xref: /openbmc/linux/drivers/gpu/drm/radeon/r100.c (revision 519424d7)
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 
29 #include <linux/firmware.h>
30 #include <linux/module.h>
31 #include <linux/pci.h>
32 #include <linux/seq_file.h>
33 #include <linux/slab.h>
34 
35 #include <drm/drm_device.h>
36 #include <drm/drm_file.h>
37 #include <drm/drm_fourcc.h>
38 #include <drm/drm_vblank.h>
39 #include <drm/radeon_drm.h>
40 
41 #include "atom.h"
42 #include "r100_reg_safe.h"
43 #include "r100d.h"
44 #include "radeon.h"
45 #include "radeon_asic.h"
46 #include "radeon_reg.h"
47 #include "rn50_reg_safe.h"
48 #include "rs100d.h"
49 #include "rv200d.h"
50 #include "rv250d.h"
51 
52 /* Firmware Names */
53 #define FIRMWARE_R100		"radeon/R100_cp.bin"
54 #define FIRMWARE_R200		"radeon/R200_cp.bin"
55 #define FIRMWARE_R300		"radeon/R300_cp.bin"
56 #define FIRMWARE_R420		"radeon/R420_cp.bin"
57 #define FIRMWARE_RS690		"radeon/RS690_cp.bin"
58 #define FIRMWARE_RS600		"radeon/RS600_cp.bin"
59 #define FIRMWARE_R520		"radeon/R520_cp.bin"
60 
61 MODULE_FIRMWARE(FIRMWARE_R100);
62 MODULE_FIRMWARE(FIRMWARE_R200);
63 MODULE_FIRMWARE(FIRMWARE_R300);
64 MODULE_FIRMWARE(FIRMWARE_R420);
65 MODULE_FIRMWARE(FIRMWARE_RS690);
66 MODULE_FIRMWARE(FIRMWARE_RS600);
67 MODULE_FIRMWARE(FIRMWARE_R520);
68 
69 #include "r100_track.h"
70 
71 /* This files gather functions specifics to:
72  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
73  * and others in some cases.
74  */
75 
76 static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
77 {
78 	if (crtc == 0) {
79 		if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
80 			return true;
81 		else
82 			return false;
83 	} else {
84 		if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
85 			return true;
86 		else
87 			return false;
88 	}
89 }
90 
91 static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
92 {
93 	u32 vline1, vline2;
94 
95 	if (crtc == 0) {
96 		vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
97 		vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
98 	} else {
99 		vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
100 		vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
101 	}
102 	if (vline1 != vline2)
103 		return true;
104 	else
105 		return false;
106 }
107 
108 /**
109  * r100_wait_for_vblank - vblank wait asic callback.
110  *
111  * @rdev: radeon_device pointer
112  * @crtc: crtc to wait for vblank on
113  *
114  * Wait for vblank on the requested crtc (r1xx-r4xx).
115  */
116 void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
117 {
118 	unsigned i = 0;
119 
120 	if (crtc >= rdev->num_crtc)
121 		return;
122 
123 	if (crtc == 0) {
124 		if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
125 			return;
126 	} else {
127 		if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
128 			return;
129 	}
130 
131 	/* depending on when we hit vblank, we may be close to active; if so,
132 	 * wait for another frame.
133 	 */
134 	while (r100_is_in_vblank(rdev, crtc)) {
135 		if (i++ % 100 == 0) {
136 			if (!r100_is_counter_moving(rdev, crtc))
137 				break;
138 		}
139 	}
140 
141 	while (!r100_is_in_vblank(rdev, crtc)) {
142 		if (i++ % 100 == 0) {
143 			if (!r100_is_counter_moving(rdev, crtc))
144 				break;
145 		}
146 	}
147 }
148 
149 /**
150  * r100_page_flip - pageflip callback.
151  *
152  * @rdev: radeon_device pointer
153  * @crtc_id: crtc to cleanup pageflip on
154  * @crtc_base: new address of the crtc (GPU MC address)
155  * @async: asynchronous flip
156  *
157  * Does the actual pageflip (r1xx-r4xx).
158  * During vblank we take the crtc lock and wait for the update_pending
159  * bit to go high, when it does, we release the lock, and allow the
160  * double buffered update to take place.
161  */
162 void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base, bool async)
163 {
164 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
165 	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
166 	int i;
167 
168 	/* Lock the graphics update lock */
169 	/* update the scanout addresses */
170 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
171 
172 	/* Wait for update_pending to go high. */
173 	for (i = 0; i < rdev->usec_timeout; i++) {
174 		if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
175 			break;
176 		udelay(1);
177 	}
178 	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
179 
180 	/* Unlock the lock, so double-buffering can take place inside vblank */
181 	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
182 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
183 
184 }
185 
186 /**
187  * r100_page_flip_pending - check if page flip is still pending
188  *
189  * @rdev: radeon_device pointer
190  * @crtc_id: crtc to check
191  *
192  * Check if the last pagefilp is still pending (r1xx-r4xx).
193  * Returns the current update pending status.
194  */
195 bool r100_page_flip_pending(struct radeon_device *rdev, int crtc_id)
196 {
197 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
198 
199 	/* Return current update_pending status: */
200 	return !!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) &
201 		RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET);
202 }
203 
204 /**
205  * r100_pm_get_dynpm_state - look up dynpm power state callback.
206  *
207  * @rdev: radeon_device pointer
208  *
209  * Look up the optimal power state based on the
210  * current state of the GPU (r1xx-r5xx).
211  * Used for dynpm only.
212  */
213 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
214 {
215 	int i;
216 	rdev->pm.dynpm_can_upclock = true;
217 	rdev->pm.dynpm_can_downclock = true;
218 
219 	switch (rdev->pm.dynpm_planned_action) {
220 	case DYNPM_ACTION_MINIMUM:
221 		rdev->pm.requested_power_state_index = 0;
222 		rdev->pm.dynpm_can_downclock = false;
223 		break;
224 	case DYNPM_ACTION_DOWNCLOCK:
225 		if (rdev->pm.current_power_state_index == 0) {
226 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
227 			rdev->pm.dynpm_can_downclock = false;
228 		} else {
229 			if (rdev->pm.active_crtc_count > 1) {
230 				for (i = 0; i < rdev->pm.num_power_states; i++) {
231 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
232 						continue;
233 					else if (i >= rdev->pm.current_power_state_index) {
234 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
235 						break;
236 					} else {
237 						rdev->pm.requested_power_state_index = i;
238 						break;
239 					}
240 				}
241 			} else
242 				rdev->pm.requested_power_state_index =
243 					rdev->pm.current_power_state_index - 1;
244 		}
245 		/* don't use the power state if crtcs are active and no display flag is set */
246 		if ((rdev->pm.active_crtc_count > 0) &&
247 		    (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
248 		     RADEON_PM_MODE_NO_DISPLAY)) {
249 			rdev->pm.requested_power_state_index++;
250 		}
251 		break;
252 	case DYNPM_ACTION_UPCLOCK:
253 		if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
254 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
255 			rdev->pm.dynpm_can_upclock = false;
256 		} else {
257 			if (rdev->pm.active_crtc_count > 1) {
258 				for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
259 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
260 						continue;
261 					else if (i <= rdev->pm.current_power_state_index) {
262 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
263 						break;
264 					} else {
265 						rdev->pm.requested_power_state_index = i;
266 						break;
267 					}
268 				}
269 			} else
270 				rdev->pm.requested_power_state_index =
271 					rdev->pm.current_power_state_index + 1;
272 		}
273 		break;
274 	case DYNPM_ACTION_DEFAULT:
275 		rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
276 		rdev->pm.dynpm_can_upclock = false;
277 		break;
278 	case DYNPM_ACTION_NONE:
279 	default:
280 		DRM_ERROR("Requested mode for not defined action\n");
281 		return;
282 	}
283 	/* only one clock mode per power state */
284 	rdev->pm.requested_clock_mode_index = 0;
285 
286 	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
287 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
288 		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
289 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
290 		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
291 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
292 		  pcie_lanes);
293 }
294 
295 /**
296  * r100_pm_init_profile - Initialize power profiles callback.
297  *
298  * @rdev: radeon_device pointer
299  *
300  * Initialize the power states used in profile mode
301  * (r1xx-r3xx).
302  * Used for profile mode only.
303  */
304 void r100_pm_init_profile(struct radeon_device *rdev)
305 {
306 	/* default */
307 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
308 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
309 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
310 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
311 	/* low sh */
312 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
313 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
314 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
315 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
316 	/* mid sh */
317 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
318 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
319 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
320 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
321 	/* high sh */
322 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
323 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
324 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
325 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
326 	/* low mh */
327 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
328 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
329 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
330 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
331 	/* mid mh */
332 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
333 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
334 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
335 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
336 	/* high mh */
337 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
338 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
339 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
340 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
341 }
342 
343 /**
344  * r100_pm_misc - set additional pm hw parameters callback.
345  *
346  * @rdev: radeon_device pointer
347  *
348  * Set non-clock parameters associated with a power state
349  * (voltage, pcie lanes, etc.) (r1xx-r4xx).
350  */
351 void r100_pm_misc(struct radeon_device *rdev)
352 {
353 	int requested_index = rdev->pm.requested_power_state_index;
354 	struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
355 	struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
356 	u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
357 
358 	if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
359 		if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
360 			tmp = RREG32(voltage->gpio.reg);
361 			if (voltage->active_high)
362 				tmp |= voltage->gpio.mask;
363 			else
364 				tmp &= ~(voltage->gpio.mask);
365 			WREG32(voltage->gpio.reg, tmp);
366 			if (voltage->delay)
367 				udelay(voltage->delay);
368 		} else {
369 			tmp = RREG32(voltage->gpio.reg);
370 			if (voltage->active_high)
371 				tmp &= ~voltage->gpio.mask;
372 			else
373 				tmp |= voltage->gpio.mask;
374 			WREG32(voltage->gpio.reg, tmp);
375 			if (voltage->delay)
376 				udelay(voltage->delay);
377 		}
378 	}
379 
380 	sclk_cntl = RREG32_PLL(SCLK_CNTL);
381 	sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
382 	sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
383 	sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
384 	sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
385 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
386 		sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
387 		if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
388 			sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
389 		else
390 			sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
391 		if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
392 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
393 		else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
394 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
395 	} else
396 		sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
397 
398 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
399 		sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
400 		if (voltage->delay) {
401 			sclk_more_cntl |= VOLTAGE_DROP_SYNC;
402 			switch (voltage->delay) {
403 			case 33:
404 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
405 				break;
406 			case 66:
407 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
408 				break;
409 			case 99:
410 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
411 				break;
412 			case 132:
413 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
414 				break;
415 			}
416 		} else
417 			sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
418 	} else
419 		sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
420 
421 	if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
422 		sclk_cntl &= ~FORCE_HDP;
423 	else
424 		sclk_cntl |= FORCE_HDP;
425 
426 	WREG32_PLL(SCLK_CNTL, sclk_cntl);
427 	WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
428 	WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
429 
430 	/* set pcie lanes */
431 	if ((rdev->flags & RADEON_IS_PCIE) &&
432 	    !(rdev->flags & RADEON_IS_IGP) &&
433 	    rdev->asic->pm.set_pcie_lanes &&
434 	    (ps->pcie_lanes !=
435 	     rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
436 		radeon_set_pcie_lanes(rdev,
437 				      ps->pcie_lanes);
438 		DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
439 	}
440 }
441 
442 /**
443  * r100_pm_prepare - pre-power state change callback.
444  *
445  * @rdev: radeon_device pointer
446  *
447  * Prepare for a power state change (r1xx-r4xx).
448  */
449 void r100_pm_prepare(struct radeon_device *rdev)
450 {
451 	struct drm_device *ddev = rdev->ddev;
452 	struct drm_crtc *crtc;
453 	struct radeon_crtc *radeon_crtc;
454 	u32 tmp;
455 
456 	/* disable any active CRTCs */
457 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
458 		radeon_crtc = to_radeon_crtc(crtc);
459 		if (radeon_crtc->enabled) {
460 			if (radeon_crtc->crtc_id) {
461 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
462 				tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
463 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
464 			} else {
465 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
466 				tmp |= RADEON_CRTC_DISP_REQ_EN_B;
467 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
468 			}
469 		}
470 	}
471 }
472 
473 /**
474  * r100_pm_finish - post-power state change callback.
475  *
476  * @rdev: radeon_device pointer
477  *
478  * Clean up after a power state change (r1xx-r4xx).
479  */
480 void r100_pm_finish(struct radeon_device *rdev)
481 {
482 	struct drm_device *ddev = rdev->ddev;
483 	struct drm_crtc *crtc;
484 	struct radeon_crtc *radeon_crtc;
485 	u32 tmp;
486 
487 	/* enable any active CRTCs */
488 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
489 		radeon_crtc = to_radeon_crtc(crtc);
490 		if (radeon_crtc->enabled) {
491 			if (radeon_crtc->crtc_id) {
492 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
493 				tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
494 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
495 			} else {
496 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
497 				tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
498 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
499 			}
500 		}
501 	}
502 }
503 
504 /**
505  * r100_gui_idle - gui idle callback.
506  *
507  * @rdev: radeon_device pointer
508  *
509  * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
510  * Returns true if idle, false if not.
511  */
512 bool r100_gui_idle(struct radeon_device *rdev)
513 {
514 	if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
515 		return false;
516 	else
517 		return true;
518 }
519 
520 /* hpd for digital panel detect/disconnect */
521 /**
522  * r100_hpd_sense - hpd sense callback.
523  *
524  * @rdev: radeon_device pointer
525  * @hpd: hpd (hotplug detect) pin
526  *
527  * Checks if a digital monitor is connected (r1xx-r4xx).
528  * Returns true if connected, false if not connected.
529  */
530 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
531 {
532 	bool connected = false;
533 
534 	switch (hpd) {
535 	case RADEON_HPD_1:
536 		if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
537 			connected = true;
538 		break;
539 	case RADEON_HPD_2:
540 		if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
541 			connected = true;
542 		break;
543 	default:
544 		break;
545 	}
546 	return connected;
547 }
548 
549 /**
550  * r100_hpd_set_polarity - hpd set polarity callback.
551  *
552  * @rdev: radeon_device pointer
553  * @hpd: hpd (hotplug detect) pin
554  *
555  * Set the polarity of the hpd pin (r1xx-r4xx).
556  */
557 void r100_hpd_set_polarity(struct radeon_device *rdev,
558 			   enum radeon_hpd_id hpd)
559 {
560 	u32 tmp;
561 	bool connected = r100_hpd_sense(rdev, hpd);
562 
563 	switch (hpd) {
564 	case RADEON_HPD_1:
565 		tmp = RREG32(RADEON_FP_GEN_CNTL);
566 		if (connected)
567 			tmp &= ~RADEON_FP_DETECT_INT_POL;
568 		else
569 			tmp |= RADEON_FP_DETECT_INT_POL;
570 		WREG32(RADEON_FP_GEN_CNTL, tmp);
571 		break;
572 	case RADEON_HPD_2:
573 		tmp = RREG32(RADEON_FP2_GEN_CNTL);
574 		if (connected)
575 			tmp &= ~RADEON_FP2_DETECT_INT_POL;
576 		else
577 			tmp |= RADEON_FP2_DETECT_INT_POL;
578 		WREG32(RADEON_FP2_GEN_CNTL, tmp);
579 		break;
580 	default:
581 		break;
582 	}
583 }
584 
585 /**
586  * r100_hpd_init - hpd setup callback.
587  *
588  * @rdev: radeon_device pointer
589  *
590  * Setup the hpd pins used by the card (r1xx-r4xx).
591  * Set the polarity, and enable the hpd interrupts.
592  */
593 void r100_hpd_init(struct radeon_device *rdev)
594 {
595 	struct drm_device *dev = rdev->ddev;
596 	struct drm_connector *connector;
597 	unsigned enable = 0;
598 
599 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
600 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
601 		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
602 			enable |= 1 << radeon_connector->hpd.hpd;
603 		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
604 	}
605 	radeon_irq_kms_enable_hpd(rdev, enable);
606 }
607 
608 /**
609  * r100_hpd_fini - hpd tear down callback.
610  *
611  * @rdev: radeon_device pointer
612  *
613  * Tear down the hpd pins used by the card (r1xx-r4xx).
614  * Disable the hpd interrupts.
