GNU Linux-libre 4.9.304-gnu1
[releases.git] / drivers / gpu / drm / radeon / r100.c
1 /*
2  * Copyright 2008 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 #include <linux/seq_file.h>
29 #include <linux/slab.h>
30 #include <drm/drmP.h>
31 #include <drm/radeon_drm.h>
32 #include "radeon_reg.h"
33 #include "radeon.h"
34 #include "radeon_asic.h"
35 #include "r100d.h"
36 #include "rs100d.h"
37 #include "rv200d.h"
38 #include "rv250d.h"
39 #include "atom.h"
40
41 #include <linux/firmware.h>
42 #include <linux/module.h>
43
44 #include "r100_reg_safe.h"
45 #include "rn50_reg_safe.h"
46
47 /* Firmware Names */
48 #define FIRMWARE_R100           "/*(DEBLOBBED)*/"
49 #define FIRMWARE_R200           "/*(DEBLOBBED)*/"
50 #define FIRMWARE_R300           "/*(DEBLOBBED)*/"
51 #define FIRMWARE_R420           "/*(DEBLOBBED)*/"
52 #define FIRMWARE_RS690          "/*(DEBLOBBED)*/"
53 #define FIRMWARE_RS600          "/*(DEBLOBBED)*/"
54 #define FIRMWARE_R520           "/*(DEBLOBBED)*/"
55
56 /*(DEBLOBBED)*/
57
58 #include "r100_track.h"
59
60 /* This files gather functions specifics to:
61  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
62  * and others in some cases.
63  */
64
65 static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
66 {
67         if (crtc == 0) {
68                 if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
69                         return true;
70                 else
71                         return false;
72         } else {
73                 if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
74                         return true;
75                 else
76                         return false;
77         }
78 }
79
80 static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
81 {
82         u32 vline1, vline2;
83
84         if (crtc == 0) {
85                 vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
86                 vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
87         } else {
88                 vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
89                 vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
90         }
91         if (vline1 != vline2)
92                 return true;
93         else
94                 return false;
95 }
96
97 /**
98  * r100_wait_for_vblank - vblank wait asic callback.
99  *
100  * @rdev: radeon_device pointer
101  * @crtc: crtc to wait for vblank on
102  *
103  * Wait for vblank on the requested crtc (r1xx-r4xx).
104  */
105 void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
106 {
107         unsigned i = 0;
108
109         if (crtc >= rdev->num_crtc)
110                 return;
111
112         if (crtc == 0) {
113                 if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
114                         return;
115         } else {
116                 if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
117                         return;
118         }
119
120         /* depending on when we hit vblank, we may be close to active; if so,
121          * wait for another frame.
122          */
123         while (r100_is_in_vblank(rdev, crtc)) {
124                 if (i++ % 100 == 0) {
125                         if (!r100_is_counter_moving(rdev, crtc))
126                                 break;
127                 }
128         }
129
130         while (!r100_is_in_vblank(rdev, crtc)) {
131                 if (i++ % 100 == 0) {
132                         if (!r100_is_counter_moving(rdev, crtc))
133                                 break;
134                 }
135         }
136 }
137
138 /**
139  * r100_page_flip - pageflip callback.
140  *
141  * @rdev: radeon_device pointer
142  * @crtc_id: crtc to cleanup pageflip on
143  * @crtc_base: new address of the crtc (GPU MC address)
144  *
145  * Does the actual pageflip (r1xx-r4xx).
146  * During vblank we take the crtc lock and wait for the update_pending
147  * bit to go high, when it does, we release the lock, and allow the
148  * double buffered update to take place.
149  */
150 void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base, bool async)
151 {
152         struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
153         u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
154         int i;
155
156         /* Lock the graphics update lock */
157         /* update the scanout addresses */
158         WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
159
160         /* Wait for update_pending to go high. */
161         for (i = 0; i < rdev->usec_timeout; i++) {
162                 if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
163                         break;
164                 udelay(1);
165         }
166         DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
167
168         /* Unlock the lock, so double-buffering can take place inside vblank */
169         tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
170         WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
171
172 }
173
174 /**
175  * r100_page_flip_pending - check if page flip is still pending
176  *
177  * @rdev: radeon_device pointer
178  * @crtc_id: crtc to check
179  *
180  * Check if the last pagefilp is still pending (r1xx-r4xx).
181  * Returns the current update pending status.
182  */
183 bool r100_page_flip_pending(struct radeon_device *rdev, int crtc_id)
184 {
185         struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
186
187         /* Return current update_pending status: */
188         return !!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) &
189                 RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET);
190 }
191
192 /**
193  * r100_pm_get_dynpm_state - look up dynpm power state callback.
194  *
195  * @rdev: radeon_device pointer
196  *
197  * Look up the optimal power state based on the
198  * current state of the GPU (r1xx-r5xx).
199  * Used for dynpm only.
200  */
201 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
202 {
203         int i;
204         rdev->pm.dynpm_can_upclock = true;
205         rdev->pm.dynpm_can_downclock = true;
206
207         switch (rdev->pm.dynpm_planned_action) {
208         case DYNPM_ACTION_MINIMUM:
209                 rdev->pm.requested_power_state_index = 0;
210                 rdev->pm.dynpm_can_downclock = false;
211                 break;
212         case DYNPM_ACTION_DOWNCLOCK:
213                 if (rdev->pm.current_power_state_index == 0) {
214                         rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
215                         rdev->pm.dynpm_can_downclock = false;
216                 } else {
217                         if (rdev->pm.active_crtc_count > 1) {
218                                 for (i = 0; i < rdev->pm.num_power_states; i++) {
219                                         if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
220                                                 continue;
221                                         else if (i >= rdev->pm.current_power_state_index) {
222                                                 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
223                                                 break;
224                                         } else {
225                                                 rdev->pm.requested_power_state_index = i;
226                                                 break;
227                                         }
228                                 }
229                         } else
230                                 rdev->pm.requested_power_state_index =
231                                         rdev->pm.current_power_state_index - 1;
232                 }
233                 /* don't use the power state if crtcs are active and no display flag is set */
234                 if ((rdev->pm.active_crtc_count > 0) &&
235                     (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
236                      RADEON_PM_MODE_NO_DISPLAY)) {
237                         rdev->pm.requested_power_state_index++;
238                 }
239                 break;
240         case DYNPM_ACTION_UPCLOCK:
241                 if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
242                         rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
243                         rdev->pm.dynpm_can_upclock = false;
244                 } else {
245                         if (rdev->pm.active_crtc_count > 1) {
246                                 for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
247                                         if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
248                                                 continue;
249                                         else if (i <= rdev->pm.current_power_state_index) {
250                                                 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
251                                                 break;
252                                         } else {
253                                                 rdev->pm.requested_power_state_index = i;
254                                                 break;
255                                         }
256                                 }
257                         } else
258                                 rdev->pm.requested_power_state_index =
259                                         rdev->pm.current_power_state_index + 1;
260                 }
261                 break;
262         case DYNPM_ACTION_DEFAULT:
263                 rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
264                 rdev->pm.dynpm_can_upclock = false;
265                 break;
266         case DYNPM_ACTION_NONE:
267         default:
268                 DRM_ERROR("Requested mode for not defined action\n");
269                 return;
270         }
271         /* only one clock mode per power state */
272         rdev->pm.requested_clock_mode_index = 0;
273
274         DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
275                   rdev->pm.power_state[rdev->pm.requested_power_state_index].
276                   clock_info[rdev->pm.requested_clock_mode_index].sclk,
277                   rdev->pm.power_state[rdev->pm.requested_power_state_index].
278                   clock_info[rdev->pm.requested_clock_mode_index].mclk,
279                   rdev->pm.power_state[rdev->pm.requested_power_state_index].
280                   pcie_lanes);
281 }
282
283 /**
284  * r100_pm_init_profile - Initialize power profiles callback.
285  *
286  * @rdev: radeon_device pointer
287  *
288  * Initialize the power states used in profile mode
289  * (r1xx-r3xx).
290  * Used for profile mode only.
291  */
292 void r100_pm_init_profile(struct radeon_device *rdev)
293 {
294         /* default */
295         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
296         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
297         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
298         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
299         /* low sh */
300         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
301         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
302         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
303         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
304         /* mid sh */
305         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
306         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
307         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
308         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
309         /* high sh */
310         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
311         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
312         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
313         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
314         /* low mh */
315         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
316         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
317         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
318         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
319         /* mid mh */
320         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
321         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
322         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
323         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
324         /* high mh */
325         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
326         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
327         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
328         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
329 }
330
331 /**
332  * r100_pm_misc - set additional pm hw parameters callback.
333  *
334  * @rdev: radeon_device pointer
335  *
336  * Set non-clock parameters associated with a power state
337  * (voltage, pcie lanes, etc.) (r1xx-r4xx).
338  */
339 void r100_pm_misc(struct radeon_device *rdev)
340 {
341         int requested_index = rdev->pm.requested_power_state_index;
342         struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
343         struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
344         u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
345
346         if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
347                 if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
348                         tmp = RREG32(voltage->gpio.reg);
349                         if (voltage->active_high)
350                                 tmp |= voltage->gpio.mask;
351                         else
352                                 tmp &= ~(voltage->gpio.mask);
353                         WREG32(voltage->gpio.reg, tmp);
354                         if (voltage->delay)
355                                 udelay(voltage->delay);
356                 } else {
357                         tmp = RREG32(voltage->gpio.reg);
358                         if (voltage->active_high)
359                                 tmp &= ~voltage->gpio.mask;
360                         else
361                                 tmp |= voltage->gpio.mask;
362                         WREG32(voltage->gpio.reg, tmp);
363                         if (voltage->delay)
364                                 udelay(voltage->delay);
365                 }
366         }
367
368         sclk_cntl = RREG32_PLL(SCLK_CNTL);
369         sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
370         sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
371         sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
372         sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
373         if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
374                 sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
375                 if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
376                         sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
377                 else
378                         sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
379                 if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
380                         sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
381                 else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
382                         sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
383         } else
384                 sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
385
386         if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
387                 sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
388                 if (voltage->delay) {
389                         sclk_more_cntl |= VOLTAGE_DROP_SYNC;
390                         switch (voltage->delay) {
391                         case 33:
392                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
393                                 break;
394                         case 66:
395                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
396                                 break;
397                         case 99:
398                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
399                                 break;
400                         case 132:
401                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
402                                 break;
403                         }
404                 } else
405                         sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
406         } else
407                 sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
408
409         if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
410                 sclk_cntl &= ~FORCE_HDP;
411         else
412                 sclk_cntl |= FORCE_HDP;
413
414         WREG32_PLL(SCLK_CNTL, sclk_cntl);
415         WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
416         WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
417
418         /* set pcie lanes */
419         if ((rdev->flags & RADEON_IS_PCIE) &&
420             !(rdev->flags & RADEON_IS_IGP) &&
421             rdev->asic->pm.set_pcie_lanes &&
422             (ps->pcie_lanes !=
423              rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
424                 radeon_set_pcie_lanes(rdev,
425                                       ps->pcie_lanes);
426                 DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
427         }
428 }
429
430 /**
431  * r100_pm_prepare - pre-power state change callback.
432  *
433  * @rdev: radeon_device pointer
434  *
435  * Prepare for a power state change (r1xx-r4xx).
436  */
437 void r100_pm_prepare(struct radeon_device *rdev)
438 {
439         struct drm_device *ddev = rdev->ddev;
440         struct drm_crtc *crtc;
441         struct radeon_crtc *radeon_crtc;
442         u32 tmp;
443
444         /* disable any active CRTCs */
445         list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
446                 radeon_crtc = to_radeon_crtc(crtc);
447                 if (radeon_crtc->enabled) {
448                         if (radeon_crtc->crtc_id) {
449                                 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
450                                 tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
451                                 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
452                         } else {
453                                 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
454                                 tmp |= RADEON_CRTC_DISP_REQ_EN_B;
455                                 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
456                         }
457                 }
458         }
459 }
460
461 /**
462  * r100_pm_finish - post-power state change callback.
463  *
464  * @rdev: radeon_device pointer
465  *
466  * Clean up after a power state change (r1xx-r4xx).
467  */
468 void r100_pm_finish(struct radeon_device *rdev)
469 {
470         struct drm_device *ddev = rdev->ddev;
471         struct drm_crtc *crtc;
472         struct radeon_crtc *radeon_crtc;
473         u32 tmp;
474
475         /* enable any active CRTCs */
476         list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
477                 radeon_crtc = to_radeon_crtc(crtc);
478                 if (radeon_crtc->enabled) {
479                         if (radeon_crtc->crtc_id) {
480                                 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
481                                 tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
482                                 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
483                         } else {
484                                 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
485                                 tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
486                                 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
487                         }
488                 }
489         }
490 }
491
492 /**
493  * r100_gui_idle - gui idle callback.
494  *
495  * @rdev: radeon_device pointer
496  *
497  * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
498  * Returns true if idle, false if not.
499  */
500 bool r100_gui_idle(struct radeon_device *rdev)
501 {
502         if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
503                 return false;
504         else
505                 return true;
506 }
507
508 /* hpd for digital panel detect/disconnect */
509 /**
510  * r100_hpd_sense - hpd sense callback.
511  *
512  * @rdev: radeon_device pointer
513  * @hpd: hpd (hotplug detect) pin
514  *
515  * Checks if a digital monitor is connected (r1xx-r4xx).
516  * Returns true if connected, false if not connected.
517  */
518 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
519 {
520         bool connected = false;
521
522         switch (hpd) {
523         case RADEON_HPD_1:
524                 if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
525                         connected = true;
526                 break;
527         case RADEON_HPD_2:
528                 if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
529                         connected = true;
530                 break;
531         default:
532                 break;
533         }
534         return connected;
535 }
536
537 /**
538  * r100_hpd_set_polarity - hpd set polarity callback.
539  *
540  * @rdev: radeon_device pointer
541  * @hpd: hpd (hotplug detect) pin
542  *
543  * Set the polarity of the hpd pin (r1xx-r4xx).
544  */
545 void r100_hpd_set_polarity(struct radeon_device *rdev,
546                            enum radeon_hpd_id hpd)
547 {
548         u32 tmp;
549         bool connected = r100_hpd_sense(rdev, hpd);
550
551         switch (hpd) {
552         case RADEON_HPD_1:
553                 tmp = RREG32(RADEON_FP_GEN_CNTL);
554                 if (connected)
555                         tmp &= ~RADEON_FP_DETECT_INT_POL;
556                 else
557                         tmp |= RADEON_FP_DETECT_INT_POL;
558                 WREG32(RADEON_FP_GEN_CNTL, tmp);
559                 break;
560         case RADEON_HPD_2:
561                 tmp = RREG32(RADEON_FP2_GEN_CNTL);
562                 if (connected)
563                         tmp &= ~RADEON_FP2_DETECT_INT_POL;
564                 else
565                         tmp |= RADEON_FP2_DETECT_INT_POL;
566                 WREG32(RADEON_FP2_GEN_CNTL, tmp);
567                 break;
568         default:
569                 break;
570         }
571 }
572
573 /**
574  * r100_hpd_init - hpd setup callback.
575  *
576  * @rdev: radeon_device pointer
577  *
578  * Setup the hpd pins used by the card (r1xx-r4xx).
579  * Set the polarity, and enable the hpd interrupts.
580  */
581 void r100_hpd_init(struct radeon_device *rdev)
582 {
583         struct drm_device *dev = rdev->ddev;
584         struct drm_connector *connector;
585         unsigned enable = 0;
586
587         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
588                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
589                 if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
590                         enable |= 1 << radeon_connector->hpd.hpd;
591                 radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
592         }
593         radeon_irq_kms_enable_hpd(rdev, enable);
594 }
595
596 /**
597  * r100_hpd_fini - hpd tear down callback.
598  *
599  * @rdev: radeon_device pointer
600  *
601  * Tear down the hpd pins used by the card (r1xx-r4xx).
602  * Disable the hpd interrupts.
