xref: /openbmc/linux/drivers/gpu/drm/radeon/r100.c (revision 0d3b051a)
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_debugfs.h>
36 #include <drm/drm_device.h>
37 #include <drm/drm_file.h>
38 #include <drm/drm_fourcc.h>
39 #include <drm/drm_vblank.h>
40 #include <drm/radeon_drm.h>
41 
42 #include "atom.h"
43 #include "r100_reg_safe.h"
44 #include "r100d.h"
45 #include "radeon.h"
46 #include "radeon_asic.h"
47 #include "radeon_reg.h"
48 #include "rn50_reg_safe.h"
49 #include "rs100d.h"
50 #include "rv200d.h"
51 #include "rv250d.h"
52 
53 /* Firmware Names */
54 #define FIRMWARE_R100		"radeon/R100_cp.bin"
55 #define FIRMWARE_R200		"radeon/R200_cp.bin"
56 #define FIRMWARE_R300		"radeon/R300_cp.bin"
57 #define FIRMWARE_R420		"radeon/R420_cp.bin"
58 #define FIRMWARE_RS690		"radeon/RS690_cp.bin"
59 #define FIRMWARE_RS600		"radeon/RS600_cp.bin"
60 #define FIRMWARE_R520		"radeon/R520_cp.bin"
61 
62 MODULE_FIRMWARE(FIRMWARE_R100);
63 MODULE_FIRMWARE(FIRMWARE_R200);
64 MODULE_FIRMWARE(FIRMWARE_R300);
65 MODULE_FIRMWARE(FIRMWARE_R420);
66 MODULE_FIRMWARE(FIRMWARE_RS690);
67 MODULE_FIRMWARE(FIRMWARE_RS600);
68 MODULE_FIRMWARE(FIRMWARE_R520);
69 
70 #include "r100_track.h"
71 
72 /* This files gather functions specifics to:
73  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
74  * and others in some cases.
75  */
76 
77 static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
78 {
79 	if (crtc == 0) {
80 		if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
81 			return true;
82 		else
83 			return false;
84 	} else {
85 		if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
86 			return true;
87 		else
88 			return false;
89 	}
90 }
91 
92 static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
93 {
94 	u32 vline1, vline2;
95 
96 	if (crtc == 0) {
97 		vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
98 		vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
99 	} else {
100 		vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
101 		vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
102 	}
103 	if (vline1 != vline2)
104 		return true;
105 	else
106 		return false;
107 }
108 
109 /**
110  * r100_wait_for_vblank - vblank wait asic callback.
111  *
112  * @rdev: radeon_device pointer
113  * @crtc: crtc to wait for vblank on
114  *
115  * Wait for vblank on the requested crtc (r1xx-r4xx).
116  */
117 void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
118 {
119 	unsigned i = 0;
120 
121 	if (crtc >= rdev->num_crtc)
122 		return;
123 
124 	if (crtc == 0) {
125 		if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
126 			return;
127 	} else {
128 		if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
129 			return;
130 	}
131 
132 	/* depending on when we hit vblank, we may be close to active; if so,
133 	 * wait for another frame.
134 	 */
135 	while (r100_is_in_vblank(rdev, crtc)) {
136 		if (i++ % 100 == 0) {
137 			if (!r100_is_counter_moving(rdev, crtc))
138 				break;
139 		}
140 	}
141 
142 	while (!r100_is_in_vblank(rdev, crtc)) {
143 		if (i++ % 100 == 0) {
144 			if (!r100_is_counter_moving(rdev, crtc))
145 				break;
146 		}
147 	}
148 }
149 
150 /**
151  * r100_page_flip - pageflip callback.
152  *
153  * @rdev: radeon_device pointer
154  * @crtc_id: crtc to cleanup pageflip on
155  * @crtc_base: new address of the crtc (GPU MC address)
156  * @async: asynchronous flip
157  *
158  * Does the actual pageflip (r1xx-r4xx).
159  * During vblank we take the crtc lock and wait for the update_pending
160  * bit to go high, when it does, we release the lock, and allow the
161  * double buffered update to take place.
162  */
163 void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base, bool async)
164 {
165 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
166 	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
167 	int i;
168 
169 	/* Lock the graphics update lock */
170 	/* update the scanout addresses */
171 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
172 
173 	/* Wait for update_pending to go high. */
174 	for (i = 0; i < rdev->usec_timeout; i++) {
175 		if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
176 			break;
177 		udelay(1);
178 	}
179 	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
180 
181 	/* Unlock the lock, so double-buffering can take place inside vblank */
182 	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
183 	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
184 
185 }
186 
187 /**
188  * r100_page_flip_pending - check if page flip is still pending
189  *
190  * @rdev: radeon_device pointer
191  * @crtc_id: crtc to check
192  *
193  * Check if the last pagefilp is still pending (r1xx-r4xx).
194  * Returns the current update pending status.
195  */
196 bool r100_page_flip_pending(struct radeon_device *rdev, int crtc_id)
197 {
198 	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
199 
200 	/* Return current update_pending status: */
201 	return !!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) &
202 		RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET);
203 }
204 
205 /**
206  * r100_pm_get_dynpm_state - look up dynpm power state callback.
207  *
208  * @rdev: radeon_device pointer
209  *
210  * Look up the optimal power state based on the
211  * current state of the GPU (r1xx-r5xx).
212  * Used for dynpm only.
213  */
214 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
215 {
216 	int i;
217 	rdev->pm.dynpm_can_upclock = true;
218 	rdev->pm.dynpm_can_downclock = true;
219 
220 	switch (rdev->pm.dynpm_planned_action) {
221 	case DYNPM_ACTION_MINIMUM:
222 		rdev->pm.requested_power_state_index = 0;
223 		rdev->pm.dynpm_can_downclock = false;
224 		break;
225 	case DYNPM_ACTION_DOWNCLOCK:
226 		if (rdev->pm.current_power_state_index == 0) {
227 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
228 			rdev->pm.dynpm_can_downclock = false;
229 		} else {
230 			if (rdev->pm.active_crtc_count > 1) {
231 				for (i = 0; i < rdev->pm.num_power_states; i++) {
232 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
233 						continue;
234 					else if (i >= rdev->pm.current_power_state_index) {
235 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
236 						break;
237 					} else {
238 						rdev->pm.requested_power_state_index = i;
239 						break;
240 					}
241 				}
242 			} else
243 				rdev->pm.requested_power_state_index =
244 					rdev->pm.current_power_state_index - 1;
245 		}
246 		/* don't use the power state if crtcs are active and no display flag is set */
247 		if ((rdev->pm.active_crtc_count > 0) &&
248 		    (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
249 		     RADEON_PM_MODE_NO_DISPLAY)) {
250 			rdev->pm.requested_power_state_index++;
251 		}
252 		break;
253 	case DYNPM_ACTION_UPCLOCK:
254 		if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
255 			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
256 			rdev->pm.dynpm_can_upclock = false;
257 		} else {
258 			if (rdev->pm.active_crtc_count > 1) {
259 				for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
260 					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
261 						continue;
262 					else if (i <= rdev->pm.current_power_state_index) {
263 						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
264 						break;
265 					} else {
266 						rdev->pm.requested_power_state_index = i;
267 						break;
268 					}
269 				}
270 			} else
271 				rdev->pm.requested_power_state_index =
272 					rdev->pm.current_power_state_index + 1;
273 		}
274 		break;
275 	case DYNPM_ACTION_DEFAULT:
276 		rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
277 		rdev->pm.dynpm_can_upclock = false;
278 		break;
279 	case DYNPM_ACTION_NONE:
280 	default:
281 		DRM_ERROR("Requested mode for not defined action\n");
282 		return;
283 	}
284 	/* only one clock mode per power state */
285 	rdev->pm.requested_clock_mode_index = 0;
286 
287 	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
288 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
289 		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
290 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
291 		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
292 		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
293 		  pcie_lanes);
294 }
295 
296 /**
297  * r100_pm_init_profile - Initialize power profiles callback.
298  *
299  * @rdev: radeon_device pointer
300  *
301  * Initialize the power states used in profile mode
302  * (r1xx-r3xx).
303  * Used for profile mode only.
304  */
305 void r100_pm_init_profile(struct radeon_device *rdev)
306 {
307 	/* default */
308 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
309 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
310 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
311 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
312 	/* low sh */
313 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
314 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
315 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
316 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
317 	/* mid sh */
318 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
319 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
320 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
321 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
322 	/* high sh */
323 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
324 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
325 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
326 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
327 	/* low mh */
328 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
329 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
330 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
331 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
332 	/* mid mh */
333 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
334 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
335 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
336 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
337 	/* high mh */
338 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
339 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
340 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
341 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
342 }
343 
344 /**
345  * r100_pm_misc - set additional pm hw parameters callback.
346  *
347  * @rdev: radeon_device pointer
348  *
349  * Set non-clock parameters associated with a power state
350  * (voltage, pcie lanes, etc.) (r1xx-r4xx).
351  */
352 void r100_pm_misc(struct radeon_device *rdev)
353 {
354 	int requested_index = rdev->pm.requested_power_state_index;
355 	struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
356 	struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
357 	u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
358 
359 	if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
360 		if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
361 			tmp = RREG32(voltage->gpio.reg);
362 			if (voltage->active_high)
363 				tmp |= voltage->gpio.mask;
364 			else
365 				tmp &= ~(voltage->gpio.mask);
366 			WREG32(voltage->gpio.reg, tmp);
367 			if (voltage->delay)
368 				udelay(voltage->delay);
369 		} else {
370 			tmp = RREG32(voltage->gpio.reg);
371 			if (voltage->active_high)
372 				tmp &= ~voltage->gpio.mask;
373 			else
374 				tmp |= voltage->gpio.mask;
375 			WREG32(voltage->gpio.reg, tmp);
376 			if (voltage->delay)
377 				udelay(voltage->delay);
378 		}
379 	}
380 
381 	sclk_cntl = RREG32_PLL(SCLK_CNTL);
382 	sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
383 	sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
384 	sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
385 	sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
386 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
387 		sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
388 		if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
389 			sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
390 		else
391 			sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
392 		if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
393 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
394 		else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
395 			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
396 	} else
397 		sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
398 
399 	if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
400 		sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
401 		if (voltage->delay) {
402 			sclk_more_cntl |= VOLTAGE_DROP_SYNC;
403 			switch (voltage->delay) {
404 			case 33:
405 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
406 				break;
407 			case 66:
408 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
409 				break;
410 			case 99:
411 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
412 				break;
413 			case 132:
414 				sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
415 				break;
416 			}
417 		} else
418 			sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
419 	} else
420 		sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
421 
422 	if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
423 		sclk_cntl &= ~FORCE_HDP;
424 	else
425 		sclk_cntl |= FORCE_HDP;
426 
427 	WREG32_PLL(SCLK_CNTL, sclk_cntl);
428 	WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
429 	WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
430 
431 	/* set pcie lanes */
432 	if ((rdev->flags & RADEON_IS_PCIE) &&
433 	    !(rdev->flags & RADEON_IS_IGP) &&
434 	    rdev->asic->pm.set_pcie_lanes &&
435 	    (ps->pcie_lanes !=
436 	     rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
437 		radeon_set_pcie_lanes(rdev,
438 				      ps->pcie_lanes);
439 		DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
440 	}
441 }
442 
443 /**
444  * r100_pm_prepare - pre-power state change callback.
445  *
446  * @rdev: radeon_device pointer
447  *
448  * Prepare for a power state change (r1xx-r4xx).
449  */
450 void r100_pm_prepare(struct radeon_device *rdev)
451 {
452 	struct drm_device *ddev = rdev->ddev;
453 	struct drm_crtc *crtc;
454 	struct radeon_crtc *radeon_crtc;
455 	u32 tmp;
456 
457 	/* disable any active CRTCs */
458 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
459 		radeon_crtc = to_radeon_crtc(crtc);
460 		if (radeon_crtc->enabled) {
461 			if (radeon_crtc->crtc_id) {
462 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
463 				tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
464 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
465 			} else {
466 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
467 				tmp |= RADEON_CRTC_DISP_REQ_EN_B;
468 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
469 			}
470 		}
471 	}
472 }
473 
474 /**
475  * r100_pm_finish - post-power state change callback.
476  *
477  * @rdev: radeon_device pointer
478  *
479  * Clean up after a power state change (r1xx-r4xx).
480  */
481 void r100_pm_finish(struct radeon_device *rdev)
482 {
483 	struct drm_device *ddev = rdev->ddev;
484 	struct drm_crtc *crtc;
485 	struct radeon_crtc *radeon_crtc;
486 	u32 tmp;
487 
488 	/* enable any active CRTCs */
489 	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
490 		radeon_crtc = to_radeon_crtc(crtc);
491 		if (radeon_crtc->enabled) {
492 			if (radeon_crtc->crtc_id) {
493 				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
494 				tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
495 				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
496 			} else {
497 				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
498 				tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
499 				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
500 			}
501 		}
502 	}
503 }
504 
505 /**
506  * r100_gui_idle - gui idle callback.
507  *
508  * @rdev: radeon_device pointer
509  *
510  * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
511  * Returns true if idle, false if not.
512  */
513 bool r100_gui_idle(struct radeon_device *rdev)
514 {
515 	if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
516 		return false;
517 	else
518 		return true;
519 }
520 
521 /* hpd for digital panel detect/disconnect */
522 /**
523  * r100_hpd_sense - hpd sense callback.
524  *
525  * @rdev: radeon_device pointer
526  * @hpd: hpd (hotplug detect) pin
527  *
528  * Checks if a digital monitor is connected (r1xx-r4xx).
529  * Returns true if connected, false if not connected.
530  */
531 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
532 {
533 	bool connected = false;
534 
535 	switch (hpd) {
536 	case RADEON_HPD_1:
537 		if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
538 			connected = true;
539 		break;
540 	case RADEON_HPD_2:
541 		if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
542 			connected = true;
543 		break;
544 	default:
545 		break;
546 	}
547 	return connected;
548 }
549 
550 /**
551  * r100_hpd_set_polarity - hpd set polarity callback.
552  *
553  * @rdev: radeon_device pointer
554  * @hpd: hpd (hotplug detect) pin
555  *
556  * Set the polarity of the hpd pin (r1xx-r4xx).
557  */
558 void r100_hpd_set_polarity(struct radeon_device *rdev,
559 			   enum radeon_hpd_id hpd)
560 {
561 	u32 tmp;
562 	bool connected = r100_hpd_sense(rdev, hpd);
563 
564 	switch (hpd) {
565 	case RADEON_HPD_1:
566 		tmp = RREG32(RADEON_FP_GEN_CNTL);
567 		if (connected)
568 			tmp &= ~RADEON_FP_DETECT_INT_POL;
569 		else
570 			tmp |= RADEON_FP_DETECT_INT_POL;
571 		WREG32(RADEON_FP_GEN_CNTL, tmp);
572 		break;
573 	case RADEON_HPD_2:
574 		tmp = RREG32(RADEON_FP2_GEN_CNTL);
575 		if (connected)
576 			tmp &= ~RADEON_FP2_DETECT_INT_POL;
577 		else
578 			tmp |= RADEON_FP2_DETECT_INT_POL;
579 		WREG32(RADEON_FP2_GEN_CNTL, tmp);
580 		break;
581 	default:
582 		break;
583 	}
584 }
585 
586 /**
587  * r100_hpd_init - hpd setup callback.
588  *
589  * @rdev: radeon_device pointer
590  *
591  * Setup the hpd pins used by the card (r1xx-r4xx).
592  * Set the polarity, and enable the hpd interrupts.
593  */
594 void r100_hpd_init(struct radeon_device *rdev)
595 {
596 	struct drm_device *dev = rdev->ddev;
597 	struct drm_connector *connector;
598 	unsigned enable = 0;
599 
600 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
601 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
602 		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
603 			enable |= 1 << radeon_connector->hpd.hpd;
604 		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
605 	}
606 	radeon_irq_kms_enable_hpd(rdev, enable);
607 }
608 
609 /**
610  * r100_hpd_fini - hpd tear down callback.
611  *
612  * @rdev: radeon_device pointer
613  *
614  * Tear down the hpd pins used by the card (r1xx-r4xx).
615  * Disable the hpd interrupts.
