		if (max_stop_req - critical_point < 4)
3490 			critical_point = 0;
3491 
3492 		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3493 			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3494 			critical_point = 0x10;
3495 		}
3496 
3497 		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3498 		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3499 		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3500 		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3501 		if ((rdev->family == CHIP_R350) &&
3502 		    (stop_req > 0x15)) {
3503 			stop_req -= 0x10;
3504 		}
3505 		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3506 		temp |= RADEON_GRPH_BUFFER_SIZE;
3507 		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3508 			  RADEON_GRPH_CRITICAL_AT_SOF |
3509 			  RADEON_GRPH_STOP_CNTL);
3510 		/*
3511 		  Write the result into the register.
3512 		*/
3513 		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3514 						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3515 
3516 #if 0
3517 		if ((rdev->family == CHIP_RS400) ||
3518 		    (rdev->family == CHIP_RS480)) {
3519 			/* attempt to program RS400 disp regs correctly ??? */
3520 			temp = RREG32(RS400_DISP1_REG_CNTL);
3521 			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3522 				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3523 			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3524 						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3525 						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3526 			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3527 			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3528 				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3529 			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3530 						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3531 						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3532 		}
3533 #endif
3534 
3535 		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3536 			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3537 			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3538 	}
3539 
3540 	if (mode2) {
3541 		u32 grph2_cntl;
3542 		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3543 
3544 		if (stop_req > max_stop_req)
3545 			stop_req = max_stop_req;
3546 
3547 		/*
3548 		  Find the drain rate of the display buffer.
3549 		*/
3550 		temp_ff.full = dfixed_const((16/pixel_bytes2));
3551 		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3552 
3553 		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3554 		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3555 		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3556 		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3557 		if ((rdev->family == CHIP_R350) &&
3558 		    (stop_req > 0x15)) {
3559 			stop_req -= 0x10;
3560 		}
3561 		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3562 		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3563 		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3564 			  RADEON_GRPH_CRITICAL_AT_SOF |
3565 			  RADEON_GRPH_STOP_CNTL);
3566 
3567 		if ((rdev->family == CHIP_RS100) ||
3568 		    (rdev->family == CHIP_RS200))
3569 			critical_point2 = 0;
3570 		else {
3571 			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3572 			temp_ff.full = dfixed_const(temp);
3573 			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3574 			if (sclk_ff.full < temp_ff.full)
3575 				temp_ff.full = sclk_ff.full;
3576 
3577 			read_return_rate.full = temp_ff.full;
3578 
3579 			if (mode1) {
3580 				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3581 				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3582 			} else {
3583 				time_disp1_drop_priority.full = 0;
3584 			}
3585 			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3586 			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3587 			crit_point_ff.full += dfixed_const_half(0);
3588 
3589 			critical_point2 = dfixed_trunc(crit_point_ff);
3590 
3591 			if (rdev->disp_priority == 2) {
3592 				critical_point2 = 0;
3593 			}
3594 
3595 			if (max_stop_req - critical_point2 < 4)
3596 				critical_point2 = 0;
3597 
3598 		}
3599 