603  */
604 void r100_hpd_fini(struct radeon_device *rdev)
605 {
606         struct drm_device *dev = rdev->ddev;
607         struct drm_connector *connector;
608         unsigned disable = 0;
609
610         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
611                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
612                 if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
613                         disable |= 1 << radeon_connector->hpd.hpd;
614         }
615         radeon_irq_kms_disable_hpd(rdev, disable);
616 }
617
618 /*
619  * PCI GART
620  */
621 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
622 {
623         /* TODO: can we do somethings here ? */
624         /* It seems hw only cache one entry so we should discard this
625          * entry otherwise if first GPU GART read hit this entry it
626          * could end up in wrong address. */
627 }
628
629 int r100_pci_gart_init(struct radeon_device *rdev)
630 {
631         int r;
632
633         if (rdev->gart.ptr) {
634                 WARN(1, "R100 PCI GART already initialized\n");
635                 return 0;
636         }
637         /* Initialize common gart structure */
638         r = radeon_gart_init(rdev);
639         if (r)
640                 return r;
641         rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
642         rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
643         rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
644         rdev->asic->gart.set_page = &r100_pci_gart_set_page;
645         return radeon_gart_table_ram_alloc(rdev);
646 }
647
648 int r100_pci_gart_enable(struct radeon_device *rdev)
649 {
650         uint32_t tmp;
651
652         /* discard memory request outside of configured range */
653         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
654         WREG32(RADEON_AIC_CNTL, tmp);
655         /* set address range for PCI address translate */
656         WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
657         WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
658         /* set PCI GART page-table base address */
659         WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
660         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
661         WREG32(RADEON_AIC_CNTL, tmp);
662         r100_pci_gart_tlb_flush(rdev);
663         DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
664                  (unsigned)(rdev->mc.gtt_size >> 20),
665                  (unsigned long long)rdev->gart.table_addr);
666         rdev->gart.ready = true;
667         return 0;
668 }
669
670 void r100_pci_gart_disable(struct radeon_device *rdev)
671 {
672         uint32_t tmp;
673
674         /* discard memory request outside of configured range */
675         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
676         WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
677         WREG32(RADEON_AIC_LO_ADDR, 0);
678         WREG32(RADEON_AIC_HI_ADDR, 0);
679 }
680
681 uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
682 {
683         return addr;
684 }
685
686 void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
687                             uint64_t entry)
688 {
689         u32 *gtt = rdev->gart.ptr;
690         gtt[i] = cpu_to_le32(lower_32_bits(entry));
691 }
692
693 void r100_pci_gart_fini(struct radeon_device *rdev)
694 {
695         radeon_gart_fini(rdev);
696         r100_pci_gart_disable(rdev);
697         radeon_gart_table_ram_free(rdev);
698 }
699
700 int r100_irq_set(struct radeon_device *rdev)
701 {
702         uint32_t tmp = 0;
703
704         if (!rdev->irq.installed) {
705                 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
706                 WREG32(R_000040_GEN_INT_CNTL, 0);
707                 return -EINVAL;
708         }
709         if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
710                 tmp |= RADEON_SW_INT_ENABLE;
711         }
712         if (rdev->irq.crtc_vblank_int[0] ||
713             atomic_read(&rdev->irq.pflip[0])) {
714                 tmp |= RADEON_CRTC_VBLANK_MASK;
715         }
716         if (rdev->irq.crtc_vblank_int[1] ||
717             atomic_read(&rdev->irq.pflip[1])) {
718                 tmp |= RADEON_CRTC2_VBLANK_MASK;
719         }
720         if (rdev->irq.hpd[0]) {
721                 tmp |= RADEON_FP_DETECT_MASK;
722         }
723         if (rdev->irq.hpd[1]) {
724                 tmp |= RADEON_FP2_DETECT_MASK;
725         }
726         WREG32(RADEON_GEN_INT_CNTL, tmp);
727
728         /* read back to post the write */
729         RREG32(RADEON_GEN_INT_CNTL);
730
731         return 0;
732 }
733
734 void r100_irq_disable(struct radeon_device *rdev)
735 {
736         u32 tmp;
737
738         WREG32(R_000040_GEN_INT_CNTL, 0);
739         /* Wait and acknowledge irq */
740         mdelay(1);
741         tmp = RREG32(R_000044_GEN_INT_STATUS);
742         WREG32(R_000044_GEN_INT_STATUS, tmp);
743 }
744
745 static uint32_t r100_irq_ack(struct radeon_device *rdev)
746 {
747         uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
748         uint32_t irq_mask = RADEON_SW_INT_TEST |
749                 RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
750                 RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
751
752         if (irqs) {
753                 WREG32(RADEON_GEN_INT_STATUS, irqs);
754         }
755         return irqs & irq_mask;
756 }
757
758 int r100_irq_process(struct radeon_device *rdev)
759 {
760         uint32_t status, msi_rearm;
761         bool queue_hotplug = false;
762
763         status = r100_irq_ack(rdev);
764         if (!status) {
765                 return IRQ_NONE;
766         }
767         if (rdev->shutdown) {
768                 return IRQ_NONE;
769         }
770         while (status) {
771                 /* SW interrupt */
772                 if (status & RADEON_SW_INT_TEST) {
773                         radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
774                 }
775                 /* Vertical blank interrupts */
776                 if (status & RADEON_CRTC_VBLANK_STAT) {
777                         if (rdev->irq.crtc_vblank_int[0]) {
778                                 drm_handle_vblank(rdev->ddev, 0);
779                                 rdev->pm.vblank_sync = true;
780                                 wake_up(&rdev->irq.vblank_queue);
781                         }
782                         if (atomic_read(&rdev->irq.pflip[0]))
783                                 radeon_crtc_handle_vblank(rdev, 0);
784                 }
785                 if (status & RADEON_CRTC2_VBLANK_STAT) {
786                         if (rdev->irq.crtc_vblank_int[1]) {
787                                 drm_handle_vblank(rdev->ddev, 1);
788                                 rdev->pm.vblank_sync = true;
789                                 wake_up(&rdev->irq.vblank_queue);
790                         }
791                         if (atomic_read(&rdev->irq.pflip[1]))
792                                 radeon_crtc_handle_vblank(rdev, 1);
793                 }
794                 if (status & RADEON_FP_DETECT_STAT) {
795                         queue_hotplug = true;
796                         DRM_DEBUG("HPD1\n");
797                 }
798                 if (status & RADEON_FP2_DETECT_STAT) {
799                         queue_hotplug = true;
800                         DRM_DEBUG("HPD2\n");
801                 }
802                 status = r100_irq_ack(rdev);
803         }
804         if (queue_hotplug)
805                 schedule_delayed_work(&rdev->hotplug_work, 0);
806         if (rdev->msi_enabled) {
807                 switch (rdev->family) {
808                 case CHIP_RS400:
809                 case CHIP_RS480:
810                         msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
811                         WREG32(RADEON_AIC_CNTL, msi_rearm);
812                         WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
813                         break;
814                 default:
815                         WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
816                         break;
817                 }
818         }
819         return IRQ_HANDLED;
820 }
821
822 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
823 {
824         if (crtc == 0)
825                 return RREG32(RADEON_CRTC_CRNT_FRAME);
826         else
827                 return RREG32(RADEON_CRTC2_CRNT_FRAME);
828 }
829
830 /**
831  * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
832  * rdev: radeon device structure
833  * ring: ring buffer struct for emitting packets
834  */
835 static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
836 {
837         radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
838         radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
839                                 RADEON_HDP_READ_BUFFER_INVALIDATE);
840         radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
841         radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
842 }
843
844 /* Who ever call radeon_fence_emit should call ring_lock and ask
845  * for enough space (today caller are ib schedule and buffer move) */
846 void r100_fence_ring_emit(struct radeon_device *rdev,
847                           struct radeon_fence *fence)
848 {
849         struct radeon_ring *ring = &rdev->ring[fence->ring];
850
851         /* We have to make sure that caches are flushed before
852          * CPU might read something from VRAM. */
853         radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
854         radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
855         radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
856         radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
857         /* Wait until IDLE & CLEAN */
858         radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
859         radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
860         r100_ring_hdp_flush(rdev, ring);
861         /* Emit fence sequence & fire IRQ */
862         radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
863         radeon_ring_write(ring, fence->seq);
864         radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
865         radeon_ring_write(ring, RADEON_SW_INT_FIRE);
866 }
867
868 bool r100_semaphore_ring_emit(struct radeon_device *rdev,
869                               struct radeon_ring *ring,
870                               struct radeon_semaphore *semaphore,
871                               bool emit_wait)
872 {
873         /* Unused on older asics, since we don't have semaphores or multiple rings */
874         BUG();
875         return false;
876 }
877
878 struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
879                                     uint64_t src_offset,
880                                     uint64_t dst_offset,
881                                     unsigned num_gpu_pages,
882                                     struct reservation_object *resv)
883 {
884         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
885         struct radeon_fence *fence;
886         uint32_t cur_pages;
887         uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
888         uint32_t pitch;
889         uint32_t stride_pixels;
890         unsigned ndw;
891         int num_loops;
892         int r = 0;
893
894         /* radeon limited to 16k stride */
895         stride_bytes &= 0x3fff;
896         /* radeon pitch is /64 */
897         pitch = stride_bytes / 64;
898         stride_pixels = stride_bytes / 4;
899         num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
900
901         /* Ask for enough room for blit + flush + fence */
902         ndw = 64 + (10 * num_loops);
903         r = radeon_ring_lock(rdev, ring, ndw);
904         if (r) {
905                 DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
906                 return ERR_PTR(-EINVAL);
907         }
908         while (num_gpu_pages > 0) {
909                 cur_pages = num_gpu_pages;
910                 if (cur_pages > 8191) {
911                         cur_pages = 8191;
912                 }
913                 num_gpu_pages -= cur_pages;
914
915                 /* pages are in Y direction - height
916                    page width in X direction - width */
917                 radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
918                 radeon_ring_write(ring,
919                                   RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
920                                   RADEON_GMC_DST_PITCH_OFFSET_CNTL |
921                                   RADEON_GMC_SRC_CLIPPING |
922                                   RADEON_GMC_DST_CLIPPING |
923                                   RADEON_GMC_BRUSH_NONE |
924                                   (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
925                                   RADEON_GMC_SRC_DATATYPE_COLOR |
926                                   RADEON_ROP3_S |
927                                   RADEON_DP_SRC_SOURCE_MEMORY |
928                                   RADEON_GMC_CLR_CMP_CNTL_DIS |
929                                   RADEON_GMC_WR_MSK_DIS);
930                 radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
931                 radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
932                 radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
933                 radeon_ring_write(ring, 0);
934                 radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
935                 radeon_ring_write(ring, num_gpu_pages);
936                 radeon_ring_write(ring, num_gpu_pages);
937                 radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
938         }
939         radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
940         radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
941         radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
942         radeon_ring_write(ring,
943                           RADEON_WAIT_2D_IDLECLEAN |
944                           RADEON_WAIT_HOST_IDLECLEAN |
945                           RADEON_WAIT_DMA_GUI_IDLE);
946         r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
947         if (r) {
948                 radeon_ring_unlock_undo(rdev, ring);
949                 return ERR_PTR(r);
950         }
951         radeon_ring_unlock_commit(rdev, ring, false);
952         return fence;
953 }
954
955 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
956 {
957         unsigned i;
958         u32 tmp;
959
960         for (i = 0; i < rdev->usec_timeout; i++) {
961                 tmp = RREG32(R_000E40_RBBM_STATUS);
962                 if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
963                         return 0;
964                 }
965                 udelay(1);
966         }
967         return -1;
968 }
969
970 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
971 {
972         int r;
973
974         r = radeon_ring_lock(rdev, ring, 2);
975         if (r) {
976                 return;
977         }
978         radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
979         radeon_ring_write(ring,
980                           RADEON_ISYNC_ANY2D_IDLE3D |
981                           RADEON_ISYNC_ANY3D_IDLE2D |
982                           RADEON_ISYNC_WAIT_IDLEGUI |
983                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
984         radeon_ring_unlock_commit(rdev, ring, false);
985 }
986
987
988 /* Load the microcode for the CP */
989 static int r100_cp_init_microcode(struct radeon_device *rdev)
990 {
991         const char *fw_name = NULL;
992         int err;
993
994         DRM_DEBUG_KMS("\n");
995
996         if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
997             (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
998             (rdev->family == CHIP_RS200)) {
999                 DRM_INFO("Loading R100 Microcode\n");
1000                 fw_name = FIRMWARE_R100;
1001         } else if ((rdev->family == CHIP_R200) ||
1002                    (rdev->family == CHIP_RV250) ||
1003                    (rdev->family == CHIP_RV280) ||
1004                    (rdev->family == CHIP_RS300)) {
1005                 DRM_INFO("Loading R200 Microcode\n");
1006                 fw_name = FIRMWARE_R200;
1007         } else if ((rdev->family == CHIP_R300) ||
1008                    (rdev->family == CHIP_R350) ||
1009                    (rdev->family == CHIP_RV350) ||
1010                    (rdev->family == CHIP_RV380) ||
1011                    (rdev->family == CHIP_RS400) ||
1012                    (rdev->family == CHIP_RS480)) {
1013                 DRM_INFO("Loading R300 Microcode\n");
1014                 fw_name = FIRMWARE_R300;
1015         } else if ((rdev->family == CHIP_R420) ||
1016                    (rdev->family == CHIP_R423) ||
1017                    (rdev->family == CHIP_RV410)) {
1018                 DRM_INFO("Loading R400 Microcode\n");
1019                 fw_name = FIRMWARE_R420;
1020         } else if ((rdev->family == CHIP_RS690) ||
1021                    (rdev->family == CHIP_RS740)) {
1022                 DRM_INFO("Loading RS690/RS740 Microcode\n");
1023                 fw_name = FIRMWARE_RS690;
1024         } else if (rdev->family == CHIP_RS600) {
1025                 DRM_INFO("Loading RS600 Microcode\n");
1026                 fw_name = FIRMWARE_RS600;
1027         } else if ((rdev->family == CHIP_RV515) ||
1028                    (rdev->family == CHIP_R520) ||
1029                    (rdev->family == CHIP_RV530) ||
1030                    (rdev->family == CHIP_R580) ||
1031                    (rdev->family == CHIP_RV560) ||
1032                    (rdev->family == CHIP_RV570)) {
1033                 DRM_INFO("Loading R500 Microcode\n");
1034                 fw_name = FIRMWARE_R520;
1035         }
1036
1037         err = reject_firmware(&rdev->me_fw, fw_name, rdev->dev);
1038         if (err) {
1039                 printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1040                        fw_name);
1041         } else if (rdev->me_fw->size % 8) {
1042                 printk(KERN_ERR
1043                        "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1044                        rdev->me_fw->size, fw_name);
1045                 err = -EINVAL;
1046                 release_firmware(rdev->me_fw);
1047                 rdev->me_fw = NULL;
1048         }
1049         return err;
1050 }
1051
1052 u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1053                       struct radeon_ring *ring)
1054 {
1055         u32 rptr;
1056
1057         if (rdev->wb.enabled)
1058                 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1059         else
1060                 rptr = RREG32(RADEON_CP_RB_RPTR);
1061
1062         return rptr;
1063 }
1064
1065 u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1066                       struct radeon_ring *ring)
1067 {
1068         return RREG32(RADEON_CP_RB_WPTR);
1069 }
1070
1071 void r100_gfx_set_wptr(struct radeon_device *rdev,
1072                        struct radeon_ring *ring)
1073 {
1074         WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1075         (void)RREG32(RADEON_CP_RB_WPTR);
1076 }
1077
1078 static void r100_cp_load_microcode(struct radeon_device *rdev)
1079 {
1080         const __be32 *fw_data;
1081         int i, size;
1082
1083         if (r100_gui_wait_for_idle(rdev)) {
1084                 printk(KERN_WARNING "Failed to wait GUI idle while "
1085                        "programming pipes. Bad things might happen.\n");
1086         }
1087
1088         if (rdev->me_fw) {
1089                 size = rdev->me_fw->size / 4;
1090                 fw_data = (const __be32 *)&rdev->me_fw->data[0];
1091                 WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1092                 for (i = 0; i < size; i += 2) {
1093                         WREG32(RADEON_CP_ME_RAM_DATAH,
1094                                be32_to_cpup(&fw_data[i]));
1095                         WREG32(RADEON_CP_ME_RAM_DATAL,
1096                                be32_to_cpup(&fw_data[i + 1]));
1097                 }
1098         }
1099 }
1100
1101 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1102 {
1103         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1104         unsigned rb_bufsz;
1105         unsigned rb_blksz;
1106         unsigned max_fetch;
1107         unsigned pre_write_timer;
1108         unsigned pre_write_limit;
1109         unsigned indirect2_start;
1110         unsigned indirect1_start;
1111         uint32_t tmp;
1112         int r;
1113
1114         if (r100_debugfs_cp_init(rdev)) {
1115                 DRM_ERROR("Failed to register debugfs file for CP !\n");
1116         }
1117         if (!rdev->me_fw) {
1118                 r = r100_cp_init_microcode(rdev);
1119                 if (r) {
1120                         DRM_ERROR("Failed to load firmware!\n");
1121                         return r;
1122                 }
1123         }
1124
1125         /* Align ring size */
1126         rb_bufsz = order_base_2(ring_size / 8);
1127         ring_size = (1 << (rb_bufsz + 1)) * 4;
1128         r100_cp_load_microcode(rdev);
1129         r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1130                              RADEON_CP_PACKET2);
1131         if (r) {
1132                 return r;
1133         }
1134         /* Each time the cp read 1024 bytes (16 dword/quadword) update
1135          * the rptr copy in system ram */
1136         rb_blksz = 9;
1137         /* cp will read 128bytes at a time (4 dwords) */
1138         max_fetch = 1;
1139         ring->align_mask = 16 - 1;
1140         /* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1141         pre_write_timer = 64;
1142         /* Force CP_RB_WPTR write if written more than one time before the
1143          * delay expire
1144          */
1145         pre_write_limit = 0;
1146         /* Setup the cp cache like this (cache size is 96 dwords) :
1147          *      RING            0  to 15
1148          *      INDIRECT1       16 to 79
1149          *      INDIRECT2       80 to 95
1150          * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1151          *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1152          *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1153          * Idea being that most of the gpu cmd will be through indirect1 buffer
1154          * so it gets the bigger cache.