615  */
616 void r100_hpd_fini(struct radeon_device *rdev)
617 {
618 	struct drm_device *dev = rdev->ddev;
619 	struct drm_connector *connector;
620 	unsigned disable = 0;
621 
622 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
623 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
624 		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
625 			disable |= 1 << radeon_connector->hpd.hpd;
626 	}
627 	radeon_irq_kms_disable_hpd(rdev, disable);
628 }
629 
630 /*
631  * PCI GART
632  */
633 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
634 {
635 	/* TODO: can we do somethings here ? */
636 	/* It seems hw only cache one entry so we should discard this
637 	 * entry otherwise if first GPU GART read hit this entry it
638 	 * could end up in wrong address. */
639 }
640 
641 int r100_pci_gart_init(struct radeon_device *rdev)
642 {
643 	int r;
644 
645 	if (rdev->gart.ptr) {
646 		WARN(1, "R100 PCI GART already initialized\n");
647 		return 0;
648 	}
649 	/* Initialize common gart structure */
650 	r = radeon_gart_init(rdev);
651 	if (r)
652 		return r;
653 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
654 	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
655 	rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
656 	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
657 	return radeon_gart_table_ram_alloc(rdev);
658 }
659 
660 int r100_pci_gart_enable(struct radeon_device *rdev)
661 {
662 	uint32_t tmp;
663 
664 	/* discard memory request outside of configured range */
665 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
666 	WREG32(RADEON_AIC_CNTL, tmp);
667 	/* set address range for PCI address translate */
668 	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
669 	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
670 	/* set PCI GART page-table base address */
671 	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
672 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
673 	WREG32(RADEON_AIC_CNTL, tmp);
674 	r100_pci_gart_tlb_flush(rdev);
675 	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
676 		 (unsigned)(rdev->mc.gtt_size >> 20),
677 		 (unsigned long long)rdev->gart.table_addr);
678 	rdev->gart.ready = true;
679 	return 0;
680 }
681 
682 void r100_pci_gart_disable(struct radeon_device *rdev)
683 {
684 	uint32_t tmp;
685 
686 	/* discard memory request outside of configured range */
687 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
688 	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
689 	WREG32(RADEON_AIC_LO_ADDR, 0);
690 	WREG32(RADEON_AIC_HI_ADDR, 0);
691 }
692 
693 uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
694 {
695 	return addr;
696 }
697 
698 void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
699 			    uint64_t entry)
700 {
701 	u32 *gtt = rdev->gart.ptr;
702 	gtt[i] = cpu_to_le32(lower_32_bits(entry));
703 }
704 
705 void r100_pci_gart_fini(struct radeon_device *rdev)
706 {
707 	radeon_gart_fini(rdev);
708 	r100_pci_gart_disable(rdev);
709 	radeon_gart_table_ram_free(rdev);
710 }
711 
712 int r100_irq_set(struct radeon_device *rdev)
713 {
714 	uint32_t tmp = 0;
715 
716 	if (!rdev->irq.installed) {
717 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
718 		WREG32(R_000040_GEN_INT_CNTL, 0);
719 		return -EINVAL;
720 	}
721 	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
722 		tmp |= RADEON_SW_INT_ENABLE;
723 	}
724 	if (rdev->irq.crtc_vblank_int[0] ||
725 	    atomic_read(&rdev->irq.pflip[0])) {
726 		tmp |= RADEON_CRTC_VBLANK_MASK;
727 	}
728 	if (rdev->irq.crtc_vblank_int[1] ||
729 	    atomic_read(&rdev->irq.pflip[1])) {
730 		tmp |= RADEON_CRTC2_VBLANK_MASK;
731 	}
732 	if (rdev->irq.hpd[0]) {
733 		tmp |= RADEON_FP_DETECT_MASK;
734 	}
735 	if (rdev->irq.hpd[1]) {
736 		tmp |= RADEON_FP2_DETECT_MASK;
737 	}
738 	WREG32(RADEON_GEN_INT_CNTL, tmp);
739 
740 	/* read back to post the write */
741 	RREG32(RADEON_GEN_INT_CNTL);
742 
743 	return 0;
744 }
745 
746 void r100_irq_disable(struct radeon_device *rdev)
747 {
748 	u32 tmp;
749 
750 	WREG32(R_000040_GEN_INT_CNTL, 0);
751 	/* Wait and acknowledge irq */
752 	mdelay(1);
753 	tmp = RREG32(R_000044_GEN_INT_STATUS);
754 	WREG32(R_000044_GEN_INT_STATUS, tmp);
755 }
756 
757 static uint32_t r100_irq_ack(struct radeon_device *rdev)
758 {
759 	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
760 	uint32_t irq_mask = RADEON_SW_INT_TEST |
761 		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
762 		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
763 
764 	if (irqs) {
765 		WREG32(RADEON_GEN_INT_STATUS, irqs);
766 	}
767 	return irqs & irq_mask;
768 }
769 
770 int r100_irq_process(struct radeon_device *rdev)
771 {
772 	uint32_t status, msi_rearm;
773 	bool queue_hotplug = false;
774 
775 	status = r100_irq_ack(rdev);
776 	if (!status) {
777 		return IRQ_NONE;
778 	}
779 	if (rdev->shutdown) {
780 		return IRQ_NONE;
781 	}
782 	while (status) {
783 		/* SW interrupt */
784 		if (status & RADEON_SW_INT_TEST) {
785 			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
786 		}
787 		/* Vertical blank interrupts */
788 		if (status & RADEON_CRTC_VBLANK_STAT) {
789 			if (rdev->irq.crtc_vblank_int[0]) {
790 				drm_handle_vblank(rdev->ddev, 0);
791 				rdev->pm.vblank_sync = true;
792 				wake_up(&rdev->irq.vblank_queue);
793 			}
794 			if (atomic_read(&rdev->irq.pflip[0]))
795 				radeon_crtc_handle_vblank(rdev, 0);
796 		}
797 		if (status & RADEON_CRTC2_VBLANK_STAT) {
798 			if (rdev->irq.crtc_vblank_int[1]) {
799 				drm_handle_vblank(rdev->ddev, 1);
800 				rdev->pm.vblank_sync = true;
801 				wake_up(&rdev->irq.vblank_queue);
802 			}
803 			if (atomic_read(&rdev->irq.pflip[1]))
804 				radeon_crtc_handle_vblank(rdev, 1);
805 		}
806 		if (status & RADEON_FP_DETECT_STAT) {
807 			queue_hotplug = true;
808 			DRM_DEBUG("HPD1\n");
809 		}
810 		if (status & RADEON_FP2_DETECT_STAT) {
811 			queue_hotplug = true;
812 			DRM_DEBUG("HPD2\n");
813 		}
814 		status = r100_irq_ack(rdev);
815 	}
816 	if (queue_hotplug)
817 		schedule_delayed_work(&rdev->hotplug_work, 0);
818 	if (rdev->msi_enabled) {
819 		switch (rdev->family) {
820 		case CHIP_RS400:
821 		case CHIP_RS480:
822 			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
823 			WREG32(RADEON_AIC_CNTL, msi_rearm);
824 			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
825 			break;
826 		default:
827 			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
828 			break;
829 		}
830 	}
831 	return IRQ_HANDLED;
832 }
833 
834 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
835 {
836 	if (crtc == 0)
837 		return RREG32(RADEON_CRTC_CRNT_FRAME);
838 	else
839 		return RREG32(RADEON_CRTC2_CRNT_FRAME);
840 }
841 
842 /**
843  * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
844  * @rdev: radeon device structure
845  * @ring: ring buffer struct for emitting packets
846  */
847 static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
848 {
849 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
850 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
851 				RADEON_HDP_READ_BUFFER_INVALIDATE);
852 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
853 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
854 }
855 
856 /* Who ever call radeon_fence_emit should call ring_lock and ask
857  * for enough space (today caller are ib schedule and buffer move) */
858 void r100_fence_ring_emit(struct radeon_device *rdev,
859 			  struct radeon_fence *fence)
860 {
861 	struct radeon_ring *ring = &rdev->ring[fence->ring];
862 
863 	/* We have to make sure that caches are flushed before
864 	 * CPU might read something from VRAM. */
865 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
866 	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
867 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
868 	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
869 	/* Wait until IDLE & CLEAN */
870 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
871 	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
872 	r100_ring_hdp_flush(rdev, ring);
873 	/* Emit fence sequence & fire IRQ */
874 	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
875 	radeon_ring_write(ring, fence->seq);
876 	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
877 	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
878 }
879 
880 bool r100_semaphore_ring_emit(struct radeon_device *rdev,
881 			      struct radeon_ring *ring,
882 			      struct radeon_semaphore *semaphore,
883 			      bool emit_wait)
884 {
885 	/* Unused on older asics, since we don't have semaphores or multiple rings */
886 	BUG();
887 	return false;
888 }
889 
890 struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
891 				    uint64_t src_offset,
892 				    uint64_t dst_offset,
893 				    unsigned num_gpu_pages,
894 				    struct dma_resv *resv)
895 {
896 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
897 	struct radeon_fence *fence;
898 	uint32_t cur_pages;
899 	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
900 	uint32_t pitch;
901 	uint32_t stride_pixels;
902 	unsigned ndw;
903 	int num_loops;
904 	int r = 0;
905 
906 	/* radeon limited to 16k stride */
907 	stride_bytes &= 0x3fff;
908 	/* radeon pitch is /64 */
909 	pitch = stride_bytes / 64;
910 	stride_pixels = stride_bytes / 4;
911 	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
912 
913 	/* Ask for enough room for blit + flush + fence */
914 	ndw = 64 + (10 * num_loops);
915 	r = radeon_ring_lock(rdev, ring, ndw);
916 	if (r) {
917 		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
918 		return ERR_PTR(-EINVAL);
919 	}
920 	while (num_gpu_pages > 0) {
921 		cur_pages = num_gpu_pages;
922 		if (cur_pages > 8191) {
923 			cur_pages = 8191;
924 		}
925 		num_gpu_pages -= cur_pages;
926 
927 		/* pages are in Y direction - height
928 		   page width in X direction - width */
929 		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
930 		radeon_ring_write(ring,
931 				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
932 				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
933 				  RADEON_GMC_SRC_CLIPPING |
934 				  RADEON_GMC_DST_CLIPPING |
935 				  RADEON_GMC_BRUSH_NONE |
936 				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
937 				  RADEON_GMC_SRC_DATATYPE_COLOR |
938 				  RADEON_ROP3_S |
939 				  RADEON_DP_SRC_SOURCE_MEMORY |
940 				  RADEON_GMC_CLR_CMP_CNTL_DIS |
941 				  RADEON_GMC_WR_MSK_DIS);
942 		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
943 		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
944 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
945 		radeon_ring_write(ring, 0);
946 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
947 		radeon_ring_write(ring, num_gpu_pages);
948 		radeon_ring_write(ring, num_gpu_pages);
949 		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
950 	}
951 	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
952 	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
953 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
954 	radeon_ring_write(ring,
955 			  RADEON_WAIT_2D_IDLECLEAN |
956 			  RADEON_WAIT_HOST_IDLECLEAN |
957 			  RADEON_WAIT_DMA_GUI_IDLE);
958 	r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
959 	if (r) {
960 		radeon_ring_unlock_undo(rdev, ring);
961 		return ERR_PTR(r);
962 	}
963 	radeon_ring_unlock_commit(rdev, ring, false);
964 	return fence;
965 }
966 
967 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
968 {
969 	unsigned i;
970 	u32 tmp;
971 
972 	for (i = 0; i < rdev->usec_timeout; i++) {
973 		tmp = RREG32(R_000E40_RBBM_STATUS);
974 		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
975 			return 0;
976 		}
977 		udelay(1);
978 	}
979 	return -1;
980 }
981 
982 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
983 {
984 	int r;
985 
986 	r = radeon_ring_lock(rdev, ring, 2);
987 	if (r) {
988 		return;
989 	}
990 	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
991 	radeon_ring_write(ring,
992 			  RADEON_ISYNC_ANY2D_IDLE3D |
993 			  RADEON_ISYNC_ANY3D_IDLE2D |
994 			  RADEON_ISYNC_WAIT_IDLEGUI |
995 			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
996 	radeon_ring_unlock_commit(rdev, ring, false);
997 }
998 
999 
1000 /* Load the microcode for the CP */
1001 static int r100_cp_init_microcode(struct radeon_device *rdev)
1002 {
1003 	const char *fw_name = NULL;
1004 	int err;
1005 
1006 	DRM_DEBUG_KMS("\n");
1007 
1008 	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
1009 	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
1010 	    (rdev->family == CHIP_RS200)) {
1011 		DRM_INFO("Loading R100 Microcode\n");
1012 		fw_name = FIRMWARE_R100;
1013 	} else if ((rdev->family == CHIP_R200) ||
1014 		   (rdev->family == CHIP_RV250) ||
1015 		   (rdev->family == CHIP_RV280) ||
1016 		   (rdev->family == CHIP_RS300)) {
1017 		DRM_INFO("Loading R200 Microcode\n");
1018 		fw_name = FIRMWARE_R200;
1019 	} else if ((rdev->family == CHIP_R300) ||
1020 		   (rdev->family == CHIP_R350) ||
1021 		   (rdev->family == CHIP_RV350) ||
1022 		   (rdev->family == CHIP_RV380) ||
1023 		   (rdev->family == CHIP_RS400) ||
1024 		   (rdev->family == CHIP_RS480)) {
1025 		DRM_INFO("Loading R300 Microcode\n");
1026 		fw_name = FIRMWARE_R300;
1027 	} else if ((rdev->family == CHIP_R420) ||
1028 		   (rdev->family == CHIP_R423) ||
1029 		   (rdev->family == CHIP_RV410)) {
1030 		DRM_INFO("Loading R400 Microcode\n");
1031 		fw_name = FIRMWARE_R420;
1032 	} else if ((rdev->family == CHIP_RS690) ||
1033 		   (rdev->family == CHIP_RS740)) {
1034 		DRM_INFO("Loading RS690/RS740 Microcode\n");
1035 		fw_name = FIRMWARE_RS690;
1036 	} else if (rdev->family == CHIP_RS600) {
1037 		DRM_INFO("Loading RS600 Microcode\n");
1038 		fw_name = FIRMWARE_RS600;
1039 	} else if ((rdev->family == CHIP_RV515) ||
1040 		   (rdev->family == CHIP_R520) ||
1041 		   (rdev->family == CHIP_RV530) ||
1042 		   (rdev->family == CHIP_R580) ||
1043 		   (rdev->family == CHIP_RV560) ||
1044 		   (rdev->family == CHIP_RV570)) {
1045 		DRM_INFO("Loading R500 Microcode\n");
1046 		fw_name = FIRMWARE_R520;
1047 	}
1048 
1049 	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1050 	if (err) {
1051 		pr_err("radeon_cp: Failed to load firmware \"%s\"\n", fw_name);
1052 	} else if (rdev->me_fw->size % 8) {
1053 		pr_err("radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1054 		       rdev->me_fw->size, fw_name);
1055 		err = -EINVAL;
1056 		release_firmware(rdev->me_fw);
1057 		rdev->me_fw = NULL;
1058 	}
1059 	return err;
1060 }
1061 
1062 u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1063 		      struct radeon_ring *ring)
1064 {
1065 	u32 rptr;
1066 
1067 	if (rdev->wb.enabled)
1068 		rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1069 	else
1070 		rptr = RREG32(RADEON_CP_RB_RPTR);
1071 
1072 	return rptr;
1073 }
1074 
1075 u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1076 		      struct radeon_ring *ring)
1077 {
1078 	return RREG32(RADEON_CP_RB_WPTR);
1079 }
1080 
1081 void r100_gfx_set_wptr(struct radeon_device *rdev,
1082 		       struct radeon_ring *ring)
1083 {
1084 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1085 	(void)RREG32(RADEON_CP_RB_WPTR);
1086 }
1087 
1088 static void r100_cp_load_microcode(struct radeon_device *rdev)
1089 {
1090 	const __be32 *fw_data;
1091 	int i, size;
1092 
1093 	if (r100_gui_wait_for_idle(rdev)) {
1094 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1095 	}
1096 
1097 	if (rdev->me_fw) {
1098 		size = rdev->me_fw->size / 4;
1099 		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1100 		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1101 		for (i = 0; i < size; i += 2) {
1102 			WREG32(RADEON_CP_ME_RAM_DATAH,
1103 			       be32_to_cpup(&fw_data[i]));
1104 			WREG32(RADEON_CP_ME_RAM_DATAL,
1105 			       be32_to_cpup(&fw_data[i + 1]));
1106 		}
1107 	}
1108 }
1109 
1110 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1111 {
1112 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1113 	unsigned rb_bufsz;
1114 	unsigned rb_blksz;
1115 	unsigned max_fetch;
1116 	unsigned pre_write_timer;
1117 	unsigned pre_write_limit;
1118 	unsigned indirect2_start;
1119 	unsigned indirect1_start;
1120 	uint32_t tmp;
1121 	int r;
1122 
1123 	r100_debugfs_cp_init(rdev);
1124 	if (!rdev->me_fw) {
1125 		r = r100_cp_init_microcode(rdev);
1126 		if (r) {
1127 			DRM_ERROR("Failed to load firmware!\n");
1128 			return r;
1129 		}
1130 	}
1131 
1132 	/* Align ring size */
1133 	rb_bufsz = order_base_2(ring_size / 8);
1134 	ring_size = (1 << (rb_bufsz + 1)) * 4;
1135 	r100_cp_load_microcode(rdev);
1136 	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1137 			     RADEON_CP_PACKET2);
1138 	if (r) {
1139 		return r;
1140 	}
1141 	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1142 	 * the rptr copy in system ram */
1143 	rb_blksz = 9;
1144 	/* cp will read 128bytes at a time (4 dwords) */
1145 	max_fetch = 1;
1146 	ring->align_mask = 16 - 1;
1147 	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1148 	pre_write_timer = 64;
1149 	/* Force CP_RB_WPTR write if written more than one time before the
1150 	 * delay expire
1151 	 */
1152 	pre_write_limit = 0;
1153 	/* Setup the cp cache like this (cache size is 96 dwords) :
1154 	 *	RING		0  to 15
1155 	 *	INDIRECT1	16 to 79
1156 	 *	INDIRECT2	80 to 95
1157 	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1158 	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1159 	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1160 	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1161 	 * so it gets the bigger cache.