616  */
617 void r100_hpd_fini(struct radeon_device *rdev)
618 {
619 	struct drm_device *dev = rdev->ddev;
620 	struct drm_connector *connector;
621 	unsigned disable = 0;
622 
623 	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
624 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
625 		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
626 			disable |= 1 << radeon_connector->hpd.hpd;
627 	}
628 	radeon_irq_kms_disable_hpd(rdev, disable);
629 }
630 
631 /*
632  * PCI GART
633  */
634 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
635 {
636 	/* TODO: can we do somethings here ? */
637 	/* It seems hw only cache one entry so we should discard this
638 	 * entry otherwise if first GPU GART read hit this entry it
639 	 * could end up in wrong address. */
640 }
641 
642 int r100_pci_gart_init(struct radeon_device *rdev)
643 {
644 	int r;
645 
646 	if (rdev->gart.ptr) {
647 		WARN(1, "R100 PCI GART already initialized\n");
648 		return 0;
649 	}
650 	/* Initialize common gart structure */
651 	r = radeon_gart_init(rdev);
652 	if (r)
653 		return r;
654 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
655 	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
656 	rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
657 	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
658 	return radeon_gart_table_ram_alloc(rdev);
659 }
660 
661 int r100_pci_gart_enable(struct radeon_device *rdev)
662 {
663 	uint32_t tmp;
664 
665 	/* discard memory request outside of configured range */
666 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
667 	WREG32(RADEON_AIC_CNTL, tmp);
668 	/* set address range for PCI address translate */
669 	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
670 	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
671 	/* set PCI GART page-table base address */
672 	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
673 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
674 	WREG32(RADEON_AIC_CNTL, tmp);
675 	r100_pci_gart_tlb_flush(rdev);
676 	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
677 		 (unsigned)(rdev->mc.gtt_size >> 20),
678 		 (unsigned long long)rdev->gart.table_addr);
679 	rdev->gart.ready = true;
680 	return 0;
681 }
682 
683 void r100_pci_gart_disable(struct radeon_device *rdev)
684 {
685 	uint32_t tmp;
686 
687 	/* discard memory request outside of configured range */
688 	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
689 	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
690 	WREG32(RADEON_AIC_LO_ADDR, 0);
691 	WREG32(RADEON_AIC_HI_ADDR, 0);
692 }
693 
694 uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
695 {
696 	return addr;
697 }
698 
699 void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
700 			    uint64_t entry)
701 {
702 	u32 *gtt = rdev->gart.ptr;
703 	gtt[i] = cpu_to_le32(lower_32_bits(entry));
704 }
705 
706 void r100_pci_gart_fini(struct radeon_device *rdev)
707 {
708 	radeon_gart_fini(rdev);
709 	r100_pci_gart_disable(rdev);
710 	radeon_gart_table_ram_free(rdev);
711 }
712 
713 int r100_irq_set(struct radeon_device *rdev)
714 {
715 	uint32_t tmp = 0;
716 
717 	if (!rdev->irq.installed) {
718 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
719 		WREG32(R_000040_GEN_INT_CNTL, 0);
720 		return -EINVAL;
721 	}
722 	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
723 		tmp |= RADEON_SW_INT_ENABLE;
724 	}
725 	if (rdev->irq.crtc_vblank_int[0] ||
726 	    atomic_read(&rdev->irq.pflip[0])) {
727 		tmp |= RADEON_CRTC_VBLANK_MASK;
728 	}
729 	if (rdev->irq.crtc_vblank_int[1] ||
730 	    atomic_read(&rdev->irq.pflip[1])) {
731 		tmp |= RADEON_CRTC2_VBLANK_MASK;
732 	}
733 	if (rdev->irq.hpd[0]) {
734 		tmp |= RADEON_FP_DETECT_MASK;
735 	}
736 	if (rdev->irq.hpd[1]) {
737 		tmp |= RADEON_FP2_DETECT_MASK;
738 	}
739 	WREG32(RADEON_GEN_INT_CNTL, tmp);
740 
741 	/* read back to post the write */
742 	RREG32(RADEON_GEN_INT_CNTL);
743 
744 	return 0;
745 }
746 
747 void r100_irq_disable(struct radeon_device *rdev)
748 {
749 	u32 tmp;
750 
751 	WREG32(R_000040_GEN_INT_CNTL, 0);
752 	/* Wait and acknowledge irq */
753 	mdelay(1);
754 	tmp = RREG32(R_000044_GEN_INT_STATUS);
755 	WREG32(R_000044_GEN_INT_STATUS, tmp);
756 }
757 
758 static uint32_t r100_irq_ack(struct radeon_device *rdev)
759 {
760 	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
761 	uint32_t irq_mask = RADEON_SW_INT_TEST |
762 		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
763 		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
764 
765 	if (irqs) {
766 		WREG32(RADEON_GEN_INT_STATUS, irqs);
767 	}
768 	return irqs & irq_mask;
769 }
770 
771 int r100_irq_process(struct radeon_device *rdev)
772 {
773 	uint32_t status, msi_rearm;
774 	bool queue_hotplug = false;
775 
776 	status = r100_irq_ack(rdev);
777 	if (!status) {
778 		return IRQ_NONE;
779 	}
780 	if (rdev->shutdown) {
781 		return IRQ_NONE;
782 	}
783 	while (status) {
784 		/* SW interrupt */
785 		if (status & RADEON_SW_INT_TEST) {
786 			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
787 		}
788 		/* Vertical blank interrupts */
789 		if (status & RADEON_CRTC_VBLANK_STAT) {
790 			if (rdev->irq.crtc_vblank_int[0]) {
791 				drm_handle_vblank(rdev->ddev, 0);
792 				rdev->pm.vblank_sync = true;
793 				wake_up(&rdev->irq.vblank_queue);
794 			}
795 			if (atomic_read(&rdev->irq.pflip[0]))
796 				radeon_crtc_handle_vblank(rdev, 0);
797 		}
798 		if (status & RADEON_CRTC2_VBLANK_STAT) {
799 			if (rdev->irq.crtc_vblank_int[1]) {
800 				drm_handle_vblank(rdev->ddev, 1);
801 				rdev->pm.vblank_sync = true;
802 				wake_up(&rdev->irq.vblank_queue);
803 			}
804 			if (atomic_read(&rdev->irq.pflip[1]))
805 				radeon_crtc_handle_vblank(rdev, 1);
806 		}
807 		if (status & RADEON_FP_DETECT_STAT) {
808 			queue_hotplug = true;
809 			DRM_DEBUG("HPD1\n");
810 		}
811 		if (status & RADEON_FP2_DETECT_STAT) {
812 			queue_hotplug = true;
813 			DRM_DEBUG("HPD2\n");
814 		}
815 		status = r100_irq_ack(rdev);
816 	}
817 	if (queue_hotplug)
818 		schedule_delayed_work(&rdev->hotplug_work, 0);
819 	if (rdev->msi_enabled) {
820 		switch (rdev->family) {
821 		case CHIP_RS400:
822 		case CHIP_RS480:
823 			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
824 			WREG32(RADEON_AIC_CNTL, msi_rearm);
825 			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
826 			break;
827 		default:
828 			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
829 			break;
830 		}
831 	}
832 	return IRQ_HANDLED;
833 }
834 
835 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
836 {
837 	if (crtc == 0)
838 		return RREG32(RADEON_CRTC_CRNT_FRAME);
839 	else
840 		return RREG32(RADEON_CRTC2_CRNT_FRAME);
841 }
842 
843 /**
844  * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
845  * @rdev: radeon device structure
846  * @ring: ring buffer struct for emitting packets
847  */
848 static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
849 {
850 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
851 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
852 				RADEON_HDP_READ_BUFFER_INVALIDATE);
853 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
854 	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
855 }
856 
857 /* Who ever call radeon_fence_emit should call ring_lock and ask
858  * for enough space (today caller are ib schedule and buffer move) */
859 void r100_fence_ring_emit(struct radeon_device *rdev,
860 			  struct radeon_fence *fence)
861 {
862 	struct radeon_ring *ring = &rdev->ring[fence->ring];
863 
864 	/* We have to make sure that caches are flushed before
865 	 * CPU might read something from VRAM. */
866 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
867 	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
868 	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
869 	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
870 	/* Wait until IDLE & CLEAN */
871 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
872 	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
873 	r100_ring_hdp_flush(rdev, ring);
874 	/* Emit fence sequence & fire IRQ */
875 	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
876 	radeon_ring_write(ring, fence->seq);
877 	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
878 	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
879 }
880 
881 bool r100_semaphore_ring_emit(struct radeon_device *rdev,
882 			      struct radeon_ring *ring,
883 			      struct radeon_semaphore *semaphore,
884 			      bool emit_wait)
885 {
886 	/* Unused on older asics, since we don't have semaphores or multiple rings */
887 	BUG();
888 	return false;
889 }
890 
891 struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
892 				    uint64_t src_offset,
893 				    uint64_t dst_offset,
894 				    unsigned num_gpu_pages,
895 				    struct dma_resv *resv)
896 {
897 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
898 	struct radeon_fence *fence;
899 	uint32_t cur_pages;
900 	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
901 	uint32_t pitch;
902 	uint32_t stride_pixels;
903 	unsigned ndw;
904 	int num_loops;
905 	int r = 0;
906 
907 	/* radeon limited to 16k stride */
908 	stride_bytes &= 0x3fff;
909 	/* radeon pitch is /64 */
910 	pitch = stride_bytes / 64;
911 	stride_pixels = stride_bytes / 4;
912 	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
913 
914 	/* Ask for enough room for blit + flush + fence */
915 	ndw = 64 + (10 * num_loops);
916 	r = radeon_ring_lock(rdev, ring, ndw);
917 	if (r) {
918 		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
919 		return ERR_PTR(-EINVAL);
920 	}
921 	while (num_gpu_pages > 0) {
922 		cur_pages = num_gpu_pages;
923 		if (cur_pages > 8191) {
924 			cur_pages = 8191;
925 		}
926 		num_gpu_pages -= cur_pages;
927 
928 		/* pages are in Y direction - height
929 		   page width in X direction - width */
930 		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
931 		radeon_ring_write(ring,
932 				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
933 				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
934 				  RADEON_GMC_SRC_CLIPPING |
935 				  RADEON_GMC_DST_CLIPPING |
936 				  RADEON_GMC_BRUSH_NONE |
937 				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
938 				  RADEON_GMC_SRC_DATATYPE_COLOR |
939 				  RADEON_ROP3_S |
940 				  RADEON_DP_SRC_SOURCE_MEMORY |
941 				  RADEON_GMC_CLR_CMP_CNTL_DIS |
942 				  RADEON_GMC_WR_MSK_DIS);
943 		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
944 		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
945 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
946 		radeon_ring_write(ring, 0);
947 		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
948 		radeon_ring_write(ring, num_gpu_pages);
949 		radeon_ring_write(ring, num_gpu_pages);
950 		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
951 	}
952 	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
953 	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
954 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
955 	radeon_ring_write(ring,
956 			  RADEON_WAIT_2D_IDLECLEAN |
957 			  RADEON_WAIT_HOST_IDLECLEAN |
958 			  RADEON_WAIT_DMA_GUI_IDLE);
959 	r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
960 	if (r) {
961 		radeon_ring_unlock_undo(rdev, ring);
962 		return ERR_PTR(r);
963 	}
964 	radeon_ring_unlock_commit(rdev, ring, false);
965 	return fence;
966 }
967 
968 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
969 {
970 	unsigned i;
971 	u32 tmp;
972 
973 	for (i = 0; i < rdev->usec_timeout; i++) {
974 		tmp = RREG32(R_000E40_RBBM_STATUS);
975 		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
976 			return 0;
977 		}
978 		udelay(1);
979 	}
980 	return -1;
981 }
982 
983 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
984 {
985 	int r;
986 
987 	r = radeon_ring_lock(rdev, ring, 2);
988 	if (r) {
989 		return;
990 	}
991 	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
992 	radeon_ring_write(ring,
993 			  RADEON_ISYNC_ANY2D_IDLE3D |
994 			  RADEON_ISYNC_ANY3D_IDLE2D |
995 			  RADEON_ISYNC_WAIT_IDLEGUI |
996 			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
997 	radeon_ring_unlock_commit(rdev, ring, false);
998 }
999 
1000 
1001 /* Load the microcode for the CP */
1002 static int r100_cp_init_microcode(struct radeon_device *rdev)
1003 {
1004 	const char *fw_name = NULL;
1005 	int err;
1006 
1007 	DRM_DEBUG_KMS("\n");
1008 
1009 	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
1010 	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
1011 	    (rdev->family == CHIP_RS200)) {
1012 		DRM_INFO("Loading R100 Microcode\n");
1013 		fw_name = FIRMWARE_R100;
1014 	} else if ((rdev->family == CHIP_R200) ||
1015 		   (rdev->family == CHIP_RV250) ||
1016 		   (rdev->family == CHIP_RV280) ||
1017 		   (rdev->family == CHIP_RS300)) {
1018 		DRM_INFO("Loading R200 Microcode\n");
1019 		fw_name = FIRMWARE_R200;
1020 	} else if ((rdev->family == CHIP_R300) ||
1021 		   (rdev->family == CHIP_R350) ||
1022 		   (rdev->family == CHIP_RV350) ||
1023 		   (rdev->family == CHIP_RV380) ||
1024 		   (rdev->family == CHIP_RS400) ||
1025 		   (rdev->family == CHIP_RS480)) {
1026 		DRM_INFO("Loading R300 Microcode\n");
1027 		fw_name = FIRMWARE_R300;
1028 	} else if ((rdev->family == CHIP_R420) ||
1029 		   (rdev->family == CHIP_R423) ||
1030 		   (rdev->family == CHIP_RV410)) {
1031 		DRM_INFO("Loading R400 Microcode\n");
1032 		fw_name = FIRMWARE_R420;
1033 	} else if ((rdev->family == CHIP_RS690) ||
1034 		   (rdev->family == CHIP_RS740)) {
1035 		DRM_INFO("Loading RS690/RS740 Microcode\n");
1036 		fw_name = FIRMWARE_RS690;
1037 	} else if (rdev->family == CHIP_RS600) {
1038 		DRM_INFO("Loading RS600 Microcode\n");
1039 		fw_name = FIRMWARE_RS600;
1040 	} else if ((rdev->family == CHIP_RV515) ||
1041 		   (rdev->family == CHIP_R520) ||
1042 		   (rdev->family == CHIP_RV530) ||
1043 		   (rdev->family == CHIP_R580) ||
1044 		   (rdev->family == CHIP_RV560) ||
1045 		   (rdev->family == CHIP_RV570)) {
1046 		DRM_INFO("Loading R500 Microcode\n");
1047 		fw_name = FIRMWARE_R520;
1048 	}
1049 
1050 	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1051 	if (err) {
1052 		pr_err("radeon_cp: Failed to load firmware \"%s\"\n", fw_name);
1053 	} else if (rdev->me_fw->size % 8) {
1054 		pr_err("radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1055 		       rdev->me_fw->size, fw_name);
1056 		err = -EINVAL;
1057 		release_firmware(rdev->me_fw);
1058 		rdev->me_fw = NULL;
1059 	}
1060 	return err;
1061 }
1062 
1063 u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1064 		      struct radeon_ring *ring)
1065 {
1066 	u32 rptr;
1067 
1068 	if (rdev->wb.enabled)
1069 		rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1070 	else
1071 		rptr = RREG32(RADEON_CP_RB_RPTR);
1072 
1073 	return rptr;
1074 }
1075 
1076 u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1077 		      struct radeon_ring *ring)
1078 {
1079 	return RREG32(RADEON_CP_RB_WPTR);
1080 }
1081 
1082 void r100_gfx_set_wptr(struct radeon_device *rdev,
1083 		       struct radeon_ring *ring)
1084 {
1085 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1086 	(void)RREG32(RADEON_CP_RB_WPTR);
1087 }
1088 
1089 static void r100_cp_load_microcode(struct radeon_device *rdev)
1090 {
1091 	const __be32 *fw_data;
1092 	int i, size;
1093 
1094 	if (r100_gui_wait_for_idle(rdev)) {
1095 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1096 	}
1097 
1098 	if (rdev->me_fw) {
1099 		size = rdev->me_fw->size / 4;
1100 		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1101 		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1102 		for (i = 0; i < size; i += 2) {
1103 			WREG32(RADEON_CP_ME_RAM_DATAH,
1104 			       be32_to_cpup(&fw_data[i]));
1105 			WREG32(RADEON_CP_ME_RAM_DATAL,
1106 			       be32_to_cpup(&fw_data[i + 1]));
1107 		}
1108 	}
1109 }
1110 
1111 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1112 {
1113 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1114 	unsigned rb_bufsz;
1115 	unsigned rb_blksz;
1116 	unsigned max_fetch;
1117 	unsigned pre_write_timer;
1118 	unsigned pre_write_limit;
1119 	unsigned indirect2_start;
1120 	unsigned indirect1_start;
1121 	uint32_t tmp;
1122 	int r;
1123 
1124 	if (r100_debugfs_cp_init(rdev)) {
1125 		DRM_ERROR("Failed to register debugfs file for CP !\n");
1126 	}
1127 	if (!rdev->me_fw) {
1128 		r = r100_cp_init_microcode(rdev);
1129 		if (r) {
1130 			DRM_ERROR("Failed to load firmware!\n");
1131 			return r;
1132 		}
1133 	}
1134 
1135 	/* Align ring size */
1136 	rb_bufsz = order_base_2(ring_size / 8);
1137 	ring_size = (1 << (rb_bufsz + 1)) * 4;
1138 	r100_cp_load_microcode(rdev);
1139 	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1140 			     RADEON_CP_PACKET2);
1141 	if (r) {
1142 		return r;
1143 	}
1144 	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1145 	 * the rptr copy in system ram */
1146 	rb_blksz = 9;
1147 	/* cp will read 128bytes at a time (4 dwords) */
1148 	max_fetch = 1;
1149 	ring->align_mask = 16 - 1;
1150 	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1151 	pre_write_timer = 64;
1152 	/* Force CP_RB_WPTR write if written more than one time before the
1153 	 * delay expire
1154 	 */
1155 	pre_write_limit = 0;
1156 	/* Setup the cp cache like this (cache size is 96 dwords) :
1157 	 *	RING		0  to 15
1158 	 *	INDIRECT1	16 to 79
1159 	 *	INDIRECT2	80 to 95
1160 	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1161 	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1162 	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1163 	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1164 	 * so it gets the bigger cache.