1155          */
1156         indirect2_start = 80;
1157         indirect1_start = 16;
1158         /* cp setup */
1159         WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1160         tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1161                REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1162                REG_SET(RADEON_MAX_FETCH, max_fetch));
1163 #ifdef __BIG_ENDIAN
1164         tmp |= RADEON_BUF_SWAP_32BIT;
1165 #endif
1166         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1167
1168         /* Set ring address */
1169         DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1170         WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1171         /* Force read & write ptr to 0 */
1172         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1173         WREG32(RADEON_CP_RB_RPTR_WR, 0);
1174         ring->wptr = 0;
1175         WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1176
1177         /* set the wb address whether it's enabled or not */
1178         WREG32(R_00070C_CP_RB_RPTR_ADDR,
1179                 S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1180         WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1181
1182         if (rdev->wb.enabled)
1183                 WREG32(R_000770_SCRATCH_UMSK, 0xff);
1184         else {
1185                 tmp |= RADEON_RB_NO_UPDATE;
1186                 WREG32(R_000770_SCRATCH_UMSK, 0);
1187         }
1188
1189         WREG32(RADEON_CP_RB_CNTL, tmp);
1190         udelay(10);
1191         /* Set cp mode to bus mastering & enable cp*/
1192         WREG32(RADEON_CP_CSQ_MODE,
1193                REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1194                REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1195         WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1196         WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1197         WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1198
1199         /* at this point everything should be setup correctly to enable master */
1200         pci_set_master(rdev->pdev);
1201
1202         radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1203         r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1204         if (r) {
1205                 DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1206                 return r;
1207         }
1208         ring->ready = true;
1209         radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1210
1211         if (!ring->rptr_save_reg /* not resuming from suspend */
1212             && radeon_ring_supports_scratch_reg(rdev, ring)) {
1213                 r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1214                 if (r) {
1215                         DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1216                         ring->rptr_save_reg = 0;
1217                 }
1218         }
1219         return 0;
1220 }
1221
1222 void r100_cp_fini(struct radeon_device *rdev)
1223 {
1224         if (r100_cp_wait_for_idle(rdev)) {
1225                 DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1226         }
1227         /* Disable ring */
1228         r100_cp_disable(rdev);
1229         radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1230         radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1231         DRM_INFO("radeon: cp finalized\n");
1232 }
1233
1234 void r100_cp_disable(struct radeon_device *rdev)
1235 {
1236         /* Disable ring */
1237         radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1238         rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1239         WREG32(RADEON_CP_CSQ_MODE, 0);
1240         WREG32(RADEON_CP_CSQ_CNTL, 0);
1241         WREG32(R_000770_SCRATCH_UMSK, 0);
1242         if (r100_gui_wait_for_idle(rdev)) {
1243                 printk(KERN_WARNING "Failed to wait GUI idle while "
1244                        "programming pipes. Bad things might happen.\n");
1245         }
1246 }
1247
1248 /*
1249  * CS functions
1250  */
1251 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1252                             struct radeon_cs_packet *pkt,
1253                             unsigned idx,
1254                             unsigned reg)
1255 {
1256         int r;
1257         u32 tile_flags = 0;
1258         u32 tmp;
1259         struct radeon_bo_list *reloc;
1260         u32 value;
1261
1262         r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1263         if (r) {
1264                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1265                           idx, reg);
1266                 radeon_cs_dump_packet(p, pkt);
1267                 return r;
1268         }
1269
1270         value = radeon_get_ib_value(p, idx);
1271         tmp = value & 0x003fffff;
1272         tmp += (((u32)reloc->gpu_offset) >> 10);
1273
1274         if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1275                 if (reloc->tiling_flags & RADEON_TILING_MACRO)
1276                         tile_flags |= RADEON_DST_TILE_MACRO;
1277                 if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1278                         if (reg == RADEON_SRC_PITCH_OFFSET) {
1279                                 DRM_ERROR("Cannot src blit from microtiled surface\n");
1280                                 radeon_cs_dump_packet(p, pkt);
1281                                 return -EINVAL;
1282                         }
1283                         tile_flags |= RADEON_DST_TILE_MICRO;
1284                 }
1285
1286                 tmp |= tile_flags;
1287                 p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1288         } else
1289                 p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1290         return 0;
1291 }
1292
1293 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1294                              struct radeon_cs_packet *pkt,
1295                              int idx)
1296 {
1297         unsigned c, i;
1298         struct radeon_bo_list *reloc;
1299         struct r100_cs_track *track;
1300         int r = 0;
1301         volatile uint32_t *ib;
1302         u32 idx_value;
1303
1304         ib = p->ib.ptr;
1305         track = (struct r100_cs_track *)p->track;
1306         c = radeon_get_ib_value(p, idx++) & 0x1F;
1307         if (c > 16) {
1308             DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1309                       pkt->opcode);
1310             radeon_cs_dump_packet(p, pkt);
1311             return -EINVAL;
1312         }
1313         track->num_arrays = c;
1314         for (i = 0; i < (c - 1); i+=2, idx+=3) {
1315                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1316                 if (r) {
1317                         DRM_ERROR("No reloc for packet3 %d\n",
1318                                   pkt->opcode);
1319                         radeon_cs_dump_packet(p, pkt);
1320                         return r;
1321                 }
1322                 idx_value = radeon_get_ib_value(p, idx);
1323                 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1324
1325                 track->arrays[i + 0].esize = idx_value >> 8;
1326                 track->arrays[i + 0].robj = reloc->robj;
1327                 track->arrays[i + 0].esize &= 0x7F;
1328                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1329                 if (r) {
1330                         DRM_ERROR("No reloc for packet3 %d\n",
1331                                   pkt->opcode);
1332                         radeon_cs_dump_packet(p, pkt);
1333                         return r;
1334                 }
1335                 ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
1336                 track->arrays[i + 1].robj = reloc->robj;
1337                 track->arrays[i + 1].esize = idx_value >> 24;
1338                 track->arrays[i + 1].esize &= 0x7F;
1339         }
1340         if (c & 1) {
1341                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1342                 if (r) {
1343                         DRM_ERROR("No reloc for packet3 %d\n",
1344                                           pkt->opcode);
1345                         radeon_cs_dump_packet(p, pkt);
1346                         return r;
1347                 }
1348                 idx_value = radeon_get_ib_value(p, idx);
1349                 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1350                 track->arrays[i + 0].robj = reloc->robj;
1351                 track->arrays[i + 0].esize = idx_value >> 8;
1352                 track->arrays[i + 0].esize &= 0x7F;
1353         }
1354         return r;
1355 }
1356
1357 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1358                           struct radeon_cs_packet *pkt,
1359                           const unsigned *auth, unsigned n,
1360                           radeon_packet0_check_t check)
1361 {
1362         unsigned reg;
1363         unsigned i, j, m;
1364         unsigned idx;
1365         int r;
1366
1367         idx = pkt->idx + 1;
1368         reg = pkt->reg;
1369         /* Check that register fall into register range
1370          * determined by the number of entry (n) in the
1371          * safe register bitmap.
1372          */
1373         if (pkt->one_reg_wr) {
1374                 if ((reg >> 7) > n) {
1375                         return -EINVAL;
1376                 }
1377         } else {
1378                 if (((reg + (pkt->count << 2)) >> 7) > n) {
1379                         return -EINVAL;
1380                 }
1381         }
1382         for (i = 0; i <= pkt->count; i++, idx++) {
1383                 j = (reg >> 7);
1384                 m = 1 << ((reg >> 2) & 31);
1385                 if (auth[j] & m) {
1386                         r = check(p, pkt, idx, reg);
1387                         if (r) {
1388                                 return r;
1389                         }
1390                 }
1391                 if (pkt->one_reg_wr) {
1392                         if (!(auth[j] & m)) {
1393                                 break;
1394                         }
1395                 } else {
1396                         reg += 4;
1397                 }
1398         }
1399         return 0;
1400 }
1401
1402 /**
1403  * r100_cs_packet_next_vline() - parse userspace VLINE packet
1404  * @parser:             parser structure holding parsing context.
1405  *
1406  * Userspace sends a special sequence for VLINE waits.
1407  * PACKET0 - VLINE_START_END + value
1408  * PACKET0 - WAIT_UNTIL +_value
1409  * RELOC (P3) - crtc_id in reloc.
1410  *
1411  * This function parses this and relocates the VLINE START END
1412  * and WAIT UNTIL packets to the correct crtc.
1413  * It also detects a switched off crtc and nulls out the
1414  * wait in that case.
1415  */
1416 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1417 {
1418         struct drm_crtc *crtc;
1419         struct radeon_crtc *radeon_crtc;
1420         struct radeon_cs_packet p3reloc, waitreloc;
1421         int crtc_id;
1422         int r;
1423         uint32_t header, h_idx, reg;
1424         volatile uint32_t *ib;
1425
1426         ib = p->ib.ptr;
1427
1428         /* parse the wait until */
1429         r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1430         if (r)
1431                 return r;
1432
1433         /* check its a wait until and only 1 count */
1434         if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1435             waitreloc.count != 0) {
1436                 DRM_ERROR("vline wait had illegal wait until segment\n");
1437                 return -EINVAL;
1438         }
1439
1440         if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1441                 DRM_ERROR("vline wait had illegal wait until\n");
1442                 return -EINVAL;
1443         }
1444
1445         /* jump over the NOP */
1446         r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1447         if (r)
1448                 return r;
1449
1450         h_idx = p->idx - 2;
1451         p->idx += waitreloc.count + 2;
1452         p->idx += p3reloc.count + 2;
1453
1454         header = radeon_get_ib_value(p, h_idx);
1455         crtc_id = radeon_get_ib_value(p, h_idx + 5);
1456         reg = R100_CP_PACKET0_GET_REG(header);
1457         crtc = drm_crtc_find(p->rdev->ddev, crtc_id);
1458         if (!crtc) {
1459                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
1460                 return -ENOENT;
1461         }
1462         radeon_crtc = to_radeon_crtc(crtc);
1463         crtc_id = radeon_crtc->crtc_id;
1464
1465         if (!crtc->enabled) {
1466                 /* if the CRTC isn't enabled - we need to nop out the wait until */
1467                 ib[h_idx + 2] = PACKET2(0);
1468                 ib[h_idx + 3] = PACKET2(0);
1469         } else if (crtc_id == 1) {
1470                 switch (reg) {
1471                 case AVIVO_D1MODE_VLINE_START_END:
1472                         header &= ~R300_CP_PACKET0_REG_MASK;
1473                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1474                         break;
1475                 case RADEON_CRTC_GUI_TRIG_VLINE:
1476                         header &= ~R300_CP_PACKET0_REG_MASK;
1477                         header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1478                         break;
1479                 default:
1480                         DRM_ERROR("unknown crtc reloc\n");
1481                         return -EINVAL;
1482                 }
1483                 ib[h_idx] = header;
1484                 ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1485         }
1486
1487         return 0;
1488 }
1489
1490 static int r100_get_vtx_size(uint32_t vtx_fmt)
1491 {
1492         int vtx_size;
1493         vtx_size = 2;
1494         /* ordered according to bits in spec */
1495         if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1496                 vtx_size++;
1497         if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1498                 vtx_size += 3;
1499         if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1500                 vtx_size++;
1501         if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1502                 vtx_size++;
1503         if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1504                 vtx_size += 3;
1505         if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1506                 vtx_size++;
1507         if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1508                 vtx_size++;
1509         if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1510                 vtx_size += 2;
1511         if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1512                 vtx_size += 2;
1513         if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1514                 vtx_size++;
1515         if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1516                 vtx_size += 2;
1517         if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1518                 vtx_size++;
1519         if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1520                 vtx_size += 2;
1521         if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1522                 vtx_size++;
1523         if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1524                 vtx_size++;
1525         /* blend weight */
1526         if (vtx_fmt & (0x7 << 15))
1527                 vtx_size += (vtx_fmt >> 15) & 0x7;
1528         if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1529                 vtx_size += 3;
1530         if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1531                 vtx_size += 2;
1532         if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1533                 vtx_size++;
1534         if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1535                 vtx_size++;
1536         if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1537                 vtx_size++;
1538         if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1539                 vtx_size++;
1540         return vtx_size;
1541 }
1542
1543 static int r100_packet0_check(struct radeon_cs_parser *p,
1544                               struct radeon_cs_packet *pkt,
1545                               unsigned idx, unsigned reg)
1546 {
1547         struct radeon_bo_list *reloc;
1548         struct r100_cs_track *track;
1549         volatile uint32_t *ib;
1550         uint32_t tmp;
1551         int r;
1552         int i, face;
1553         u32 tile_flags = 0;
1554         u32 idx_value;
1555
1556         ib = p->ib.ptr;
1557         track = (struct r100_cs_track *)p->track;
1558
1559         idx_value = radeon_get_ib_value(p, idx);
1560
1561         switch (reg) {
1562         case RADEON_CRTC_GUI_TRIG_VLINE:
1563                 r = r100_cs_packet_parse_vline(p);
1564                 if (r) {
1565                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1566                                   idx, reg);
1567                         radeon_cs_dump_packet(p, pkt);
1568                         return r;
1569                 }
1570                 break;
1571                 /* FIXME: only allow PACKET3 blit? easier to check for out of
1572                  * range access */
1573         case RADEON_DST_PITCH_OFFSET:
1574         case RADEON_SRC_PITCH_OFFSET:
1575                 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1576                 if (r)
1577                         return r;
1578                 break;
1579         case RADEON_RB3D_DEPTHOFFSET:
1580                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1581                 if (r) {
1582                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1583                                   idx, reg);
1584                         radeon_cs_dump_packet(p, pkt);
1585                         return r;
1586                 }
1587                 track->zb.robj = reloc->robj;
1588                 track->zb.offset = idx_value;
1589                 track->zb_dirty = true;
1590                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1591                 break;
1592         case RADEON_RB3D_COLOROFFSET:
1593                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1594                 if (r) {
1595                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1596                                   idx, reg);
1597                         radeon_cs_dump_packet(p, pkt);
1598                         return r;
1599                 }
1600                 track->cb[0].robj = reloc->robj;
1601                 track->cb[0].offset = idx_value;
1602                 track->cb_dirty = true;
1603                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1604                 break;
1605         case RADEON_PP_TXOFFSET_0:
1606         case RADEON_PP_TXOFFSET_1:
1607         case RADEON_PP_TXOFFSET_2:
1608                 i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1609                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1610                 if (r) {
1611                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1612                                   idx, reg);
1613                         radeon_cs_dump_packet(p, pkt);
1614                         return r;
1615                 }
1616                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1617                         if (reloc->tiling_flags & RADEON_TILING_MACRO)
1618                                 tile_flags |= RADEON_TXO_MACRO_TILE;
1619                         if (reloc->tiling_flags & RADEON_TILING_MICRO)
1620                                 tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1621
1622                         tmp = idx_value & ~(0x7 << 2);
1623                         tmp |= tile_flags;
1624                         ib[idx] = tmp + ((u32)reloc->gpu_offset);
1625                 } else
1626                         ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1627                 track->textures[i].robj = reloc->robj;
1628                 track->tex_dirty = true;
1629                 break;
1630         case RADEON_PP_CUBIC_OFFSET_T0_0:
1631         case RADEON_PP_CUBIC_OFFSET_T0_1:
1632         case RADEON_PP_CUBIC_OFFSET_T0_2:
1633         case RADEON_PP_CUBIC_OFFSET_T0_3:
1634         case RADEON_PP_CUBIC_OFFSET_T0_4:
1635                 i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1636                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1637                 if (r) {
1638                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1639                                   idx, reg);
1640                         radeon_cs_dump_packet(p, pkt);
1641                         return r;
1642                 }
1643                 track->textures[0].cube_info[i].offset = idx_value;
1644                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1645                 track->textures[0].cube_info[i].robj = reloc->robj;
1646                 track->tex_dirty = true;
1647                 break;
1648         case RADEON_PP_CUBIC_OFFSET_T1_0:
1649         case RADEON_PP_CUBIC_OFFSET_T1_1:
1650         case RADEON_PP_CUBIC_OFFSET_T1_2:
1651         case RADEON_PP_CUBIC_OFFSET_T1_3:
1652         case RADEON_PP_CUBIC_OFFSET_T1_4:
1653                 i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1654                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1655                 if (r) {
1656                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1657                                   idx, reg);
1658                         radeon_cs_dump_packet(p, pkt);
1659                         return r;
1660                 }