1162 	 */
1163 	indirect2_start = 80;
1164 	indirect1_start = 16;
1165 	/* cp setup */
1166 	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1167 	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1168 	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1169 	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1170 #ifdef __BIG_ENDIAN
1171 	tmp |= RADEON_BUF_SWAP_32BIT;
1172 #endif
1173 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1174 
1175 	/* Set ring address */
1176 	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1177 	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1178 	/* Force read & write ptr to 0 */
1179 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1180 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1181 	ring->wptr = 0;
1182 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1183 
1184 	/* set the wb address whether it's enabled or not */
1185 	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1186 		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1187 	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1188 
1189 	if (rdev->wb.enabled)
1190 		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1191 	else {
1192 		tmp |= RADEON_RB_NO_UPDATE;
1193 		WREG32(R_000770_SCRATCH_UMSK, 0);
1194 	}
1195 
1196 	WREG32(RADEON_CP_RB_CNTL, tmp);
1197 	udelay(10);
1198 	/* Set cp mode to bus mastering & enable cp*/
1199 	WREG32(RADEON_CP_CSQ_MODE,
1200 	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1201 	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1202 	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1203 	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1204 	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1205 
1206 	/* at this point everything should be setup correctly to enable master */
1207 	pci_set_master(rdev->pdev);
1208 
1209 	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1210 	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1211 	if (r) {
1212 		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1213 		return r;
1214 	}
1215 	ring->ready = true;
1216 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1217 
1218 	if (!ring->rptr_save_reg /* not resuming from suspend */
1219 	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1220 		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1221 		if (r) {
1222 			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1223 			ring->rptr_save_reg = 0;
1224 		}
1225 	}
1226 	return 0;
1227 }
1228 
1229 void r100_cp_fini(struct radeon_device *rdev)
1230 {
1231 	if (r100_cp_wait_for_idle(rdev)) {
1232 		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1233 	}
1234 	/* Disable ring */
1235 	r100_cp_disable(rdev);
1236 	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1237 	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1238 	DRM_INFO("radeon: cp finalized\n");
1239 }
1240 
1241 void r100_cp_disable(struct radeon_device *rdev)
1242 {
1243 	/* Disable ring */
1244 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1245 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1246 	WREG32(RADEON_CP_CSQ_MODE, 0);
1247 	WREG32(RADEON_CP_CSQ_CNTL, 0);
1248 	WREG32(R_000770_SCRATCH_UMSK, 0);
1249 	if (r100_gui_wait_for_idle(rdev)) {
1250 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1251 	}
1252 }
1253 
1254 /*
1255  * CS functions
1256  */
1257 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1258 			    struct radeon_cs_packet *pkt,
1259 			    unsigned idx,
1260 			    unsigned reg)
1261 {
1262 	int r;
1263 	u32 tile_flags = 0;
1264 	u32 tmp;
1265 	struct radeon_bo_list *reloc;
1266 	u32 value;
1267 
1268 	r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1269 	if (r) {
1270 		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1271 			  idx, reg);
1272 		radeon_cs_dump_packet(p, pkt);
1273 		return r;
1274 	}
1275 
1276 	value = radeon_get_ib_value(p, idx);
1277 	tmp = value & 0x003fffff;
1278 	tmp += (((u32)reloc->gpu_offset) >> 10);
1279 
1280 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1281 		if (reloc->tiling_flags & RADEON_TILING_MACRO)
1282 			tile_flags |= RADEON_DST_TILE_MACRO;
1283 		if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1284 			if (reg == RADEON_SRC_PITCH_OFFSET) {
1285 				DRM_ERROR("Cannot src blit from microtiled surface\n");
1286 				radeon_cs_dump_packet(p, pkt);
1287 				return -EINVAL;
1288 			}
1289 			tile_flags |= RADEON_DST_TILE_MICRO;
1290 		}
1291 
1292 		tmp |= tile_flags;
1293 		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1294 	} else
1295 		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1296 	return 0;
1297 }
1298 
1299 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1300 			     struct radeon_cs_packet *pkt,
1301 			     int idx)
1302 {
1303 	unsigned c, i;
1304 	struct radeon_bo_list *reloc;
1305 	struct r100_cs_track *track;
1306 	int r = 0;
1307 	volatile uint32_t *ib;
1308 	u32 idx_value;
1309 
1310 	ib = p->ib.ptr;
1311 	track = (struct r100_cs_track *)p->track;
1312 	c = radeon_get_ib_value(p, idx++) & 0x1F;
1313 	if (c > 16) {
1314 	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1315 		      pkt->opcode);
1316 	    radeon_cs_dump_packet(p, pkt);
1317 	    return -EINVAL;
1318 	}
1319 	track->num_arrays = c;
1320 	for (i = 0; i < (c - 1); i+=2, idx+=3) {
1321 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1322 		if (r) {
1323 			DRM_ERROR("No reloc for packet3 %d\n",
1324 				  pkt->opcode);
1325 			radeon_cs_dump_packet(p, pkt);
1326 			return r;
1327 		}
1328 		idx_value = radeon_get_ib_value(p, idx);
1329 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1330 
1331 		track->arrays[i + 0].esize = idx_value >> 8;
1332 		track->arrays[i + 0].robj = reloc->robj;
1333 		track->arrays[i + 0].esize &= 0x7F;
1334 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1335 		if (r) {
1336 			DRM_ERROR("No reloc for packet3 %d\n",
1337 				  pkt->opcode);
1338 			radeon_cs_dump_packet(p, pkt);
1339 			return r;
1340 		}
1341 		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
1342 		track->arrays[i + 1].robj = reloc->robj;
1343 		track->arrays[i + 1].esize = idx_value >> 24;
1344 		track->arrays[i + 1].esize &= 0x7F;
1345 	}
1346 	if (c & 1) {
1347 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1348 		if (r) {
1349 			DRM_ERROR("No reloc for packet3 %d\n",
1350 					  pkt->opcode);
1351 			radeon_cs_dump_packet(p, pkt);
1352 			return r;
1353 		}
1354 		idx_value = radeon_get_ib_value(p, idx);
1355 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1356 		track->arrays[i + 0].robj = reloc->robj;
1357 		track->arrays[i + 0].esize = idx_value >> 8;
1358 		track->arrays[i + 0].esize &= 0x7F;
1359 	}
1360 	return r;
1361 }
1362 
1363 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1364 			  struct radeon_cs_packet *pkt,
1365 			  const unsigned *auth, unsigned n,
1366 			  radeon_packet0_check_t check)
1367 {
1368 	unsigned reg;
1369 	unsigned i, j, m;
1370 	unsigned idx;
1371 	int r;
1372 
1373 	idx = pkt->idx + 1;
1374 	reg = pkt->reg;
1375 	/* Check that register fall into register range
1376 	 * determined by the number of entry (n) in the
1377 	 * safe register bitmap.
1378 	 */
1379 	if (pkt->one_reg_wr) {
1380 		if ((reg >> 7) > n) {
1381 			return -EINVAL;
1382 		}
1383 	} else {
1384 		if (((reg + (pkt->count << 2)) >> 7) > n) {
1385 			return -EINVAL;
1386 		}
1387 	}
1388 	for (i = 0; i <= pkt->count; i++, idx++) {
1389 		j = (reg >> 7);
1390 		m = 1 << ((reg >> 2) & 31);
1391 		if (auth[j] & m) {
1392 			r = check(p, pkt, idx, reg);
1393 			if (r) {
1394 				return r;
1395 			}
1396 		}
1397 		if (pkt->one_reg_wr) {
1398 			if (!(auth[j] & m)) {
1399 				break;
1400 			}
1401 		} else {
1402 			reg += 4;
1403 		}
1404 	}
1405 	return 0;
1406 }
1407 
1408 /**
1409  * r100_cs_packet_parse_vline() - parse userspace VLINE packet
1410  * @p:		parser structure holding parsing context.
1411  *
1412  * Userspace sends a special sequence for VLINE waits.
1413  * PACKET0 - VLINE_START_END + value
1414  * PACKET0 - WAIT_UNTIL +_value
1415  * RELOC (P3) - crtc_id in reloc.
1416  *
1417  * This function parses this and relocates the VLINE START END
1418  * and WAIT UNTIL packets to the correct crtc.
1419  * It also detects a switched off crtc and nulls out the
1420  * wait in that case.
1421  */
1422 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1423 {
1424 	struct drm_crtc *crtc;
1425 	struct radeon_crtc *radeon_crtc;
1426 	struct radeon_cs_packet p3reloc, waitreloc;
1427 	int crtc_id;
1428 	int r;
1429 	uint32_t header, h_idx, reg;
1430 	volatile uint32_t *ib;
1431 
1432 	ib = p->ib.ptr;
1433 
1434 	/* parse the wait until */
1435 	r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1436 	if (r)
1437 		return r;
1438 
1439 	/* check its a wait until and only 1 count */
1440 	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1441 	    waitreloc.count != 0) {
1442 		DRM_ERROR("vline wait had illegal wait until segment\n");
1443 		return -EINVAL;
1444 	}
1445 
1446 	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1447 		DRM_ERROR("vline wait had illegal wait until\n");
1448 		return -EINVAL;
1449 	}
1450 
1451 	/* jump over the NOP */
1452 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1453 	if (r)
1454 		return r;
1455 
1456 	h_idx = p->idx - 2;
1457 	p->idx += waitreloc.count + 2;
1458 	p->idx += p3reloc.count + 2;
1459 
1460 	header = radeon_get_ib_value(p, h_idx);
1461 	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1462 	reg = R100_CP_PACKET0_GET_REG(header);
1463 	crtc = drm_crtc_find(p->rdev->ddev, p->filp, crtc_id);
1464 	if (!crtc) {
1465 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1466 		return -ENOENT;
1467 	}
1468 	radeon_crtc = to_radeon_crtc(crtc);
1469 	crtc_id = radeon_crtc->crtc_id;
1470 
1471 	if (!crtc->enabled) {
1472 		/* if the CRTC isn't enabled - we need to nop out the wait until */
1473 		ib[h_idx + 2] = PACKET2(0);
1474 		ib[h_idx + 3] = PACKET2(0);
1475 	} else if (crtc_id == 1) {
1476 		switch (reg) {
1477 		case AVIVO_D1MODE_VLINE_START_END:
1478 			header &= ~R300_CP_PACKET0_REG_MASK;
1479 			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1480 			break;
1481 		case RADEON_CRTC_GUI_TRIG_VLINE:
1482 			header &= ~R300_CP_PACKET0_REG_MASK;
1483 			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1484 			break;
1485 		default:
1486 			DRM_ERROR("unknown crtc reloc\n");
1487 			return -EINVAL;
1488 		}
1489 		ib[h_idx] = header;
1490 		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1491 	}
1492 
1493 	return 0;
1494 }
1495 
1496 static int r100_get_vtx_size(uint32_t vtx_fmt)
1497 {
1498 	int vtx_size;
1499 	vtx_size = 2;
1500 	/* ordered according to bits in spec */
1501 	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1502 		vtx_size++;
1503 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1504 		vtx_size += 3;
1505 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1506 		vtx_size++;
1507 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1508 		vtx_size++;
1509 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1510 		vtx_size += 3;
1511 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1512 		vtx_size++;
1513 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1514 		vtx_size++;
1515 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1516 		vtx_size += 2;
1517 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1518 		vtx_size += 2;
1519 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1520 		vtx_size++;
1521 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1522 		vtx_size += 2;
1523 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1524 		vtx_size++;
1525 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1526 		vtx_size += 2;
1527 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1528 		vtx_size++;
1529 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1530 		vtx_size++;
1531 	/* blend weight */
1532 	if (vtx_fmt & (0x7 << 15))
1533 		vtx_size += (vtx_fmt >> 15) & 0x7;
1534 	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1535 		vtx_size += 3;
1536 	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1537 		vtx_size += 2;
1538 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1539 		vtx_size++;
1540 	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1541 		vtx_size++;
1542 	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1543 		vtx_size++;
1544 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1545 		vtx_size++;
1546 	return vtx_size;
1547 }
1548 
1549 static int r100_packet0_check(struct radeon_cs_parser *p,
1550 			      struct radeon_cs_packet *pkt,
1551 			      unsigned idx, unsigned reg)
1552 {
1553 	struct radeon_bo_list *reloc;
1554 	struct r100_cs_track *track;
1555 	volatile uint32_t *ib;
1556 	uint32_t tmp;
1557 	int r;
1558 	int i, face;
1559 	u32 tile_flags = 0;
1560 	u32 idx_value;
1561 
1562 	ib = p->ib.ptr;
1563 	track = (struct r100_cs_track *)p->track;
1564 
1565 	idx_value = radeon_get_ib_value(p, idx);
1566 
1567 	switch (reg) {
1568 	case RADEON_CRTC_GUI_TRIG_VLINE:
1569 		r = r100_cs_packet_parse_vline(p);
1570 		if (r) {
1571 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1572 				  idx, reg);
1573 			radeon_cs_dump_packet(p, pkt);
1574 			return r;
1575 		}
1576 		break;
1577 		/* FIXME: only allow PACKET3 blit? easier to check for out of
1578 		 * range access */
1579 	case RADEON_DST_PITCH_OFFSET:
1580 	case RADEON_SRC_PITCH_OFFSET:
1581 		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1582 		if (r)
1583 			return r;
1584 		break;
1585 	case RADEON_RB3D_DEPTHOFFSET:
1586 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1587 		if (r) {
1588 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1589 				  idx, reg);
1590 			radeon_cs_dump_packet(p, pkt);
1591 			return r;
1592 		}
1593 		track->zb.robj = reloc->robj;
1594 		track->zb.offset = idx_value;
1595 		track->zb_dirty = true;
1596 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1597 		break;
1598 	case RADEON_RB3D_COLOROFFSET:
1599 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1600 		if (r) {
1601 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1602 				  idx, reg);
1603 			radeon_cs_dump_packet(p, pkt);
1604 			return r;
1605 		}
1606 		track->cb[0].robj = reloc->robj;
1607 		track->cb[0].offset = idx_value;
1608 		track->cb_dirty = true;
1609 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1610 		break;
1611 	case RADEON_PP_TXOFFSET_0:
1612 	case RADEON_PP_TXOFFSET_1:
1613 	case RADEON_PP_TXOFFSET_2:
1614 		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1615 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1616 		if (r) {
1617 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1618 				  idx, reg);
1619 			radeon_cs_dump_packet(p, pkt);
1620 			return r;
1621 		}
1622 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1623 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1624 				tile_flags |= RADEON_TXO_MACRO_TILE;
1625 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1626 				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1627 
1628 			tmp = idx_value & ~(0x7 << 2);
1629 			tmp |= tile_flags;
1630 			ib[idx] = tmp + ((u32)reloc->gpu_offset);
1631 		} else
1632 			ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1633 		track->textures[i].robj = reloc->robj;
1634 		track->tex_dirty = true;
1635 		break;
1636 	case RADEON_PP_CUBIC_OFFSET_T0_0:
1637 	case RADEON_PP_CUBIC_OFFSET_T0_1:
1638 	case RADEON_PP_CUBIC_OFFSET_T0_2:
1639 	case RADEON_PP_CUBIC_OFFSET_T0_3:
1640 	case RADEON_PP_CUBIC_OFFSET_T0_4:
1641 		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1642 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1643 		if (r) {
1644 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1645 				  idx, reg);
1646 			radeon_cs_dump_packet(p, pkt);
1647 			return r;
1648 		}
1649 		track->textures[0].cube_info[i].offset = idx_value;