1165 	 */
1166 	indirect2_start = 80;
1167 	indirect1_start = 16;
1168 	/* cp setup */
1169 	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1170 	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1171 	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1172 	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1173 #ifdef __BIG_ENDIAN
1174 	tmp |= RADEON_BUF_SWAP_32BIT;
1175 #endif
1176 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1177 
1178 	/* Set ring address */
1179 	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1180 	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1181 	/* Force read & write ptr to 0 */
1182 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1183 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1184 	ring->wptr = 0;
1185 	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1186 
1187 	/* set the wb address whether it's enabled or not */
1188 	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1189 		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1190 	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1191 
1192 	if (rdev->wb.enabled)
1193 		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1194 	else {
1195 		tmp |= RADEON_RB_NO_UPDATE;
1196 		WREG32(R_000770_SCRATCH_UMSK, 0);
1197 	}
1198 
1199 	WREG32(RADEON_CP_RB_CNTL, tmp);
1200 	udelay(10);
1201 	/* Set cp mode to bus mastering & enable cp*/
1202 	WREG32(RADEON_CP_CSQ_MODE,
1203 	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1204 	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1205 	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1206 	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1207 	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1208 
1209 	/* at this point everything should be setup correctly to enable master */
1210 	pci_set_master(rdev->pdev);
1211 
1212 	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1213 	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1214 	if (r) {
1215 		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1216 		return r;
1217 	}
1218 	ring->ready = true;
1219 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1220 
1221 	if (!ring->rptr_save_reg /* not resuming from suspend */
1222 	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1223 		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1224 		if (r) {
1225 			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1226 			ring->rptr_save_reg = 0;
1227 		}
1228 	}
1229 	return 0;
1230 }
1231 
1232 void r100_cp_fini(struct radeon_device *rdev)
1233 {
1234 	if (r100_cp_wait_for_idle(rdev)) {
1235 		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1236 	}
1237 	/* Disable ring */
1238 	r100_cp_disable(rdev);
1239 	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1240 	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1241 	DRM_INFO("radeon: cp finalized\n");
1242 }
1243 
1244 void r100_cp_disable(struct radeon_device *rdev)
1245 {
1246 	/* Disable ring */
1247 	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1248 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1249 	WREG32(RADEON_CP_CSQ_MODE, 0);
1250 	WREG32(RADEON_CP_CSQ_CNTL, 0);
1251 	WREG32(R_000770_SCRATCH_UMSK, 0);
1252 	if (r100_gui_wait_for_idle(rdev)) {
1253 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1254 	}
1255 }
1256 
1257 /*
1258  * CS functions
1259  */
1260 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1261 			    struct radeon_cs_packet *pkt,
1262 			    unsigned idx,
1263 			    unsigned reg)
1264 {
1265 	int r;
1266 	u32 tile_flags = 0;
1267 	u32 tmp;
1268 	struct radeon_bo_list *reloc;
1269 	u32 value;
1270 
1271 	r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1272 	if (r) {
1273 		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1274 			  idx, reg);
1275 		radeon_cs_dump_packet(p, pkt);
1276 		return r;
1277 	}
1278 
1279 	value = radeon_get_ib_value(p, idx);
1280 	tmp = value & 0x003fffff;
1281 	tmp += (((u32)reloc->gpu_offset) >> 10);
1282 
1283 	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1284 		if (reloc->tiling_flags & RADEON_TILING_MACRO)
1285 			tile_flags |= RADEON_DST_TILE_MACRO;
1286 		if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1287 			if (reg == RADEON_SRC_PITCH_OFFSET) {
1288 				DRM_ERROR("Cannot src blit from microtiled surface\n");
1289 				radeon_cs_dump_packet(p, pkt);
1290 				return -EINVAL;
1291 			}
1292 			tile_flags |= RADEON_DST_TILE_MICRO;
1293 		}
1294 
1295 		tmp |= tile_flags;
1296 		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1297 	} else
1298 		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1299 	return 0;
1300 }
1301 
1302 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1303 			     struct radeon_cs_packet *pkt,
1304 			     int idx)
1305 {
1306 	unsigned c, i;
1307 	struct radeon_bo_list *reloc;
1308 	struct r100_cs_track *track;
1309 	int r = 0;
1310 	volatile uint32_t *ib;
1311 	u32 idx_value;
1312 
1313 	ib = p->ib.ptr;
1314 	track = (struct r100_cs_track *)p->track;
1315 	c = radeon_get_ib_value(p, idx++) & 0x1F;
1316 	if (c > 16) {
1317 	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1318 		      pkt->opcode);
1319 	    radeon_cs_dump_packet(p, pkt);
1320 	    return -EINVAL;
1321 	}
1322 	track->num_arrays = c;
1323 	for (i = 0; i < (c - 1); i+=2, idx+=3) {
1324 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1325 		if (r) {
1326 			DRM_ERROR("No reloc for packet3 %d\n",
1327 				  pkt->opcode);
1328 			radeon_cs_dump_packet(p, pkt);
1329 			return r;
1330 		}
1331 		idx_value = radeon_get_ib_value(p, idx);
1332 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1333 
1334 		track->arrays[i + 0].esize = idx_value >> 8;
1335 		track->arrays[i + 0].robj = reloc->robj;
1336 		track->arrays[i + 0].esize &= 0x7F;
1337 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1338 		if (r) {
1339 			DRM_ERROR("No reloc for packet3 %d\n",
1340 				  pkt->opcode);
1341 			radeon_cs_dump_packet(p, pkt);
1342 			return r;
1343 		}
1344 		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
1345 		track->arrays[i + 1].robj = reloc->robj;
1346 		track->arrays[i + 1].esize = idx_value >> 24;
1347 		track->arrays[i + 1].esize &= 0x7F;
1348 	}
1349 	if (c & 1) {
1350 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1351 		if (r) {
1352 			DRM_ERROR("No reloc for packet3 %d\n",
1353 					  pkt->opcode);
1354 			radeon_cs_dump_packet(p, pkt);
1355 			return r;
1356 		}
1357 		idx_value = radeon_get_ib_value(p, idx);
1358 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1359 		track->arrays[i + 0].robj = reloc->robj;
1360 		track->arrays[i + 0].esize = idx_value >> 8;
1361 		track->arrays[i + 0].esize &= 0x7F;
1362 	}
1363 	return r;
1364 }
1365 
1366 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1367 			  struct radeon_cs_packet *pkt,
1368 			  const unsigned *auth, unsigned n,
1369 			  radeon_packet0_check_t check)
1370 {
1371 	unsigned reg;
1372 	unsigned i, j, m;
1373 	unsigned idx;
1374 	int r;
1375 
1376 	idx = pkt->idx + 1;
1377 	reg = pkt->reg;
1378 	/* Check that register fall into register range
1379 	 * determined by the number of entry (n) in the
1380 	 * safe register bitmap.
1381 	 */
1382 	if (pkt->one_reg_wr) {
1383 		if ((reg >> 7) > n) {
1384 			return -EINVAL;
1385 		}
1386 	} else {
1387 		if (((reg + (pkt->count << 2)) >> 7) > n) {
1388 			return -EINVAL;
1389 		}
1390 	}
1391 	for (i = 0; i <= pkt->count; i++, idx++) {
1392 		j = (reg >> 7);
1393 		m = 1 << ((reg >> 2) & 31);
1394 		if (auth[j] & m) {
1395 			r = check(p, pkt, idx, reg);
1396 			if (r) {
1397 				return r;
1398 			}
1399 		}
1400 		if (pkt->one_reg_wr) {
1401 			if (!(auth[j] & m)) {
1402 				break;
1403 			}
1404 		} else {
1405 			reg += 4;
1406 		}
1407 	}
1408 	return 0;
1409 }
1410 
1411 /**
1412  * r100_cs_packet_next_vline() - parse userspace VLINE packet
1413  * @p:		parser structure holding parsing context.
1414  *
1415  * Userspace sends a special sequence for VLINE waits.
1416  * PACKET0 - VLINE_START_END + value
1417  * PACKET0 - WAIT_UNTIL +_value
1418  * RELOC (P3) - crtc_id in reloc.
1419  *
1420  * This function parses this and relocates the VLINE START END
1421  * and WAIT UNTIL packets to the correct crtc.
1422  * It also detects a switched off crtc and nulls out the
1423  * wait in that case.
1424  */
1425 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1426 {
1427 	struct drm_crtc *crtc;
1428 	struct radeon_crtc *radeon_crtc;
1429 	struct radeon_cs_packet p3reloc, waitreloc;
1430 	int crtc_id;
1431 	int r;
1432 	uint32_t header, h_idx, reg;
1433 	volatile uint32_t *ib;
1434 
1435 	ib = p->ib.ptr;
1436 
1437 	/* parse the wait until */
1438 	r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1439 	if (r)
1440 		return r;
1441 
1442 	/* check its a wait until and only 1 count */
1443 	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1444 	    waitreloc.count != 0) {
1445 		DRM_ERROR("vline wait had illegal wait until segment\n");
1446 		return -EINVAL;
1447 	}
1448 
1449 	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1450 		DRM_ERROR("vline wait had illegal wait until\n");
1451 		return -EINVAL;
1452 	}
1453 
1454 	/* jump over the NOP */
1455 	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1456 	if (r)
1457 		return r;
1458 
1459 	h_idx = p->idx - 2;
1460 	p->idx += waitreloc.count + 2;
1461 	p->idx += p3reloc.count + 2;
1462 
1463 	header = radeon_get_ib_value(p, h_idx);
1464 	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1465 	reg = R100_CP_PACKET0_GET_REG(header);
1466 	crtc = drm_crtc_find(p->rdev->ddev, p->filp, crtc_id);
1467 	if (!crtc) {
1468 		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1469 		return -ENOENT;
1470 	}
1471 	radeon_crtc = to_radeon_crtc(crtc);
1472 	crtc_id = radeon_crtc->crtc_id;
1473 
1474 	if (!crtc->enabled) {
1475 		/* if the CRTC isn't enabled - we need to nop out the wait until */
1476 		ib[h_idx + 2] = PACKET2(0);
1477 		ib[h_idx + 3] = PACKET2(0);
1478 	} else if (crtc_id == 1) {
1479 		switch (reg) {
1480 		case AVIVO_D1MODE_VLINE_START_END:
1481 			header &= ~R300_CP_PACKET0_REG_MASK;
1482 			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1483 			break;
1484 		case RADEON_CRTC_GUI_TRIG_VLINE:
1485 			header &= ~R300_CP_PACKET0_REG_MASK;
1486 			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1487 			break;
1488 		default:
1489 			DRM_ERROR("unknown crtc reloc\n");
1490 			return -EINVAL;
1491 		}
1492 		ib[h_idx] = header;
1493 		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1494 	}
1495 
1496 	return 0;
1497 }
1498 
1499 static int r100_get_vtx_size(uint32_t vtx_fmt)
1500 {
1501 	int vtx_size;
1502 	vtx_size = 2;
1503 	/* ordered according to bits in spec */
1504 	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1505 		vtx_size++;
1506 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1507 		vtx_size += 3;
1508 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1509 		vtx_size++;
1510 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1511 		vtx_size++;
1512 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1513 		vtx_size += 3;
1514 	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1515 		vtx_size++;
1516 	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1517 		vtx_size++;
1518 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1519 		vtx_size += 2;
1520 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1521 		vtx_size += 2;
1522 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1523 		vtx_size++;
1524 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1525 		vtx_size += 2;
1526 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1527 		vtx_size++;
1528 	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1529 		vtx_size += 2;
1530 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1531 		vtx_size++;
1532 	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1533 		vtx_size++;
1534 	/* blend weight */
1535 	if (vtx_fmt & (0x7 << 15))
1536 		vtx_size += (vtx_fmt >> 15) & 0x7;
1537 	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1538 		vtx_size += 3;
1539 	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1540 		vtx_size += 2;
1541 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1542 		vtx_size++;
1543 	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1544 		vtx_size++;
1545 	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1546 		vtx_size++;
1547 	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1548 		vtx_size++;
1549 	return vtx_size;
1550 }
1551 
1552 static int r100_packet0_check(struct radeon_cs_parser *p,
1553 			      struct radeon_cs_packet *pkt,
1554 			      unsigned idx, unsigned reg)
1555 {
1556 	struct radeon_bo_list *reloc;
1557 	struct r100_cs_track *track;
1558 	volatile uint32_t *ib;
1559 	uint32_t tmp;
1560 	int r;
1561 	int i, face;
1562 	u32 tile_flags = 0;
1563 	u32 idx_value;
1564 
1565 	ib = p->ib.ptr;
1566 	track = (struct r100_cs_track *)p->track;
1567 
1568 	idx_value = radeon_get_ib_value(p, idx);
1569 
1570 	switch (reg) {
1571 	case RADEON_CRTC_GUI_TRIG_VLINE:
1572 		r = r100_cs_packet_parse_vline(p);
1573 		if (r) {
1574 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1575 				  idx, reg);
1576 			radeon_cs_dump_packet(p, pkt);
1577 			return r;
1578 		}
1579 		break;
1580 		/* FIXME: only allow PACKET3 blit? easier to check for out of
1581 		 * range access */
1582 	case RADEON_DST_PITCH_OFFSET:
1583 	case RADEON_SRC_PITCH_OFFSET:
1584 		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1585 		if (r)
1586 			return r;
1587 		break;
1588 	case RADEON_RB3D_DEPTHOFFSET:
1589 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1590 		if (r) {
1591 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1592 				  idx, reg);
1593 			radeon_cs_dump_packet(p, pkt);
1594 			return r;
1595 		}
1596 		track->zb.robj = reloc->robj;
1597 		track->zb.offset = idx_value;
1598 		track->zb_dirty = true;
1599 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1600 		break;
1601 	case RADEON_RB3D_COLOROFFSET:
1602 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1603 		if (r) {
1604 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1605 				  idx, reg);
1606 			radeon_cs_dump_packet(p, pkt);
1607 			return r;
1608 		}
1609 		track->cb[0].robj = reloc->robj;
1610 		track->cb[0].offset = idx_value;
1611 		track->cb_dirty = true;
1612 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1613 		break;
1614 	case RADEON_PP_TXOFFSET_0:
1615 	case RADEON_PP_TXOFFSET_1:
1616 	case RADEON_PP_TXOFFSET_2:
1617 		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1618 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1619 		if (r) {
1620 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1621 				  idx, reg);
1622 			radeon_cs_dump_packet(p, pkt);
1623 			return r;
1624 		}
1625 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1626 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1627 				tile_flags |= RADEON_TXO_MACRO_TILE;
1628 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1629 				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1630 
1631 			tmp = idx_value & ~(0x7 << 2);
1632 			tmp |= tile_flags;
1633 			ib[idx] = tmp + ((u32)reloc->gpu_offset);
1634 		} else
1635 			ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1636 		track->textures[i].robj = reloc->robj;
1637 		track->tex_dirty = true;
1638 		break;
1639 	case RADEON_PP_CUBIC_OFFSET_T0_0:
1640 	case RADEON_PP_CUBIC_OFFSET_T0_1:
1641 	case RADEON_PP_CUBIC_OFFSET_T0_2:
1642 	case RADEON_PP_CUBIC_OFFSET_T0_3:
1643 	case RADEON_PP_CUBIC_OFFSET_T0_4:
1644 		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1645 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1646 		if (r) {
1647 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1648 				  idx, reg);
1649 			radeon_cs_dump_packet(p, pkt);
1650 			return r;
1651 		}
1652 		track->textures[0].cube_info[i].offset = idx_value;