3600 		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3601 			/* some R300 cards have problem with this set to 0 */
3602 			critical_point2 = 0x10;
3603 		}
3604 
3605 		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3606 						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3607 
3608 		if ((rdev->family == CHIP_RS400) ||
3609 		    (rdev->family == CHIP_RS480)) {
3610 #if 0
3611 			/* attempt to program RS400 disp2 regs correctly ??? */
3612 			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3613 			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3614 				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3615 			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3616 						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3617 						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3618 			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3619 			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3620 				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3621 			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3622 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3623 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3624 #endif
3625 			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3626 			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3627 			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3628 			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3629 		}
3630 
3631 		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3632 			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3633 	}
3634 }
3635 
3636 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3637 {
3638 	uint32_t scratch;
3639 	uint32_t tmp = 0;
3640 	unsigned i;
3641 	int r;
3642 
3643 	r = radeon_scratch_get(rdev, &scratch);
3644 	if (r) {
3645 		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3646 		return r;
3647 	}
3648 	WREG32(scratch, 0xCAFEDEAD);
3649 	r = radeon_ring_lock(rdev, ring, 2);
3650 	if (r) {
3651 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3652 		radeon_scratch_free(rdev, scratch);
3653 		return r;
3654 	}
3655 	radeon_ring_write(ring, PACKET0(scratch, 0));
3656 	radeon_ring_write(ring, 0xDEADBEEF);
3657 	radeon_ring_unlock_commit(rdev, ring, false);
3658 	for (i = 0; i < rdev->usec_timeout; i++) {
3659 		tmp = RREG32(scratch);
3660 		if (tmp == 0xDEADBEEF) {
3661 			break;
3662 		}
3663 		DRM_UDELAY(1);
3664 	}
3665 	if (i < rdev->usec_timeout) {
3666 		DRM_INFO("ring test succeeded in %d usecs\n", i);
3667 	} else {
3668 		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3669 			  scratch, tmp);
3670 		r = -EINVAL;
3671 	}
3672 	radeon_scratch_free(rdev, scratch);
3673 	return r;
3674 }
3675 
3676 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3677 {
3678 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3679 
3680 	if (ring->rptr_save_reg) {
3681 		u32 next_rptr = ring->wptr + 2 + 3;
3682 		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3683 		radeon_ring_write(ring, next_rptr);
3684 	}
3685 
3686 	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3687 	radeon_ring_write(ring, ib->gpu_addr);
3688 	radeon_ring_write(ring, ib->length_dw);
3689 }
3690 
3691 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3692 {
3693 	struct radeon_ib ib;
3694 	uint32_t scratch;
3695 	uint32_t tmp = 0;
3696 	unsigned i;
3697 	int r;
3698 
3699 	r = radeon_scratch_get(rdev, &scratch);
3700 	if (r) {
3701 		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3702 		return r;
3703 	}
3704 	WREG32(scratch, 0xCAFEDEAD);
3705 	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3706 	if (r) {
3707 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3708 		goto free_scratch;
3709 	}
3710 	ib.ptr[0] = PACKET0(scratch, 0);
3711 	ib.ptr[1] = 0xDEADBEEF;
3712 	ib.ptr[2] = PACKET2(0);
3713 	ib.ptr[3] = PACKET2(0);
3714 	ib.ptr[4] = PACKET2(0);
3715 	ib.ptr[5] = PACKET2(0);
3716 	ib.ptr[6] = PACKET2(0);
3717 	ib.ptr[7] = PACKET2(0);
3718 	ib.length_dw = 8;
3719 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
3720 	if (r) {
3721 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3722 		goto free_ib;
3723 	}
3724 	r = radeon_fence_wait(ib.fence, false);
3725 	if (r) {
3726 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3727 		goto free_ib;