1650 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1651 		track->textures[0].cube_info[i].robj = reloc->robj;
1652 		track->tex_dirty = true;
1653 		break;
1654 	case RADEON_PP_CUBIC_OFFSET_T1_0:
1655 	case RADEON_PP_CUBIC_OFFSET_T1_1:
1656 	case RADEON_PP_CUBIC_OFFSET_T1_2:
1657 	case RADEON_PP_CUBIC_OFFSET_T1_3:
1658 	case RADEON_PP_CUBIC_OFFSET_T1_4:
1659 		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1660 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1661 		if (r) {
1662 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1663 				  idx, reg);
1664 			radeon_cs_dump_packet(p, pkt);
1665 			return r;
1666 		}
1667 		track->textures[1].cube_info[i].offset = idx_value;
1668 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1669 		track->textures[1].cube_info[i].robj = reloc->robj;
1670 		track->tex_dirty = true;
1671 		break;
1672 	case RADEON_PP_CUBIC_OFFSET_T2_0:
1673 	case RADEON_PP_CUBIC_OFFSET_T2_1:
1674 	case RADEON_PP_CUBIC_OFFSET_T2_2:
1675 	case RADEON_PP_CUBIC_OFFSET_T2_3:
1676 	case RADEON_PP_CUBIC_OFFSET_T2_4:
1677 		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1678 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1679 		if (r) {
1680 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1681 				  idx, reg);
1682 			radeon_cs_dump_packet(p, pkt);
1683 			return r;
1684 		}
1685 		track->textures[2].cube_info[i].offset = idx_value;
1686 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1687 		track->textures[2].cube_info[i].robj = reloc->robj;
1688 		track->tex_dirty = true;
1689 		break;
1690 	case RADEON_RE_WIDTH_HEIGHT:
1691 		track->maxy = ((idx_value >> 16) & 0x7FF);
1692 		track->cb_dirty = true;
1693 		track->zb_dirty = true;
1694 		break;
1695 	case RADEON_RB3D_COLORPITCH:
1696 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1697 		if (r) {
1698 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1699 				  idx, reg);
1700 			radeon_cs_dump_packet(p, pkt);
1701 			return r;
1702 		}
1703 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1704 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1705 				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1706 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1707 				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1708 
1709 			tmp = idx_value & ~(0x7 << 16);
1710 			tmp |= tile_flags;
1711 			ib[idx] = tmp;
1712 		} else
1713 			ib[idx] = idx_value;
1714 
1715 		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1716 		track->cb_dirty = true;
1717 		break;
1718 	case RADEON_RB3D_DEPTHPITCH:
1719 		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1720 		track->zb_dirty = true;
1721 		break;
1722 	case RADEON_RB3D_CNTL:
1723 		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1724 		case 7:
1725 		case 8:
1726 		case 9:
1727 		case 11:
1728 		case 12:
1729 			track->cb[0].cpp = 1;
1730 			break;
1731 		case 3:
1732 		case 4:
1733 		case 15:
1734 			track->cb[0].cpp = 2;
1735 			break;
1736 		case 6:
1737 			track->cb[0].cpp = 4;
1738 			break;
1739 		default:
1740 			DRM_ERROR("Invalid color buffer format (%d) !\n",
1741 				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1742 			return -EINVAL;
1743 		}
1744 		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1745 		track->cb_dirty = true;
1746 		track->zb_dirty = true;
1747 		break;
1748 	case RADEON_RB3D_ZSTENCILCNTL:
1749 		switch (idx_value & 0xf) {
1750 		case 0:
1751 			track->zb.cpp = 2;
1752 			break;
1753 		case 2:
1754 		case 3:
1755 		case 4:
1756 		case 5:
1757 		case 9:
1758 		case 11:
1759 			track->zb.cpp = 4;
1760 			break;
1761 		default:
1762 			break;
1763 		}
1764 		track->zb_dirty = true;
1765 		break;
1766 	case RADEON_RB3D_ZPASS_ADDR:
1767 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1768 		if (r) {
1769 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1770 				  idx, reg);
1771 			radeon_cs_dump_packet(p, pkt);
1772 			return r;
1773 		}
1774 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1775 		break;
1776 	case RADEON_PP_CNTL:
1777 		{
1778 			uint32_t temp = idx_value >> 4;
1779 			for (i = 0; i < track->num_texture; i++)
1780 				track->textures[i].enabled = !!(temp & (1 << i));
1781 			track->tex_dirty = true;
1782 		}
1783 		break;
1784 	case RADEON_SE_VF_CNTL:
1785 		track->vap_vf_cntl = idx_value;
1786 		break;
1787 	case RADEON_SE_VTX_FMT:
1788 		track->vtx_size = r100_get_vtx_size(idx_value);
1789 		break;
1790 	case RADEON_PP_TEX_SIZE_0:
1791 	case RADEON_PP_TEX_SIZE_1:
1792 	case RADEON_PP_TEX_SIZE_2:
1793 		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1794 		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1795 		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1796 		track->tex_dirty = true;
1797 		break;
1798 	case RADEON_PP_TEX_PITCH_0:
1799 	case RADEON_PP_TEX_PITCH_1:
1800 	case RADEON_PP_TEX_PITCH_2:
1801 		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1802 		track->textures[i].pitch = idx_value + 32;
1803 		track->tex_dirty = true;
1804 		break;
1805 	case RADEON_PP_TXFILTER_0:
1806 	case RADEON_PP_TXFILTER_1:
1807 	case RADEON_PP_TXFILTER_2:
1808 		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1809 		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1810 						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1811 		tmp = (idx_value >> 23) & 0x7;
1812 		if (tmp == 2 || tmp == 6)
1813 			track->textures[i].roundup_w = false;
1814 		tmp = (idx_value >> 27) & 0x7;
1815 		if (tmp == 2 || tmp == 6)
1816 			track->textures[i].roundup_h = false;
1817 		track->tex_dirty = true;
1818 		break;
1819 	case RADEON_PP_TXFORMAT_0:
1820 	case RADEON_PP_TXFORMAT_1:
1821 	case RADEON_PP_TXFORMAT_2:
1822 		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1823 		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1824 			track->textures[i].use_pitch = true;
1825 		} else {
1826 			track->textures[i].use_pitch = false;
1827 			track->textures[i].width = 1 << ((idx_value & RADEON_TXFORMAT_WIDTH_MASK) >> RADEON_TXFORMAT_WIDTH_SHIFT);
1828 			track->textures[i].height = 1 << ((idx_value & RADEON_TXFORMAT_HEIGHT_MASK) >> RADEON_TXFORMAT_HEIGHT_SHIFT);
1829 		}
1830 		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1831 			track->textures[i].tex_coord_type = 2;
1832 		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1833 		case RADEON_TXFORMAT_I8:
1834 		case RADEON_TXFORMAT_RGB332:
1835 		case RADEON_TXFORMAT_Y8:
1836 			track->textures[i].cpp = 1;
1837 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1838 			break;
1839 		case RADEON_TXFORMAT_AI88:
1840 		case RADEON_TXFORMAT_ARGB1555:
1841 		case RADEON_TXFORMAT_RGB565:
1842 		case RADEON_TXFORMAT_ARGB4444:
1843 		case RADEON_TXFORMAT_VYUY422:
1844 		case RADEON_TXFORMAT_YVYU422:
1845 		case RADEON_TXFORMAT_SHADOW16:
1846 		case RADEON_TXFORMAT_LDUDV655:
1847 		case RADEON_TXFORMAT_DUDV88:
1848 			track->textures[i].cpp = 2;
1849 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1850 			break;
1851 		case RADEON_TXFORMAT_ARGB8888:
1852 		case RADEON_TXFORMAT_RGBA8888:
1853 		case RADEON_TXFORMAT_SHADOW32:
1854 		case RADEON_TXFORMAT_LDUDUV8888:
1855 			track->textures[i].cpp = 4;
1856 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1857 			break;
1858 		case RADEON_TXFORMAT_DXT1:
1859 			track->textures[i].cpp = 1;
1860 			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1861 			break;
1862 		case RADEON_TXFORMAT_DXT23:
1863 		case RADEON_TXFORMAT_DXT45:
1864 			track->textures[i].cpp = 1;
1865 			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1866 			break;
1867 		}
1868 		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1869 		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1870 		track->tex_dirty = true;
1871 		break;
1872 	case RADEON_PP_CUBIC_FACES_0:
1873 	case RADEON_PP_CUBIC_FACES_1:
1874 	case RADEON_PP_CUBIC_FACES_2:
1875 		tmp = idx_value;
1876 		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1877 		for (face = 0; face < 4; face++) {
1878 			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1879 			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1880 		}
1881 		track->tex_dirty = true;
1882 		break;
1883 	default:
1884 		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
1885 		return -EINVAL;
1886 	}
1887 	return 0;
1888 }
1889 
1890 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1891 					 struct radeon_cs_packet *pkt,
1892 					 struct radeon_bo *robj)
1893 {
1894 	unsigned idx;
1895 	u32 value;
1896 	idx = pkt->idx + 1;
1897 	value = radeon_get_ib_value(p, idx + 2);
1898 	if ((value + 1) > radeon_bo_size(robj)) {
1899 		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1900 			  "(need %u have %lu) !\n",
1901 			  value + 1,
1902 			  radeon_bo_size(robj));
1903 		return -EINVAL;
1904 	}
1905 	return 0;
1906 }
1907 
1908 static int r100_packet3_check(struct radeon_cs_parser *p,
1909 			      struct radeon_cs_packet *pkt)
1910 {
1911 	struct radeon_bo_list *reloc;
1912 	struct r100_cs_track *track;
1913 	unsigned idx;
1914 	volatile uint32_t *ib;
1915 	int r;
1916 
1917 	ib = p->ib.ptr;
1918 	idx = pkt->idx + 1;
1919 	track = (struct r100_cs_track *)p->track;
1920 	switch (pkt->opcode) {
1921 	case PACKET3_3D_LOAD_VBPNTR:
1922 		r = r100_packet3_load_vbpntr(p, pkt, idx);
1923 		if (r)
1924 			return r;
1925 		break;
1926 	case PACKET3_INDX_BUFFER:
1927 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1928 		if (r) {
1929 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1930 			radeon_cs_dump_packet(p, pkt);
1931 			return r;
1932 		}
1933 		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1934 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1935 		if (r) {
1936 			return r;
1937 		}
1938 		break;
1939 	case 0x23:
1940 		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1941 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1942 		if (r) {
1943 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1944 			radeon_cs_dump_packet(p, pkt);
1945 			return r;
1946 		}
1947 		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1948 		track->num_arrays = 1;
1949 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1950 
1951 		track->arrays[0].robj = reloc->robj;
1952 		track->arrays[0].esize = track->vtx_size;
1953 
1954 		track->max_indx = radeon_get_ib_value(p, idx+1);
1955 
1956 		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1957 		track->immd_dwords = pkt->count - 1;
1958 		r = r100_cs_track_check(p->rdev, track);
1959 		if (r)
1960 			return r;
1961 		break;
1962 	case PACKET3_3D_DRAW_IMMD:
1963 		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1964 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1965 			return -EINVAL;
1966 		}
1967 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1968 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1969 		track->immd_dwords = pkt->count - 1;
1970 		r = r100_cs_track_check(p->rdev, track);
1971 		if (r)
1972 			return r;
1973 		break;
1974 		/* triggers drawing using in-packet vertex data */
1975 	case PACKET3_3D_DRAW_IMMD_2:
1976 		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1977 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1978 			return -EINVAL;
1979 		}
1980 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1981 		track->immd_dwords = pkt->count;
1982 		r = r100_cs_track_check(p->rdev, track);
1983 		if (r)
1984 			return r;
1985 		break;
1986 		/* triggers drawing using in-packet vertex data */
1987 	case PACKET3_3D_DRAW_VBUF_2:
1988 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1989 		r = r100_cs_track_check(p->rdev, track);
1990 		if (r)
1991 			return r;
1992 		break;
1993 		/* triggers drawing of vertex buffers setup elsewhere */
1994 	case PACKET3_3D_DRAW_INDX_2:
1995 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1996 		r = r100_cs_track_check(p->rdev, track);
1997 		if (r)
1998 			return r;
1999 		break;
2000 		/* triggers drawing using indices to vertex buffer */
2001 	case PACKET3_3D_DRAW_VBUF:
2002 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2003 		r = r100_cs_track_check(p->rdev, track);
2004 		if (r)
2005 			return r;
2006 		break;
2007 		/* triggers drawing of vertex buffers setup elsewhere */
2008 	case PACKET3_3D_DRAW_INDX:
2009 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2010 		r = r100_cs_track_check(p->rdev, track);
2011 		if (r)
2012 			return r;
2013 		break;
2014 		/* triggers drawing using indices to vertex buffer */
2015 	case PACKET3_3D_CLEAR_HIZ:
2016 	case PACKET3_3D_CLEAR_ZMASK:
2017 		if (p->rdev->hyperz_filp != p->filp)
2018 			return -EINVAL;
2019 		break;
2020 	case PACKET3_NOP:
2021 		break;
2022 	default:
2023 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2024 		return -EINVAL;
2025 	}
2026 	return 0;
2027 }
2028 
2029 int r100_cs_parse(struct radeon_cs_parser *p)
2030 {
2031 	struct radeon_cs_packet pkt;
2032 	struct r100_cs_track *track;
2033 	int r;
2034 
2035 	track = kzalloc(sizeof(*track), GFP_KERNEL);
2036 	if (!track)
2037 		return -ENOMEM;
2038 	r100_cs_track_clear(p->rdev, track);
2039 	p->track = track;
2040 	do {
2041 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2042 		if (r) {
2043 			return r;
2044 		}
2045 		p->idx += pkt.count + 2;
2046 		switch (pkt.type) {
2047 		case RADEON_PACKET_TYPE0:
2048 			if (p->rdev->family >= CHIP_R200)
2049 				r = r100_cs_parse_packet0(p, &pkt,
2050 					p->rdev->config.r100.reg_safe_bm,
2051 					p->rdev->config.r100.reg_safe_bm_size,
2052 					&r200_packet0_check);
2053 			else
2054 				r = r100_cs_parse_packet0(p, &pkt,
2055 					p->rdev->config.r100.reg_safe_bm,
2056 					p->rdev->config.r100.reg_safe_bm_size,
2057 					&r100_packet0_check);
2058 			break;
2059 		case RADEON_PACKET_TYPE2:
2060 			break;
2061 		case RADEON_PACKET_TYPE3:
2062 			r = r100_packet3_check(p, &pkt);
2063 			break;
2064 		default:
2065 			DRM_ERROR("Unknown packet type %d !\n",
2066 				  pkt.type);
2067 			return -EINVAL;
2068 		}
2069 		if (r)
2070 			return r;
2071 	} while (p->idx < p->chunk_ib->length_dw);
2072 	return 0;
2073 }
2074 
2075 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2076 {
2077 	DRM_ERROR("pitch                      %d\n", t->pitch);
2078 	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2079 	DRM_ERROR("width                      %d\n", t->width);
2080 	DRM_ERROR("width_11                   %d\n", t->width_11);
2081 	DRM_ERROR("height                     %d\n", t->height);
2082 	DRM_ERROR("height_11                  %d\n", t->height_11);
2083 	DRM_ERROR("num levels                 %d\n", t->num_levels);
2084 	DRM_ERROR("depth                      %d\n", t->txdepth);
2085 	DRM_ERROR("bpp                        %d\n", t->cpp);
2086 	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2087 	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2088 	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2089 	DRM_ERROR("compress format            %d\n", t->compress_format);
2090 }
2091 
2092 static int r100_track_compress_size(int compress_format, int w, int h)
2093 {
2094 	int block_width, block_height, block_bytes;
2095 	int wblocks, hblocks;
2096 	int min_wblocks;
2097 	int sz;
2098 
2099 	block_width = 4;
2100 	block_height = 4;
2101 
2102 	switch (compress_format) {
2103 	case R100_TRACK_COMP_DXT1:
2104 		block_bytes = 8;
2105 		min_wblocks = 4;
2106 		break;
2107 	default:
2108 	case R100_TRACK_COMP_DXT35:
2109 		block_bytes = 16;
2110 		min_wblocks = 2;
2111 		break;
2112 	}
2113 
2114 	hblocks = (h + block_height - 1) / block_height;
2115 	wblocks = (w + block_width - 1) / block_width;
2116 	if (wblocks < min_wblocks)
2117 		wblocks = min_wblocks;
2118 	sz = wblocks * hblocks * block_bytes;
2119 	return sz;
2120 }
2121 
2122 static int r100_cs_track_cube(struct radeon_device *rdev,
2123 			      struct r100_cs_track *track, unsigned idx)
2124 {
2125 	unsigned face, w, h;
2126 	struct radeon_bo *cube_robj;
2127 	unsigned long size;
2128 	unsigned compress_format = track->textures[idx].compress_format;
2129 
2130 	for (face = 0; face < 5; face++) {