1653 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1654 		track->textures[0].cube_info[i].robj = reloc->robj;
1655 		track->tex_dirty = true;
1656 		break;
1657 	case RADEON_PP_CUBIC_OFFSET_T1_0:
1658 	case RADEON_PP_CUBIC_OFFSET_T1_1:
1659 	case RADEON_PP_CUBIC_OFFSET_T1_2:
1660 	case RADEON_PP_CUBIC_OFFSET_T1_3:
1661 	case RADEON_PP_CUBIC_OFFSET_T1_4:
1662 		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1663 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1664 		if (r) {
1665 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1666 				  idx, reg);
1667 			radeon_cs_dump_packet(p, pkt);
1668 			return r;
1669 		}
1670 		track->textures[1].cube_info[i].offset = idx_value;
1671 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1672 		track->textures[1].cube_info[i].robj = reloc->robj;
1673 		track->tex_dirty = true;
1674 		break;
1675 	case RADEON_PP_CUBIC_OFFSET_T2_0:
1676 	case RADEON_PP_CUBIC_OFFSET_T2_1:
1677 	case RADEON_PP_CUBIC_OFFSET_T2_2:
1678 	case RADEON_PP_CUBIC_OFFSET_T2_3:
1679 	case RADEON_PP_CUBIC_OFFSET_T2_4:
1680 		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1681 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1682 		if (r) {
1683 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1684 				  idx, reg);
1685 			radeon_cs_dump_packet(p, pkt);
1686 			return r;
1687 		}
1688 		track->textures[2].cube_info[i].offset = idx_value;
1689 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1690 		track->textures[2].cube_info[i].robj = reloc->robj;
1691 		track->tex_dirty = true;
1692 		break;
1693 	case RADEON_RE_WIDTH_HEIGHT:
1694 		track->maxy = ((idx_value >> 16) & 0x7FF);
1695 		track->cb_dirty = true;
1696 		track->zb_dirty = true;
1697 		break;
1698 	case RADEON_RB3D_COLORPITCH:
1699 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1700 		if (r) {
1701 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1702 				  idx, reg);
1703 			radeon_cs_dump_packet(p, pkt);
1704 			return r;
1705 		}
1706 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1707 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1708 				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1709 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1710 				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1711 
1712 			tmp = idx_value & ~(0x7 << 16);
1713 			tmp |= tile_flags;
1714 			ib[idx] = tmp;
1715 		} else
1716 			ib[idx] = idx_value;
1717 
1718 		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1719 		track->cb_dirty = true;
1720 		break;
1721 	case RADEON_RB3D_DEPTHPITCH:
1722 		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1723 		track->zb_dirty = true;
1724 		break;
1725 	case RADEON_RB3D_CNTL:
1726 		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1727 		case 7:
1728 		case 8:
1729 		case 9:
1730 		case 11:
1731 		case 12:
1732 			track->cb[0].cpp = 1;
1733 			break;
1734 		case 3:
1735 		case 4:
1736 		case 15:
1737 			track->cb[0].cpp = 2;
1738 			break;
1739 		case 6:
1740 			track->cb[0].cpp = 4;
1741 			break;
1742 		default:
1743 			DRM_ERROR("Invalid color buffer format (%d) !\n",
1744 				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1745 			return -EINVAL;
1746 		}
1747 		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1748 		track->cb_dirty = true;
1749 		track->zb_dirty = true;
1750 		break;
1751 	case RADEON_RB3D_ZSTENCILCNTL:
1752 		switch (idx_value & 0xf) {
1753 		case 0:
1754 			track->zb.cpp = 2;
1755 			break;
1756 		case 2:
1757 		case 3:
1758 		case 4:
1759 		case 5:
1760 		case 9:
1761 		case 11:
1762 			track->zb.cpp = 4;
1763 			break;
1764 		default:
1765 			break;
1766 		}
1767 		track->zb_dirty = true;
1768 		break;
1769 	case RADEON_RB3D_ZPASS_ADDR:
1770 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1771 		if (r) {
1772 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1773 				  idx, reg);
1774 			radeon_cs_dump_packet(p, pkt);
1775 			return r;
1776 		}
1777 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1778 		break;
1779 	case RADEON_PP_CNTL:
1780 		{
1781 			uint32_t temp = idx_value >> 4;
1782 			for (i = 0; i < track->num_texture; i++)
1783 				track->textures[i].enabled = !!(temp & (1 << i));
1784 			track->tex_dirty = true;
1785 		}
1786 		break;
1787 	case RADEON_SE_VF_CNTL:
1788 		track->vap_vf_cntl = idx_value;
1789 		break;
1790 	case RADEON_SE_VTX_FMT:
1791 		track->vtx_size = r100_get_vtx_size(idx_value);
1792 		break;
1793 	case RADEON_PP_TEX_SIZE_0:
1794 	case RADEON_PP_TEX_SIZE_1:
1795 	case RADEON_PP_TEX_SIZE_2:
1796 		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1797 		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1798 		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1799 		track->tex_dirty = true;
1800 		break;
1801 	case RADEON_PP_TEX_PITCH_0:
1802 	case RADEON_PP_TEX_PITCH_1:
1803 	case RADEON_PP_TEX_PITCH_2:
1804 		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1805 		track->textures[i].pitch = idx_value + 32;
1806 		track->tex_dirty = true;
1807 		break;
1808 	case RADEON_PP_TXFILTER_0:
1809 	case RADEON_PP_TXFILTER_1:
1810 	case RADEON_PP_TXFILTER_2:
1811 		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1812 		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1813 						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1814 		tmp = (idx_value >> 23) & 0x7;
1815 		if (tmp == 2 || tmp == 6)
1816 			track->textures[i].roundup_w = false;
1817 		tmp = (idx_value >> 27) & 0x7;
1818 		if (tmp == 2 || tmp == 6)
1819 			track->textures[i].roundup_h = false;
1820 		track->tex_dirty = true;
1821 		break;
1822 	case RADEON_PP_TXFORMAT_0:
1823 	case RADEON_PP_TXFORMAT_1:
1824 	case RADEON_PP_TXFORMAT_2:
1825 		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1826 		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1827 			track->textures[i].use_pitch = true;
1828 		} else {
1829 			track->textures[i].use_pitch = false;
1830 			track->textures[i].width = 1 << ((idx_value & RADEON_TXFORMAT_WIDTH_MASK) >> RADEON_TXFORMAT_WIDTH_SHIFT);
1831 			track->textures[i].height = 1 << ((idx_value & RADEON_TXFORMAT_HEIGHT_MASK) >> RADEON_TXFORMAT_HEIGHT_SHIFT);
1832 		}
1833 		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1834 			track->textures[i].tex_coord_type = 2;
1835 		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1836 		case RADEON_TXFORMAT_I8:
1837 		case RADEON_TXFORMAT_RGB332:
1838 		case RADEON_TXFORMAT_Y8:
1839 			track->textures[i].cpp = 1;
1840 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1841 			break;
1842 		case RADEON_TXFORMAT_AI88:
1843 		case RADEON_TXFORMAT_ARGB1555:
1844 		case RADEON_TXFORMAT_RGB565:
1845 		case RADEON_TXFORMAT_ARGB4444:
1846 		case RADEON_TXFORMAT_VYUY422:
1847 		case RADEON_TXFORMAT_YVYU422:
1848 		case RADEON_TXFORMAT_SHADOW16:
1849 		case RADEON_TXFORMAT_LDUDV655:
1850 		case RADEON_TXFORMAT_DUDV88:
1851 			track->textures[i].cpp = 2;
1852 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1853 			break;
1854 		case RADEON_TXFORMAT_ARGB8888:
1855 		case RADEON_TXFORMAT_RGBA8888:
1856 		case RADEON_TXFORMAT_SHADOW32:
1857 		case RADEON_TXFORMAT_LDUDUV8888:
1858 			track->textures[i].cpp = 4;
1859 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1860 			break;
1861 		case RADEON_TXFORMAT_DXT1:
1862 			track->textures[i].cpp = 1;
1863 			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1864 			break;
1865 		case RADEON_TXFORMAT_DXT23:
1866 		case RADEON_TXFORMAT_DXT45:
1867 			track->textures[i].cpp = 1;
1868 			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1869 			break;
1870 		}
1871 		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1872 		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1873 		track->tex_dirty = true;
1874 		break;
1875 	case RADEON_PP_CUBIC_FACES_0:
1876 	case RADEON_PP_CUBIC_FACES_1:
1877 	case RADEON_PP_CUBIC_FACES_2:
1878 		tmp = idx_value;
1879 		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1880 		for (face = 0; face < 4; face++) {
1881 			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1882 			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1883 		}
1884 		track->tex_dirty = true;
1885 		break;
1886 	default:
1887 		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
1888 		return -EINVAL;
1889 	}
1890 	return 0;
1891 }
1892 
1893 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1894 					 struct radeon_cs_packet *pkt,
1895 					 struct radeon_bo *robj)
1896 {
1897 	unsigned idx;
1898 	u32 value;
1899 	idx = pkt->idx + 1;
1900 	value = radeon_get_ib_value(p, idx + 2);
1901 	if ((value + 1) > radeon_bo_size(robj)) {
1902 		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1903 			  "(need %u have %lu) !\n",
1904 			  value + 1,
1905 			  radeon_bo_size(robj));
1906 		return -EINVAL;
1907 	}
1908 	return 0;
1909 }
1910 
1911 static int r100_packet3_check(struct radeon_cs_parser *p,
1912 			      struct radeon_cs_packet *pkt)
1913 {
1914 	struct radeon_bo_list *reloc;
1915 	struct r100_cs_track *track;
1916 	unsigned idx;
1917 	volatile uint32_t *ib;
1918 	int r;
1919 
1920 	ib = p->ib.ptr;
1921 	idx = pkt->idx + 1;
1922 	track = (struct r100_cs_track *)p->track;
1923 	switch (pkt->opcode) {
1924 	case PACKET3_3D_LOAD_VBPNTR:
1925 		r = r100_packet3_load_vbpntr(p, pkt, idx);
1926 		if (r)
1927 			return r;
1928 		break;
1929 	case PACKET3_INDX_BUFFER:
1930 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1931 		if (r) {
1932 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1933 			radeon_cs_dump_packet(p, pkt);
1934 			return r;
1935 		}
1936 		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1937 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1938 		if (r) {
1939 			return r;
1940 		}
1941 		break;
1942 	case 0x23:
1943 		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1944 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1945 		if (r) {
1946 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1947 			radeon_cs_dump_packet(p, pkt);
1948 			return r;
1949 		}
1950 		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1951 		track->num_arrays = 1;
1952 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1953 
1954 		track->arrays[0].robj = reloc->robj;
1955 		track->arrays[0].esize = track->vtx_size;
1956 
1957 		track->max_indx = radeon_get_ib_value(p, idx+1);
1958 
1959 		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1960 		track->immd_dwords = pkt->count - 1;
1961 		r = r100_cs_track_check(p->rdev, track);
1962 		if (r)
1963 			return r;
1964 		break;
1965 	case PACKET3_3D_DRAW_IMMD:
1966 		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1967 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1968 			return -EINVAL;
1969 		}
1970 		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1971 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1972 		track->immd_dwords = pkt->count - 1;
1973 		r = r100_cs_track_check(p->rdev, track);
1974 		if (r)
1975 			return r;
1976 		break;
1977 		/* triggers drawing using in-packet vertex data */
1978 	case PACKET3_3D_DRAW_IMMD_2:
1979 		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1980 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1981 			return -EINVAL;
1982 		}
1983 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1984 		track->immd_dwords = pkt->count;
1985 		r = r100_cs_track_check(p->rdev, track);
1986 		if (r)
1987 			return r;
1988 		break;
1989 		/* triggers drawing using in-packet vertex data */
1990 	case PACKET3_3D_DRAW_VBUF_2:
1991 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1992 		r = r100_cs_track_check(p->rdev, track);
1993 		if (r)
1994 			return r;
1995 		break;
1996 		/* triggers drawing of vertex buffers setup elsewhere */
1997 	case PACKET3_3D_DRAW_INDX_2:
1998 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1999 		r = r100_cs_track_check(p->rdev, track);
2000 		if (r)
2001 			return r;
2002 		break;
2003 		/* triggers drawing using indices to vertex buffer */
2004 	case PACKET3_3D_DRAW_VBUF:
2005 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2006 		r = r100_cs_track_check(p->rdev, track);
2007 		if (r)
2008 			return r;
2009 		break;
2010 		/* triggers drawing of vertex buffers setup elsewhere */
2011 	case PACKET3_3D_DRAW_INDX:
2012 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2013 		r = r100_cs_track_check(p->rdev, track);
2014 		if (r)
2015 			return r;
2016 		break;
2017 		/* triggers drawing using indices to vertex buffer */
2018 	case PACKET3_3D_CLEAR_HIZ:
2019 	case PACKET3_3D_CLEAR_ZMASK:
2020 		if (p->rdev->hyperz_filp != p->filp)
2021 			return -EINVAL;
2022 		break;
2023 	case PACKET3_NOP:
2024 		break;
2025 	default:
2026 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2027 		return -EINVAL;
2028 	}
2029 	return 0;
2030 }
2031 
2032 int r100_cs_parse(struct radeon_cs_parser *p)
2033 {
2034 	struct radeon_cs_packet pkt;
2035 	struct r100_cs_track *track;
2036 	int r;
2037 
2038 	track = kzalloc(sizeof(*track), GFP_KERNEL);
2039 	if (!track)
2040 		return -ENOMEM;
2041 	r100_cs_track_clear(p->rdev, track);
2042 	p->track = track;
2043 	do {
2044 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2045 		if (r) {
2046 			return r;
2047 		}
2048 		p->idx += pkt.count + 2;
2049 		switch (pkt.type) {
2050 		case RADEON_PACKET_TYPE0:
2051 			if (p->rdev->family >= CHIP_R200)
2052 				r = r100_cs_parse_packet0(p, &pkt,
2053 					p->rdev->config.r100.reg_safe_bm,
2054 					p->rdev->config.r100.reg_safe_bm_size,
2055 					&r200_packet0_check);
2056 			else
2057 				r = r100_cs_parse_packet0(p, &pkt,
2058 					p->rdev->config.r100.reg_safe_bm,
2059 					p->rdev->config.r100.reg_safe_bm_size,
2060 					&r100_packet0_check);
2061 			break;
2062 		case RADEON_PACKET_TYPE2:
2063 			break;
2064 		case RADEON_PACKET_TYPE3:
2065 			r = r100_packet3_check(p, &pkt);
2066 			break;
2067 		default:
2068 			DRM_ERROR("Unknown packet type %d !\n",
2069 				  pkt.type);
2070 			return -EINVAL;
2071 		}
2072 		if (r)
2073 			return r;
2074 	} while (p->idx < p->chunk_ib->length_dw);
2075 	return 0;
2076 }
2077 
2078 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2079 {
2080 	DRM_ERROR("pitch                      %d\n", t->pitch);
2081 	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2082 	DRM_ERROR("width                      %d\n", t->width);
2083 	DRM_ERROR("width_11                   %d\n", t->width_11);
2084 	DRM_ERROR("height                     %d\n", t->height);
2085 	DRM_ERROR("height_11                  %d\n", t->height_11);
2086 	DRM_ERROR("num levels                 %d\n", t->num_levels);
2087 	DRM_ERROR("depth                      %d\n", t->txdepth);
2088 	DRM_ERROR("bpp                        %d\n", t->cpp);
2089 	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2090 	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2091 	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2092 	DRM_ERROR("compress format            %d\n", t->compress_format);
2093 }
2094 
2095 static int r100_track_compress_size(int compress_format, int w, int h)
2096 {
2097 	int block_width, block_height, block_bytes;
2098 	int wblocks, hblocks;
2099 	int min_wblocks;
2100 	int sz;
2101 
2102 	block_width = 4;
2103 	block_height = 4;
2104 
2105 	switch (compress_format) {
2106 	case R100_TRACK_COMP_DXT1:
2107 		block_bytes = 8;
2108 		min_wblocks = 4;
2109 		break;
2110 	default:
2111 	case R100_TRACK_COMP_DXT35:
2112 		block_bytes = 16;
2113 		min_wblocks = 2;
2114 		break;
2115 	}
2116 
2117 	hblocks = (h + block_height - 1) / block_height;
2118 	wblocks = (w + block_width - 1) / block_width;
2119 	if (wblocks < min_wblocks)
2120 		wblocks = min_wblocks;
2121 	sz = wblocks * hblocks * block_bytes;
2122 	return sz;
2123 }
2124 
2125 static int r100_cs_track_cube(struct radeon_device *rdev,
2126 			      struct r100_cs_track *track, unsigned idx)
2127 {
2128 	unsigned face, w, h;
2129 	struct radeon_bo *cube_robj;
2130 	unsigned long size;
2131 	unsigned compress_format = track->textures[idx].compress_format;
2132 
2133 	for (face = 0; face < 5; face++) {