1661                 track->textures[1].cube_info[i].offset = idx_value;
1662                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1663                 track->textures[1].cube_info[i].robj = reloc->robj;
1664                 track->tex_dirty = true;
1665                 break;
1666         case RADEON_PP_CUBIC_OFFSET_T2_0:
1667         case RADEON_PP_CUBIC_OFFSET_T2_1:
1668         case RADEON_PP_CUBIC_OFFSET_T2_2:
1669         case RADEON_PP_CUBIC_OFFSET_T2_3:
1670         case RADEON_PP_CUBIC_OFFSET_T2_4:
1671                 i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1672                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1673                 if (r) {
1674                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1675                                   idx, reg);
1676                         radeon_cs_dump_packet(p, pkt);
1677                         return r;
1678                 }
1679                 track->textures[2].cube_info[i].offset = idx_value;
1680                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1681                 track->textures[2].cube_info[i].robj = reloc->robj;
1682                 track->tex_dirty = true;
1683                 break;
1684         case RADEON_RE_WIDTH_HEIGHT:
1685                 track->maxy = ((idx_value >> 16) & 0x7FF);
1686                 track->cb_dirty = true;
1687                 track->zb_dirty = true;
1688                 break;
1689         case RADEON_RB3D_COLORPITCH:
1690                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1691                 if (r) {
1692                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1693                                   idx, reg);
1694                         radeon_cs_dump_packet(p, pkt);
1695                         return r;
1696                 }
1697                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1698                         if (reloc->tiling_flags & RADEON_TILING_MACRO)
1699                                 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1700                         if (reloc->tiling_flags & RADEON_TILING_MICRO)
1701                                 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1702
1703                         tmp = idx_value & ~(0x7 << 16);
1704                         tmp |= tile_flags;
1705                         ib[idx] = tmp;
1706                 } else
1707                         ib[idx] = idx_value;
1708
1709                 track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1710                 track->cb_dirty = true;
1711                 break;
1712         case RADEON_RB3D_DEPTHPITCH:
1713                 track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1714                 track->zb_dirty = true;
1715                 break;
1716         case RADEON_RB3D_CNTL:
1717                 switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1718                 case 7:
1719                 case 8:
1720                 case 9:
1721                 case 11:
1722                 case 12:
1723                         track->cb[0].cpp = 1;
1724                         break;
1725                 case 3:
1726                 case 4:
1727                 case 15:
1728                         track->cb[0].cpp = 2;
1729                         break;
1730                 case 6:
1731                         track->cb[0].cpp = 4;
1732                         break;
1733                 default:
1734                         DRM_ERROR("Invalid color buffer format (%d) !\n",
1735                                   ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1736                         return -EINVAL;
1737                 }
1738                 track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1739                 track->cb_dirty = true;
1740                 track->zb_dirty = true;
1741                 break;
1742         case RADEON_RB3D_ZSTENCILCNTL:
1743                 switch (idx_value & 0xf) {
1744                 case 0:
1745                         track->zb.cpp = 2;
1746                         break;
1747                 case 2:
1748                 case 3:
1749                 case 4:
1750                 case 5:
1751                 case 9:
1752                 case 11:
1753                         track->zb.cpp = 4;
1754                         break;
1755                 default:
1756                         break;
1757                 }
1758                 track->zb_dirty = true;
1759                 break;
1760         case RADEON_RB3D_ZPASS_ADDR:
1761                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1762                 if (r) {
1763                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1764                                   idx, reg);
1765                         radeon_cs_dump_packet(p, pkt);
1766                         return r;
1767                 }
1768                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1769                 break;
1770         case RADEON_PP_CNTL:
1771                 {
1772                         uint32_t temp = idx_value >> 4;
1773                         for (i = 0; i < track->num_texture; i++)
1774                                 track->textures[i].enabled = !!(temp & (1 << i));
1775                         track->tex_dirty = true;
1776                 }
1777                 break;
1778         case RADEON_SE_VF_CNTL:
1779                 track->vap_vf_cntl = idx_value;
1780                 break;
1781         case RADEON_SE_VTX_FMT:
1782                 track->vtx_size = r100_get_vtx_size(idx_value);
1783                 break;
1784         case RADEON_PP_TEX_SIZE_0:
1785         case RADEON_PP_TEX_SIZE_1:
1786         case RADEON_PP_TEX_SIZE_2:
1787                 i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1788                 track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1789                 track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1790                 track->tex_dirty = true;
1791                 break;
1792         case RADEON_PP_TEX_PITCH_0:
1793         case RADEON_PP_TEX_PITCH_1:
1794         case RADEON_PP_TEX_PITCH_2:
1795                 i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1796                 track->textures[i].pitch = idx_value + 32;
1797                 track->tex_dirty = true;
1798                 break;
1799         case RADEON_PP_TXFILTER_0:
1800         case RADEON_PP_TXFILTER_1:
1801         case RADEON_PP_TXFILTER_2:
1802                 i = (reg - RADEON_PP_TXFILTER_0) / 24;
1803                 track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1804                                                  >> RADEON_MAX_MIP_LEVEL_SHIFT);
1805                 tmp = (idx_value >> 23) & 0x7;
1806                 if (tmp == 2 || tmp == 6)
1807                         track->textures[i].roundup_w = false;
1808                 tmp = (idx_value >> 27) & 0x7;
1809                 if (tmp == 2 || tmp == 6)
1810                         track->textures[i].roundup_h = false;
1811                 track->tex_dirty = true;
1812                 break;
1813         case RADEON_PP_TXFORMAT_0:
1814         case RADEON_PP_TXFORMAT_1:
1815         case RADEON_PP_TXFORMAT_2:
1816                 i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1817                 if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1818                         track->textures[i].use_pitch = 1;
1819                 } else {
1820                         track->textures[i].use_pitch = 0;
1821                         track->textures[i].width = 1 << ((idx_value & RADEON_TXFORMAT_WIDTH_MASK) >> RADEON_TXFORMAT_WIDTH_SHIFT);
1822                         track->textures[i].height = 1 << ((idx_value & RADEON_TXFORMAT_HEIGHT_MASK) >> RADEON_TXFORMAT_HEIGHT_SHIFT);
1823                 }
1824                 if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1825                         track->textures[i].tex_coord_type = 2;
1826                 switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1827                 case RADEON_TXFORMAT_I8:
1828                 case RADEON_TXFORMAT_RGB332:
1829                 case RADEON_TXFORMAT_Y8:
1830                         track->textures[i].cpp = 1;
1831                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1832                         break;
1833                 case RADEON_TXFORMAT_AI88:
1834                 case RADEON_TXFORMAT_ARGB1555:
1835                 case RADEON_TXFORMAT_RGB565:
1836                 case RADEON_TXFORMAT_ARGB4444:
1837                 case RADEON_TXFORMAT_VYUY422:
1838                 case RADEON_TXFORMAT_YVYU422:
1839                 case RADEON_TXFORMAT_SHADOW16:
1840                 case RADEON_TXFORMAT_LDUDV655:
1841                 case RADEON_TXFORMAT_DUDV88:
1842                         track->textures[i].cpp = 2;
1843                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1844                         break;
1845                 case RADEON_TXFORMAT_ARGB8888:
1846                 case RADEON_TXFORMAT_RGBA8888:
1847                 case RADEON_TXFORMAT_SHADOW32:
1848                 case RADEON_TXFORMAT_LDUDUV8888:
1849                         track->textures[i].cpp = 4;
1850                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1851                         break;
1852                 case RADEON_TXFORMAT_DXT1:
1853                         track->textures[i].cpp = 1;
1854                         track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1855                         break;
1856                 case RADEON_TXFORMAT_DXT23:
1857                 case RADEON_TXFORMAT_DXT45:
1858                         track->textures[i].cpp = 1;
1859                         track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1860                         break;
1861                 }
1862                 track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1863                 track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1864                 track->tex_dirty = true;
1865                 break;
1866         case RADEON_PP_CUBIC_FACES_0:
1867         case RADEON_PP_CUBIC_FACES_1:
1868         case RADEON_PP_CUBIC_FACES_2:
1869                 tmp = idx_value;
1870                 i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1871                 for (face = 0; face < 4; face++) {
1872                         track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1873                         track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1874                 }
1875                 track->tex_dirty = true;
1876                 break;
1877         default:
1878                 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1879                        reg, idx);
1880                 return -EINVAL;
1881         }
1882         return 0;
1883 }
1884
1885 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1886                                          struct radeon_cs_packet *pkt,
1887                                          struct radeon_bo *robj)
1888 {
1889         unsigned idx;
1890         u32 value;
1891         idx = pkt->idx + 1;
1892         value = radeon_get_ib_value(p, idx + 2);
1893         if ((value + 1) > radeon_bo_size(robj)) {
1894                 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1895                           "(need %u have %lu) !\n",
1896                           value + 1,
1897                           radeon_bo_size(robj));
1898                 return -EINVAL;
1899         }
1900         return 0;
1901 }
1902
1903 static int r100_packet3_check(struct radeon_cs_parser *p,
1904                               struct radeon_cs_packet *pkt)
1905 {
1906         struct radeon_bo_list *reloc;
1907         struct r100_cs_track *track;
1908         unsigned idx;
1909         volatile uint32_t *ib;
1910         int r;
1911
1912         ib = p->ib.ptr;
1913         idx = pkt->idx + 1;
1914         track = (struct r100_cs_track *)p->track;
1915         switch (pkt->opcode) {
1916         case PACKET3_3D_LOAD_VBPNTR:
1917                 r = r100_packet3_load_vbpntr(p, pkt, idx);
1918                 if (r)
1919                         return r;
1920                 break;
1921         case PACKET3_INDX_BUFFER:
1922                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1923                 if (r) {
1924                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1925                         radeon_cs_dump_packet(p, pkt);
1926                         return r;
1927                 }
1928                 ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1929                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1930                 if (r) {
1931                         return r;
1932                 }
1933                 break;
1934         case 0x23:
1935                 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1936                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1937                 if (r) {
1938                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1939                         radeon_cs_dump_packet(p, pkt);
1940                         return r;
1941                 }
1942                 ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1943                 track->num_arrays = 1;
1944                 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1945
1946                 track->arrays[0].robj = reloc->robj;
1947                 track->arrays[0].esize = track->vtx_size;
1948
1949                 track->max_indx = radeon_get_ib_value(p, idx+1);
1950
1951                 track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1952                 track->immd_dwords = pkt->count - 1;
1953                 r = r100_cs_track_check(p->rdev, track);
1954                 if (r)
1955                         return r;
1956                 break;
1957         case PACKET3_3D_DRAW_IMMD:
1958                 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1959                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1960                         return -EINVAL;
1961                 }
1962                 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1963                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1964                 track->immd_dwords = pkt->count - 1;
1965                 r = r100_cs_track_check(p->rdev, track);
1966                 if (r)
1967                         return r;
1968                 break;
1969                 /* triggers drawing using in-packet vertex data */
1970         case PACKET3_3D_DRAW_IMMD_2:
1971                 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1972                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1973                         return -EINVAL;
1974                 }
1975                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1976                 track->immd_dwords = pkt->count;
1977                 r = r100_cs_track_check(p->rdev, track);
1978                 if (r)
1979                         return r;
1980                 break;
1981                 /* triggers drawing using in-packet vertex data */
1982         case PACKET3_3D_DRAW_VBUF_2:
1983                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1984                 r = r100_cs_track_check(p->rdev, track);
1985                 if (r)
1986                         return r;
1987                 break;
1988                 /* triggers drawing of vertex buffers setup elsewhere */
1989         case PACKET3_3D_DRAW_INDX_2:
1990                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1991                 r = r100_cs_track_check(p->rdev, track);
1992                 if (r)
1993                         return r;
1994                 break;
1995                 /* triggers drawing using indices to vertex buffer */
1996         case PACKET3_3D_DRAW_VBUF:
1997                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1998                 r = r100_cs_track_check(p->rdev, track);
1999                 if (r)
2000                         return r;
2001                 break;
2002                 /* triggers drawing of vertex buffers setup elsewhere */
2003         case PACKET3_3D_DRAW_INDX:
2004                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2005                 r = r100_cs_track_check(p->rdev, track);
2006                 if (r)
2007                         return r;
2008                 break;
2009                 /* triggers drawing using indices to vertex buffer */
2010         case PACKET3_3D_CLEAR_HIZ:
2011         case PACKET3_3D_CLEAR_ZMASK:
2012                 if (p->rdev->hyperz_filp != p->filp)
2013                         return -EINVAL;
2014                 break;
2015         case PACKET3_NOP:
2016                 break;
2017         default:
2018                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2019                 return -EINVAL;
2020         }
2021         return 0;
2022 }
2023
2024 int r100_cs_parse(struct radeon_cs_parser *p)
2025 {
2026         struct radeon_cs_packet pkt;
2027         struct r100_cs_track *track;
2028         int r;
2029
2030         track = kzalloc(sizeof(*track), GFP_KERNEL);
2031         if (!track)
2032                 return -ENOMEM;
2033         r100_cs_track_clear(p->rdev, track);
2034         p->track = track;
2035         do {
2036                 r = radeon_cs_packet_parse(p, &pkt, p->idx);
2037                 if (r) {
2038                         return r;
2039                 }
2040                 p->idx += pkt.count + 2;
2041                 switch (pkt.type) {
2042                 case RADEON_PACKET_TYPE0:
2043                         if (p->rdev->family >= CHIP_R200)
2044                                 r = r100_cs_parse_packet0(p, &pkt,
2045                                         p->rdev->config.r100.reg_safe_bm,
2046                                         p->rdev->config.r100.reg_safe_bm_size,
2047                                         &r200_packet0_check);
2048                         else
2049                                 r = r100_cs_parse_packet0(p, &pkt,
2050                                         p->rdev->config.r100.reg_safe_bm,
2051                                         p->rdev->config.r100.reg_safe_bm_size,
2052                                         &r100_packet0_check);
2053                         break;
2054                 case RADEON_PACKET_TYPE2:
2055                         break;
2056                 case RADEON_PACKET_TYPE3:
2057                         r = r100_packet3_check(p, &pkt);
2058                         break;
2059                 default:
2060                         DRM_ERROR("Unknown packet type %d !\n",
2061                                   pkt.type);
2062                         return -EINVAL;
2063                 }
2064                 if (r)
2065                         return r;
2066         } while (p->idx < p->chunk_ib->length_dw);
2067         return 0;
2068 }
2069
2070 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2071 {
2072         DRM_ERROR("pitch                      %d\n", t->pitch);
2073         DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2074         DRM_ERROR("width                      %d\n", t->width);
2075         DRM_ERROR("width_11                   %d\n", t->width_11);
2076         DRM_ERROR("height                     %d\n", t->height);
2077         DRM_ERROR("height_11                  %d\n", t->height_11);
2078         DRM_ERROR("num levels                 %d\n", t->num_levels);
2079         DRM_ERROR("depth                      %d\n", t->txdepth);
2080         DRM_ERROR("bpp                        %d\n", t->cpp);
2081         DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2082         DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2083         DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2084         DRM_ERROR("compress format            %d\n", t->compress_format);
2085 }
2086
2087 static int r100_track_compress_size(int compress_format, int w, int h)
2088 {
2089         int block_width, block_height, block_bytes;
2090         int wblocks, hblocks;
2091         int min_wblocks;
2092         int sz;
2093
2094         block_width = 4;
2095         block_height = 4;
2096
2097         switch (compress_format) {
2098         case R100_TRACK_COMP_DXT1:
2099                 block_bytes = 8;
2100                 min_wblocks = 4;
2101                 break;
2102         default:
2103         case R100_TRACK_COMP_DXT35:
2104                 block_bytes = 16;
2105                 min_wblocks = 2;
2106                 break;
2107         }
2108
2109         hblocks = (h + block_height - 1) / block_height;
2110         wblocks = (w + block_width - 1) / block_width;
2111         if (wblocks < min_wblocks)
2112                 wblocks = min_wblocks;
2113         sz = wblocks * hblocks * block_bytes;
2114         return sz;
2115 }
2116
2117 static int r100_cs_track_cube(struct radeon_device *rdev,
2118                               struct r100_cs_track *track, unsigned idx)
2119 {
2120         unsigned face, w, h;
2121         struct radeon_bo *cube_robj;
2122         unsigned long size;
2123         unsigned compress_format = track->textures[idx].compress_format;
2124
2125         for (face = 0; face < 5; face++) {
2126                 cube_robj = track->textures[idx].cube_info[face].robj;
2127                 w = track->textures[idx].cube_info[face].width;
2128                 h = track->textures[idx].cube_info[face].height;
2129
2130                 if (compress_format) {
2131                         size = r100_track_compress_size(compress_format, w, h);