2131 		cube_robj = track->textures[idx].cube_info[face].robj;
2132 		w = track->textures[idx].cube_info[face].width;
2133 		h = track->textures[idx].cube_info[face].height;
2134 
2135 		if (compress_format) {
2136 			size = r100_track_compress_size(compress_format, w, h);
2137 		} else
2138 			size = w * h;
2139 		size *= track->textures[idx].cpp;
2140 
2141 		size += track->textures[idx].cube_info[face].offset;
2142 
2143 		if (size > radeon_bo_size(cube_robj)) {
2144 			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2145 				  size, radeon_bo_size(cube_robj));
2146 			r100_cs_track_texture_print(&track->textures[idx]);
2147 			return -1;
2148 		}
2149 	}
2150 	return 0;
2151 }
2152 
2153 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2154 				       struct r100_cs_track *track)
2155 {
2156 	struct radeon_bo *robj;
2157 	unsigned long size;
2158 	unsigned u, i, w, h, d;
2159 	int ret;
2160 
2161 	for (u = 0; u < track->num_texture; u++) {
2162 		if (!track->textures[u].enabled)
2163 			continue;
2164 		if (track->textures[u].lookup_disable)
2165 			continue;
2166 		robj = track->textures[u].robj;
2167 		if (robj == NULL) {
2168 			DRM_ERROR("No texture bound to unit %u\n", u);
2169 			return -EINVAL;
2170 		}
2171 		size = 0;
2172 		for (i = 0; i <= track->textures[u].num_levels; i++) {
2173 			if (track->textures[u].use_pitch) {
2174 				if (rdev->family < CHIP_R300)
2175 					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2176 				else
2177 					w = track->textures[u].pitch / (1 << i);
2178 			} else {
2179 				w = track->textures[u].width;
2180 				if (rdev->family >= CHIP_RV515)
2181 					w |= track->textures[u].width_11;
2182 				w = w / (1 << i);
2183 				if (track->textures[u].roundup_w)
2184 					w = roundup_pow_of_two(w);
2185 			}
2186 			h = track->textures[u].height;
2187 			if (rdev->family >= CHIP_RV515)
2188 				h |= track->textures[u].height_11;
2189 			h = h / (1 << i);
2190 			if (track->textures[u].roundup_h)
2191 				h = roundup_pow_of_two(h);
2192 			if (track->textures[u].tex_coord_type == 1) {
2193 				d = (1 << track->textures[u].txdepth) / (1 << i);
2194 				if (!d)
2195 					d = 1;
2196 			} else {
2197 				d = 1;
2198 			}
2199 			if (track->textures[u].compress_format) {
2200 
2201 				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2202 				/* compressed textures are block based */
2203 			} else
2204 				size += w * h * d;
2205 		}
2206 		size *= track->textures[u].cpp;
2207 
2208 		switch (track->textures[u].tex_coord_type) {
2209 		case 0:
2210 		case 1:
2211 			break;
2212 		case 2:
2213 			if (track->separate_cube) {
2214 				ret = r100_cs_track_cube(rdev, track, u);
2215 				if (ret)
2216 					return ret;
2217 			} else
2218 				size *= 6;
2219 			break;
2220 		default:
2221 			DRM_ERROR("Invalid texture coordinate type %u for unit "
2222 				  "%u\n", track->textures[u].tex_coord_type, u);
2223 			return -EINVAL;
2224 		}
2225 		if (size > radeon_bo_size(robj)) {
2226 			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2227 				  "%lu\n", u, size, radeon_bo_size(robj));
2228 			r100_cs_track_texture_print(&track->textures[u]);
2229 			return -EINVAL;
2230 		}
2231 	}
2232 	return 0;
2233 }
2234 
2235 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2236 {
2237 	unsigned i;
2238 	unsigned long size;
2239 	unsigned prim_walk;
2240 	unsigned nverts;
2241 	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2242 
2243 	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2244 	    !track->blend_read_enable)
2245 		num_cb = 0;
2246 
2247 	for (i = 0; i < num_cb; i++) {
2248 		if (track->cb[i].robj == NULL) {
2249 			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2250 			return -EINVAL;
2251 		}
2252 		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2253 		size += track->cb[i].offset;
2254 		if (size > radeon_bo_size(track->cb[i].robj)) {
2255 			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2256 				  "(need %lu have %lu) !\n", i, size,
2257 				  radeon_bo_size(track->cb[i].robj));
2258 			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2259 				  i, track->cb[i].pitch, track->cb[i].cpp,
2260 				  track->cb[i].offset, track->maxy);
2261 			return -EINVAL;
2262 		}
2263 	}
2264 	track->cb_dirty = false;
2265 
2266 	if (track->zb_dirty && track->z_enabled) {
2267 		if (track->zb.robj == NULL) {
2268 			DRM_ERROR("[drm] No buffer for z buffer !\n");
2269 			return -EINVAL;
2270 		}
2271 		size = track->zb.pitch * track->zb.cpp * track->maxy;
2272 		size += track->zb.offset;
2273 		if (size > radeon_bo_size(track->zb.robj)) {
2274 			DRM_ERROR("[drm] Buffer too small for z buffer "
2275 				  "(need %lu have %lu) !\n", size,
2276 				  radeon_bo_size(track->zb.robj));
2277 			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2278 				  track->zb.pitch, track->zb.cpp,
2279 				  track->zb.offset, track->maxy);
2280 			return -EINVAL;
2281 		}
2282 	}
2283 	track->zb_dirty = false;
2284 
2285 	if (track->aa_dirty && track->aaresolve) {
2286 		if (track->aa.robj == NULL) {
2287 			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2288 			return -EINVAL;
2289 		}
2290 		/* I believe the format comes from colorbuffer0. */
2291 		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2292 		size += track->aa.offset;
2293 		if (size > radeon_bo_size(track->aa.robj)) {
2294 			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2295 				  "(need %lu have %lu) !\n", i, size,
2296 				  radeon_bo_size(track->aa.robj));
2297 			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2298 				  i, track->aa.pitch, track->cb[0].cpp,
2299 				  track->aa.offset, track->maxy);
2300 			return -EINVAL;
2301 		}
2302 	}
2303 	track->aa_dirty = false;
2304 
2305 	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2306 	if (track->vap_vf_cntl & (1 << 14)) {
2307 		nverts = track->vap_alt_nverts;
2308 	} else {
2309 		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2310 	}
2311 	switch (prim_walk) {
2312 	case 1:
2313 		for (i = 0; i < track->num_arrays; i++) {
2314 			size = track->arrays[i].esize * track->max_indx * 4;
2315 			if (track->arrays[i].robj == NULL) {
2316 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2317 					  "bound\n", prim_walk, i);
2318 				return -EINVAL;
2319 			}
2320 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2321 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2322 					"need %lu dwords have %lu dwords\n",
2323 					prim_walk, i, size >> 2,
2324 					radeon_bo_size(track->arrays[i].robj)
2325 					>> 2);
2326 				DRM_ERROR("Max indices %u\n", track->max_indx);
2327 				return -EINVAL;
2328 			}
2329 		}
2330 		break;
2331 	case 2:
2332 		for (i = 0; i < track->num_arrays; i++) {
2333 			size = track->arrays[i].esize * (nverts - 1) * 4;
2334 			if (track->arrays[i].robj == NULL) {
2335 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2336 					  "bound\n", prim_walk, i);
2337 				return -EINVAL;
2338 			}
2339 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2340 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2341 					"need %lu dwords have %lu dwords\n",
2342 					prim_walk, i, size >> 2,
2343 					radeon_bo_size(track->arrays[i].robj)
2344 					>> 2);
2345 				return -EINVAL;
2346 			}
2347 		}
2348 		break;
2349 	case 3:
2350 		size = track->vtx_size * nverts;
2351 		if (size != track->immd_dwords) {
2352 			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2353 				  track->immd_dwords, size);
2354 			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2355 				  nverts, track->vtx_size);
2356 			return -EINVAL;
2357 		}
2358 		break;
2359 	default:
2360 		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2361 			  prim_walk);
2362 		return -EINVAL;
2363 	}
2364 
2365 	if (track->tex_dirty) {
2366 		track->tex_dirty = false;
2367 		return r100_cs_track_texture_check(rdev, track);
2368 	}
2369 	return 0;
2370 }
2371 
2372 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2373 {
2374 	unsigned i, face;
2375 
2376 	track->cb_dirty = true;
2377 	track->zb_dirty = true;
2378 	track->tex_dirty = true;
2379 	track->aa_dirty = true;
2380 
2381 	if (rdev->family < CHIP_R300) {
2382 		track->num_cb = 1;
2383 		if (rdev->family <= CHIP_RS200)
2384 			track->num_texture = 3;
2385 		else
2386 			track->num_texture = 6;
2387 		track->maxy = 2048;
2388 		track->separate_cube = true;
2389 	} else {
2390 		track->num_cb = 4;
2391 		track->num_texture = 16;
2392 		track->maxy = 4096;
2393 		track->separate_cube = false;
2394 		track->aaresolve = false;
2395 		track->aa.robj = NULL;
2396 	}
2397 
2398 	for (i = 0; i < track->num_cb; i++) {
2399 		track->cb[i].robj = NULL;
2400 		track->cb[i].pitch = 8192;
2401 		track->cb[i].cpp = 16;
2402 		track->cb[i].offset = 0;
2403 	}
2404 	track->z_enabled = true;
2405 	track->zb.robj = NULL;
2406 	track->zb.pitch = 8192;
2407 	track->zb.cpp = 4;
2408 	track->zb.offset = 0;
2409 	track->vtx_size = 0x7F;
2410 	track->immd_dwords = 0xFFFFFFFFUL;
2411 	track->num_arrays = 11;
2412 	track->max_indx = 0x00FFFFFFUL;
2413 	for (i = 0; i < track->num_arrays; i++) {
2414 		track->arrays[i].robj = NULL;
2415 		track->arrays[i].esize = 0x7F;
2416 	}
2417 	for (i = 0; i < track->num_texture; i++) {
2418 		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2419 		track->textures[i].pitch = 16536;
2420 		track->textures[i].width = 16536;
2421 		track->textures[i].height = 16536;
2422 		track->textures[i].width_11 = 1 << 11;
2423 		track->textures[i].height_11 = 1 << 11;
2424 		track->textures[i].num_levels = 12;
2425 		if (rdev->family <= CHIP_RS200) {
2426 			track->textures[i].tex_coord_type = 0;
2427 			track->textures[i].txdepth = 0;
2428 		} else {
2429 			track->textures[i].txdepth = 16;
2430 			track->textures[i].tex_coord_type = 1;
2431 		}
2432 		track->textures[i].cpp = 64;
2433 		track->textures[i].robj = NULL;
2434 		/* CS IB emission code makes sure texture unit are disabled */
2435 		track->textures[i].enabled = false;
2436 		track->textures[i].lookup_disable = false;
2437 		track->textures[i].roundup_w = true;
2438 		track->textures[i].roundup_h = true;
2439 		if (track->separate_cube)
2440 			for (face = 0; face < 5; face++) {
2441 				track->textures[i].cube_info[face].robj = NULL;
2442 				track->textures[i].cube_info[face].width = 16536;
2443 				track->textures[i].cube_info[face].height = 16536;
2444 				track->textures[i].cube_info[face].offset = 0;
2445 			}
2446 	}
2447 }
2448 
2449 /*
2450  * Global GPU functions
2451  */
2452 static void r100_errata(struct radeon_device *rdev)
2453 {
2454 	rdev->pll_errata = 0;
2455 
2456 	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2457 		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2458 	}
2459 
2460 	if (rdev->family == CHIP_RV100 ||
2461 	    rdev->family == CHIP_RS100 ||
2462 	    rdev->family == CHIP_RS200) {
2463 		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2464 	}
2465 }
2466 
2467 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2468 {
2469 	unsigned i;
2470 	uint32_t tmp;
2471 
2472 	for (i = 0; i < rdev->usec_timeout; i++) {
2473 		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2474 		if (tmp >= n) {
2475 			return 0;
2476 		}
2477 		udelay(1);
2478 	}
2479 	return -1;
2480 }
2481 
2482 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2483 {
2484 	unsigned i;
2485 	uint32_t tmp;
2486 
2487 	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2488 		pr_warn("radeon: wait for empty RBBM fifo failed! Bad things might happen.\n");
2489 	}
2490 	for (i = 0; i < rdev->usec_timeout; i++) {
2491 		tmp = RREG32(RADEON_RBBM_STATUS);
2492 		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2493 			return 0;
2494 		}
2495 		udelay(1);
2496 	}
2497 	return -1;
2498 }
2499 
2500 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2501 {
2502 	unsigned i;
2503 	uint32_t tmp;
2504 
2505 	for (i = 0; i < rdev->usec_timeout; i++) {
2506 		/* read MC_STATUS */
2507 		tmp = RREG32(RADEON_MC_STATUS);
2508 		if (tmp & RADEON_MC_IDLE) {
2509 			return 0;
2510 		}
2511 		udelay(1);
2512 	}
2513 	return -1;
2514 }
2515 
2516 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2517 {
2518 	u32 rbbm_status;
2519 
2520 	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2521 	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2522 		radeon_ring_lockup_update(rdev, ring);
2523 		return false;
2524 	}
2525 	return radeon_ring_test_lockup(rdev, ring);
2526 }
2527 
2528 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2529 void r100_enable_bm(struct radeon_device *rdev)
2530 {
2531 	uint32_t tmp;
2532 	/* Enable bus mastering */
2533 	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2534 	WREG32(RADEON_BUS_CNTL, tmp);
2535 }
2536 
2537 void r100_bm_disable(struct radeon_device *rdev)
2538 {
2539 	u32 tmp;
2540 
2541 	/* disable bus mastering */
2542 	tmp = RREG32(R_000030_BUS_CNTL);
2543 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2544 	mdelay(1);
2545 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2546 	mdelay(1);
2547 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2548 	tmp = RREG32(RADEON_BUS_CNTL);
2549 	mdelay(1);
2550 	pci_clear_master(rdev->pdev);
2551 	mdelay(1);
2552 }
2553 
2554 int r100_asic_reset(struct radeon_device *rdev, bool hard)
2555 {
2556 	struct r100_mc_save save;
2557 	u32 status, tmp;
2558 	int ret = 0;
2559 
2560 	status = RREG32(R_000E40_RBBM_STATUS);
2561 	if (!G_000E40_GUI_ACTIVE(status)) {
2562 		return 0;
2563 	}
2564 	r100_mc_stop(rdev, &save);
2565 	status = RREG32(R_000E40_RBBM_STATUS);
2566 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2567 	/* stop CP */
2568 	WREG32(RADEON_CP_CSQ_CNTL, 0);
2569 	tmp = RREG32(RADEON_CP_RB_CNTL);
2570 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2571 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2572 	WREG32(RADEON_CP_RB_WPTR, 0);
2573 	WREG32(RADEON_CP_RB_CNTL, tmp);
2574 	/* save PCI state */
2575 	pci_save_state(rdev->pdev);
2576 	/* disable bus mastering */
2577 	r100_bm_disable(rdev);
2578 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2579 					S_0000F0_SOFT_RESET_RE(1) |
2580 					S_0000F0_SOFT_RESET_PP(1) |
2581 					S_0000F0_SOFT_RESET_RB(1));
2582 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2583 	mdelay(500);
2584 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2585 	mdelay(1);
2586 	status = RREG32(R_000E40_RBBM_STATUS);
2587 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2588 	/* reset CP */
2589 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2590 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2591 	mdelay(500);
2592 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2593 	mdelay(1);
2594 	status = RREG32(R_000E40_RBBM_STATUS);
2595 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2596 	/* restore PCI & busmastering */
2597 	pci_restore_state(rdev->pdev);
2598 	r100_enable_bm(rdev);
2599 	/* Check if GPU is idle */
2600 	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2601 		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2602 		dev_err(rdev->dev, "failed to reset GPU\n");
2603 		ret = -1;
2604 	} else
2605 		dev_info(rdev->dev, "GPU reset succeed\n");
2606 	r100_mc_resume(rdev, &save);
2607 	return ret;
2608 }
2609 
2610 void r100_set_common_regs(struct radeon_device *rdev)
2611 {
2612 	bool force_dac2 = false;
2613 	u32 tmp;
2614 
2615 	/* set these so they don't interfere with anything */
2616 	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2617 	WREG32(RADEON_SUBPIC_CNTL, 0);
2618 	WREG32(RADEON_VIPH_CONTROL, 0);
2619 	WREG32(RADEON_I2C_CNTL_1, 0);
2620 	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2621 	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2622 	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2623 
2624 	/* always set up dac2 on rn50 and some rv100 as lots
2625 	 * of servers seem to wire it up to a VGA port but
2626 	 * don't report it in the bios connector
2627 	 * table.