2134 		cube_robj = track->textures[idx].cube_info[face].robj;
2135 		w = track->textures[idx].cube_info[face].width;
2136 		h = track->textures[idx].cube_info[face].height;
2137 
2138 		if (compress_format) {
2139 			size = r100_track_compress_size(compress_format, w, h);
2140 		} else
2141 			size = w * h;
2142 		size *= track->textures[idx].cpp;
2143 
2144 		size += track->textures[idx].cube_info[face].offset;
2145 
2146 		if (size > radeon_bo_size(cube_robj)) {
2147 			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2148 				  size, radeon_bo_size(cube_robj));
2149 			r100_cs_track_texture_print(&track->textures[idx]);
2150 			return -1;
2151 		}
2152 	}
2153 	return 0;
2154 }
2155 
2156 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2157 				       struct r100_cs_track *track)
2158 {
2159 	struct radeon_bo *robj;
2160 	unsigned long size;
2161 	unsigned u, i, w, h, d;
2162 	int ret;
2163 
2164 	for (u = 0; u < track->num_texture; u++) {
2165 		if (!track->textures[u].enabled)
2166 			continue;
2167 		if (track->textures[u].lookup_disable)
2168 			continue;
2169 		robj = track->textures[u].robj;
2170 		if (robj == NULL) {
2171 			DRM_ERROR("No texture bound to unit %u\n", u);
2172 			return -EINVAL;
2173 		}
2174 		size = 0;
2175 		for (i = 0; i <= track->textures[u].num_levels; i++) {
2176 			if (track->textures[u].use_pitch) {
2177 				if (rdev->family < CHIP_R300)
2178 					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2179 				else
2180 					w = track->textures[u].pitch / (1 << i);
2181 			} else {
2182 				w = track->textures[u].width;
2183 				if (rdev->family >= CHIP_RV515)
2184 					w |= track->textures[u].width_11;
2185 				w = w / (1 << i);
2186 				if (track->textures[u].roundup_w)
2187 					w = roundup_pow_of_two(w);
2188 			}
2189 			h = track->textures[u].height;
2190 			if (rdev->family >= CHIP_RV515)
2191 				h |= track->textures[u].height_11;
2192 			h = h / (1 << i);
2193 			if (track->textures[u].roundup_h)
2194 				h = roundup_pow_of_two(h);
2195 			if (track->textures[u].tex_coord_type == 1) {
2196 				d = (1 << track->textures[u].txdepth) / (1 << i);
2197 				if (!d)
2198 					d = 1;
2199 			} else {
2200 				d = 1;
2201 			}
2202 			if (track->textures[u].compress_format) {
2203 
2204 				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2205 				/* compressed textures are block based */
2206 			} else
2207 				size += w * h * d;
2208 		}
2209 		size *= track->textures[u].cpp;
2210 
2211 		switch (track->textures[u].tex_coord_type) {
2212 		case 0:
2213 		case 1:
2214 			break;
2215 		case 2:
2216 			if (track->separate_cube) {
2217 				ret = r100_cs_track_cube(rdev, track, u);
2218 				if (ret)
2219 					return ret;
2220 			} else
2221 				size *= 6;
2222 			break;
2223 		default:
2224 			DRM_ERROR("Invalid texture coordinate type %u for unit "
2225 				  "%u\n", track->textures[u].tex_coord_type, u);
2226 			return -EINVAL;
2227 		}
2228 		if (size > radeon_bo_size(robj)) {
2229 			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2230 				  "%lu\n", u, size, radeon_bo_size(robj));
2231 			r100_cs_track_texture_print(&track->textures[u]);
2232 			return -EINVAL;
2233 		}
2234 	}
2235 	return 0;
2236 }
2237 
2238 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2239 {
2240 	unsigned i;
2241 	unsigned long size;
2242 	unsigned prim_walk;
2243 	unsigned nverts;
2244 	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2245 
2246 	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2247 	    !track->blend_read_enable)
2248 		num_cb = 0;
2249 
2250 	for (i = 0; i < num_cb; i++) {
2251 		if (track->cb[i].robj == NULL) {
2252 			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2253 			return -EINVAL;
2254 		}
2255 		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2256 		size += track->cb[i].offset;
2257 		if (size > radeon_bo_size(track->cb[i].robj)) {
2258 			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2259 				  "(need %lu have %lu) !\n", i, size,
2260 				  radeon_bo_size(track->cb[i].robj));
2261 			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2262 				  i, track->cb[i].pitch, track->cb[i].cpp,
2263 				  track->cb[i].offset, track->maxy);
2264 			return -EINVAL;
2265 		}
2266 	}
2267 	track->cb_dirty = false;
2268 
2269 	if (track->zb_dirty && track->z_enabled) {
2270 		if (track->zb.robj == NULL) {
2271 			DRM_ERROR("[drm] No buffer for z buffer !\n");
2272 			return -EINVAL;
2273 		}
2274 		size = track->zb.pitch * track->zb.cpp * track->maxy;
2275 		size += track->zb.offset;
2276 		if (size > radeon_bo_size(track->zb.robj)) {
2277 			DRM_ERROR("[drm] Buffer too small for z buffer "
2278 				  "(need %lu have %lu) !\n", size,
2279 				  radeon_bo_size(track->zb.robj));
2280 			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2281 				  track->zb.pitch, track->zb.cpp,
2282 				  track->zb.offset, track->maxy);
2283 			return -EINVAL;
2284 		}
2285 	}
2286 	track->zb_dirty = false;
2287 
2288 	if (track->aa_dirty && track->aaresolve) {
2289 		if (track->aa.robj == NULL) {
2290 			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2291 			return -EINVAL;
2292 		}
2293 		/* I believe the format comes from colorbuffer0. */
2294 		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2295 		size += track->aa.offset;
2296 		if (size > radeon_bo_size(track->aa.robj)) {
2297 			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2298 				  "(need %lu have %lu) !\n", i, size,
2299 				  radeon_bo_size(track->aa.robj));
2300 			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2301 				  i, track->aa.pitch, track->cb[0].cpp,
2302 				  track->aa.offset, track->maxy);
2303 			return -EINVAL;
2304 		}
2305 	}
2306 	track->aa_dirty = false;
2307 
2308 	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2309 	if (track->vap_vf_cntl & (1 << 14)) {
2310 		nverts = track->vap_alt_nverts;
2311 	} else {
2312 		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2313 	}
2314 	switch (prim_walk) {
2315 	case 1:
2316 		for (i = 0; i < track->num_arrays; i++) {
2317 			size = track->arrays[i].esize * track->max_indx * 4;
2318 			if (track->arrays[i].robj == NULL) {
2319 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2320 					  "bound\n", prim_walk, i);
2321 				return -EINVAL;
2322 			}
2323 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2324 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2325 					"need %lu dwords have %lu dwords\n",
2326 					prim_walk, i, size >> 2,
2327 					radeon_bo_size(track->arrays[i].robj)
2328 					>> 2);
2329 				DRM_ERROR("Max indices %u\n", track->max_indx);
2330 				return -EINVAL;
2331 			}
2332 		}
2333 		break;
2334 	case 2:
2335 		for (i = 0; i < track->num_arrays; i++) {
2336 			size = track->arrays[i].esize * (nverts - 1) * 4;
2337 			if (track->arrays[i].robj == NULL) {
2338 				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2339 					  "bound\n", prim_walk, i);
2340 				return -EINVAL;
2341 			}
2342 			if (size > radeon_bo_size(track->arrays[i].robj)) {
2343 				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2344 					"need %lu dwords have %lu dwords\n",
2345 					prim_walk, i, size >> 2,
2346 					radeon_bo_size(track->arrays[i].robj)
2347 					>> 2);
2348 				return -EINVAL;
2349 			}
2350 		}
2351 		break;
2352 	case 3:
2353 		size = track->vtx_size * nverts;
2354 		if (size != track->immd_dwords) {
2355 			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2356 				  track->immd_dwords, size);
2357 			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2358 				  nverts, track->vtx_size);
2359 			return -EINVAL;
2360 		}
2361 		break;
2362 	default:
2363 		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2364 			  prim_walk);
2365 		return -EINVAL;
2366 	}
2367 
2368 	if (track->tex_dirty) {
2369 		track->tex_dirty = false;
2370 		return r100_cs_track_texture_check(rdev, track);
2371 	}
2372 	return 0;
2373 }
2374 
2375 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2376 {
2377 	unsigned i, face;
2378 
2379 	track->cb_dirty = true;
2380 	track->zb_dirty = true;
2381 	track->tex_dirty = true;
2382 	track->aa_dirty = true;
2383 
2384 	if (rdev->family < CHIP_R300) {
2385 		track->num_cb = 1;
2386 		if (rdev->family <= CHIP_RS200)
2387 			track->num_texture = 3;
2388 		else
2389 			track->num_texture = 6;
2390 		track->maxy = 2048;
2391 		track->separate_cube = true;
2392 	} else {
2393 		track->num_cb = 4;
2394 		track->num_texture = 16;
2395 		track->maxy = 4096;
2396 		track->separate_cube = false;
2397 		track->aaresolve = false;
2398 		track->aa.robj = NULL;
2399 	}
2400 
2401 	for (i = 0; i < track->num_cb; i++) {
2402 		track->cb[i].robj = NULL;
2403 		track->cb[i].pitch = 8192;
2404 		track->cb[i].cpp = 16;
2405 		track->cb[i].offset = 0;
2406 	}
2407 	track->z_enabled = true;
2408 	track->zb.robj = NULL;
2409 	track->zb.pitch = 8192;
2410 	track->zb.cpp = 4;
2411 	track->zb.offset = 0;
2412 	track->vtx_size = 0x7F;
2413 	track->immd_dwords = 0xFFFFFFFFUL;
2414 	track->num_arrays = 11;
2415 	track->max_indx = 0x00FFFFFFUL;
2416 	for (i = 0; i < track->num_arrays; i++) {
2417 		track->arrays[i].robj = NULL;
2418 		track->arrays[i].esize = 0x7F;
2419 	}
2420 	for (i = 0; i < track->num_texture; i++) {
2421 		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2422 		track->textures[i].pitch = 16536;
2423 		track->textures[i].width = 16536;
2424 		track->textures[i].height = 16536;
2425 		track->textures[i].width_11 = 1 << 11;
2426 		track->textures[i].height_11 = 1 << 11;
2427 		track->textures[i].num_levels = 12;
2428 		if (rdev->family <= CHIP_RS200) {
2429 			track->textures[i].tex_coord_type = 0;
2430 			track->textures[i].txdepth = 0;
2431 		} else {
2432 			track->textures[i].txdepth = 16;
2433 			track->textures[i].tex_coord_type = 1;
2434 		}
2435 		track->textures[i].cpp = 64;
2436 		track->textures[i].robj = NULL;
2437 		/* CS IB emission code makes sure texture unit are disabled */
2438 		track->textures[i].enabled = false;
2439 		track->textures[i].lookup_disable = false;
2440 		track->textures[i].roundup_w = true;
2441 		track->textures[i].roundup_h = true;
2442 		if (track->separate_cube)
2443 			for (face = 0; face < 5; face++) {
2444 				track->textures[i].cube_info[face].robj = NULL;
2445 				track->textures[i].cube_info[face].width = 16536;
2446 				track->textures[i].cube_info[face].height = 16536;
2447 				track->textures[i].cube_info[face].offset = 0;
2448 			}
2449 	}
2450 }
2451 
2452 /*
2453  * Global GPU functions
2454  */
2455 static void r100_errata(struct radeon_device *rdev)
2456 {
2457 	rdev->pll_errata = 0;
2458 
2459 	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2460 		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2461 	}
2462 
2463 	if (rdev->family == CHIP_RV100 ||
2464 	    rdev->family == CHIP_RS100 ||
2465 	    rdev->family == CHIP_RS200) {
2466 		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2467 	}
2468 }
2469 
2470 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2471 {
2472 	unsigned i;
2473 	uint32_t tmp;
2474 
2475 	for (i = 0; i < rdev->usec_timeout; i++) {
2476 		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2477 		if (tmp >= n) {
2478 			return 0;
2479 		}
2480 		udelay(1);
2481 	}
2482 	return -1;
2483 }
2484 
2485 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2486 {
2487 	unsigned i;
2488 	uint32_t tmp;
2489 
2490 	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2491 		pr_warn("radeon: wait for empty RBBM fifo failed! Bad things might happen.\n");
2492 	}
2493 	for (i = 0; i < rdev->usec_timeout; i++) {
2494 		tmp = RREG32(RADEON_RBBM_STATUS);
2495 		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2496 			return 0;
2497 		}
2498 		udelay(1);
2499 	}
2500 	return -1;
2501 }
2502 
2503 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2504 {
2505 	unsigned i;
2506 	uint32_t tmp;
2507 
2508 	for (i = 0; i < rdev->usec_timeout; i++) {
2509 		/* read MC_STATUS */
2510 		tmp = RREG32(RADEON_MC_STATUS);
2511 		if (tmp & RADEON_MC_IDLE) {
2512 			return 0;
2513 		}
2514 		udelay(1);
2515 	}
2516 	return -1;
2517 }
2518 
2519 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2520 {
2521 	u32 rbbm_status;
2522 
2523 	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2524 	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2525 		radeon_ring_lockup_update(rdev, ring);
2526 		return false;
2527 	}
2528 	return radeon_ring_test_lockup(rdev, ring);
2529 }
2530 
2531 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2532 void r100_enable_bm(struct radeon_device *rdev)
2533 {
2534 	uint32_t tmp;
2535 	/* Enable bus mastering */
2536 	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2537 	WREG32(RADEON_BUS_CNTL, tmp);
2538 }
2539 
2540 void r100_bm_disable(struct radeon_device *rdev)
2541 {
2542 	u32 tmp;
2543 
2544 	/* disable bus mastering */
2545 	tmp = RREG32(R_000030_BUS_CNTL);
2546 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2547 	mdelay(1);
2548 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2549 	mdelay(1);
2550 	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2551 	tmp = RREG32(RADEON_BUS_CNTL);
2552 	mdelay(1);
2553 	pci_clear_master(rdev->pdev);
2554 	mdelay(1);
2555 }
2556 
2557 int r100_asic_reset(struct radeon_device *rdev, bool hard)
2558 {
2559 	struct r100_mc_save save;
2560 	u32 status, tmp;
2561 	int ret = 0;
2562 
2563 	status = RREG32(R_000E40_RBBM_STATUS);
2564 	if (!G_000E40_GUI_ACTIVE(status)) {
2565 		return 0;
2566 	}
2567 	r100_mc_stop(rdev, &save);
2568 	status = RREG32(R_000E40_RBBM_STATUS);
2569 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2570 	/* stop CP */
2571 	WREG32(RADEON_CP_CSQ_CNTL, 0);
2572 	tmp = RREG32(RADEON_CP_RB_CNTL);
2573 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2574 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2575 	WREG32(RADEON_CP_RB_WPTR, 0);
2576 	WREG32(RADEON_CP_RB_CNTL, tmp);
2577 	/* save PCI state */
2578 	pci_save_state(rdev->pdev);
2579 	/* disable bus mastering */
2580 	r100_bm_disable(rdev);
2581 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2582 					S_0000F0_SOFT_RESET_RE(1) |
2583 					S_0000F0_SOFT_RESET_PP(1) |
2584 					S_0000F0_SOFT_RESET_RB(1));
2585 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2586 	mdelay(500);
2587 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2588 	mdelay(1);
2589 	status = RREG32(R_000E40_RBBM_STATUS);
2590 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2591 	/* reset CP */
2592 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2593 	RREG32(R_0000F0_RBBM_SOFT_RESET);
2594 	mdelay(500);
2595 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2596 	mdelay(1);
2597 	status = RREG32(R_000E40_RBBM_STATUS);
2598 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2599 	/* restore PCI & busmastering */
2600 	pci_restore_state(rdev->pdev);
2601 	r100_enable_bm(rdev);
2602 	/* Check if GPU is idle */
2603 	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2604 		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2605 		dev_err(rdev->dev, "failed to reset GPU\n");
2606 		ret = -1;
2607 	} else
2608 		dev_info(rdev->dev, "GPU reset succeed\n");
2609 	r100_mc_resume(rdev, &save);
2610 	return ret;
2611 }
2612 
2613 void r100_set_common_regs(struct radeon_device *rdev)
2614 {
2615 	struct drm_device *dev = rdev->ddev;
2616 	bool force_dac2 = false;
2617 	u32 tmp;
2618 
2619 	/* set these so they don't interfere with anything */
2620 	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2621 	WREG32(RADEON_SUBPIC_CNTL, 0);
2622 	WREG32(RADEON_VIPH_CONTROL, 0);
2623 	WREG32(RADEON_I2C_CNTL_1, 0);
2624 	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2625 	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2626 	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2627 
2628 	/* always set up dac2 on rn50 and some rv100 as lots
2629 	 * of servers seem to wire it up to a VGA port but
2630 	 * don't report it in the bios connector
2631 	 * table.