2132                 } else
2133                         size = w * h;
2134                 size *= track->textures[idx].cpp;
2135
2136                 size += track->textures[idx].cube_info[face].offset;
2137
2138                 if (size > radeon_bo_size(cube_robj)) {
2139                         DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2140                                   size, radeon_bo_size(cube_robj));
2141                         r100_cs_track_texture_print(&track->textures[idx]);
2142                         return -1;
2143                 }
2144         }
2145         return 0;
2146 }
2147
2148 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2149                                        struct r100_cs_track *track)
2150 {
2151         struct radeon_bo *robj;
2152         unsigned long size;
2153         unsigned u, i, w, h, d;
2154         int ret;
2155
2156         for (u = 0; u < track->num_texture; u++) {
2157                 if (!track->textures[u].enabled)
2158                         continue;
2159                 if (track->textures[u].lookup_disable)
2160                         continue;
2161                 robj = track->textures[u].robj;
2162                 if (robj == NULL) {
2163                         DRM_ERROR("No texture bound to unit %u\n", u);
2164                         return -EINVAL;
2165                 }
2166                 size = 0;
2167                 for (i = 0; i <= track->textures[u].num_levels; i++) {
2168                         if (track->textures[u].use_pitch) {
2169                                 if (rdev->family < CHIP_R300)
2170                                         w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2171                                 else
2172                                         w = track->textures[u].pitch / (1 << i);
2173                         } else {
2174                                 w = track->textures[u].width;
2175                                 if (rdev->family >= CHIP_RV515)
2176                                         w |= track->textures[u].width_11;
2177                                 w = w / (1 << i);
2178                                 if (track->textures[u].roundup_w)
2179                                         w = roundup_pow_of_two(w);
2180                         }
2181                         h = track->textures[u].height;
2182                         if (rdev->family >= CHIP_RV515)
2183                                 h |= track->textures[u].height_11;
2184                         h = h / (1 << i);
2185                         if (track->textures[u].roundup_h)
2186                                 h = roundup_pow_of_two(h);
2187                         if (track->textures[u].tex_coord_type == 1) {
2188                                 d = (1 << track->textures[u].txdepth) / (1 << i);
2189                                 if (!d)
2190                                         d = 1;
2191                         } else {
2192                                 d = 1;
2193                         }
2194                         if (track->textures[u].compress_format) {
2195
2196                                 size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2197                                 /* compressed textures are block based */
2198                         } else
2199                                 size += w * h * d;
2200                 }
2201                 size *= track->textures[u].cpp;
2202
2203                 switch (track->textures[u].tex_coord_type) {
2204                 case 0:
2205                 case 1:
2206                         break;
2207                 case 2:
2208                         if (track->separate_cube) {
2209                                 ret = r100_cs_track_cube(rdev, track, u);
2210                                 if (ret)
2211                                         return ret;
2212                         } else
2213                                 size *= 6;
2214                         break;
2215                 default:
2216                         DRM_ERROR("Invalid texture coordinate type %u for unit "
2217                                   "%u\n", track->textures[u].tex_coord_type, u);
2218                         return -EINVAL;
2219                 }
2220                 if (size > radeon_bo_size(robj)) {
2221                         DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2222                                   "%lu\n", u, size, radeon_bo_size(robj));
2223                         r100_cs_track_texture_print(&track->textures[u]);
2224                         return -EINVAL;
2225                 }
2226         }
2227         return 0;
2228 }
2229
2230 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2231 {
2232         unsigned i;
2233         unsigned long size;
2234         unsigned prim_walk;
2235         unsigned nverts;
2236         unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2237
2238         if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2239             !track->blend_read_enable)
2240                 num_cb = 0;
2241
2242         for (i = 0; i < num_cb; i++) {
2243                 if (track->cb[i].robj == NULL) {
2244                         DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2245                         return -EINVAL;
2246                 }
2247                 size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2248                 size += track->cb[i].offset;
2249                 if (size > radeon_bo_size(track->cb[i].robj)) {
2250                         DRM_ERROR("[drm] Buffer too small for color buffer %d "
2251                                   "(need %lu have %lu) !\n", i, size,
2252                                   radeon_bo_size(track->cb[i].robj));
2253                         DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2254                                   i, track->cb[i].pitch, track->cb[i].cpp,
2255                                   track->cb[i].offset, track->maxy);
2256                         return -EINVAL;
2257                 }
2258         }
2259         track->cb_dirty = false;
2260
2261         if (track->zb_dirty && track->z_enabled) {
2262                 if (track->zb.robj == NULL) {
2263                         DRM_ERROR("[drm] No buffer for z buffer !\n");
2264                         return -EINVAL;
2265                 }
2266                 size = track->zb.pitch * track->zb.cpp * track->maxy;
2267                 size += track->zb.offset;
2268                 if (size > radeon_bo_size(track->zb.robj)) {
2269                         DRM_ERROR("[drm] Buffer too small for z buffer "
2270                                   "(need %lu have %lu) !\n", size,
2271                                   radeon_bo_size(track->zb.robj));
2272                         DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2273                                   track->zb.pitch, track->zb.cpp,
2274                                   track->zb.offset, track->maxy);
2275                         return -EINVAL;
2276                 }
2277         }
2278         track->zb_dirty = false;
2279
2280         if (track->aa_dirty && track->aaresolve) {
2281                 if (track->aa.robj == NULL) {
2282                         DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2283                         return -EINVAL;
2284                 }
2285                 /* I believe the format comes from colorbuffer0. */
2286                 size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2287                 size += track->aa.offset;
2288                 if (size > radeon_bo_size(track->aa.robj)) {
2289                         DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2290                                   "(need %lu have %lu) !\n", i, size,
2291                                   radeon_bo_size(track->aa.robj));
2292                         DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2293                                   i, track->aa.pitch, track->cb[0].cpp,
2294                                   track->aa.offset, track->maxy);
2295                         return -EINVAL;
2296                 }
2297         }
2298         track->aa_dirty = false;
2299
2300         prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2301         if (track->vap_vf_cntl & (1 << 14)) {
2302                 nverts = track->vap_alt_nverts;
2303         } else {
2304                 nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2305         }
2306         switch (prim_walk) {
2307         case 1:
2308                 for (i = 0; i < track->num_arrays; i++) {
2309                         size = track->arrays[i].esize * track->max_indx * 4;
2310                         if (track->arrays[i].robj == NULL) {
2311                                 DRM_ERROR("(PW %u) Vertex array %u no buffer "
2312                                           "bound\n", prim_walk, i);
2313                                 return -EINVAL;
2314                         }
2315                         if (size > radeon_bo_size(track->arrays[i].robj)) {
2316                                 dev_err(rdev->dev, "(PW %u) Vertex array %u "
2317                                         "need %lu dwords have %lu dwords\n",
2318                                         prim_walk, i, size >> 2,
2319                                         radeon_bo_size(track->arrays[i].robj)
2320                                         >> 2);
2321                                 DRM_ERROR("Max indices %u\n", track->max_indx);
2322                                 return -EINVAL;
2323                         }
2324                 }
2325                 break;
2326         case 2:
2327                 for (i = 0; i < track->num_arrays; i++) {
2328                         size = track->arrays[i].esize * (nverts - 1) * 4;
2329                         if (track->arrays[i].robj == NULL) {
2330                                 DRM_ERROR("(PW %u) Vertex array %u no buffer "
2331                                           "bound\n", prim_walk, i);
2332                                 return -EINVAL;
2333                         }
2334                         if (size > radeon_bo_size(track->arrays[i].robj)) {
2335                                 dev_err(rdev->dev, "(PW %u) Vertex array %u "
2336                                         "need %lu dwords have %lu dwords\n",
2337                                         prim_walk, i, size >> 2,
2338                                         radeon_bo_size(track->arrays[i].robj)
2339                                         >> 2);
2340                                 return -EINVAL;
2341                         }
2342                 }
2343                 break;
2344         case 3:
2345                 size = track->vtx_size * nverts;
2346                 if (size != track->immd_dwords) {
2347                         DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2348                                   track->immd_dwords, size);
2349                         DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2350                                   nverts, track->vtx_size);
2351                         return -EINVAL;
2352                 }
2353                 break;
2354         default:
2355                 DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2356                           prim_walk);
2357                 return -EINVAL;
2358         }
2359
2360         if (track->tex_dirty) {
2361                 track->tex_dirty = false;
2362                 return r100_cs_track_texture_check(rdev, track);
2363         }
2364         return 0;
2365 }
2366
2367 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2368 {
2369         unsigned i, face;
2370
2371         track->cb_dirty = true;
2372         track->zb_dirty = true;
2373         track->tex_dirty = true;
2374         track->aa_dirty = true;
2375
2376         if (rdev->family < CHIP_R300) {
2377                 track->num_cb = 1;
2378                 if (rdev->family <= CHIP_RS200)
2379                         track->num_texture = 3;
2380                 else
2381                         track->num_texture = 6;
2382                 track->maxy = 2048;
2383                 track->separate_cube = 1;
2384         } else {
2385                 track->num_cb = 4;
2386                 track->num_texture = 16;
2387                 track->maxy = 4096;
2388                 track->separate_cube = 0;
2389                 track->aaresolve = false;
2390                 track->aa.robj = NULL;
2391         }
2392
2393         for (i = 0; i < track->num_cb; i++) {
2394                 track->cb[i].robj = NULL;
2395                 track->cb[i].pitch = 8192;
2396                 track->cb[i].cpp = 16;
2397                 track->cb[i].offset = 0;
2398         }
2399         track->z_enabled = true;
2400         track->zb.robj = NULL;
2401         track->zb.pitch = 8192;
2402         track->zb.cpp = 4;
2403         track->zb.offset = 0;
2404         track->vtx_size = 0x7F;
2405         track->immd_dwords = 0xFFFFFFFFUL;
2406         track->num_arrays = 11;
2407         track->max_indx = 0x00FFFFFFUL;
2408         for (i = 0; i < track->num_arrays; i++) {
2409                 track->arrays[i].robj = NULL;
2410                 track->arrays[i].esize = 0x7F;
2411         }
2412         for (i = 0; i < track->num_texture; i++) {
2413                 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2414                 track->textures[i].pitch = 16536;
2415                 track->textures[i].width = 16536;
2416                 track->textures[i].height = 16536;
2417                 track->textures[i].width_11 = 1 << 11;
2418                 track->textures[i].height_11 = 1 << 11;
2419                 track->textures[i].num_levels = 12;
2420                 if (rdev->family <= CHIP_RS200) {
2421                         track->textures[i].tex_coord_type = 0;
2422                         track->textures[i].txdepth = 0;
2423                 } else {
2424                         track->textures[i].txdepth = 16;
2425                         track->textures[i].tex_coord_type = 1;
2426                 }
2427                 track->textures[i].cpp = 64;
2428                 track->textures[i].robj = NULL;
2429                 /* CS IB emission code makes sure texture unit are disabled */
2430                 track->textures[i].enabled = false;
2431                 track->textures[i].lookup_disable = false;
2432                 track->textures[i].roundup_w = true;
2433                 track->textures[i].roundup_h = true;
2434                 if (track->separate_cube)
2435                         for (face = 0; face < 5; face++) {
2436                                 track->textures[i].cube_info[face].robj = NULL;
2437                                 track->textures[i].cube_info[face].width = 16536;
2438                                 track->textures[i].cube_info[face].height = 16536;
2439                                 track->textures[i].cube_info[face].offset = 0;
2440                         }
2441         }
2442 }
2443
2444 /*
2445  * Global GPU functions
2446  */
2447 static void r100_errata(struct radeon_device *rdev)
2448 {
2449         rdev->pll_errata = 0;
2450
2451         if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2452                 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2453         }
2454
2455         if (rdev->family == CHIP_RV100 ||
2456             rdev->family == CHIP_RS100 ||
2457             rdev->family == CHIP_RS200) {
2458                 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2459         }
2460 }
2461
2462 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2463 {
2464         unsigned i;
2465         uint32_t tmp;
2466
2467         for (i = 0; i < rdev->usec_timeout; i++) {
2468                 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2469                 if (tmp >= n) {
2470                         return 0;
2471                 }
2472                 DRM_UDELAY(1);
2473         }
2474         return -1;
2475 }
2476
2477 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2478 {
2479         unsigned i;
2480         uint32_t tmp;
2481
2482         if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2483                 printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
2484                        " Bad things might happen.\n");
2485         }
2486         for (i = 0; i < rdev->usec_timeout; i++) {
2487                 tmp = RREG32(RADEON_RBBM_STATUS);
2488                 if (!(tmp & RADEON_RBBM_ACTIVE)) {
2489                         return 0;
2490                 }
2491                 DRM_UDELAY(1);
2492         }
2493         return -1;
2494 }
2495
2496 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2497 {
2498         unsigned i;
2499         uint32_t tmp;
2500
2501         for (i = 0; i < rdev->usec_timeout; i++) {
2502                 /* read MC_STATUS */
2503                 tmp = RREG32(RADEON_MC_STATUS);
2504                 if (tmp & RADEON_MC_IDLE) {
2505                         return 0;
2506                 }
2507                 DRM_UDELAY(1);
2508         }
2509         return -1;
2510 }
2511
2512 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2513 {
2514         u32 rbbm_status;
2515
2516         rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2517         if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2518                 radeon_ring_lockup_update(rdev, ring);
2519                 return false;
2520         }
2521         return radeon_ring_test_lockup(rdev, ring);
2522 }
2523
2524 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2525 void r100_enable_bm(struct radeon_device *rdev)
2526 {
2527         uint32_t tmp;
2528         /* Enable bus mastering */
2529         tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2530         WREG32(RADEON_BUS_CNTL, tmp);
2531 }
2532
2533 void r100_bm_disable(struct radeon_device *rdev)
2534 {
2535         u32 tmp;
2536
2537         /* disable bus mastering */
2538         tmp = RREG32(R_000030_BUS_CNTL);
2539         WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2540         mdelay(1);
2541         WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2542         mdelay(1);
2543         WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2544         tmp = RREG32(RADEON_BUS_CNTL);
2545         mdelay(1);
2546         pci_clear_master(rdev->pdev);
2547         mdelay(1);
2548 }
2549
2550 int r100_asic_reset(struct radeon_device *rdev, bool hard)
2551 {
2552         struct r100_mc_save save;
2553         u32 status, tmp;
2554         int ret = 0;
2555
2556         status = RREG32(R_000E40_RBBM_STATUS);
2557         if (!G_000E40_GUI_ACTIVE(status)) {
2558                 return 0;
2559         }
2560         r100_mc_stop(rdev, &save);
2561         status = RREG32(R_000E40_RBBM_STATUS);
2562         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2563         /* stop CP */
2564         WREG32(RADEON_CP_CSQ_CNTL, 0);
2565         tmp = RREG32(RADEON_CP_RB_CNTL);
2566         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2567         WREG32(RADEON_CP_RB_RPTR_WR, 0);
2568         WREG32(RADEON_CP_RB_WPTR, 0);
2569         WREG32(RADEON_CP_RB_CNTL, tmp);
2570         /* save PCI state */
2571         pci_save_state(rdev->pdev);
2572         /* disable bus mastering */
2573         r100_bm_disable(rdev);
2574         WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2575                                         S_0000F0_SOFT_RESET_RE(1) |
2576                                         S_0000F0_SOFT_RESET_PP(1) |
2577                                         S_0000F0_SOFT_RESET_RB(1));
2578         RREG32(R_0000F0_RBBM_SOFT_RESET);
2579         mdelay(500);
2580         WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2581         mdelay(1);
2582         status = RREG32(R_000E40_RBBM_STATUS);
2583         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2584         /* reset CP */
2585         WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2586         RREG32(R_0000F0_RBBM_SOFT_RESET);
2587         mdelay(500);
2588         WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2589         mdelay(1);
2590         status = RREG32(R_000E40_RBBM_STATUS);
2591         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2592         /* restore PCI & busmastering */
2593         pci_restore_state(rdev->pdev);
2594         r100_enable_bm(rdev);
2595         /* Check if GPU is idle */
2596         if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2597                 G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2598                 dev_err(rdev->dev, "failed to reset GPU\n");
2599                 ret = -1;
2600         } else
2601                 dev_info(rdev->dev, "GPU reset succeed\n");
2602         r100_mc_resume(rdev, &save);
2603         return ret;
2604 }
2605
2606 void r100_set_common_regs(struct radeon_device *rdev)
2607 {
2608         struct drm_device *dev = rdev->ddev;
2609         bool force_dac2 = false;
2610         u32 tmp;
2611
2612         /* set these so they don't interfere with anything */
2613         WREG32(RADEON_OV0_SCALE_CNTL, 0);
2614         WREG32(RADEON_SUBPIC_CNTL, 0);
2615         WREG32(RADEON_VIPH_CONTROL, 0);
2616         WREG32(RADEON_I2C_CNTL_1, 0);
2617         WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2618         WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2619         WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2620
2621         /* always set up dac2 on rn50 and some rv100 as lots
2622          * of servers seem to wire it up to a VGA port but
2623          * don't report it in the bios connector
2624          * table.