2628 	 */
2629 	switch (rdev->pdev->device) {
2630 		/* RN50 */
2631 	case 0x515e:
2632 	case 0x5969:
2633 		force_dac2 = true;
2634 		break;
2635 		/* RV100*/
2636 	case 0x5159:
2637 	case 0x515a:
2638 		/* DELL triple head servers */
2639 		if ((rdev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2640 		    ((rdev->pdev->subsystem_device == 0x016c) ||
2641 		     (rdev->pdev->subsystem_device == 0x016d) ||
2642 		     (rdev->pdev->subsystem_device == 0x016e) ||
2643 		     (rdev->pdev->subsystem_device == 0x016f) ||
2644 		     (rdev->pdev->subsystem_device == 0x0170) ||
2645 		     (rdev->pdev->subsystem_device == 0x017d) ||
2646 		     (rdev->pdev->subsystem_device == 0x017e) ||
2647 		     (rdev->pdev->subsystem_device == 0x0183) ||
2648 		     (rdev->pdev->subsystem_device == 0x018a) ||
2649 		     (rdev->pdev->subsystem_device == 0x019a)))
2650 			force_dac2 = true;
2651 		break;
2652 	}
2653 
2654 	if (force_dac2) {
2655 		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2656 		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2657 		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2658 
2659 		/* For CRT on DAC2, don't turn it on if BIOS didn't
2660 		   enable it, even it's detected.
2661 		*/
2662 
2663 		/* force it to crtc0 */
2664 		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2665 		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2666 		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2667 
2668 		/* set up the TV DAC */
2669 		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2670 				 RADEON_TV_DAC_STD_MASK |
2671 				 RADEON_TV_DAC_RDACPD |
2672 				 RADEON_TV_DAC_GDACPD |
2673 				 RADEON_TV_DAC_BDACPD |
2674 				 RADEON_TV_DAC_BGADJ_MASK |
2675 				 RADEON_TV_DAC_DACADJ_MASK);
2676 		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2677 				RADEON_TV_DAC_NHOLD |
2678 				RADEON_TV_DAC_STD_PS2 |
2679 				(0x58 << 16));
2680 
2681 		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2682 		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2683 		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2684 	}
2685 
2686 	/* switch PM block to ACPI mode */
2687 	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2688 	tmp &= ~RADEON_PM_MODE_SEL;
2689 	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2690 
2691 }
2692 
2693 /*
2694  * VRAM info
2695  */
2696 static void r100_vram_get_type(struct radeon_device *rdev)
2697 {
2698 	uint32_t tmp;
2699 
2700 	rdev->mc.vram_is_ddr = false;
2701 	if (rdev->flags & RADEON_IS_IGP)
2702 		rdev->mc.vram_is_ddr = true;
2703 	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2704 		rdev->mc.vram_is_ddr = true;
2705 	if ((rdev->family == CHIP_RV100) ||
2706 	    (rdev->family == CHIP_RS100) ||
2707 	    (rdev->family == CHIP_RS200)) {
2708 		tmp = RREG32(RADEON_MEM_CNTL);
2709 		if (tmp & RV100_HALF_MODE) {
2710 			rdev->mc.vram_width = 32;
2711 		} else {
2712 			rdev->mc.vram_width = 64;
2713 		}
2714 		if (rdev->flags & RADEON_SINGLE_CRTC) {
2715 			rdev->mc.vram_width /= 4;
2716 			rdev->mc.vram_is_ddr = true;
2717 		}
2718 	} else if (rdev->family <= CHIP_RV280) {
2719 		tmp = RREG32(RADEON_MEM_CNTL);
2720 		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2721 			rdev->mc.vram_width = 128;
2722 		} else {
2723 			rdev->mc.vram_width = 64;
2724 		}
2725 	} else {
2726 		/* newer IGPs */
2727 		rdev->mc.vram_width = 128;
2728 	}
2729 }
2730 
2731 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2732 {
2733 	u32 aper_size;
2734 	u8 byte;
2735 
2736 	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2737 
2738 	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2739 	 * that is has the 2nd generation multifunction PCI interface
2740 	 */
2741 	if (rdev->family == CHIP_RV280 ||
2742 	    rdev->family >= CHIP_RV350) {
2743 		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2744 		       ~RADEON_HDP_APER_CNTL);
2745 		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2746 		return aper_size * 2;
2747 	}
2748 
2749 	/* Older cards have all sorts of funny issues to deal with. First
2750 	 * check if it's a multifunction card by reading the PCI config
2751 	 * header type... Limit those to one aperture size
2752 	 */
2753 	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2754 	if (byte & 0x80) {
2755 		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2756 		DRM_INFO("Limiting VRAM to one aperture\n");
2757 		return aper_size;
2758 	}
2759 
2760 	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2761 	 * have set it up. We don't write this as it's broken on some ASICs but
2762 	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2763 	 */
2764 	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2765 		return aper_size * 2;
2766 	return aper_size;
2767 }
2768 
2769 void r100_vram_init_sizes(struct radeon_device *rdev)
2770 {
2771 	u64 config_aper_size;
2772 
2773 	/* work out accessible VRAM */
2774 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2775 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2776 	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2777 	/* FIXME we don't use the second aperture yet when we could use it */
2778 	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2779 		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2780 	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2781 	if (rdev->flags & RADEON_IS_IGP) {
2782 		uint32_t tom;
2783 		/* read NB_TOM to get the amount of ram stolen for the GPU */
2784 		tom = RREG32(RADEON_NB_TOM);
2785 		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2786 		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2787 		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2788 	} else {
2789 		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2790 		/* Some production boards of m6 will report 0
2791 		 * if it's 8 MB
2792 		 */
2793 		if (rdev->mc.real_vram_size == 0) {
2794 			rdev->mc.real_vram_size = 8192 * 1024;
2795 			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2796 		}
2797 		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2798 		 * Novell bug 204882 + along with lots of ubuntu ones
2799 		 */
2800 		if (rdev->mc.aper_size > config_aper_size)
2801 			config_aper_size = rdev->mc.aper_size;
2802 
2803 		if (config_aper_size > rdev->mc.real_vram_size)
2804 			rdev->mc.mc_vram_size = config_aper_size;
2805 		else
2806 			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2807 	}
2808 }
2809 
2810 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2811 {
2812 	uint32_t temp;
2813 
2814 	temp = RREG32(RADEON_CONFIG_CNTL);
2815 	if (!state) {
2816 		temp &= ~RADEON_CFG_VGA_RAM_EN;
2817 		temp |= RADEON_CFG_VGA_IO_DIS;
2818 	} else {
2819 		temp &= ~RADEON_CFG_VGA_IO_DIS;
2820 	}
2821 	WREG32(RADEON_CONFIG_CNTL, temp);
2822 }
2823 
2824 static void r100_mc_init(struct radeon_device *rdev)
2825 {
2826 	u64 base;
2827 
2828 	r100_vram_get_type(rdev);
2829 	r100_vram_init_sizes(rdev);
2830 	base = rdev->mc.aper_base;
2831 	if (rdev->flags & RADEON_IS_IGP)
2832 		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2833 	radeon_vram_location(rdev, &rdev->mc, base);
2834 	rdev->mc.gtt_base_align = 0;
2835 	if (!(rdev->flags & RADEON_IS_AGP))
2836 		radeon_gtt_location(rdev, &rdev->mc);
2837 	radeon_update_bandwidth_info(rdev);
2838 }
2839 
2840 
2841 /*
2842  * Indirect registers accessor
2843  */
2844 void r100_pll_errata_after_index(struct radeon_device *rdev)
2845 {
2846 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2847 		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2848 		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2849 	}
2850 }
2851 
2852 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2853 {
2854 	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2855 	 * or the chip could hang on a subsequent access
2856 	 */
2857 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2858 		mdelay(5);
2859 	}
2860 
2861 	/* This function is required to workaround a hardware bug in some (all?)
2862 	 * revisions of the R300.  This workaround should be called after every
2863 	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2864 	 * may not be correct.
2865 	 */
2866 	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2867 		uint32_t save, tmp;
2868 
2869 		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2870 		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2871 		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2872 		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2873 		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2874 	}
2875 }
2876 
2877 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2878 {
2879 	unsigned long flags;
2880 	uint32_t data;
2881 
2882 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2883 	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2884 	r100_pll_errata_after_index(rdev);
2885 	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2886 	r100_pll_errata_after_data(rdev);
2887 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2888 	return data;
2889 }
2890 
2891 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2892 {
2893 	unsigned long flags;
2894 
2895 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2896 	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2897 	r100_pll_errata_after_index(rdev);
2898 	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2899 	r100_pll_errata_after_data(rdev);
2900 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2901 }
2902 
2903 static void r100_set_safe_registers(struct radeon_device *rdev)
2904 {
2905 	if (ASIC_IS_RN50(rdev)) {
2906 		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2907 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2908 	} else if (rdev->family < CHIP_R200) {
2909 		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2910 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2911 	} else {
2912 		r200_set_safe_registers(rdev);
2913 	}
2914 }
2915 
2916 /*
2917  * Debugfs info
2918  */
2919 #if defined(CONFIG_DEBUG_FS)
2920 static int r100_debugfs_rbbm_info_show(struct seq_file *m, void *unused)
2921 {
2922 	struct radeon_device *rdev = (struct radeon_device *)m->private;
2923 	uint32_t reg, value;
2924 	unsigned i;
2925 
2926 	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2927 	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2928 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2929 	for (i = 0; i < 64; i++) {
2930 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2931 		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2932 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2933 		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2934 		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2935 	}
2936 	return 0;
2937 }
2938 
2939 static int r100_debugfs_cp_ring_info_show(struct seq_file *m, void *unused)
2940 {
2941 	struct radeon_device *rdev = (struct radeon_device *)m->private;
2942 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2943 	uint32_t rdp, wdp;
2944 	unsigned count, i, j;
2945 
2946 	radeon_ring_free_size(rdev, ring);
2947 	rdp = RREG32(RADEON_CP_RB_RPTR);
2948 	wdp = RREG32(RADEON_CP_RB_WPTR);
2949 	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2950 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2951 	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2952 	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2953 	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2954 	seq_printf(m, "%u dwords in ring\n", count);
2955 	if (ring->ready) {
2956 		for (j = 0; j <= count; j++) {
2957 			i = (rdp + j) & ring->ptr_mask;
2958 			seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2959 		}
2960 	}
2961 	return 0;
2962 }
2963 
2964 
2965 static int r100_debugfs_cp_csq_fifo_show(struct seq_file *m, void *unused)
2966 {
2967 	struct radeon_device *rdev = (struct radeon_device *)m->private;
2968 	uint32_t csq_stat, csq2_stat, tmp;
2969 	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2970 	unsigned i;
2971 
2972 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2973 	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2974 	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2975 	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2976 	r_rptr = (csq_stat >> 0) & 0x3ff;
2977 	r_wptr = (csq_stat >> 10) & 0x3ff;
2978 	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2979 	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2980 	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2981 	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2982 	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2983 	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2984 	seq_printf(m, "Ring rptr %u\n", r_rptr);
2985 	seq_printf(m, "Ring wptr %u\n", r_wptr);
2986 	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2987 	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2988 	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2989 	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2990 	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
2991 	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
2992 	seq_printf(m, "Ring fifo:\n");
2993 	for (i = 0; i < 256; i++) {
2994 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2995 		tmp = RREG32(RADEON_CP_CSQ_DATA);
2996 		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
2997 	}
2998 	seq_printf(m, "Indirect1 fifo:\n");
2999 	for (i = 256; i <= 512; i++) {
3000 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3001 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3002 		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3003 	}
3004 	seq_printf(m, "Indirect2 fifo:\n");
3005 	for (i = 640; i < ib1_wptr; i++) {
3006 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3007 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3008 		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3009 	}
3010 	return 0;
3011 }
3012 
3013 static int r100_debugfs_mc_info_show(struct seq_file *m, void *unused)
3014 {
3015 	struct radeon_device *rdev = (struct radeon_device *)m->private;
3016 	uint32_t tmp;
3017 
3018 	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3019 	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3020 	tmp = RREG32(RADEON_MC_FB_LOCATION);
3021 	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3022 	tmp = RREG32(RADEON_BUS_CNTL);
3023 	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3024 	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3025 	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3026 	tmp = RREG32(RADEON_AGP_BASE);
3027 	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3028 	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3029 	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3030 	tmp = RREG32(0x01D0);
3031 	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3032 	tmp = RREG32(RADEON_AIC_LO_ADDR);
3033 	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3034 	tmp = RREG32(RADEON_AIC_HI_ADDR);
3035 	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3036 	tmp = RREG32(0x01E4);
3037 	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3038 	return 0;
3039 }
3040 
3041 DEFINE_SHOW_ATTRIBUTE(r100_debugfs_rbbm_info);
3042 DEFINE_SHOW_ATTRIBUTE(r100_debugfs_cp_ring_info);
3043 DEFINE_SHOW_ATTRIBUTE(r100_debugfs_cp_csq_fifo);
3044 DEFINE_SHOW_ATTRIBUTE(r100_debugfs_mc_info);
3045 
3046 #endif
3047 
3048 void  r100_debugfs_rbbm_init(struct radeon_device *rdev)
3049 {
3050 #if defined(CONFIG_DEBUG_FS)
3051 	struct dentry *root = rdev->ddev->primary->debugfs_root;
3052 
3053 	debugfs_create_file("r100_rbbm_info", 0444, root, rdev,
3054 			    &r100_debugfs_rbbm_info_fops);
3055 #endif
3056 }
3057 
3058 void r100_debugfs_cp_init(struct radeon_device *rdev)
3059 {
3060 #if defined(CONFIG_DEBUG_FS)
3061 	struct dentry *root = rdev->ddev->primary->debugfs_root;
3062 
3063 	debugfs_create_file("r100_cp_ring_info", 0444, root, rdev,
3064 			    &r100_debugfs_cp_ring_info_fops);
3065 	debugfs_create_file("r100_cp_csq_fifo", 0444, root, rdev,
3066 			    &r100_debugfs_cp_csq_fifo_fops);
3067 #endif
3068 }
3069 
3070 void  r100_debugfs_mc_info_init(struct radeon_device *rdev)
3071 {
3072 #if defined(CONFIG_DEBUG_FS)
3073 	struct dentry *root = rdev->ddev->primary->debugfs_root;