2632 	 */
2633 	switch (dev->pdev->device) {
2634 		/* RN50 */
2635 	case 0x515e:
2636 	case 0x5969:
2637 		force_dac2 = true;
2638 		break;
2639 		/* RV100*/
2640 	case 0x5159:
2641 	case 0x515a:
2642 		/* DELL triple head servers */
2643 		if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2644 		    ((dev->pdev->subsystem_device == 0x016c) ||
2645 		     (dev->pdev->subsystem_device == 0x016d) ||
2646 		     (dev->pdev->subsystem_device == 0x016e) ||
2647 		     (dev->pdev->subsystem_device == 0x016f) ||
2648 		     (dev->pdev->subsystem_device == 0x0170) ||
2649 		     (dev->pdev->subsystem_device == 0x017d) ||
2650 		     (dev->pdev->subsystem_device == 0x017e) ||
2651 		     (dev->pdev->subsystem_device == 0x0183) ||
2652 		     (dev->pdev->subsystem_device == 0x018a) ||
2653 		     (dev->pdev->subsystem_device == 0x019a)))
2654 			force_dac2 = true;
2655 		break;
2656 	}
2657 
2658 	if (force_dac2) {
2659 		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2660 		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2661 		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2662 
2663 		/* For CRT on DAC2, don't turn it on if BIOS didn't
2664 		   enable it, even it's detected.
2665 		*/
2666 
2667 		/* force it to crtc0 */
2668 		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2669 		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2670 		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2671 
2672 		/* set up the TV DAC */
2673 		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2674 				 RADEON_TV_DAC_STD_MASK |
2675 				 RADEON_TV_DAC_RDACPD |
2676 				 RADEON_TV_DAC_GDACPD |
2677 				 RADEON_TV_DAC_BDACPD |
2678 				 RADEON_TV_DAC_BGADJ_MASK |
2679 				 RADEON_TV_DAC_DACADJ_MASK);
2680 		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2681 				RADEON_TV_DAC_NHOLD |
2682 				RADEON_TV_DAC_STD_PS2 |
2683 				(0x58 << 16));
2684 
2685 		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2686 		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2687 		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2688 	}
2689 
2690 	/* switch PM block to ACPI mode */
2691 	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2692 	tmp &= ~RADEON_PM_MODE_SEL;
2693 	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2694 
2695 }
2696 
2697 /*
2698  * VRAM info
2699  */
2700 static void r100_vram_get_type(struct radeon_device *rdev)
2701 {
2702 	uint32_t tmp;
2703 
2704 	rdev->mc.vram_is_ddr = false;
2705 	if (rdev->flags & RADEON_IS_IGP)
2706 		rdev->mc.vram_is_ddr = true;
2707 	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2708 		rdev->mc.vram_is_ddr = true;
2709 	if ((rdev->family == CHIP_RV100) ||
2710 	    (rdev->family == CHIP_RS100) ||
2711 	    (rdev->family == CHIP_RS200)) {
2712 		tmp = RREG32(RADEON_MEM_CNTL);
2713 		if (tmp & RV100_HALF_MODE) {
2714 			rdev->mc.vram_width = 32;
2715 		} else {
2716 			rdev->mc.vram_width = 64;
2717 		}
2718 		if (rdev->flags & RADEON_SINGLE_CRTC) {
2719 			rdev->mc.vram_width /= 4;
2720 			rdev->mc.vram_is_ddr = true;
2721 		}
2722 	} else if (rdev->family <= CHIP_RV280) {
2723 		tmp = RREG32(RADEON_MEM_CNTL);
2724 		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2725 			rdev->mc.vram_width = 128;
2726 		} else {
2727 			rdev->mc.vram_width = 64;
2728 		}
2729 	} else {
2730 		/* newer IGPs */
2731 		rdev->mc.vram_width = 128;
2732 	}
2733 }
2734 
2735 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2736 {
2737 	u32 aper_size;
2738 	u8 byte;
2739 
2740 	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2741 
2742 	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2743 	 * that is has the 2nd generation multifunction PCI interface
2744 	 */
2745 	if (rdev->family == CHIP_RV280 ||
2746 	    rdev->family >= CHIP_RV350) {
2747 		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2748 		       ~RADEON_HDP_APER_CNTL);
2749 		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2750 		return aper_size * 2;
2751 	}
2752 
2753 	/* Older cards have all sorts of funny issues to deal with. First
2754 	 * check if it's a multifunction card by reading the PCI config
2755 	 * header type... Limit those to one aperture size
2756 	 */
2757 	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2758 	if (byte & 0x80) {
2759 		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2760 		DRM_INFO("Limiting VRAM to one aperture\n");
2761 		return aper_size;
2762 	}
2763 
2764 	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2765 	 * have set it up. We don't write this as it's broken on some ASICs but
2766 	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2767 	 */
2768 	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2769 		return aper_size * 2;
2770 	return aper_size;
2771 }
2772 
2773 void r100_vram_init_sizes(struct radeon_device *rdev)
2774 {
2775 	u64 config_aper_size;
2776 
2777 	/* work out accessible VRAM */
2778 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2779 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2780 	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2781 	/* FIXME we don't use the second aperture yet when we could use it */
2782 	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2783 		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2784 	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2785 	if (rdev->flags & RADEON_IS_IGP) {
2786 		uint32_t tom;
2787 		/* read NB_TOM to get the amount of ram stolen for the GPU */
2788 		tom = RREG32(RADEON_NB_TOM);
2789 		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2790 		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2791 		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2792 	} else {
2793 		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2794 		/* Some production boards of m6 will report 0
2795 		 * if it's 8 MB
2796 		 */
2797 		if (rdev->mc.real_vram_size == 0) {
2798 			rdev->mc.real_vram_size = 8192 * 1024;
2799 			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2800 		}
2801 		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2802 		 * Novell bug 204882 + along with lots of ubuntu ones
2803 		 */
2804 		if (rdev->mc.aper_size > config_aper_size)
2805 			config_aper_size = rdev->mc.aper_size;
2806 
2807 		if (config_aper_size > rdev->mc.real_vram_size)
2808 			rdev->mc.mc_vram_size = config_aper_size;
2809 		else
2810 			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2811 	}
2812 }
2813 
2814 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2815 {
2816 	uint32_t temp;
2817 
2818 	temp = RREG32(RADEON_CONFIG_CNTL);
2819 	if (!state) {
2820 		temp &= ~RADEON_CFG_VGA_RAM_EN;
2821 		temp |= RADEON_CFG_VGA_IO_DIS;
2822 	} else {
2823 		temp &= ~RADEON_CFG_VGA_IO_DIS;
2824 	}
2825 	WREG32(RADEON_CONFIG_CNTL, temp);
2826 }
2827 
2828 static void r100_mc_init(struct radeon_device *rdev)
2829 {
2830 	u64 base;
2831 
2832 	r100_vram_get_type(rdev);
2833 	r100_vram_init_sizes(rdev);
2834 	base = rdev->mc.aper_base;
2835 	if (rdev->flags & RADEON_IS_IGP)
2836 		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2837 	radeon_vram_location(rdev, &rdev->mc, base);
2838 	rdev->mc.gtt_base_align = 0;
2839 	if (!(rdev->flags & RADEON_IS_AGP))
2840 		radeon_gtt_location(rdev, &rdev->mc);
2841 	radeon_update_bandwidth_info(rdev);
2842 }
2843 
2844 
2845 /*
2846  * Indirect registers accessor
2847  */
2848 void r100_pll_errata_after_index(struct radeon_device *rdev)
2849 {
2850 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2851 		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2852 		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2853 	}
2854 }
2855 
2856 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2857 {
2858 	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2859 	 * or the chip could hang on a subsequent access
2860 	 */
2861 	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2862 		mdelay(5);
2863 	}
2864 
2865 	/* This function is required to workaround a hardware bug in some (all?)
2866 	 * revisions of the R300.  This workaround should be called after every
2867 	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2868 	 * may not be correct.
2869 	 */
2870 	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2871 		uint32_t save, tmp;
2872 
2873 		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2874 		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2875 		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2876 		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2877 		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2878 	}
2879 }
2880 
2881 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2882 {
2883 	unsigned long flags;
2884 	uint32_t data;
2885 
2886 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2887 	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2888 	r100_pll_errata_after_index(rdev);
2889 	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2890 	r100_pll_errata_after_data(rdev);
2891 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2892 	return data;
2893 }
2894 
2895 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2896 {
2897 	unsigned long flags;
2898 
2899 	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2900 	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2901 	r100_pll_errata_after_index(rdev);
2902 	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2903 	r100_pll_errata_after_data(rdev);
2904 	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2905 }
2906 
2907 static void r100_set_safe_registers(struct radeon_device *rdev)
2908 {
2909 	if (ASIC_IS_RN50(rdev)) {
2910 		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2911 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2912 	} else if (rdev->family < CHIP_R200) {
2913 		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2914 		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2915 	} else {
2916 		r200_set_safe_registers(rdev);
2917 	}
2918 }
2919 
2920 /*
2921  * Debugfs info
2922  */
2923 #if defined(CONFIG_DEBUG_FS)
2924 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2925 {
2926 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2927 	struct drm_device *dev = node->minor->dev;
2928 	struct radeon_device *rdev = dev->dev_private;
2929 	uint32_t reg, value;
2930 	unsigned i;
2931 
2932 	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2933 	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2934 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2935 	for (i = 0; i < 64; i++) {
2936 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2937 		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2938 		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2939 		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2940 		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2941 	}
2942 	return 0;
2943 }
2944 
2945 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2946 {
2947 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2948 	struct drm_device *dev = node->minor->dev;
2949 	struct radeon_device *rdev = dev->dev_private;
2950 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2951 	uint32_t rdp, wdp;
2952 	unsigned count, i, j;
2953 
2954 	radeon_ring_free_size(rdev, ring);
2955 	rdp = RREG32(RADEON_CP_RB_RPTR);
2956 	wdp = RREG32(RADEON_CP_RB_WPTR);
2957 	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2958 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2959 	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2960 	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2961 	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2962 	seq_printf(m, "%u dwords in ring\n", count);
2963 	if (ring->ready) {
2964 		for (j = 0; j <= count; j++) {
2965 			i = (rdp + j) & ring->ptr_mask;
2966 			seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2967 		}
2968 	}
2969 	return 0;
2970 }
2971 
2972 
2973 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2974 {
2975 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2976 	struct drm_device *dev = node->minor->dev;
2977 	struct radeon_device *rdev = dev->dev_private;
2978 	uint32_t csq_stat, csq2_stat, tmp;
2979 	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2980 	unsigned i;
2981 
2982 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2983 	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2984 	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2985 	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2986 	r_rptr = (csq_stat >> 0) & 0x3ff;
2987 	r_wptr = (csq_stat >> 10) & 0x3ff;
2988 	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2989 	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2990 	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2991 	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2992 	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2993 	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2994 	seq_printf(m, "Ring rptr %u\n", r_rptr);
2995 	seq_printf(m, "Ring wptr %u\n", r_wptr);
2996 	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2997 	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2998 	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2999 	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
3000 	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
3001 	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
3002 	seq_printf(m, "Ring fifo:\n");
3003 	for (i = 0; i < 256; i++) {
3004 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3005 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3006 		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3007 	}
3008 	seq_printf(m, "Indirect1 fifo:\n");
3009 	for (i = 256; i <= 512; i++) {
3010 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3011 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3012 		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3013 	}
3014 	seq_printf(m, "Indirect2 fifo:\n");
3015 	for (i = 640; i < ib1_wptr; i++) {
3016 		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3017 		tmp = RREG32(RADEON_CP_CSQ_DATA);
3018 		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3019 	}
3020 	return 0;
3021 }
3022 
3023 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
3024 {
3025 	struct drm_info_node *node = (struct drm_info_node *) m->private;
3026 	struct drm_device *dev = node->minor->dev;
3027 	struct radeon_device *rdev = dev->dev_private;
3028 	uint32_t tmp;
3029 
3030 	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3031 	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3032 	tmp = RREG32(RADEON_MC_FB_LOCATION);
3033 	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3034 	tmp = RREG32(RADEON_BUS_CNTL);
3035 	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3036 	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3037 	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3038 	tmp = RREG32(RADEON_AGP_BASE);
3039 	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3040 	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3041 	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3042 	tmp = RREG32(0x01D0);
3043 	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3044 	tmp = RREG32(RADEON_AIC_LO_ADDR);
3045 	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3046 	tmp = RREG32(RADEON_AIC_HI_ADDR);
3047 	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3048 	tmp = RREG32(0x01E4);
3049 	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3050 	return 0;
3051 }
3052 
3053 static struct drm_info_list r100_debugfs_rbbm_list[] = {
3054 	{"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3055 };
3056 
3057 static struct drm_info_list r100_debugfs_cp_list[] = {
3058 	{"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3059 	{"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3060 };
3061 
3062 static struct drm_info_list r100_debugfs_mc_info_list[] = {
3063 	{"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3064 };
3065 #endif
3066 
3067 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3068 {
3069 #if defined(CONFIG_DEBUG_FS)
3070 	return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3071 #else
3072 	return 0;
3073 #endif
3074 }
3075 
3076 int r100_debugfs_cp_init(struct radeon_device *rdev)
3077 {
3078 #if defined(CONFIG_DEBUG_FS)
3079 	return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3080 #else
3081 	return 0;
3082 #endif