2625          */
2626         switch (dev->pdev->device) {
2627                 /* RN50 */
2628         case 0x515e:
2629         case 0x5969:
2630                 force_dac2 = true;
2631                 break;
2632                 /* RV100*/
2633         case 0x5159:
2634         case 0x515a:
2635                 /* DELL triple head servers */
2636                 if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2637                     ((dev->pdev->subsystem_device == 0x016c) ||
2638                      (dev->pdev->subsystem_device == 0x016d) ||
2639                      (dev->pdev->subsystem_device == 0x016e) ||
2640                      (dev->pdev->subsystem_device == 0x016f) ||
2641                      (dev->pdev->subsystem_device == 0x0170) ||
2642                      (dev->pdev->subsystem_device == 0x017d) ||
2643                      (dev->pdev->subsystem_device == 0x017e) ||
2644                      (dev->pdev->subsystem_device == 0x0183) ||
2645                      (dev->pdev->subsystem_device == 0x018a) ||
2646                      (dev->pdev->subsystem_device == 0x019a)))
2647                         force_dac2 = true;
2648                 break;
2649         }
2650
2651         if (force_dac2) {
2652                 u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2653                 u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2654                 u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2655
2656                 /* For CRT on DAC2, don't turn it on if BIOS didn't
2657                    enable it, even it's detected.
2658                 */
2659
2660                 /* force it to crtc0 */
2661                 dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2662                 dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2663                 disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2664
2665                 /* set up the TV DAC */
2666                 tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2667                                  RADEON_TV_DAC_STD_MASK |
2668                                  RADEON_TV_DAC_RDACPD |
2669                                  RADEON_TV_DAC_GDACPD |
2670                                  RADEON_TV_DAC_BDACPD |
2671                                  RADEON_TV_DAC_BGADJ_MASK |
2672                                  RADEON_TV_DAC_DACADJ_MASK);
2673                 tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2674                                 RADEON_TV_DAC_NHOLD |
2675                                 RADEON_TV_DAC_STD_PS2 |
2676                                 (0x58 << 16));
2677
2678                 WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2679                 WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2680                 WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2681         }
2682
2683         /* switch PM block to ACPI mode */
2684         tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2685         tmp &= ~RADEON_PM_MODE_SEL;
2686         WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2687
2688 }
2689
2690 /*
2691  * VRAM info
2692  */
2693 static void r100_vram_get_type(struct radeon_device *rdev)
2694 {
2695         uint32_t tmp;
2696
2697         rdev->mc.vram_is_ddr = false;
2698         if (rdev->flags & RADEON_IS_IGP)
2699                 rdev->mc.vram_is_ddr = true;
2700         else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2701                 rdev->mc.vram_is_ddr = true;
2702         if ((rdev->family == CHIP_RV100) ||
2703             (rdev->family == CHIP_RS100) ||
2704             (rdev->family == CHIP_RS200)) {
2705                 tmp = RREG32(RADEON_MEM_CNTL);
2706                 if (tmp & RV100_HALF_MODE) {
2707                         rdev->mc.vram_width = 32;
2708                 } else {
2709                         rdev->mc.vram_width = 64;
2710                 }
2711                 if (rdev->flags & RADEON_SINGLE_CRTC) {
2712                         rdev->mc.vram_width /= 4;
2713                         rdev->mc.vram_is_ddr = true;
2714                 }
2715         } else if (rdev->family <= CHIP_RV280) {
2716                 tmp = RREG32(RADEON_MEM_CNTL);
2717                 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2718                         rdev->mc.vram_width = 128;
2719                 } else {
2720                         rdev->mc.vram_width = 64;
2721                 }
2722         } else {
2723                 /* newer IGPs */
2724                 rdev->mc.vram_width = 128;
2725         }
2726 }
2727
2728 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2729 {
2730         u32 aper_size;
2731         u8 byte;
2732
2733         aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2734
2735         /* Set HDP_APER_CNTL only on cards that are known not to be broken,
2736          * that is has the 2nd generation multifunction PCI interface
2737          */
2738         if (rdev->family == CHIP_RV280 ||
2739             rdev->family >= CHIP_RV350) {
2740                 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2741                        ~RADEON_HDP_APER_CNTL);
2742                 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2743                 return aper_size * 2;
2744         }
2745
2746         /* Older cards have all sorts of funny issues to deal with. First
2747          * check if it's a multifunction card by reading the PCI config
2748          * header type... Limit those to one aperture size
2749          */
2750         pci_read_config_byte(rdev->pdev, 0xe, &byte);
2751         if (byte & 0x80) {
2752                 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2753                 DRM_INFO("Limiting VRAM to one aperture\n");
2754                 return aper_size;
2755         }
2756
2757         /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2758          * have set it up. We don't write this as it's broken on some ASICs but
2759          * we expect the BIOS to have done the right thing (might be too optimistic...)
2760          */
2761         if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2762                 return aper_size * 2;
2763         return aper_size;
2764 }
2765
2766 void r100_vram_init_sizes(struct radeon_device *rdev)
2767 {
2768         u64 config_aper_size;
2769
2770         /* work out accessible VRAM */
2771         rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2772         rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2773         rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2774         /* FIXME we don't use the second aperture yet when we could use it */
2775         if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2776                 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2777         config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2778         if (rdev->flags & RADEON_IS_IGP) {
2779                 uint32_t tom;
2780                 /* read NB_TOM to get the amount of ram stolen for the GPU */
2781                 tom = RREG32(RADEON_NB_TOM);
2782                 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2783                 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2784                 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2785         } else {
2786                 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2787                 /* Some production boards of m6 will report 0
2788                  * if it's 8 MB
2789                  */
2790                 if (rdev->mc.real_vram_size == 0) {
2791                         rdev->mc.real_vram_size = 8192 * 1024;
2792                         WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2793                 }
2794                 /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM - 
2795                  * Novell bug 204882 + along with lots of ubuntu ones
2796                  */
2797                 if (rdev->mc.aper_size > config_aper_size)
2798                         config_aper_size = rdev->mc.aper_size;
2799
2800                 if (config_aper_size > rdev->mc.real_vram_size)
2801                         rdev->mc.mc_vram_size = config_aper_size;
2802                 else
2803                         rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2804         }
2805 }
2806
2807 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2808 {
2809         uint32_t temp;
2810
2811         temp = RREG32(RADEON_CONFIG_CNTL);
2812         if (state == false) {
2813                 temp &= ~RADEON_CFG_VGA_RAM_EN;
2814                 temp |= RADEON_CFG_VGA_IO_DIS;
2815         } else {
2816                 temp &= ~RADEON_CFG_VGA_IO_DIS;
2817         }
2818         WREG32(RADEON_CONFIG_CNTL, temp);
2819 }
2820
2821 static void r100_mc_init(struct radeon_device *rdev)
2822 {
2823         u64 base;
2824
2825         r100_vram_get_type(rdev);
2826         r100_vram_init_sizes(rdev);
2827         base = rdev->mc.aper_base;
2828         if (rdev->flags & RADEON_IS_IGP)
2829                 base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2830         radeon_vram_location(rdev, &rdev->mc, base);
2831         rdev->mc.gtt_base_align = 0;
2832         if (!(rdev->flags & RADEON_IS_AGP))
2833                 radeon_gtt_location(rdev, &rdev->mc);
2834         radeon_update_bandwidth_info(rdev);
2835 }
2836
2837
2838 /*
2839  * Indirect registers accessor
2840  */
2841 void r100_pll_errata_after_index(struct radeon_device *rdev)
2842 {
2843         if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2844                 (void)RREG32(RADEON_CLOCK_CNTL_DATA);
2845                 (void)RREG32(RADEON_CRTC_GEN_CNTL);
2846         }
2847 }
2848
2849 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2850 {
2851         /* This workarounds is necessary on RV100, RS100 and RS200 chips
2852          * or the chip could hang on a subsequent access
2853          */
2854         if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2855                 mdelay(5);
2856         }
2857
2858         /* This function is required to workaround a hardware bug in some (all?)
2859          * revisions of the R300.  This workaround should be called after every
2860          * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2861          * may not be correct.