3728 	}
3729 	for (i = 0; i < rdev->usec_timeout; i++) {
3730 		tmp = RREG32(scratch);
3731 		if (tmp == 0xDEADBEEF) {
3732 			break;
3733 		}
3734 		DRM_UDELAY(1);
3735 	}
3736 	if (i < rdev->usec_timeout) {
3737 		DRM_INFO("ib test succeeded in %u usecs\n", i);
3738 	} else {
3739 		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3740 			  scratch, tmp);
3741 		r = -EINVAL;
3742 	}
3743 free_ib:
3744 	radeon_ib_free(rdev, &ib);
3745 free_scratch:
3746 	radeon_scratch_free(rdev, scratch);
3747 	return r;
3748 }
3749 
3750 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3751 {
3752 	/* Shutdown CP we shouldn't need to do that but better be safe than
3753 	 * sorry
3754 	 */
3755 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3756 	WREG32(R_000740_CP_CSQ_CNTL, 0);
3757 
3758 	/* Save few CRTC registers */
3759 	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3760 	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3761 	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3762 	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3763 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3764 		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3765 		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3766 	}
3767 
3768 	/* Disable VGA aperture access */
3769 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3770 	/* Disable cursor, overlay, crtc */
3771 	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3772 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3773 					S_000054_CRTC_DISPLAY_DIS(1));
3774 	WREG32(R_000050_CRTC_GEN_CNTL,
3775 			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3776 			S_000050_CRTC_DISP_REQ_EN_B(1));
3777 	WREG32(R_000420_OV0_SCALE_CNTL,
3778 		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3779 	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3780 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3781 		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3782 						S_000360_CUR2_LOCK(1));
3783 		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3784 			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3785 			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3786 			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3787 		WREG32(R_000360_CUR2_OFFSET,
3788 			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3789 	}
3790 }
3791 
3792 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3793 {
3794 	/* Update base address for crtc */
3795 	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3796 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3797 		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3798 	}
3799 	/* Restore CRTC registers */
3800 	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3801 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3802 	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3803 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3804 		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3805 	}
3806 }
3807 
3808 void r100_vga_render_disable(struct radeon_device *rdev)
3809 {
3810 	u32 tmp;
3811 
3812 	tmp = RREG8(R_0003C2_GENMO_WT);
3813 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3814 }
3815 
3816 static void r100_debugfs(struct radeon_device *rdev)
3817 {
3818 	int r;
3819 
3820 	r = r100_debugfs_mc_info_init(rdev);
3821 	if (r)
3822 		dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3823 }
3824 
3825 static void r100_mc_program(struct radeon_device *rdev)
3826 {
3827 	struct r100_mc_save save;
3828 
3829 	/* Stops all mc clients */
3830 	r100_mc_stop(rdev, &save);
3831 	if (rdev->flags & RADEON_IS_AGP) {
3832 		WREG32(R_00014C_MC_AGP_LOCATION,
3833 			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3834 			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3835 		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3836 		if (rdev->family > CHIP_RV200)
3837 			WREG32(R_00015C_AGP_BASE_2,
3838 				upper_32_bits(rdev->mc.agp_base) & 0xff);
3839 	} else {
3840 		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3841 		WREG32(R_000170_AGP_BASE, 0);
3842 		if (rdev->family > CHIP_RV200)
3843 			WREG32(R_00015C_AGP_BASE_2, 0);
3844 	}
3845 	/* Wait for mc idle */