3083 }
3084 
3085 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3086 {
3087 #if defined(CONFIG_DEBUG_FS)
3088 	return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3089 #else
3090 	return 0;
3091 #endif
3092 }
3093 
3094 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3095 			 uint32_t tiling_flags, uint32_t pitch,
3096 			 uint32_t offset, uint32_t obj_size)
3097 {
3098 	int surf_index = reg * 16;
3099 	int flags = 0;
3100 
3101 	if (rdev->family <= CHIP_RS200) {
3102 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3103 				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3104 			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3105 		if (tiling_flags & RADEON_TILING_MACRO)
3106 			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3107 		/* setting pitch to 0 disables tiling */
3108 		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3109 				== 0)
3110 			pitch = 0;
3111 	} else if (rdev->family <= CHIP_RV280) {
3112 		if (tiling_flags & (RADEON_TILING_MACRO))
3113 			flags |= R200_SURF_TILE_COLOR_MACRO;
3114 		if (tiling_flags & RADEON_TILING_MICRO)
3115 			flags |= R200_SURF_TILE_COLOR_MICRO;
3116 	} else {
3117 		if (tiling_flags & RADEON_TILING_MACRO)
3118 			flags |= R300_SURF_TILE_MACRO;
3119 		if (tiling_flags & RADEON_TILING_MICRO)
3120 			flags |= R300_SURF_TILE_MICRO;
3121 	}
3122 
3123 	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3124 		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3125 	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3126 		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3127 
3128 	/* r100/r200 divide by 16 */
3129 	if (rdev->family < CHIP_R300)
3130 		flags |= pitch / 16;
3131 	else
3132 		flags |= pitch / 8;
3133 
3134 
3135 	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3136 	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3137 	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3138 	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3139 	return 0;
3140 }
3141 
3142 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3143 {
3144 	int surf_index = reg * 16;
3145 	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3146 }
3147 
3148 void r100_bandwidth_update(struct radeon_device *rdev)
3149 {
3150 	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3151 	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3152 	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
3153 	fixed20_12 crit_point_ff = {0};
3154 	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3155 	fixed20_12 memtcas_ff[8] = {
3156 		dfixed_init(1),
3157 		dfixed_init(2),
3158 		dfixed_init(3),
3159 		dfixed_init(0),
3160 		dfixed_init_half(1),
3161 		dfixed_init_half(2),
3162 		dfixed_init(0),
3163 	};
3164 	fixed20_12 memtcas_rs480_ff[8] = {
3165 		dfixed_init(0),
3166 		dfixed_init(1),
3167 		dfixed_init(2),
3168 		dfixed_init(3),
3169 		dfixed_init(0),
3170 		dfixed_init_half(1),
3171 		dfixed_init_half(2),
3172 		dfixed_init_half(3),
3173 	};
3174 	fixed20_12 memtcas2_ff[8] = {
3175 		dfixed_init(0),
3176 		dfixed_init(1),
3177 		dfixed_init(2),
3178 		dfixed_init(3),
3179 		dfixed_init(4),
3180 		dfixed_init(5),
3181 		dfixed_init(6),
3182 		dfixed_init(7),
3183 	};
3184 	fixed20_12 memtrbs[8] = {
3185 		dfixed_init(1),
3186 		dfixed_init_half(1),
3187 		dfixed_init(2),
3188 		dfixed_init_half(2),
3189 		dfixed_init(3),
3190 		dfixed_init_half(3),
3191 		dfixed_init(4),
3192 		dfixed_init_half(4)
3193 	};
3194 	fixed20_12 memtrbs_r4xx[8] = {
3195 		dfixed_init(4),
3196 		dfixed_init(5),
3197 		dfixed_init(6),
3198 		dfixed_init(7),
3199 		dfixed_init(8),
3200 		dfixed_init(9),
3201 		dfixed_init(10),
3202 		dfixed_init(11)
3203 	};
3204 	fixed20_12 min_mem_eff;
3205 	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3206 	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3207 	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate = {0},
3208 		disp_drain_rate2, read_return_rate;
3209 	fixed20_12 time_disp1_drop_priority;
3210 	int c;
3211 	int cur_size = 16;       /* in octawords */
3212 	int critical_point = 0, critical_point2;
3213 /* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3214 	int stop_req, max_stop_req;
3215 	struct drm_display_mode *mode1 = NULL;
3216 	struct drm_display_mode *mode2 = NULL;
3217 	uint32_t pixel_bytes1 = 0;
3218 	uint32_t pixel_bytes2 = 0;
3219 
3220 	/* Guess line buffer size to be 8192 pixels */
3221 	u32 lb_size = 8192;
3222 
3223 	if (!rdev->mode_info.mode_config_initialized)
3224 		return;
3225 
3226 	radeon_update_display_priority(rdev);
3227 
3228 	if (rdev->mode_info.crtcs[0]->base.enabled) {
3229 		const struct drm_framebuffer *fb =
3230 			rdev->mode_info.crtcs[0]->base.primary->fb;
3231 
3232 		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3233 		pixel_bytes1 = fb->format->cpp[0];
3234 	}
3235 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3236 		if (rdev->mode_info.crtcs[1]->base.enabled) {
3237 			const struct drm_framebuffer *fb =
3238 				rdev->mode_info.crtcs[1]->base.primary->fb;
3239 
3240 			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3241 			pixel_bytes2 = fb->format->cpp[0];
3242 		}
3243 	}
3244 
3245 	min_mem_eff.full = dfixed_const_8(0);
3246 	/* get modes */
3247 	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3248 		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3249 		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3250 		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3251 		/* check crtc enables */
3252 		if (mode2)
3253 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3254 		if (mode1)
3255 			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3256 		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3257 	}
3258 
3259 	/*
3260 	 * determine is there is enough bw for current mode
3261 	 */
3262 	sclk_ff = rdev->pm.sclk;
3263 	mclk_ff = rdev->pm.mclk;
3264 
3265 	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3266 	temp_ff.full = dfixed_const(temp);
3267 	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3268 
3269 	pix_clk.full = 0;
3270 	pix_clk2.full = 0;
3271 	peak_disp_bw.full = 0;
3272 	if (mode1) {
3273 		temp_ff.full = dfixed_const(1000);
3274 		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3275 		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3276 		temp_ff.full = dfixed_const(pixel_bytes1);
3277 		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3278 	}
3279 	if (mode2) {
3280 		temp_ff.full = dfixed_const(1000);
3281 		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3282 		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3283 		temp_ff.full = dfixed_const(pixel_bytes2);
3284 		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3285 	}
3286 
3287 	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3288 	if (peak_disp_bw.full >= mem_bw.full) {
3289 		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3290 			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3291 	}
3292 
3293 	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3294 	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3295 	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3296 		mem_trcd = ((temp >> 2) & 0x3) + 1;
3297 		mem_trp  = ((temp & 0x3)) + 1;
3298 		mem_tras = ((temp & 0x70) >> 4) + 1;
3299 	} else if (rdev->family == CHIP_R300 ||
3300 		   rdev->family == CHIP_R350) { /* r300, r350 */
3301 		mem_trcd = (temp & 0x7) + 1;
3302 		mem_trp = ((temp >> 8) & 0x7) + 1;
3303 		mem_tras = ((temp >> 11) & 0xf) + 4;
3304 	} else if (rdev->family == CHIP_RV350 ||
3305 		   rdev->family == CHIP_RV380) {
3306 		/* rv3x0 */
3307 		mem_trcd = (temp & 0x7) + 3;
3308 		mem_trp = ((temp >> 8) & 0x7) + 3;
3309 		mem_tras = ((temp >> 11) & 0xf) + 6;
3310 	} else if (rdev->family == CHIP_R420 ||
3311 		   rdev->family == CHIP_R423 ||
3312 		   rdev->family == CHIP_RV410) {
3313 		/* r4xx */
3314 		mem_trcd = (temp & 0xf) + 3;
3315 		if (mem_trcd > 15)
3316 			mem_trcd = 15;
3317 		mem_trp = ((temp >> 8) & 0xf) + 3;
3318 		if (mem_trp > 15)
3319 			mem_trp = 15;
3320 		mem_tras = ((temp >> 12) & 0x1f) + 6;
3321 		if (mem_tras > 31)
3322 			mem_tras = 31;
3323 	} else { /* RV200, R200 */
3324 		mem_trcd = (temp & 0x7) + 1;
3325 		mem_trp = ((temp >> 8) & 0x7) + 1;
3326 		mem_tras = ((temp >> 12) & 0xf) + 4;
3327 	}
3328 	/* convert to FF */
3329 	trcd_ff.full = dfixed_const(mem_trcd);
3330 	trp_ff.full = dfixed_const(mem_trp);
3331 	tras_ff.full = dfixed_const(mem_tras);
3332 
3333 	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3334 	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3335 	data = (temp & (7 << 20)) >> 20;
3336 	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3337 		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3338 			tcas_ff = memtcas_rs480_ff[data];
3339 		else
3340 			tcas_ff = memtcas_ff[data];
3341 	} else
3342 		tcas_ff = memtcas2_ff[data];
3343 
3344 	if (rdev->family == CHIP_RS400 ||
3345 	    rdev->family == CHIP_RS480) {
3346 		/* extra cas latency stored in bits 23-25 0-4 clocks */
3347 		data = (temp >> 23) & 0x7;
3348 		if (data < 5)
3349 			tcas_ff.full += dfixed_const(data);
3350 	}
3351 
3352 	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3353 		/* on the R300, Tcas is included in Trbs.
3354 		 */
3355 		temp = RREG32(RADEON_MEM_CNTL);
3356 		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3357 		if (data == 1) {
3358 			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3359 				temp = RREG32(R300_MC_IND_INDEX);
3360 				temp &= ~R300_MC_IND_ADDR_MASK;
3361 				temp |= R300_MC_READ_CNTL_CD_mcind;
3362 				WREG32(R300_MC_IND_INDEX, temp);
3363 				temp = RREG32(R300_MC_IND_DATA);
3364 				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3365 			} else {
3366 				temp = RREG32(R300_MC_READ_CNTL_AB);
3367 				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3368 			}
3369 		} else {
3370 			temp = RREG32(R300_MC_READ_CNTL_AB);
3371 			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3372 		}
3373 		if (rdev->family == CHIP_RV410 ||
3374 		    rdev->family == CHIP_R420 ||
3375 		    rdev->family == CHIP_R423)
3376 			trbs_ff = memtrbs_r4xx[data];
3377 		else
3378 			trbs_ff = memtrbs[data];
3379 		tcas_ff.full += trbs_ff.full;
3380 	}
3381 
3382 	sclk_eff_ff.full = sclk_ff.full;
3383 
3384 	if (rdev->flags & RADEON_IS_AGP) {
3385 		fixed20_12 agpmode_ff;
3386 		agpmode_ff.full = dfixed_const(radeon_agpmode);
3387 		temp_ff.full = dfixed_const_666(16);
3388 		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3389 	}
3390 	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3391 
3392 	if (ASIC_IS_R300(rdev)) {
3393 		sclk_delay_ff.full = dfixed_const(250);
3394 	} else {
3395 		if ((rdev->family == CHIP_RV100) ||
3396 		    rdev->flags & RADEON_IS_IGP) {
3397 			if (rdev->mc.vram_is_ddr)
3398 				sclk_delay_ff.full = dfixed_const(41);
3399 			else
3400 				sclk_delay_ff.full = dfixed_const(33);
3401 		} else {
3402 			if (rdev->mc.vram_width == 128)
3403 				sclk_delay_ff.full = dfixed_const(57);
3404 			else
3405 				sclk_delay_ff.full = dfixed_const(41);
3406 		}
3407 	}
3408 
3409 	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3410 
3411 	if (rdev->mc.vram_is_ddr) {
3412 		if (rdev->mc.vram_width == 32) {
3413 			k1.full = dfixed_const(40);
3414 			c  = 3;
3415 		} else {
3416 			k1.full = dfixed_const(20);
3417 			c  = 1;
3418 		}
3419 	} else {
3420 		k1.full = dfixed_const(40);
3421 		c  = 3;
3422 	}
3423 
3424 	temp_ff.full = dfixed_const(2);
3425 	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3426 	temp_ff.full = dfixed_const(c);
3427 	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3428 	temp_ff.full = dfixed_const(4);
3429 	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3430 	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3431 	mc_latency_mclk.full += k1.full;
3432 
3433 	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3434 	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3435 
3436 	/*
3437 	  HW cursor time assuming worst case of full size colour cursor.
3438 	*/
3439 	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3440 	temp_ff.full += trcd_ff.full;
3441 	if (temp_ff.full < tras_ff.full)
3442 		temp_ff.full = tras_ff.full;
3443 	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3444 
3445 	temp_ff.full = dfixed_const(cur_size);
3446 	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3447 	/*
3448 	  Find the total latency for the display data.
3449 	*/
3450 	disp_latency_overhead.full = dfixed_const(8);
3451 	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3452 	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3453 	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3454 
3455 	if (mc_latency_mclk.full > mc_latency_sclk.full)
3456 		disp_latency.full = mc_latency_mclk.full;
3457 	else
3458 		disp_latency.full = mc_latency_sclk.full;
3459 
3460 	/* setup Max GRPH_STOP_REQ default value */
3461 	if (ASIC_IS_RV100(rdev))
3462 		max_stop_req = 0x5c;
3463 	else
3464 		max_stop_req = 0x7c;
3465 
3466 	if (mode1) {
3467 		/*  CRTC1
3468 		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3469 		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3470 		*/
3471 		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3472 
3473 		if (stop_req > max_stop_req)
3474 			stop_req = max_stop_req;
3475 
3476 		/*
3477 		  Find the drain rate of the display buffer.
3478 		*/
3479 		temp_ff.full = dfixed_const((16/pixel_bytes1));
3480 		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3481 
3482 		/*
3483 		  Find the critical point of the display buffer.
3484 		*/
3485 		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3486 		crit_point_ff.full += dfixed_const_half(0);
3487 
3488 		critical_point = dfixed_trunc(crit_point_ff);
3489 
3490 		if (rdev->disp_priority == 2) {
3491 			critical_point = 0;
3492 		}
3493 
3494 		/*
3495 		  The critical point should never be above max_stop_req-4.  Setting
3496 		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3497 		*/
3498 		if (max_stop_req - critical_point < 4)
3499 			critical_point = 0;
3500 
3501 		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3502 			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3503 			critical_point = 0x10;
3504 		}
3505 
3506 		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3507 		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3508 		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3509 		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3510 		if ((rdev->family == CHIP_R350) &&
3511 		    (stop_req > 0x15)) {
3512 			stop_req -= 0x10;
3513 		}
3514 		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3515 		temp |= RADEON_GRPH_BUFFER_SIZE;
3516 		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3517 			  RADEON_GRPH_CRITICAL_AT_SOF |
3518 			  RADEON_GRPH_STOP_CNTL);
3519 		/*
3520 		  Write the result into the register.
3521 		*/
3522 		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3523 						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3524 
3525 #if 0
3526 		if ((rdev->family == CHIP_RS400) ||
3527 		    (rdev->family == CHIP_RS480)) {
3528 			/* attempt to program RS400 disp regs correctly ??? */
3529 			temp = RREG32(RS400_DISP1_REG_CNTL);
3530 			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3531 				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3532 			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3533 						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3534 						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3535 			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3536 			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3537 				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3538 			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3539 						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3540 						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3541 		}
3542 #endif
3543 
3544 		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3545 			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3546 			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3547 	}
3548 
3549 	if (mode2) {
3550 		u32 grph2_cntl;
3551 		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3552 
3553 		if (stop_req > max_stop_req)
3554 			stop_req = max_stop_req;