2862          */
2863         if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2864                 uint32_t save, tmp;
2865
2866                 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2867                 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2868                 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2869                 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2870                 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2871         }
2872 }
2873
2874 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2875 {
2876         unsigned long flags;
2877         uint32_t data;
2878
2879         spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2880         WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2881         r100_pll_errata_after_index(rdev);
2882         data = RREG32(RADEON_CLOCK_CNTL_DATA);
2883         r100_pll_errata_after_data(rdev);
2884         spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2885         return data;
2886 }
2887
2888 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2889 {
2890         unsigned long flags;
2891
2892         spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2893         WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2894         r100_pll_errata_after_index(rdev);
2895         WREG32(RADEON_CLOCK_CNTL_DATA, v);
2896         r100_pll_errata_after_data(rdev);
2897         spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2898 }
2899
2900 static void r100_set_safe_registers(struct radeon_device *rdev)
2901 {
2902         if (ASIC_IS_RN50(rdev)) {
2903                 rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2904                 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2905         } else if (rdev->family < CHIP_R200) {
2906                 rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2907                 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2908         } else {
2909                 r200_set_safe_registers(rdev);
2910         }
2911 }
2912
2913 /*
2914  * Debugfs info
2915  */
2916 #if defined(CONFIG_DEBUG_FS)
2917 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2918 {
2919         struct drm_info_node *node = (struct drm_info_node *) m->private;
2920         struct drm_device *dev = node->minor->dev;
2921         struct radeon_device *rdev = dev->dev_private;
2922         uint32_t reg, value;
2923         unsigned i;
2924
2925         seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2926         seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2927         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2928         for (i = 0; i < 64; i++) {
2929                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2930                 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2931                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2932                 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2933                 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2934         }
2935         return 0;
2936 }
2937
2938 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2939 {
2940         struct drm_info_node *node = (struct drm_info_node *) m->private;
2941         struct drm_device *dev = node->minor->dev;
2942         struct radeon_device *rdev = dev->dev_private;
2943         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2944         uint32_t rdp, wdp;
2945         unsigned count, i, j;
2946
2947         radeon_ring_free_size(rdev, ring);
2948         rdp = RREG32(RADEON_CP_RB_RPTR);
2949         wdp = RREG32(RADEON_CP_RB_WPTR);
2950         count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2951         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2952         seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2953         seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2954         seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2955         seq_printf(m, "%u dwords in ring\n", count);
2956         if (ring->ready) {
2957                 for (j = 0; j <= count; j++) {
2958                         i = (rdp + j) & ring->ptr_mask;
2959                         seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2960                 }
2961         }
2962         return 0;
2963 }
2964
2965
2966 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2967 {
2968         struct drm_info_node *node = (struct drm_info_node *) m->private;
2969         struct drm_device *dev = node->minor->dev;
2970         struct radeon_device *rdev = dev->dev_private;
2971         uint32_t csq_stat, csq2_stat, tmp;
2972         unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2973         unsigned i;
2974
2975         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2976         seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2977         csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2978         csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2979         r_rptr = (csq_stat >> 0) & 0x3ff;
2980         r_wptr = (csq_stat >> 10) & 0x3ff;
2981         ib1_rptr = (csq_stat >> 20) & 0x3ff;
2982         ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2983         ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2984         ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2985         seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2986         seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2987         seq_printf(m, "Ring rptr %u\n", r_rptr);
2988         seq_printf(m, "Ring wptr %u\n", r_wptr);
2989         seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2990         seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2991         seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2992         seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2993         /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
2994          * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
2995         seq_printf(m, "Ring fifo:\n");
2996         for (i = 0; i < 256; i++) {
2997                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2998                 tmp = RREG32(RADEON_CP_CSQ_DATA);
2999                 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3000         }
3001         seq_printf(m, "Indirect1 fifo:\n");
3002         for (i = 256; i <= 512; i++) {
3003                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3004                 tmp = RREG32(RADEON_CP_CSQ_DATA);
3005                 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3006         }
3007         seq_printf(m, "Indirect2 fifo:\n");
3008         for (i = 640; i < ib1_wptr; i++) {
3009                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3010                 tmp = RREG32(RADEON_CP_CSQ_DATA);
3011                 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3012         }
3013         return 0;
3014 }
3015
3016 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
3017 {
3018         struct drm_info_node *node = (struct drm_info_node *) m->private;
3019         struct drm_device *dev = node->minor->dev;
3020         struct radeon_device *rdev = dev->dev_private;
3021         uint32_t tmp;
3022
3023         tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3024         seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3025         tmp = RREG32(RADEON_MC_FB_LOCATION);
3026         seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3027         tmp = RREG32(RADEON_BUS_CNTL);
3028         seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3029         tmp = RREG32(RADEON_MC_AGP_LOCATION);
3030         seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3031         tmp = RREG32(RADEON_AGP_BASE);
3032         seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3033         tmp = RREG32(RADEON_HOST_PATH_CNTL);
3034         seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3035         tmp = RREG32(0x01D0);
3036         seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3037         tmp = RREG32(RADEON_AIC_LO_ADDR);
3038         seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3039         tmp = RREG32(RADEON_AIC_HI_ADDR);
3040         seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3041         tmp = RREG32(0x01E4);
3042         seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3043         return 0;
3044 }
3045
3046 static struct drm_info_list r100_debugfs_rbbm_list[] = {
3047         {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3048 };
3049
3050 static struct drm_info_list r100_debugfs_cp_list[] = {
3051         {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3052         {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3053 };
3054
3055 static struct drm_info_list r100_debugfs_mc_info_list[] = {
3056         {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3057 };
3058 #endif
3059
3060 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3061 {
3062 #if defined(CONFIG_DEBUG_FS)
3063         return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3064 #else
3065         return 0;
3066 #endif
3067 }
3068
3069 int r100_debugfs_cp_init(struct radeon_device *rdev)
3070 {
3071 #if defined(CONFIG_DEBUG_FS)
3072         return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3073 #else
3074         return 0;
3075 #endif
3076 }
3077
3078 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3079 {
3080 #if defined(CONFIG_DEBUG_FS)
3081         return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3082 #else
3083         return 0;
3084 #endif
3085 }
3086
3087 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3088                          uint32_t tiling_flags, uint32_t pitch,
3089                          uint32_t offset, uint32_t obj_size)
3090 {
3091         int surf_index = reg * 16;
3092         int flags = 0;
3093
3094         if (rdev->family <= CHIP_RS200) {
3095                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3096                                  == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3097                         flags |= RADEON_SURF_TILE_COLOR_BOTH;
3098                 if (tiling_flags & RADEON_TILING_MACRO)
3099                         flags |= RADEON_SURF_TILE_COLOR_MACRO;
3100                 /* setting pitch to 0 disables tiling */
3101                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3102                                 == 0)
3103                         pitch = 0;
3104         } else if (rdev->family <= CHIP_RV280) {
3105                 if (tiling_flags & (RADEON_TILING_MACRO))
3106                         flags |= R200_SURF_TILE_COLOR_MACRO;
3107                 if (tiling_flags & RADEON_TILING_MICRO)
3108                         flags |= R200_SURF_TILE_COLOR_MICRO;
3109         } else {
3110                 if (tiling_flags & RADEON_TILING_MACRO)
3111                         flags |= R300_SURF_TILE_MACRO;
3112                 if (tiling_flags & RADEON_TILING_MICRO)
3113                         flags |= R300_SURF_TILE_MICRO;
3114         }
3115
3116         if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3117                 flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3118         if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3119                 flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3120
3121         /* r100/r200 divide by 16 */
3122         if (rdev->family < CHIP_R300)
3123                 flags |= pitch / 16;
3124         else
3125                 flags |= pitch / 8;
3126
3127
3128         DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3129         WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3130         WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3131         WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3132         return 0;
3133 }
3134
3135 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3136 {
3137         int surf_index = reg * 16;
3138         WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3139 }
3140
3141 void r100_bandwidth_update(struct radeon_device *rdev)
3142 {
3143         fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3144         fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3145         fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
3146         fixed20_12 crit_point_ff = {0};
3147         uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3148         fixed20_12 memtcas_ff[8] = {
3149                 dfixed_init(1),
3150                 dfixed_init(2),
3151                 dfixed_init(3),
3152                 dfixed_init(0),
3153                 dfixed_init_half(1),
3154                 dfixed_init_half(2),
3155                 dfixed_init(0),
3156         };
3157         fixed20_12 memtcas_rs480_ff[8] = {
3158                 dfixed_init(0),
3159                 dfixed_init(1),
3160                 dfixed_init(2),
3161                 dfixed_init(3),
3162                 dfixed_init(0),
3163                 dfixed_init_half(1),
3164                 dfixed_init_half(2),
3165                 dfixed_init_half(3),
3166         };
3167         fixed20_12 memtcas2_ff[8] = {
3168                 dfixed_init(0),
3169                 dfixed_init(1),
3170                 dfixed_init(2),
3171                 dfixed_init(3),
3172                 dfixed_init(4),
3173                 dfixed_init(5),
3174                 dfixed_init(6),
3175                 dfixed_init(7),
3176         };
3177         fixed20_12 memtrbs[8] = {
3178                 dfixed_init(1),
3179                 dfixed_init_half(1),
3180                 dfixed_init(2),
3181                 dfixed_init_half(2),
3182                 dfixed_init(3),
3183                 dfixed_init_half(3),
3184                 dfixed_init(4),
3185                 dfixed_init_half(4)
3186         };
3187         fixed20_12 memtrbs_r4xx[8] = {
3188                 dfixed_init(4),
3189                 dfixed_init(5),
3190                 dfixed_init(6),
3191                 dfixed_init(7),
3192                 dfixed_init(8),
3193                 dfixed_init(9),
3194                 dfixed_init(10),
3195                 dfixed_init(11)
3196         };
3197         fixed20_12 min_mem_eff;
3198         fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3199         fixed20_12 cur_latency_mclk, cur_latency_sclk;
3200         fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate = {0},
3201                 disp_drain_rate2, read_return_rate;
3202         fixed20_12 time_disp1_drop_priority;
3203         int c;
3204         int cur_size = 16;       /* in octawords */
3205         int critical_point = 0, critical_point2;
3206 /*      uint32_t read_return_rate, time_disp1_drop_priority; */
3207         int stop_req, max_stop_req;
3208         struct drm_display_mode *mode1 = NULL;
3209         struct drm_display_mode *mode2 = NULL;
3210         uint32_t pixel_bytes1 = 0;
3211         uint32_t pixel_bytes2 = 0;
3212
3213         /* Guess line buffer size to be 8192 pixels */
3214         u32 lb_size = 8192;
3215
3216         if (!rdev->mode_info.mode_config_initialized)
3217                 return;
3218
3219         radeon_update_display_priority(rdev);
3220
3221         if (rdev->mode_info.crtcs[0]->base.enabled) {
3222                 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3223                 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.primary->fb->bits_per_pixel / 8;
3224         }
3225         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3226                 if (rdev->mode_info.crtcs[1]->base.enabled) {
3227                         mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3228                         pixel_bytes2 = rdev->mode_info.crtcs[1]->base.primary->fb->bits_per_pixel / 8;
3229                 }
3230         }
3231
3232         min_mem_eff.full = dfixed_const_8(0);
3233         /* get modes */
3234         if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3235                 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3236                 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3237                 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3238                 /* check crtc enables */
3239                 if (mode2)
3240                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3241                 if (mode1)
3242                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3243                 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3244         }
3245
3246         /*
3247          * determine is there is enough bw for current mode
3248          */
3249         sclk_ff = rdev->pm.sclk;
3250         mclk_ff = rdev->pm.mclk;
3251
3252         temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3253         temp_ff.full = dfixed_const(temp);
3254         mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3255
3256         pix_clk.full = 0;
3257         pix_clk2.full = 0;
3258         peak_disp_bw.full = 0;
3259         if (mode1) {
3260                 temp_ff.full = dfixed_const(1000);
3261                 pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3262                 pix_clk.full = dfixed_div(pix_clk, temp_ff);
3263                 temp_ff.full = dfixed_const(pixel_bytes1);
3264                 peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3265         }
3266         if (mode2) {
3267                 temp_ff.full = dfixed_const(1000);
3268                 pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3269                 pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3270                 temp_ff.full = dfixed_const(pixel_bytes2);
3271                 peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3272         }
3273
3274         mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3275         if (peak_disp_bw.full >= mem_bw.full) {
3276                 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3277                           "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3278         }
3279
3280         /*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3281         temp = RREG32(RADEON_MEM_TIMING_CNTL);
3282         if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3283                 mem_trcd = ((temp >> 2) & 0x3) + 1;
3284                 mem_trp  = ((temp & 0x3)) + 1;
3285                 mem_tras = ((temp & 0x70) >> 4) + 1;
3286         } else if (rdev->family == CHIP_R300 ||
3287                    rdev->family == CHIP_R350) { /* r300, r350 */
3288                 mem_trcd = (temp & 0x7) + 1;
3289                 mem_trp = ((temp >> 8) & 0x7) + 1;
3290                 mem_tras = ((temp >> 11) & 0xf) + 4;
3291         } else if (rdev->family == CHIP_RV350 ||
3292                    rdev->family <= CHIP_RV380) {
3293                 /* rv3x0 */
3294                 mem_trcd = (temp & 0x7) + 3;
3295                 mem_trp = ((temp >> 8) & 0x7) + 3;
3296                 mem_tras = ((temp >> 11) & 0xf) + 6;
3297         } else if (rdev->family == CHIP_R420 ||
3298                    rdev->family == CHIP_R423 ||
3299                    rdev->family == CHIP_RV410) {
3300                 /* r4xx */
3301                 mem_trcd = (temp & 0xf) + 3;
3302                 if (mem_trcd > 15)
3303                         mem_trcd = 15;
3304                 mem_trp = ((temp >> 8) & 0xf) + 3;
3305                 if (mem_trp > 15)
3306                         mem_trp = 15;
3307                 mem_tras = ((temp >> 12) & 0x1f) + 6;
3308                 if (mem_tras > 31)
3309                         mem_tras = 31;
3310         } else { /* RV200, R200 */
3311                 mem_trcd = (temp & 0x7) + 1;
3312                 mem_trp = ((temp >> 8) & 0x7) + 1;
3313                 mem_tras = ((temp >> 12) & 0xf) + 4;
3314         }
3315         /* convert to FF */
3316         trcd_ff.full = dfixed_const(mem_trcd);
3317         trp_ff.full = dfixed_const(mem_trp);
3318         tras_ff.full = dfixed_const(mem_tras);
3319
3320         /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3321         temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3322         data = (temp & (7 << 20)) >> 20;
3323         if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3324                 if (rdev->family == CHIP_RS480) /* don't think rs400 */
3325                         tcas_ff = memtcas_rs480_ff[data];
3326                 else
3327                         tcas_ff = memtcas_ff[data];
3328         } else
3329                 tcas_ff = memtcas2_ff[data];
3330
3331         if (rdev->family == CHIP_RS400 ||
3332             rdev->family == CHIP_RS480) {
3333                 /* extra cas latency stored in bits 23-25 0-4 clocks */
3334                 data = (temp >> 23) & 0x7;
3335                 if (data < 5)
3336                         tcas_ff.full += dfixed_const(data);
3337         }
3338
3339         if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3340                 /* on the R300, Tcas is included in Trbs.
3341                  */
3342                 temp = RREG32(RADEON_MEM_CNTL);
3343                 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3344                 if (data == 1) {
3345                         if (R300_MEM_USE_CD_CH_ONLY & temp) {
3346                                 temp = RREG32(R300_MC_IND_INDEX);
3347                                 temp &= ~R300_MC_IND_ADDR_MASK;
3348                                 temp |= R300_MC_READ_CNTL_CD_mcind;
3349                                 WREG32(R300_MC_IND_INDEX, temp);
3350                                 temp = RREG32(R300_MC_IND_DATA);
3351                                 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3352                         } else {
3353                                 temp = RREG32(R300_MC_READ_CNTL_AB);
3354                                 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3355                         }
3356                 } else {
3357                         temp = RREG32(R300_MC_READ_CNTL_AB);
3358                         data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3359                 }
3360                 if (rdev->family == CHIP_RV410 ||
3361                     rdev->family == CHIP_R420 ||
3362                     rdev->family == CHIP_R423)
3363                         trbs_ff = memtrbs_r4xx[data];
3364                 else
3365                         trbs_ff = memtrbs[data];
3366                 tcas_ff.full += trbs_ff.full;
3367         }
3368
3369         sclk_eff_ff.full = sclk_ff.full;
3370
3371         if (rdev->flags & RADEON_IS_AGP) {
3372                 fixed20_12 agpmode_ff;
3373                 agpmode_ff.full = dfixed_const(radeon_agpmode);
3374                 temp_ff.full = dfixed_const_666(16);
3375                 sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3376         }
3377         /* TODO PCIE lanes may affect this - agpmode == 16?? */
3378
3379         if (ASIC_IS_R300(rdev)) {
3380                 sclk_delay_ff.full = dfixed_const(250);
3381         } else {
3382                 if ((rdev->family == CHIP_RV100) ||
3383                     rdev->flags & RADEON_IS_IGP) {
3384                         if (rdev->mc.vram_is_ddr)
3385                                 sclk_delay_ff.full = dfixed_const(41);
3386                         else
3387                                 sclk_delay_ff.full = dfixed_const(33);
3388                 } else {
3389                         if (rdev->mc.vram_width == 128)
3390                                 sclk_delay_ff.full = dfixed_const(57);
3391                         else
3392                                 sclk_delay_ff.full = dfixed_const(41);
3393                 }
3394         }
3395
3396         mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3397
3398         if (rdev->mc.vram_is_ddr) {
3399                 if (rdev->mc.vram_width == 32) {
3400                         k1.full = dfixed_const(40);
3401                         c  = 3;
3402                 } else {
3403                         k1.full = dfixed_const(20);
3404                         c  = 1;
3405                 }
3406         } else {
3407                 k1.full = dfixed_const(40);
3408                 c  = 3;
3409         }
3410
3411         temp_ff.full = dfixed_const(2);
3412         mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3413         temp_ff.full = dfixed_const(c);
3414         mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3415         temp_ff.full = dfixed_const(4);
3416         mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3417         mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3418         mc_latency_mclk.full += k1.full;
3419
3420         mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3421         mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3422
3423         /*
3424           HW cursor time assuming worst case of full size colour cursor.
3425         */
3426         temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3427         temp_ff.full += trcd_ff.full;
3428         if (temp_ff.full < tras_ff.full)
3429                 temp_ff.full = tras_ff.full;
3430         cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3431
3432         temp_ff.full = dfixed_const(cur_size);
3433         cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3434         /*
3435           Find the total latency for the display data.
3436         */
3437         disp_latency_overhead.full = dfixed_const(8);
3438         disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3439         mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3440         mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3441
3442         if (mc_latency_mclk.full > mc_latency_sclk.full)
3443                 disp_latency.full = mc_latency_mclk.full;
3444         else
3445                 disp_latency.full = mc_latency_sclk.full;
3446
3447         /* setup Max GRPH_STOP_REQ default value */
3448         if (ASIC_IS_RV100(rdev))
3449                 max_stop_req = 0x5c;
3450         else
3451                 max_stop_req = 0x7c;
3452
3453         if (mode1) {
3454                 /*  CRTC1
3455                     Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3456                     GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3457                 */
3458                 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3459
3460                 if (stop_req > max_stop_req)
3461                         stop_req = max_stop_req;