3846 	if (r100_mc_wait_for_idle(rdev))
3847 		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3848 	/* Program MC, should be a 32bits limited address space */
3849 	WREG32(R_000148_MC_FB_LOCATION,
3850 		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3851 		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3852 	r100_mc_resume(rdev, &save);
3853 }
3854 
3855 static void r100_clock_startup(struct radeon_device *rdev)
3856 {
3857 	u32 tmp;
3858 
3859 	if (radeon_dynclks != -1 && radeon_dynclks)
3860 		radeon_legacy_set_clock_gating(rdev, 1);
3861 	/* We need to force on some of the block */
3862 	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3863 	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3864 	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3865 		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3866 	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3867 }
3868 
3869 static int r100_startup(struct radeon_device *rdev)
3870 {
3871 	int r;
3872 
3873 	/* set common regs */
3874 	r100_set_common_regs(rdev);
3875 	/* program mc */
3876 	r100_mc_program(rdev);
3877 	/* Resume clock */
3878 	r100_clock_startup(rdev);
3879 	/* Initialize GART (initialize after TTM so we can allocate
3880 	 * memory through TTM but finalize after TTM) */
3881 	r100_enable_bm(rdev);
3882 	if (rdev->flags & RADEON_IS_PCI) {
3883 		r = r100_pci_gart_enable(rdev);
3884 		if (r)
3885 			return r;
3886 	}
3887 
3888 	/* allocate wb buffer */
3889 	r = radeon_wb_init(rdev);
3890 	if (r)
3891 		return r;
3892 
3893 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3894 	if (r) {
3895 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3896 		return r;
3897 	}
3898 
3899 	/* Enable IRQ */
3900 	if (!rdev->irq.installed) {
3901 		r = radeon_irq_kms_init(rdev);
3902 		if (r)
3903 			return r;
3904 	}
3905 
3906 	r100_irq_set(rdev);
3907 	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3908 	/* 1M ring buffer */
3909 	r = r100_cp_init(rdev, 1024 * 1024);
3910 	if (r) {
3911 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3912 		return r;
3913 	}
3914 
3915 	r = radeon_ib_pool_init(rdev);
3916 	if (r) {
3917 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3918 		return r;
3919 	}
3920 
3921 	return 0;
3922 }
3923 
3924 int r100_resume(struct radeon_device *rdev)
3925 {
3926 	int r;
3927 
3928 	/* Make sur GART are not working */
3929 	if (rdev->flags & RADEON_IS_PCI)
3930 		r100_pci_gart_disable(rdev);
3931 	/* Resume clock before doing reset */
3932 	r100_clock_startup(rdev);
3933 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3934 	if (radeon_asic_reset(rdev)) {
3935 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3936 			RREG32(R_000E40_RBBM_STATUS),
3937 			RREG32(R_0007C0_CP_STAT));
3938 	}
3939 	/* post */
3940 	radeon_combios_asic_init(rdev->ddev);
3941 	/* Resume clock after posting */
3942 	r100_clock_startup(rdev);
3943 	/* Initialize surface registers */
3944 	radeon_surface_init(rdev);
3945 
3946 	rdev->accel_working = true;
3947 	r = r100_startup(rdev);
3948 	if (r) {
3949 		rdev->accel_working = false;
3950 	}
3951 	return r;
3952 }
3953 
3954 int r100_suspend(struct radeon_device *rdev)
3955 {
3956 	radeon_pm_suspend(rdev);
3957 	r100_cp_disable(rdev);
3958 	radeon_wb_disable(rdev);
3959 	r100_irq_disable(rdev);
3960 	if (rdev->flags & RADEON_IS_PCI)
3961 		r100_pci_gart_disable(rdev);
3962 	return 0;
3963 }
3964 
3965 void r100_fini(struct radeon_device *rdev)
3966 {
3967 	radeon_pm_fini(rdev);
3968 	r100_cp_fini(rdev);
3969 	radeon_wb_fini(rdev);
3970 	radeon_ib_pool_fini(rdev);
3971 	radeon_gem_fini(rdev);
3972 	if (rdev->flags & RADEON_IS_PCI)
3973 		r100_pci_gart_fini(rdev);
3974 	radeon_agp_fini(rdev);
3975 	radeon_irq_kms_fini(rdev);
3976 	radeon_fence_driver_fini(rdev);
3977 	radeon_bo_fini(rdev);
3978 	radeon_atombios_fini(rdev);
3979 	kfree(rdev->bios);
3980 	rdev->bios = NULL;
3981 }
3982 
3983 /*
3984  * Due to how kexec works, it can leave the hw fully initialised when it
3985  * boots the new kernel. However doing our init sequence with the CP and
3986  * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3987  * do some quick sanity checks and restore sane values to avoid this
3988  * problem.