3555 
3556 		/*
3557 		  Find the drain rate of the display buffer.
3558 		*/
3559 		temp_ff.full = dfixed_const((16/pixel_bytes2));
3560 		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3561 
3562 		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3563 		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3564 		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3565 		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3566 		if ((rdev->family == CHIP_R350) &&
3567 		    (stop_req > 0x15)) {
3568 			stop_req -= 0x10;
3569 		}
3570 		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3571 		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3572 		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3573 			  RADEON_GRPH_CRITICAL_AT_SOF |
3574 			  RADEON_GRPH_STOP_CNTL);
3575 
3576 		if ((rdev->family == CHIP_RS100) ||
3577 		    (rdev->family == CHIP_RS200))
3578 			critical_point2 = 0;
3579 		else {
3580 			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3581 			temp_ff.full = dfixed_const(temp);
3582 			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3583 			if (sclk_ff.full < temp_ff.full)
3584 				temp_ff.full = sclk_ff.full;
3585 
3586 			read_return_rate.full = temp_ff.full;
3587 
3588 			if (mode1) {
3589 				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3590 				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3591 			} else {
3592 				time_disp1_drop_priority.full = 0;
3593 			}
3594 			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3595 			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3596 			crit_point_ff.full += dfixed_const_half(0);
3597 
3598 			critical_point2 = dfixed_trunc(crit_point_ff);
3599 
3600 			if (rdev->disp_priority == 2) {
3601 				critical_point2 = 0;
3602 			}
3603 
3604 			if (max_stop_req - critical_point2 < 4)
3605 				critical_point2 = 0;
3606 
3607 		}
3608 
3609 		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3610 			/* some R300 cards have problem with this set to 0 */
3611 			critical_point2 = 0x10;
3612 		}
3613 
3614 		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3615 						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3616 
3617 		if ((rdev->family == CHIP_RS400) ||
3618 		    (rdev->family == CHIP_RS480)) {
3619 #if 0
3620 			/* attempt to program RS400 disp2 regs correctly ??? */
3621 			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3622 			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3623 				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3624 			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3625 						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3626 						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3627 			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3628 			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3629 				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3630 			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3631 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3632 						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3633 #endif
3634 			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3635 			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3636 			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3637 			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3638 		}
3639 
3640 		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3641 			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3642 	}
3643 
3644 	/* Save number of lines the linebuffer leads before the scanout */
3645 	if (mode1)
3646 	    rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
3647 
3648 	if (mode2)
3649 	    rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
3650 }
3651 
3652 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3653 {
3654 	uint32_t scratch;
3655 	uint32_t tmp = 0;
3656 	unsigned i;
3657 	int r;
3658 
3659 	r = radeon_scratch_get(rdev, &scratch);
3660 	if (r) {
3661 		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3662 		return r;
3663 	}
3664 	WREG32(scratch, 0xCAFEDEAD);
3665 	r = radeon_ring_lock(rdev, ring, 2);
3666 	if (r) {
3667 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3668 		radeon_scratch_free(rdev, scratch);
3669 		return r;
3670 	}
3671 	radeon_ring_write(ring, PACKET0(scratch, 0));
3672 	radeon_ring_write(ring, 0xDEADBEEF);
3673 	radeon_ring_unlock_commit(rdev, ring, false);
3674 	for (i = 0; i < rdev->usec_timeout; i++) {
3675 		tmp = RREG32(scratch);
3676 		if (tmp == 0xDEADBEEF) {
3677 			break;
3678 		}
3679 		udelay(1);
3680 	}
3681 	if (i < rdev->usec_timeout) {
3682 		DRM_INFO("ring test succeeded in %d usecs\n", i);
3683 	} else {
3684 		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3685 			  scratch, tmp);
3686 		r = -EINVAL;
3687 	}
3688 	radeon_scratch_free(rdev, scratch);
3689 	return r;
3690 }
3691 
3692 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3693 {
3694 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3695 
3696 	if (ring->rptr_save_reg) {
3697 		u32 next_rptr = ring->wptr + 2 + 3;
3698 		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3699 		radeon_ring_write(ring, next_rptr);
3700 	}
3701 
3702 	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3703 	radeon_ring_write(ring, ib->gpu_addr);
3704 	radeon_ring_write(ring, ib->length_dw);
3705 }
3706 
3707 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3708 {
3709 	struct radeon_ib ib;
3710 	uint32_t scratch;
3711 	uint32_t tmp = 0;
3712 	unsigned i;
3713 	int r;
3714 
3715 	r = radeon_scratch_get(rdev, &scratch);
3716 	if (r) {
3717 		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3718 		return r;
3719 	}
3720 	WREG32(scratch, 0xCAFEDEAD);
3721 	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3722 	if (r) {
3723 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3724 		goto free_scratch;
3725 	}
3726 	ib.ptr[0] = PACKET0(scratch, 0);
3727 	ib.ptr[1] = 0xDEADBEEF;
3728 	ib.ptr[2] = PACKET2(0);
3729 	ib.ptr[3] = PACKET2(0);
3730 	ib.ptr[4] = PACKET2(0);
3731 	ib.ptr[5] = PACKET2(0);
3732 	ib.ptr[6] = PACKET2(0);
3733 	ib.ptr[7] = PACKET2(0);
3734 	ib.length_dw = 8;
3735 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
3736 	if (r) {
3737 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3738 		goto free_ib;
3739 	}
3740 	r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
3741 		RADEON_USEC_IB_TEST_TIMEOUT));
3742 	if (r < 0) {
3743 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3744 		goto free_ib;
3745 	} else if (r == 0) {
3746 		DRM_ERROR("radeon: fence wait timed out.\n");
3747 		r = -ETIMEDOUT;
3748 		goto free_ib;
3749 	}
3750 	r = 0;
3751 	for (i = 0; i < rdev->usec_timeout; i++) {
3752 		tmp = RREG32(scratch);
3753 		if (tmp == 0xDEADBEEF) {
3754 			break;
3755 		}
3756 		udelay(1);
3757 	}
3758 	if (i < rdev->usec_timeout) {
3759 		DRM_INFO("ib test succeeded in %u usecs\n", i);
3760 	} else {
3761 		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3762 			  scratch, tmp);
3763 		r = -EINVAL;
3764 	}
3765 free_ib:
3766 	radeon_ib_free(rdev, &ib);
3767 free_scratch:
3768 	radeon_scratch_free(rdev, scratch);
3769 	return r;
3770 }
3771 
3772 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3773 {
3774 	/* Shutdown CP we shouldn't need to do that but better be safe than
3775 	 * sorry
3776 	 */
3777 	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3778 	WREG32(R_000740_CP_CSQ_CNTL, 0);
3779 
3780 	/* Save few CRTC registers */
3781 	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3782 	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3783 	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3784 	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3785 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3786 		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3787 		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3788 	}
3789 
3790 	/* Disable VGA aperture access */
3791 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3792 	/* Disable cursor, overlay, crtc */
3793 	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3794 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3795 					S_000054_CRTC_DISPLAY_DIS(1));
3796 	WREG32(R_000050_CRTC_GEN_CNTL,
3797 			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3798 			S_000050_CRTC_DISP_REQ_EN_B(1));
3799 	WREG32(R_000420_OV0_SCALE_CNTL,
3800 		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3801 	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3802 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3803 		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3804 						S_000360_CUR2_LOCK(1));
3805 		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3806 			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3807 			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3808 			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3809 		WREG32(R_000360_CUR2_OFFSET,
3810 			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3811 	}
3812 }
3813 
3814 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3815 {
3816 	/* Update base address for crtc */
3817 	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3818 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3819 		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3820 	}
3821 	/* Restore CRTC registers */
3822 	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3823 	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3824 	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3825 	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3826 		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3827 	}
3828 }
3829 
3830 void r100_vga_render_disable(struct radeon_device *rdev)
3831 {
3832 	u32 tmp;
3833 
3834 	tmp = RREG8(R_0003C2_GENMO_WT);
3835 	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3836 }
3837 
3838 static void r100_debugfs(struct radeon_device *rdev)
3839 {
3840 	int r;
3841 
3842 	r = r100_debugfs_mc_info_init(rdev);
3843 	if (r)
3844 		dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3845 }
3846 
3847 static void r100_mc_program(struct radeon_device *rdev)
3848 {
3849 	struct r100_mc_save save;
3850 
3851 	/* Stops all mc clients */
3852 	r100_mc_stop(rdev, &save);
3853 	if (rdev->flags & RADEON_IS_AGP) {
3854 		WREG32(R_00014C_MC_AGP_LOCATION,
3855 			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3856 			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3857 		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3858 		if (rdev->family > CHIP_RV200)
3859 			WREG32(R_00015C_AGP_BASE_2,
3860 				upper_32_bits(rdev->mc.agp_base) & 0xff);
3861 	} else {
3862 		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3863 		WREG32(R_000170_AGP_BASE, 0);
3864 		if (rdev->family > CHIP_RV200)
3865 			WREG32(R_00015C_AGP_BASE_2, 0);
3866 	}
3867 	/* Wait for mc idle */
3868 	if (r100_mc_wait_for_idle(rdev))
3869 		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3870 	/* Program MC, should be a 32bits limited address space */
3871 	WREG32(R_000148_MC_FB_LOCATION,
3872 		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3873 		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3874 	r100_mc_resume(rdev, &save);
3875 }
3876 
3877 static void r100_clock_startup(struct radeon_device *rdev)
3878 {
3879 	u32 tmp;
3880 
3881 	if (radeon_dynclks != -1 && radeon_dynclks)
3882 		radeon_legacy_set_clock_gating(rdev, 1);
3883 	/* We need to force on some of the block */
3884 	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3885 	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3886 	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3887 		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3888 	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3889 }
3890 
3891 static int r100_startup(struct radeon_device *rdev)
3892 {
3893 	int r;
3894 
3895 	/* set common regs */
3896 	r100_set_common_regs(rdev);
3897 	/* program mc */
3898 	r100_mc_program(rdev);
3899 	/* Resume clock */
3900 	r100_clock_startup(rdev);
3901 	/* Initialize GART (initialize after TTM so we can allocate
3902 	 * memory through TTM but finalize after TTM) */
3903 	r100_enable_bm(rdev);
3904 	if (rdev->flags & RADEON_IS_PCI) {
3905 		r = r100_pci_gart_enable(rdev);
3906 		if (r)
3907 			return r;
3908 	}
3909 
3910 	/* allocate wb buffer */
3911 	r = radeon_wb_init(rdev);
3912 	if (r)
3913 		return r;
3914 
3915 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3916 	if (r) {
3917 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3918 		return r;
3919 	}
3920 
3921 	/* Enable IRQ */
3922 	if (!rdev->irq.installed) {
3923 		r = radeon_irq_kms_init(rdev);
3924 		if (r)
3925 			return r;
3926 	}
3927 
3928 	r100_irq_set(rdev);
3929 	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3930 	/* 1M ring buffer */
3931 	r = r100_cp_init(rdev, 1024 * 1024);
3932 	if (r) {
3933 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3934 		return r;
3935 	}
3936 
3937 	r = radeon_ib_pool_init(rdev);
3938 	if (r) {
3939 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3940 		return r;
3941 	}
3942 
3943 	return 0;
3944 }
3945 
3946 int r100_resume(struct radeon_device *rdev)
3947 {
3948 	int r;
3949 
3950 	/* Make sur GART are not working */
3951 	if (rdev->flags & RADEON_IS_PCI)
3952 		r100_pci_gart_disable(rdev);
3953 	/* Resume clock before doing reset */
3954 	r100_clock_startup(rdev);
3955 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3956 	if (radeon_asic_reset(rdev)) {
3957 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3958 			RREG32(R_000E40_RBBM_STATUS),
3959 			RREG32(R_0007C0_CP_STAT));
3960 	}
3961 	/* post */
3962 	radeon_combios_asic_init(rdev->ddev);
3963 	/* Resume clock after posting */
3964 	r100_clock_startup(rdev);
3965 	/* Initialize surface registers */
3966 	radeon_surface_init(rdev);
3967 
3968 	rdev->accel_working = true;
3969 	r = r100_startup(rdev);
3970 	if (r) {
3971 		rdev->accel_working = false;
3972 	}
3973 	return r;
3974 }
3975 
3976 int r100_suspend(struct radeon_device *rdev)
3977 {
3978 	radeon_pm_suspend(rdev);
3979 	r100_cp_disable(rdev);
3980 	radeon_wb_disable(rdev);
3981 	r100_irq_disable(rdev);
3982 	if (rdev->flags & RADEON_IS_PCI)
3983 		r100_pci_gart_disable(rdev);
3984 	return 0;
3985 }
3986 
3987 void r100_fini(struct radeon_device *rdev)
3988 {
3989 	radeon_pm_fini(rdev);
3990 	r100_cp_fini(rdev);
3991 	radeon_wb_fini(rdev);
3992 	radeon_ib_pool_fini(rdev);
3993 	radeon_gem_fini(rdev);
3994 	if (rdev->flags & RADEON_IS_PCI)
3995 		r100_pci_gart_fini(rdev);
3996 	radeon_agp_fini(rdev);
3997 	radeon_irq_kms_fini(rdev);
3998 	radeon_fence_driver_fini(rdev);
3999 	radeon_bo_fini(rdev);
4000 	radeon_atombios_fini(rdev);
4001 	kfree(rdev->bios);
4002 	rdev->bios = NULL;
4003 }
4004 
4005 /*
4006  * Due to how kexec works, it can leave the hw fully initialised when it
4007  * boots the new kernel. However doing our init sequence with the CP and
4008  * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
4009  * do some quick sanity checks and restore sane values to avoid this
4010  * problem.
4011  */
4012 void r100_restore_sanity(struct radeon_device *rdev)
4013 {
4014 	u32 tmp;
4015 
4016 	tmp = RREG32(RADEON_CP_CSQ_CNTL);
4017 	if (tmp) {
4018 		WREG32(RADEON_CP_CSQ_CNTL, 0);
4019 	}
4020 	tmp = RREG32(RADEON_CP_RB_CNTL);
4021 	if (tmp) {
4022 		WREG32(RADEON_CP_RB_CNTL, 0);
4023 	}
4024 	tmp = RREG32(RADEON_SCRATCH_UMSK);
4025 	if (tmp) {
4026 		WREG32(RADEON_SCRATCH_UMSK, 0);
4027 	}
4028 }
4029 
4030 int r100_init(struct radeon_device *rdev)
4031 {
4032 	int r;
4033 
4034 	/* Register debugfs file specific to this group of asics */
4035 	r100_debugfs(rdev);
4036 	/* Disable VGA */
4037 	r100_vga_render_disable(rdev);
4038 	/* Initialize scratch registers */
4039 	radeon_scratch_init(rdev);
4040 	/* Initialize surface registers */
4041 	radeon_surface_init(rdev);
4042 	/* sanity check some register to avoid hangs like after kexec */
4043 	r100_restore_sanity(rdev);
4044 	/* TODO: disable VGA need to use VGA request */
4045 	/* BIOS*/
4046 	if (!radeon_get_bios(rdev)) {
4047 		if (ASIC_IS_AVIVO(rdev))
4048 			return -EINVAL;
4049 	}
4050 	if (rdev->is_atom_bios) {
4051 		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4052 		return -EINVAL;
4053 	} else {
4054 		r = radeon_combios_init(rdev);
4055 		if (r)
4056 			return r;
4057 	}
4058 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4059 	if (radeon_asic_reset(rdev)) {
4060 		dev_warn(rdev->dev,
4061 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4062 			RREG32(R_000E40_RBBM_STATUS),
4063 			RREG32(R_0007C0_CP_STAT));
4064 	}
4065 	/* check if cards are posted or not */
4066 	if (radeon_boot_test_post_card(rdev) == false)
4067 		return -EINVAL;
4068 	/* Set asic errata */
4069 	r100_errata(rdev);
4070 	/* Initialize clocks */
4071 	radeon_get_clock_info(rdev->ddev);
4072 	/* initialize AGP */
4073 	if (rdev->flags & RADEON_IS_AGP) {
4074 		r = radeon_agp_init(rdev);
4075 		if (r) {
4076 			radeon_agp_disable(rdev);
4077 		}
4078 	}
4079 	/* initialize VRAM */
4080 	r100_mc_init(rdev);
4081 	/* Fence driver */
4082 	r = radeon_fence_driver_init(rdev);
4083 	if (r)
4084 		return r;
4085 	/* Memory manager */
4086 	r = radeon_bo_init(rdev);
4087 	if (r)
4088 		return r;
4089 	if (rdev->flags & RADEON_IS_PCI) {
4090 		r = r100_pci_gart_init(rdev);
4091 		if (r)
4092 			return r;
4093 	}
4094 	r100_set_safe_registers(rdev);
4095 
4096 	/* Initialize power management */
4097 	radeon_pm_init(rdev);
4098 
4099 	rdev->accel_working = true;
4100 	r = r100_startup(rdev);
4101 	if (r) {
4102 		/* Somethings want wront with the accel init stop accel */
4103 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4104 		r100_cp_fini(rdev);
4105 		radeon_wb_fini(rdev);
4106 		radeon_ib_pool_fini(rdev);
4107 		radeon_irq_kms_fini(rdev);
4108 		if (rdev->flags & RADEON_IS_PCI)
4109 			r100_pci_gart_fini(rdev);
4110 		rdev->accel_working = false;
4111 	}
4112 	return 0;
4113 }
4114 
4115 uint32_t r100_mm_rreg_slow(struct radeon_device *rdev, uint32_t reg)
4116 {
4117 	unsigned long flags;
4118 	uint32_t ret;
4119 
4120 	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4121 	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4122 	ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4123 	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4124 	return ret;
4125 }
4126 
4127 void r100_mm_wreg_slow(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4128 {
4129 	unsigned long flags;
4130 
4131 	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4132 	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4133 	writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4134 	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4135 }
4136 
4137 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4138 {
4139 	if (reg < rdev->rio_mem_size)
4140 		return ioread32(rdev->rio_mem + reg);
4141 	else {
4142 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4143 		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4144 	}
4145 }
4146 
4147 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4148 {
4149 	if (reg < rdev->rio_mem_size)
4150 		iowrite32(v, rdev->rio_mem + reg);
4151 	else {
4152 		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4153 		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4154 	}
4155 }
4156