3074 
3075 	debugfs_create_file("r100_mc_info", 0444, root, rdev,
3076 			    &r100_debugfs_mc_info_fops);
3077 #endif
3078 }
3079 
3080 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3081 			 uint32_t tiling_flags, uint32_t pitch,
3082 			 uint32_t offset, uint32_t obj_size)
3083 {
3084 	int surf_index = reg * 16;
3085 	int flags = 0;
3086 
3087 	if (rdev->family <= CHIP_RS200) {
3088 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3089 				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3090 			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3091 		if (tiling_flags & RADEON_TILING_MACRO)
3092 			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3093 		/* setting pitch to 0 disables tiling */
3094 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3095 				== 0)
3096 			pitch = 0;
3097 	} else if (rdev->family <= CHIP_RV280) {
3098 		if (tiling_flags & (RADEON_TILING_MACRO))
3099 			flags |= R200_SURF_TILE_COLOR_MACRO;
3100 		if (tiling_flags & RADEON_TILING_MICRO)
3101 			flags |= R200_SURF_TILE_COLOR_MICRO;
3102 	} else {
3103 		if (tiling_flags & RADEON_TILING_MACRO)
3104 			flags |= R300_SURF_TILE_MACRO;
3105 		if (tiling_flags & RADEON_TILING_MICRO)
3106 			flags |= R300_SURF_TILE_MICRO;
3107 	}
3108 
3109 	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3110 		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3111 	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3112 		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3113 
3114 	/* r100/r200 divide by 16 */
3115 	if (rdev->family < CHIP_R300)
3116 		flags |= pitch / 16;
3117 	else
3118 		flags |= pitch / 8;
3119 
3120 
3121 	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3122 	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3123 	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3124 	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3125 	return 0;
3126 }
3127 
3128 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3129 {
3130 	int surf_index = reg * 16;
3131 	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3132 }
3133 
3134 void r100_bandwidth_update(struct radeon_device *rdev)
3135 {
3136 	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3137 	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3138 	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
3139 	fixed20_12 crit_point_ff = {0};
3140 	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3141 	fixed20_12 memtcas_ff[8] = {
3142 		dfixed_init(1),
3143 		dfixed_init(2),
3144 		dfixed_init(3),
3145 		dfixed_init(0),
3146 		dfixed_init_half(1),
3147 		dfixed_init_half(2),
3148 		dfixed_init(0),
3149 	};
3150 	fixed20_12 memtcas_rs480_ff[8] = {
3151 		dfixed_init(0),
3152 		dfixed_init(1),
3153 		dfixed_init(2),
3154 		dfixed_init(3),
3155 		dfixed_init(0),
3156 		dfixed_init_half(1),
3157 		dfixed_init_half(2),
3158 		dfixed_init_half(3),
3159 	};
3160 	fixed20_12 memtcas2_ff[8] = {
3161 		dfixed_init(0),
3162 		dfixed_init(1),
3163 		dfixed_init(2),
3164 		dfixed_init(3),
3165 		dfixed_init(4),
3166 		dfixed_init(5),
3167 		dfixed_init(6),
3168 		dfixed_init(7),
3169 	};
3170 	fixed20_12 memtrbs[8] = {
3171 		dfixed_init(1),
3172 		dfixed_init_half(1),
3173 		dfixed_init(2),
3174 		dfixed_init_half(2),
3175 		dfixed_init(3),
3176 		dfixed_init_half(3),
3177 		dfixed_init(4),
3178 		dfixed_init_half(4)
3179 	};
3180 	fixed20_12 memtrbs_r4xx[8] = {
3181 		dfixed_init(4),
3182 		dfixed_init(5),
3183 		dfixed_init(6),
3184 		dfixed_init(7),
3185 		dfixed_init(8),
3186 		dfixed_init(9),
3187 		dfixed_init(10),
3188 		dfixed_init(11)
3189 	};
3190 	fixed20_12 min_mem_eff;
3191 	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3192 	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3193 	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate = {0},
3194 		disp_drain_rate2, read_return_rate;
3195 	fixed20_12 time_disp1_drop_priority;
3196 	int c;
3197 	int cur_size = 16;       /* in octawords */
3198 	int critical_point = 0, critical_point2;
3199 /* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3200 	int stop_req, max_stop_req;
3201 	struct drm_display_mode *mode1 = NULL;
3202 	struct drm_display_mode *mode2 = NULL;
3203 	uint32_t pixel_bytes1 = 0;
3204 	uint32_t pixel_bytes2 = 0;
3205 
3206 	/* Guess line buffer size to be 8192 pixels */
3207 	u32 lb_size = 8192;
3208 
3209 	if (!rdev->mode_info.mode_config_initialized)
3210 		return;
3211 
3212 	radeon_update_display_priority(rdev);
3213 
3214 	if (rdev->mode_info.crtcs[0]->base.enabled) {
3215 		const struct drm_framebuffer *fb =
3216 			rdev->mode_info.crtcs[0]->base.primary->fb;
3217 
3218 		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3219 		pixel_bytes1 = fb->format->cpp[0];
3220 	}
3221 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3222 		if (rdev->mode_info.crtcs[1]->base.enabled) {
3223 			const struct drm_framebuffer *fb =
3224 				rdev->mode_info.crtcs[1]->base.primary->fb;
3225 
3226 			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3227 			pixel_bytes2 = fb->format->cpp[0];
3228 		}
3229 	}
3230 
3231 	min_mem_eff.full = dfixed_const_8(0);
3232 	/* get modes */
3233 	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3234 		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3235 		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3236 		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3237 		/* check crtc enables */
3238 		if (mode2)
3239 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3240 		if (mode1)
3241 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3242 		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3243 	}
3244 
3245 	/*
3246 	 * determine is there is enough bw for current mode
3247 	 */
3248 	sclk_ff = rdev->pm.sclk;
3249 	mclk_ff = rdev->pm.mclk;
3250 
3251 	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3252 	temp_ff.full = dfixed_const(temp);
3253 	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3254 
3255 	pix_clk.full = 0;
3256 	pix_clk2.full = 0;
3257 	peak_disp_bw.full = 0;
3258 	if (mode1) {
3259 		temp_ff.full = dfixed_const(1000);
3260 		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3261 		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3262 		temp_ff.full = dfixed_const(pixel_bytes1);
3263 		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3264 	}
3265 	if (mode2) {
3266 		temp_ff.full = dfixed_const(1000);
3267 		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3268 		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3269 		temp_ff.full = dfixed_const(pixel_bytes2);
3270 		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3271 	}
3272 
3273 	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3274 	if (peak_disp_bw.full >= mem_bw.full) {
3275 		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3276 			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3277 	}
3278 
3279 	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3280 	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3281 	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3282 		mem_trcd = ((temp >> 2) & 0x3) + 1;
3283 		mem_trp  = ((temp & 0x3)) + 1;
3284 		mem_tras = ((temp & 0x70) >> 4) + 1;
3285 	} else if (rdev->family == CHIP_R300 ||
3286 		   rdev->family == CHIP_R350) { /* r300, r350 */
3287 		mem_trcd = (temp & 0x7) + 1;
3288 		mem_trp = ((temp >> 8) & 0x7) + 1;
3289 		mem_tras = ((temp >> 11) & 0xf) + 4;
3290 	} else if (rdev->family == CHIP_RV350 ||
3291 		   rdev->family == CHIP_RV380) {
3292 		/* rv3x0 */
3293 		mem_trcd = (temp & 0x7) + 3;
3294 		mem_trp = ((temp >> 8) & 0x7) + 3;
3295 		mem_tras = ((temp >> 11) & 0xf) + 6;
3296 	} else if (rdev->family == CHIP_R420 ||
3297 		   rdev->family == CHIP_R423 ||
3298 		   rdev->family == CHIP_RV410) {
3299 		/* r4xx */
3300 		mem_trcd = (temp & 0xf) + 3;
3301 		if (mem_trcd > 15)
3302 			mem_trcd = 15;
3303 		mem_trp = ((temp >> 8) & 0xf) + 3;
3304 		if (mem_trp > 15)
3305 			mem_trp = 15;
3306 		mem_tras = ((temp >> 12) & 0x1f) + 6;
3307 		if (mem_tras > 31)
3308 			mem_tras = 31;
3309 	} else { /* RV200, R200 */
3310 		mem_trcd = (temp & 0x7) + 1;
3311 		mem_trp = ((temp >> 8) & 0x7) + 1;
3312 		mem_tras = ((temp >> 12) & 0xf) + 4;
3313 	}
3314 	/* convert to FF */
3315 	trcd_ff.full = dfixed_const(mem_trcd);
3316 	trp_ff.full = dfixed_const(mem_trp);
3317 	tras_ff.full = dfixed_const(mem_tras);
3318 
3319 	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3320 	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3321 	data = (temp & (7 << 20)) >> 20;
3322 	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3323 		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3324 			tcas_ff = memtcas_rs480_ff[data];
3325 		else
3326 			tcas_ff = memtcas_ff[data];
3327 	} else
3328 		tcas_ff = memtcas2_ff[data];
3329 
3330 	if (rdev->family == CHIP_RS400 ||
3331 	    rdev->family == CHIP_RS480) {
3332 		/* extra cas latency stored in bits 23-25 0-4 clocks */
3333 		data = (temp >> 23) & 0x7;
3334 		if (data < 5)
3335 			tcas_ff.full += dfixed_const(data);
3336 	}
3337 
3338 	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3339 		/* on the R300, Tcas is included in Trbs.
3340 		 */
3341 		temp = RREG32(RADEON_MEM_CNTL);
3342 		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3343 		if (data == 1) {
3344 			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3345 				temp = RREG32(R300_MC_IND_INDEX);
3346 				temp &= ~R300_MC_IND_ADDR_MASK;
3347 				temp |= R300_MC_READ_CNTL_CD_mcind;
3348 				WREG32(R300_MC_IND_INDEX, temp);
3349 				temp = RREG32(R300_MC_IND_DATA);
3350 				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3351 			} else {
3352 				temp = RREG32(R300_MC_READ_CNTL_AB);
3353 				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3354 			}
3355 		} else {
3356 			temp = RREG32(R300_MC_READ_CNTL_AB);
3357 			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3358 		}
3359 		if (rdev->family == CHIP_RV410 ||
3360 		    rdev->family == CHIP_R420 ||
3361 		    rdev->family == CHIP_R423)
3362 			trbs_ff = memtrbs_r4xx[data];
3363 		else
3364 			trbs_ff = memtrbs[data];
3365 		tcas_ff.full += trbs_ff.full;
3366 	}
3367 
3368 	sclk_eff_ff.full = sclk_ff.full;
3369 
3370 	if (rdev->flags & RADEON_IS_AGP) {
3371 		fixed20_12 agpmode_ff;
3372 		agpmode_ff.full = dfixed_const(radeon_agpmode);
3373 		temp_ff.full = dfixed_const_666(16);
3374 		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3375 	}
3376 	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3377 
3378 	if (ASIC_IS_R300(rdev)) {
3379 		sclk_delay_ff.full = dfixed_const(250);
3380 	} else {
3381 		if ((rdev->family == CHIP_RV100) ||
3382 		    rdev->flags & RADEON_IS_IGP) {
3383 			if (rdev->mc.vram_is_ddr)
3384 				sclk_delay_ff.full = dfixed_const(41);
3385 			else
3386 				sclk_delay_ff.full = dfixed_const(33);
3387 		} else {
3388 			if (rdev->mc.vram_width == 128)
3389 				sclk_delay_ff.full = dfixed_const(57);
3390 			else
3391 				sclk_delay_ff.full = dfixed_const(41);
3392 		}
3393 	}
3394 
3395 	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3396 
3397 	if (rdev->mc.vram_is_ddr) {
3398 		if (rdev->mc.vram_width == 32) {
3399 			k1.full = dfixed_const(40);
3400 			c  = 3;
3401 		} else {
3402 			k1.full = dfixed_const(20);
3403 			c  = 1;
3404 		}
3405 	} else {
3406 		k1.full = dfixed_const(40);
3407 		c  = 3;
3408 	}
3409 
3410 	temp_ff.full = dfixed_const(2);
3411 	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3412 	temp_ff.full = dfixed_const(c);
3413 	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3414 	temp_ff.full = dfixed_const(4);
3415 	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3416 	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3417 	mc_latency_mclk.full += k1.full;
3418 
3419 	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3420 	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3421 
3422 	/*
3423 	  HW cursor time assuming worst case of full size colour cursor.
3424 	*/
3425 	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3426 	temp_ff.full += trcd_ff.full;
3427 	if (temp_ff.full < tras_ff.full)
3428 		temp_ff.full = tras_ff.full;
3429 	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3430 
3431 	temp_ff.full = dfixed_const(cur_size);
3432 	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3433 	/*
3434 	  Find the total latency for the display data.
3435 	*/
3436 	disp_latency_overhead.full = dfixed_const(8);
3437 	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3438 	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3439 	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3440 
3441 	if (mc_latency_mclk.full > mc_latency_sclk.full)
3442 		disp_latency.full = mc_latency_mclk.full;
3443 	else
3444 		disp_latency.full = mc_latency_sclk.full;
3445 
3446 	/* setup Max GRPH_STOP_REQ default value */
3447 	if (ASIC_IS_RV100(rdev))
3448 		max_stop_req = 0x5c;
3449 	else
3450 		max_stop_req = 0x7c;
3451 
3452 	if (mode1) {
3453 		/*  CRTC1
3454 		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3455 		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3456 		*/
3457 		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3458 
3459 		if (stop_req > max_stop_req)
3460 			stop_req = max_stop_req;
3461 
3462 		/*
3463 		  Find the drain rate of the display buffer.
3464 		*/
3465 		temp_ff.full = dfixed_const((16/pixel_bytes1));
3466 		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3467 
3468 		/*
3469 		  Find the critical point of the display buffer.
3470 		*/
3471 		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3472 		crit_point_ff.full += dfixed_const_half(0);
3473 
3474 		critical_point = dfixed_trunc(crit_point_ff);
3475 
3476 		if (rdev->disp_priority == 2) {
3477 			critical_point = 0;
3478 		}
3479 
3480 		/*
3481 		  The critical point should never be above max_stop_req-4.  Setting
3482 		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3483 		*/
3484 		if (max_stop_req - critical_point < 4)
3485 			critical_point = 0;
3486 
3487 		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3488 			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3489 			critical_point = 0x10;
3490 		}
3491 
3492 		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3493 		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3494 		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3495 		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3496 		if ((rdev->family == CHIP_R350) &&
3497 		    (stop_req > 0x15)) {
3498 			stop_req -= 0x10;
3499 		}
3500 		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3501 		temp |= RADEON_GRPH_BUFFER_SIZE;
3502 		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3503 			  RADEON_GRPH_CRITICAL_AT_SOF |
3504 			  RADEON_GRPH_STOP_CNTL);
3505 		/*
3506 		  Write the result into the register.
3507 		*/
3508 		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3509 						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3510 
3511 #if 0
3512 		if ((rdev->family == CHIP_RS400) ||
3513 		    (rdev->family == CHIP_RS480)) {
3514 			/* attempt to program RS400 disp regs correctly ??? */
3515 			temp = RREG32(RS400_DISP1_REG_CNTL);
3516 			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3517 				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3518 			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3519 						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3520 						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3521 			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3522 			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3523 				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3524 			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3525 						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3526 						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3527 		}
3528 #endif
3529 
3530 		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3531 			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3532 			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3533 	}
3534 
3535 	if (mode2) {
3536 		u32 grph2_cntl;
3537 		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3538 
3539 		if (stop_req > max_stop_req)
3540 			stop_req = max_stop_req;