3989  */
3990 void r100_restore_sanity(struct radeon_device *rdev)
3991 {
3992 	u32 tmp;
3993 
3994 	tmp = RREG32(RADEON_CP_CSQ_CNTL);
3995 	if (tmp) {
3996 		WREG32(RADEON_CP_CSQ_CNTL, 0);
3997 	}
3998 	tmp = RREG32(RADEON_CP_RB_CNTL);
3999 	if (tmp) {
4000 		WREG32(RADEON_CP_RB_CNTL, 0);
4001 	}
4002 	tmp = RREG32(RADEON_SCRATCH_UMSK);
4003 	if (tmp) {
4004 		WREG32(RADEON_SCRATCH_UMSK, 0);
4005 	}
4006 }
4007 
4008 int r100_init(struct radeon_device *rdev)
4009 {
4010 	int r;
4011 
4012 	/* Register debugfs file specific to this group of asics */
4013 	r100_debugfs(rdev);
4014 	/* Disable VGA */
4015 	r100_vga_render_disable(rdev);
4016 	/* Initialize scratch registers */
4017 	radeon_scratch_init(rdev);
4018 	/* Initialize surface registers */
4019 	radeon_surface_init(rdev);
4020 	/* sanity check some register to avoid hangs like after kexec */
4021 	r100_restore_sanity(rdev);
4022 	/* TODO: disable VGA need to use VGA request */
4023 	/* BIOS*/
4024 	if (!radeon_get_bios(rdev)) {
4025 		if (ASIC_IS_AVIVO(rdev))
4026 			return -EINVAL;
4027 	}
4028 	if (rdev->is_atom_bios) {
4029 		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4030 		return -EINVAL;
4031 	} else {
4032 		r = radeon_combios_init(rdev);
4033 		if (r)
4034 			return r;
4035 	}
4036 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4037 	if (radeon_asic_reset(rdev)) {
4038 		dev_warn(rdev->dev,
4039 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4040 			RREG32(R_000E40_RBBM_STATUS),
4041 			RREG32(R_0007C0_CP_STAT));
4042 	}
4043 	/* check if cards are posted or not */
4044 	if (radeon_boot_test_post_card(rdev) == false)
4045 		return -EINVAL;
4046 	/* Set asic errata */
4047 	r100_errata(rdev);
4048 	/* Initialize clocks */
4049 	radeon_get_clock_info(rdev->ddev);
4050 	/* initialize AGP */
4051 	if (rdev->flags & RADEON_IS_AGP) {
4052 		r = radeon_agp_init(rdev);
4053 		if (r) {
4054 			radeon_agp_disable(rdev);
4055 		}
4056 	}
4057 	/* initialize VRAM */
4058 	r100_mc_init(rdev);
4059 	/* Fence driver */
4060 	r = radeon_fence_driver_init(rdev);
4061 	if (r)
4062 		return r;
4063 	/* Memory manager */
4064 	r = radeon_bo_init(rdev);
4065 	if (r)
4066 		return r;
4067 	if (rdev->flags & RADEON_IS_PCI) {
4068 		r = r100_pci_gart_init(rdev);
4069 		if (r)
4070 			return r;
4071 	}
4072 	r100_set_safe_registers(rdev);
4073 
4074 	/* Initialize power management */
4075 	radeon_pm_init(rdev);
4076 
4077 	rdev->accel_working = true;
4078 	r = r100_startup(rdev);
4079 	if (r) {
4080 		/* Somethings want wront with the accel init stop accel */
4081 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4082 		r100_cp_fini(rdev);
4083 		radeon_wb_fini(rdev);
4084 		radeon_ib_pool_fini(rdev);
4085 		radeon_irq_kms_fini(rdev);
4086 		if (rdev->flags & RADEON_IS_PCI)
4087 			r100_pci_gart_fini(rdev);
4088 		rdev->accel_working = false;
4089 	}
4090 	return 0;
4091 }
4092 
4093 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4094 {
4095 	if (reg < rdev->rio_mem_size)
4096 		return ioread32(rdev->rio_mem + reg);
4097 	else {
4098 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4099 		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4100 	}
4101 }
4102 
4103 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4104 {
4105 	if (reg < rdev->rio_mem_size)
4106 		iowrite32(v, rdev->rio_mem + reg);
4107 	else {
4108 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4109 		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4110 	}
4111 }
4112