3541 
3542 		/*
3543 		  Find the drain rate of the display buffer.
3544 		*/
3545 		temp_ff.full = dfixed_const((16/pixel_bytes2));
3546 		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3547 
3548 		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3549 		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3550 		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3551 		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3552 		if ((rdev->family == CHIP_R350) &&
3553 		    (stop_req > 0x15)) {
3554 			stop_req -= 0x10;
3555 		}
3556 		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3557 		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3558 		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3559 			  RADEON_GRPH_CRITICAL_AT_SOF |
3560 			  RADEON_GRPH_STOP_CNTL);
3561 
3562 		if ((rdev->family == CHIP_RS100) ||
3563 		    (rdev->family == CHIP_RS200))
3564 			critical_point2 = 0;
3565 		else {
3566 			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3567 			temp_ff.full = dfixed_const(temp);
3568 			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3569 			if (sclk_ff.full < temp_ff.full)
3570 				temp_ff.full = sclk_ff.full;
3571 
3572 			read_return_rate.full = temp_ff.full;
3573 
3574 			if (mode1) {
3575 				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3576 				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3577 			} else {
3578 				time_disp1_drop_priority.full = 0;
3579 			}
3580 			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3581 			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3582 			crit_point_ff.full += dfixed_const_half(0);
3583 
3584 			critical_point2 = dfixed_trunc(crit_point_ff);
3585 
3586 			if (rdev->disp_priority == 2) {
3587 				critical_point2 = 0;
3588 			}
3589 
3590 			if (max_stop_req - critical_point2 < 4)
3591 				critical_point2 = 0;
3592 
3593 		}
3594 
3595 		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3596 			/* some R300 cards have problem with this set to 0 */
3597 			critical_point2 = 0x10;
3598 		}
3599 
3600 		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3601 						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3602 
3603 		if ((rdev->family == CHIP_RS400) ||
3604 		    (rdev->family == CHIP_RS480)) {
3605 #if 0
3606 			/* attempt to program RS400 disp2 regs correctly ??? */
3607 			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3608 			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3609 				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3610 			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3611 						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3612 						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3613 			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3614 			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3615 				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3616 			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3617 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3618 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3619 #endif
3620 			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3621 			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3622 			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3623 			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3624 		}
3625 
3626 		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3627 			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3628 	}
3629 
3630 	/* Save number of lines the linebuffer leads before the scanout */
3631 	if (mode1)
3632 	    rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
3633 
3634 	if (mode2)
3635 	    rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
3636 }
3637 
3638 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3639 {
3640 	uint32_t scratch;
3641 	uint32_t tmp = 0;
3642 	unsigned i;
3643 	int r;
3644 
3645 	r = radeon_scratch_get(rdev, &scratch);
3646 	if (r) {
3647 		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3648 		return r;
3649 	}
3650 	WREG32(scratch, 0xCAFEDEAD);
3651 	r = radeon_ring_lock(rdev, ring, 2);
3652 	if (r) {
3653 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3654 		radeon_scratch_free(rdev, scratch);
3655 		return r;
3656 	}
3657 	radeon_ring_write(ring, PACKET0(scratch, 0));
3658 	radeon_ring_write(ring, 0xDEADBEEF);
3659 	radeon_ring_unlock_commit(rdev, ring, false);
3660 	for (i = 0; i < rdev->usec_timeout; i++) {
3661 		tmp = RREG32(scratch);
3662 		if (tmp == 0xDEADBEEF) {
3663 			break;
3664 		}
3665 		udelay(1);
3666 	}
3667 	if (i < rdev->usec_timeout) {
3668 		DRM_INFO("ring test succeeded in %d usecs\n", i);
3669 	} else {
3670 		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3671 			  scratch, tmp);
3672 		r = -EINVAL;
3673 	}
3674 	radeon_scratch_free(rdev, scratch);
3675 	return r;
3676 }
3677 
3678 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3679 {
3680 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3681 
3682 	if (ring->rptr_save_reg) {
3683 		u32 next_rptr = ring->wptr + 2 + 3;
3684 		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3685 		radeon_ring_write(ring, next_rptr);
3686 	}
3687 
3688 	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3689 	radeon_ring_write(ring, ib->gpu_addr);
3690 	radeon_ring_write(ring, ib->length_dw);
3691 }
3692 
3693 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3694 {
3695 	struct radeon_ib ib;
3696 	uint32_t scratch;
3697 	uint32_t tmp = 0;
3698 	unsigned i;
3699 	int r;
3700 
3701 	r = radeon_scratch_get(rdev, &scratch);
3702 	if (r) {
3703 		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3704 		return r;
3705 	}
3706 	WREG32(scratch, 0xCAFEDEAD);
3707 	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3708 	if (r) {
3709 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3710 		goto free_scratch;
3711 	}
3712 	ib.ptr[0] = PACKET0(scratch, 0);
3713 	ib.ptr[1] = 0xDEADBEEF;
3714 	ib.ptr[2] = PACKET2(0);
3715 	ib.ptr[3] = PACKET2(0);
3716 	ib.ptr[4] = PACKET2(0);
3717 	ib.ptr[5] = PACKET2(0);
3718 	ib.ptr[6] = PACKET2(0);
3719 	ib.ptr[7] = PACKET2(0);
3720 	ib.length_dw = 8;
3721 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
3722 	if (r) {
3723 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3724 		goto free_ib;
3725 	}
3726 	r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
3727 		RADEON_USEC_IB_TEST_TIMEOUT));
3728 	if (r < 0) {
3729 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3730 		goto free_ib;
3731 	} else if (r == 0) {
3732 		DRM_ERROR("radeon: fence wait timed out.\n");
3733 		r = -ETIMEDOUT;
3734 		goto free_ib;
3735 	}
3736 	r = 0;
3737 	for (i = 0; i < rdev->usec_timeout; i++) {
3738 		tmp = RREG32(scratch);
3739 		if (tmp == 0xDEADBEEF) {
3740 			break;
3741 		}
3742 		udelay(1);
3743 	}
3744 	if (i < rdev->usec_timeout) {
3745 		DRM_INFO("ib test succeeded in %u usecs\n", i);
3746 	} else {
3747 		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3748 			  scratch, tmp);
3749 		r = -EINVAL;
3750 	}
3751 free_ib:
3752 	radeon_ib_free(rdev, &ib);
3753 free_scratch:
3754 	radeon_scratch_free(rdev, scratch);
3755 	return r;
3756 }
3757 
3758 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3759 {
3760 	/* Shutdown CP we shouldn't need to do that but better be safe than
3761 	 * sorry
3762 	 */
3763 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3764 	WREG32(R_000740_CP_CSQ_CNTL, 0);
3765 
3766 	/* Save few CRTC registers */
3767 	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3768 	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3769 	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3770 	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3771 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3772 		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3773 		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3774 	}
3775 
3776 	/* Disable VGA aperture access */
3777 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3778 	/* Disable cursor, overlay, crtc */
3779 	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3780 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3781 					S_000054_CRTC_DISPLAY_DIS(1));
3782 	WREG32(R_000050_CRTC_GEN_CNTL,
3783 			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3784 			S_000050_CRTC_DISP_REQ_EN_B(1));
3785 	WREG32(R_000420_OV0_SCALE_CNTL,
3786 		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3787 	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3788 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3789 		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3790 						S_000360_CUR2_LOCK(1));
3791 		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3792 			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3793 			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3794 			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3795 		WREG32(R_000360_CUR2_OFFSET,
3796 			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3797 	}
3798 }
3799 
3800 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3801 {
3802 	/* Update base address for crtc */
3803 	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3804 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3805 		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3806 	}
3807 	/* Restore CRTC registers */
3808 	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3809 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3810 	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3811 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3812 		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3813 	}
3814 }
3815 
3816 void r100_vga_render_disable(struct radeon_device *rdev)
3817 {
3818 	u32 tmp;
3819 
3820 	tmp = RREG8(R_0003C2_GENMO_WT);
3821 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3822 }
3823 
3824 static void r100_mc_program(struct radeon_device *rdev)
3825 {
3826 	struct r100_mc_save save;
3827 
3828 	/* Stops all mc clients */
3829 	r100_mc_stop(rdev, &save);
3830 	if (rdev->flags & RADEON_IS_AGP) {
3831 		WREG32(R_00014C_MC_AGP_LOCATION,
3832 			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3833 			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3834 		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3835 		if (rdev->family > CHIP_RV200)
3836 			WREG32(R_00015C_AGP_BASE_2,
3837 				upper_32_bits(rdev->mc.agp_base) & 0xff);
3838 	} else {
3839 		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3840 		WREG32(R_000170_AGP_BASE, 0);
3841 		if (rdev->family > CHIP_RV200)
3842 			WREG32(R_00015C_AGP_BASE_2, 0);
3843 	}
3844 	/* Wait for mc idle */
3845 	if (r100_mc_wait_for_idle(rdev))
3846 		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3847 	/* Program MC, should be a 32bits limited address space */
3848 	WREG32(R_000148_MC_FB_LOCATION,
3849 		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3850 		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3851 	r100_mc_resume(rdev, &save);
3852 }
3853 
3854 static void r100_clock_startup(struct radeon_device *rdev)
3855 {
3856 	u32 tmp;
3857 
3858 	if (radeon_dynclks != -1 && radeon_dynclks)
3859 		radeon_legacy_set_clock_gating(rdev, 1);
3860 	/* We need to force on some of the block */
3861 	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3862 	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3863 	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3864 		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3865 	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3866 }
3867 
3868 static int r100_startup(struct radeon_device *rdev)
3869 {
3870 	int r;
3871 
3872 	/* set common regs */
3873 	r100_set_common_regs(rdev);
3874 	/* program mc */
3875 	r100_mc_program(rdev);
3876 	/* Resume clock */
3877 	r100_clock_startup(rdev);
3878 	/* Initialize GART (initialize after TTM so we can allocate
3879 	 * memory through TTM but finalize after TTM) */
3880 	r100_enable_bm(rdev);
3881 	if (rdev->flags & RADEON_IS_PCI) {
3882 		r = r100_pci_gart_enable(rdev);
3883 		if (r)
3884 			return r;
3885 	}
3886 
3887 	/* allocate wb buffer */
3888 	r = radeon_wb_init(rdev);
3889 	if (r)
3890 		return r;
3891 
3892 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3893 	if (r) {
3894 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3895 		return r;
3896 	}
3897 
3898 	/* Enable IRQ */
3899 	if (!rdev->irq.installed) {
3900 		r = radeon_irq_kms_init(rdev);
3901 		if (r)
3902 			return r;
3903 	}
3904 
3905 	r100_irq_set(rdev);
3906 	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3907 	/* 1M ring buffer */
3908 	r = r100_cp_init(rdev, 1024 * 1024);
3909 	if (r) {
3910 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3911 		return r;
3912 	}
3913 
3914 	r = radeon_ib_pool_init(rdev);
3915 	if (r) {
3916 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3917 		return r;
3918 	}
3919 
3920 	return 0;
3921 }
3922 
3923 int r100_resume(struct radeon_device *rdev)
3924 {
3925 	int r;
3926 
3927 	/* Make sur GART are not working */
3928 	if (rdev->flags & RADEON_IS_PCI)
3929 		r100_pci_gart_disable(rdev);
3930 	/* Resume clock before doing reset */
3931 	r100_clock_startup(rdev);
3932 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3933 	if (radeon_asic_reset(rdev)) {
3934 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3935 			RREG32(R_000E40_RBBM_STATUS),
3936 			RREG32(R_0007C0_CP_STAT));
3937 	}
3938 	/* post */
3939 	radeon_combios_asic_init(rdev->ddev);
3940 	/* Resume clock after posting */
3941 	r100_clock_startup(rdev);
3942 	/* Initialize surface registers */
3943 	radeon_surface_init(rdev);
3944 
3945 	rdev->accel_working = true;
3946 	r = r100_startup(rdev);
3947 	if (r) {
3948 		rdev->accel_working = false;
3949 	}
3950 	return r;
3951 }
3952 
3953 int r100_suspend(struct radeon_device *rdev)
3954 {
3955 	radeon_pm_suspend(rdev);
3956 	r100_cp_disable(rdev);
3957 	radeon_wb_disable(rdev);
3958 	r100_irq_disable(rdev);
3959 	if (rdev->flags & RADEON_IS_PCI)
3960 		r100_pci_gart_disable(rdev);
3961 	return 0;
3962 }
3963 
3964 void r100_fini(struct radeon_device *rdev)
3965 {
3966 	radeon_pm_fini(rdev);
3967 	r100_cp_fini(rdev);
3968 	radeon_wb_fini(rdev);
3969 	radeon_ib_pool_fini(rdev);
3970 	radeon_gem_fini(rdev);
3971 	if (rdev->flags & RADEON_IS_PCI)
3972 		r100_pci_gart_fini(rdev);
3973 	radeon_agp_fini(rdev);
3974 	radeon_irq_kms_fini(rdev);
3975 	radeon_fence_driver_fini(rdev);
3976 	radeon_bo_fini(rdev);
3977 	radeon_atombios_fini(rdev);
3978 	kfree(rdev->bios);
3979 	rdev->bios = NULL;
3980 }
3981 
3982 /*
3983  * Due to how kexec works, it can leave the hw fully initialised when it
3984  * boots the new kernel. However doing our init sequence with the CP and
3985  * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3986  * do some quick sanity checks and restore sane values to avoid this
3987  * problem.
3988  */
3989 void r100_restore_sanity(struct radeon_device *rdev)
3990 {
3991 	u32 tmp;
3992 
3993 	tmp = RREG32(RADEON_CP_CSQ_CNTL);
3994 	if (tmp) {
3995 		WREG32(RADEON_CP_CSQ_CNTL, 0);
3996 	}
3997 	tmp = RREG32(RADEON_CP_RB_CNTL);
3998 	if (tmp) {
3999 		WREG32(RADEON_CP_RB_CNTL, 0);
4000 	}
4001 	tmp = RREG32(RADEON_SCRATCH_UMSK);
4002 	if (tmp) {
4003 		WREG32(RADEON_SCRATCH_UMSK, 0);
4004 	}
4005 }
4006 
4007 int r100_init(struct radeon_device *rdev)
4008 {
4009 	int r;
4010 
4011 	/* Register debugfs file specific to this group of asics */
4012 	r100_debugfs_mc_info_init(rdev);
4013 	/* Disable VGA */
4014 	r100_vga_render_disable(rdev);
4015 	/* Initialize scratch registers */
4016 	radeon_scratch_init(rdev);
4017 	/* Initialize surface registers */
4018 	radeon_surface_init(rdev);
4019 	/* sanity check some register to avoid hangs like after kexec */
4020 	r100_restore_sanity(rdev);
4021 	/* TODO: disable VGA need to use VGA request */
4022 	/* BIOS*/
4023 	if (!radeon_get_bios(rdev)) {
4024 		if (ASIC_IS_AVIVO(rdev))
4025 			return -EINVAL;
4026 	}
4027 	if (rdev->is_atom_bios) {
4028 		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4029 		return -EINVAL;
4030 	} else {
4031 		r = radeon_combios_init(rdev);
4032 		if (r)
4033 			return r;
4034 	}
4035 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4036 	if (radeon_asic_reset(rdev)) {
4037 		dev_warn(rdev->dev,
4038 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4039 			RREG32(R_000E40_RBBM_STATUS),
4040 			RREG32(R_0007C0_CP_STAT));
4041 	}
4042 	/* check if cards are posted or not */
4043 	if (radeon_boot_test_post_card(rdev) == false)
4044 		return -EINVAL;
4045 	/* Set asic errata */
4046 	r100_errata(rdev);
4047 	/* Initialize clocks */
4048 	radeon_get_clock_info(rdev->ddev);
4049 	/* initialize AGP */
4050 	if (rdev->flags & RADEON_IS_AGP) {
4051 		r = radeon_agp_init(rdev);
4052 		if (r) {
4053 			radeon_agp_disable(rdev);
4054 		}
4055 	}
4056 	/* initialize VRAM */
4057 	r100_mc_init(rdev);
4058 	/* Fence driver */
4059 	radeon_fence_driver_init(rdev);
4060 	/* Memory manager */
4061 	r = radeon_bo_init(rdev);
4062 	if (r)
4063 		return r;
4064 	if (rdev->flags & RADEON_IS_PCI) {
4065 		r = r100_pci_gart_init(rdev);
4066 		if (r)
4067 			return r;
4068 	}
4069 	r100_set_safe_registers(rdev);
4070 
4071 	/* Initialize power management */
4072 	radeon_pm_init(rdev);
4073 
4074 	rdev->accel_working = true;
4075 	r = r100_startup(rdev);
4076 	if (r) {
4077 		/* Somethings want wront with the accel init stop accel */
4078 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4079 		r100_cp_fini(rdev);
4080 		radeon_wb_fini(rdev);
4081 		radeon_ib_pool_fini(rdev);
4082 		radeon_irq_kms_fini(rdev);
4083 		if (rdev->flags & RADEON_IS_PCI)
4084 			r100_pci_gart_fini(rdev);
4085 		rdev->accel_working = false;
4086 	}
4087 	return 0;
4088 }
4089 
4090 uint32_t r100_mm_rreg_slow(struct radeon_device *rdev, uint32_t reg)
4091 {
4092 	unsigned long flags;
4093 	uint32_t ret;
4094 
4095 	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4096 	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4097 	ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4098 	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4099 	return ret;
4100 }
4101 
4102 void r100_mm_wreg_slow(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4103 {
4104 	unsigned long flags;
4105 
4106 	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4107 	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4108 	writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4109 	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4110 }
4111 
4112 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4113 {
4114 	if (reg < rdev->rio_mem_size)
4115 		return ioread32(rdev->rio_mem + reg);
4116 	else {
4117 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4118 		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4119 	}
4120 }
4121 
4122 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4123 {
4124 	if (reg < rdev->rio_mem_size)
4125 		iowrite32(v, rdev->rio_mem + reg);
4126 	else {
4127 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4128 		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4129 	}
4130 }
4131