3462
3463                 /*
3464                   Find the drain rate of the display buffer.
3465                 */
3466                 temp_ff.full = dfixed_const((16/pixel_bytes1));
3467                 disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3468
3469                 /*
3470                   Find the critical point of the display buffer.
3471                 */
3472                 crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3473                 crit_point_ff.full += dfixed_const_half(0);
3474
3475                 critical_point = dfixed_trunc(crit_point_ff);
3476
3477                 if (rdev->disp_priority == 2) {
3478                         critical_point = 0;
3479                 }
3480
3481                 /*
3482                   The critical point should never be above max_stop_req-4.  Setting
3483                   GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3484                 */
3485                 if (max_stop_req - critical_point < 4)
3486                         critical_point = 0;
3487
3488                 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3489                         /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3490                         critical_point = 0x10;
3491                 }
3492
3493                 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3494                 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3495                 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3496                 temp &= ~(RADEON_GRPH_START_REQ_MASK);
3497                 if ((rdev->family == CHIP_R350) &&
3498                     (stop_req > 0x15)) {
3499                         stop_req -= 0x10;
3500                 }
3501                 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3502                 temp |= RADEON_GRPH_BUFFER_SIZE;
3503                 temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3504                           RADEON_GRPH_CRITICAL_AT_SOF |
3505                           RADEON_GRPH_STOP_CNTL);
3506                 /*
3507                   Write the result into the register.
3508                 */
3509                 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3510                                                        (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3511
3512 #if 0
3513                 if ((rdev->family == CHIP_RS400) ||
3514                     (rdev->family == CHIP_RS480)) {
3515                         /* attempt to program RS400 disp regs correctly ??? */
3516                         temp = RREG32(RS400_DISP1_REG_CNTL);
3517                         temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3518                                   RS400_DISP1_STOP_REQ_LEVEL_MASK);
3519                         WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3520                                                        (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3521                                                        (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3522                         temp = RREG32(RS400_DMIF_MEM_CNTL1);
3523                         temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3524                                   RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3525                         WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3526                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3527                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3528                 }
3529 #endif
3530
3531                 DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3532                           /*      (unsigned int)info->SavedReg->grph_buffer_cntl, */
3533                           (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3534         }
3535
3536         if (mode2) {
3537                 u32 grph2_cntl;
3538                 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3539
3540                 if (stop_req > max_stop_req)
3541                         stop_req = max_stop_req;
3542
3543                 /*
3544                   Find the drain rate of the display buffer.
3545                 */
3546                 temp_ff.full = dfixed_const((16/pixel_bytes2));
3547                 disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3548
3549                 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3550                 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3551                 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3552                 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3553                 if ((rdev->family == CHIP_R350) &&
3554                     (stop_req > 0x15)) {
3555                         stop_req -= 0x10;
3556                 }
3557                 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3558                 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3559                 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3560                           RADEON_GRPH_CRITICAL_AT_SOF |
3561                           RADEON_GRPH_STOP_CNTL);
3562
3563                 if ((rdev->family == CHIP_RS100) ||
3564                     (rdev->family == CHIP_RS200))
3565                         critical_point2 = 0;
3566                 else {
3567                         temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3568                         temp_ff.full = dfixed_const(temp);
3569                         temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3570                         if (sclk_ff.full < temp_ff.full)
3571                                 temp_ff.full = sclk_ff.full;
3572
3573                         read_return_rate.full = temp_ff.full;
3574
3575                         if (mode1) {
3576                                 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3577                                 time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3578                         } else {
3579                                 time_disp1_drop_priority.full = 0;
3580                         }
3581                         crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3582                         crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3583                         crit_point_ff.full += dfixed_const_half(0);
3584
3585                         critical_point2 = dfixed_trunc(crit_point_ff);
3586
3587                         if (rdev->disp_priority == 2) {
3588                                 critical_point2 = 0;
3589                         }
3590
3591                         if (max_stop_req - critical_point2 < 4)
3592                                 critical_point2 = 0;
3593
3594                 }
3595
3596                 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3597                         /* some R300 cards have problem with this set to 0 */
3598                         critical_point2 = 0x10;
3599                 }
3600
3601                 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3602                                                   (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3603
3604                 if ((rdev->family == CHIP_RS400) ||
3605                     (rdev->family == CHIP_RS480)) {
3606 #if 0
3607                         /* attempt to program RS400 disp2 regs correctly ??? */
3608                         temp = RREG32(RS400_DISP2_REQ_CNTL1);
3609                         temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3610                                   RS400_DISP2_STOP_REQ_LEVEL_MASK);
3611                         WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3612                                                        (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3613                                                        (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3614                         temp = RREG32(RS400_DISP2_REQ_CNTL2);
3615                         temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3616                                   RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3617                         WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3618                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3619                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3620 #endif
3621                         WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3622                         WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3623                         WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3624                         WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3625                 }
3626
3627                 DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3628                           (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3629         }
3630
3631         /* Save number of lines the linebuffer leads before the scanout */
3632         if (mode1)
3633             rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
3634
3635         if (mode2)
3636             rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
3637 }
3638
3639 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3640 {
3641         uint32_t scratch;
3642         uint32_t tmp = 0;
3643         unsigned i;
3644         int r;
3645
3646         r = radeon_scratch_get(rdev, &scratch);
3647         if (r) {
3648                 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3649                 return r;
3650         }
3651         WREG32(scratch, 0xCAFEDEAD);
3652         r = radeon_ring_lock(rdev, ring, 2);
3653         if (r) {
3654                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3655                 radeon_scratch_free(rdev, scratch);
3656                 return r;
3657         }
3658         radeon_ring_write(ring, PACKET0(scratch, 0));
3659         radeon_ring_write(ring, 0xDEADBEEF);
3660         radeon_ring_unlock_commit(rdev, ring, false);
3661         for (i = 0; i < rdev->usec_timeout; i++) {
3662                 tmp = RREG32(scratch);
3663                 if (tmp == 0xDEADBEEF) {
3664                         break;
3665                 }
3666                 DRM_UDELAY(1);
3667         }
3668         if (i < rdev->usec_timeout) {
3669                 DRM_INFO("ring test succeeded in %d usecs\n", i);
3670         } else {
3671                 DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3672                           scratch, tmp);
3673                 r = -EINVAL;
3674         }
3675         radeon_scratch_free(rdev, scratch);
3676         return r;
3677 }
3678
3679 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3680 {
3681         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3682
3683         if (ring->rptr_save_reg) {
3684                 u32 next_rptr = ring->wptr + 2 + 3;
3685                 radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3686                 radeon_ring_write(ring, next_rptr);
3687         }
3688
3689         radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3690         radeon_ring_write(ring, ib->gpu_addr);
3691         radeon_ring_write(ring, ib->length_dw);
3692 }
3693
3694 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3695 {
3696         struct radeon_ib ib;
3697         uint32_t scratch;
3698         uint32_t tmp = 0;
3699         unsigned i;
3700         int r;
3701
3702         r = radeon_scratch_get(rdev, &scratch);
3703         if (r) {
3704                 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3705                 return r;
3706         }
3707         WREG32(scratch, 0xCAFEDEAD);
3708         r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3709         if (r) {
3710                 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3711                 goto free_scratch;
3712         }
3713         ib.ptr[0] = PACKET0(scratch, 0);
3714         ib.ptr[1] = 0xDEADBEEF;
3715         ib.ptr[2] = PACKET2(0);
3716         ib.ptr[3] = PACKET2(0);
3717         ib.ptr[4] = PACKET2(0);
3718         ib.ptr[5] = PACKET2(0);
3719         ib.ptr[6] = PACKET2(0);
3720         ib.ptr[7] = PACKET2(0);
3721         ib.length_dw = 8;
3722         r = radeon_ib_schedule(rdev, &ib, NULL, false);
3723         if (r) {
3724                 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3725                 goto free_ib;
3726         }
3727         r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
3728                 RADEON_USEC_IB_TEST_TIMEOUT));
3729         if (r < 0) {
3730                 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3731                 goto free_ib;
3732         } else if (r == 0) {
3733                 DRM_ERROR("radeon: fence wait timed out.\n");
3734                 r = -ETIMEDOUT;
3735                 goto free_ib;
3736         }
3737         r = 0;
3738         for (i = 0; i < rdev->usec_timeout; i++) {
3739                 tmp = RREG32(scratch);
3740                 if (tmp == 0xDEADBEEF) {
3741                         break;
3742                 }
3743                 DRM_UDELAY(1);
3744         }
3745         if (i < rdev->usec_timeout) {
3746                 DRM_INFO("ib test succeeded in %u usecs\n", i);
3747         } else {
3748                 DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3749                           scratch, tmp);
3750                 r = -EINVAL;
3751         }
3752 free_ib:
3753         radeon_ib_free(rdev, &ib);
3754 free_scratch:
3755         radeon_scratch_free(rdev, scratch);
3756         return r;
3757 }
3758
3759 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3760 {
3761         /* Shutdown CP we shouldn't need to do that but better be safe than
3762          * sorry
3763          */
3764         rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3765         WREG32(R_000740_CP_CSQ_CNTL, 0);
3766
3767         /* Save few CRTC registers */
3768         save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3769         save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3770         save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3771         save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3772         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3773                 save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3774                 save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3775         }
3776
3777         /* Disable VGA aperture access */
3778         WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3779         /* Disable cursor, overlay, crtc */
3780         WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3781         WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3782                                         S_000054_CRTC_DISPLAY_DIS(1));
3783         WREG32(R_000050_CRTC_GEN_CNTL,
3784                         (C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3785                         S_000050_CRTC_DISP_REQ_EN_B(1));
3786         WREG32(R_000420_OV0_SCALE_CNTL,
3787                 C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3788         WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3789         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3790                 WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3791                                                 S_000360_CUR2_LOCK(1));
3792                 WREG32(R_0003F8_CRTC2_GEN_CNTL,
3793                         (C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3794                         S_0003F8_CRTC2_DISPLAY_DIS(1) |
3795                         S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3796                 WREG32(R_000360_CUR2_OFFSET,
3797                         C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3798         }
3799 }
3800
3801 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3802 {
3803         /* Update base address for crtc */
3804         WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3805         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3806                 WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3807         }
3808         /* Restore CRTC registers */
3809         WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3810         WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3811         WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3812         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3813                 WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3814         }
3815 }
3816
3817 void r100_vga_render_disable(struct radeon_device *rdev)
3818 {
3819         u32 tmp;
3820
3821         tmp = RREG8(R_0003C2_GENMO_WT);
3822         WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3823 }
3824
3825 static void r100_debugfs(struct radeon_device *rdev)
3826 {
3827         int r;
3828
3829         r = r100_debugfs_mc_info_init(rdev);
3830         if (r)
3831                 dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3832 }
3833
3834 static void r100_mc_program(struct radeon_device *rdev)
3835 {
3836         struct r100_mc_save save;
3837
3838         /* Stops all mc clients */
3839         r100_mc_stop(rdev, &save);
3840         if (rdev->flags & RADEON_IS_AGP) {
3841                 WREG32(R_00014C_MC_AGP_LOCATION,
3842                         S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3843                         S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3844                 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3845                 if (rdev->family > CHIP_RV200)
3846                         WREG32(R_00015C_AGP_BASE_2,
3847                                 upper_32_bits(rdev->mc.agp_base) & 0xff);
3848         } else {
3849                 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3850                 WREG32(R_000170_AGP_BASE, 0);
3851                 if (rdev->family > CHIP_RV200)
3852                         WREG32(R_00015C_AGP_BASE_2, 0);
3853         }
3854         /* Wait for mc idle */
3855         if (r100_mc_wait_for_idle(rdev))
3856                 dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3857         /* Program MC, should be a 32bits limited address space */
3858         WREG32(R_000148_MC_FB_LOCATION,
3859                 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3860                 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3861         r100_mc_resume(rdev, &save);
3862 }
3863
3864 static void r100_clock_startup(struct radeon_device *rdev)
3865 {
3866         u32 tmp;
3867
3868         if (radeon_dynclks != -1 && radeon_dynclks)
3869                 radeon_legacy_set_clock_gating(rdev, 1);
3870         /* We need to force on some of the block */
3871         tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3872         tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3873         if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3874                 tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3875         WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3876 }
3877
3878 static int r100_startup(struct radeon_device *rdev)
3879 {
3880         int r;
3881
3882         /* set common regs */
3883         r100_set_common_regs(rdev);
3884         /* program mc */
3885         r100_mc_program(rdev);
3886         /* Resume clock */
3887         r100_clock_startup(rdev);
3888         /* Initialize GART (initialize after TTM so we can allocate
3889          * memory through TTM but finalize after TTM) */
3890         r100_enable_bm(rdev);
3891         if (rdev->flags & RADEON_IS_PCI) {
3892                 r = r100_pci_gart_enable(rdev);
3893                 if (r)
3894                         return r;
3895         }
3896
3897         /* allocate wb buffer */
3898         r = radeon_wb_init(rdev);
3899         if (r)
3900                 return r;
3901
3902         r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3903         if (r) {
3904                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3905                 return r;
3906         }
3907
3908         /* Enable IRQ */
3909         if (!rdev->irq.installed) {
3910                 r = radeon_irq_kms_init(rdev);
3911                 if (r)
3912                         return r;
3913         }
3914
3915         r100_irq_set(rdev);
3916         rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3917         /* 1M ring buffer */
3918         r = r100_cp_init(rdev, 1024 * 1024);
3919         if (r) {
3920                 dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3921                 return r;
3922         }
3923
3924         r = radeon_ib_pool_init(rdev);
3925         if (r) {
3926                 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3927                 return r;
3928         }
3929
3930         return 0;
3931 }
3932
3933 int r100_resume(struct radeon_device *rdev)
3934 {
3935         int r;
3936
3937         /* Make sur GART are not working */
3938         if (rdev->flags & RADEON_IS_PCI)
3939                 r100_pci_gart_disable(rdev);
3940         /* Resume clock before doing reset */
3941         r100_clock_startup(rdev);
3942         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
3943         if (radeon_asic_reset(rdev)) {
3944                 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3945                         RREG32(R_000E40_RBBM_STATUS),
3946                         RREG32(R_0007C0_CP_STAT));
3947         }
3948         /* post */
3949         radeon_combios_asic_init(rdev->ddev);
3950         /* Resume clock after posting */
3951         r100_clock_startup(rdev);
3952         /* Initialize surface registers */
3953         radeon_surface_init(rdev);
3954
3955         rdev->accel_working = true;
3956         r = r100_startup(rdev);
3957         if (r) {
3958                 rdev->accel_working = false;
3959         }
3960         return r;
3961 }
3962
3963 int r100_suspend(struct radeon_device *rdev)
3964 {
3965         radeon_pm_suspend(rdev);
3966         r100_cp_disable(rdev);
3967         radeon_wb_disable(rdev);
3968         r100_irq_disable(rdev);
3969         if (rdev->flags & RADEON_IS_PCI)
3970                 r100_pci_gart_disable(rdev);
3971         return 0;
3972 }
3973
3974 void r100_fini(struct radeon_device *rdev)
3975 {
3976         radeon_pm_fini(rdev);
3977         r100_cp_fini(rdev);
3978         radeon_wb_fini(rdev);
3979         radeon_ib_pool_fini(rdev);
3980         radeon_gem_fini(rdev);
3981         if (rdev->flags & RADEON_IS_PCI)
3982                 r100_pci_gart_fini(rdev);
3983         radeon_agp_fini(rdev);
3984         radeon_irq_kms_fini(rdev);
3985         radeon_fence_driver_fini(rdev);
3986         radeon_bo_fini(rdev);
3987         radeon_atombios_fini(rdev);
3988         kfree(rdev->bios);
3989         rdev->bios = NULL;
3990 }
3991
3992 /*
3993  * Due to how kexec works, it can leave the hw fully initialised when it
3994  * boots the new kernel. However doing our init sequence with the CP and
3995  * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3996  * do some quick sanity checks and restore sane values to avoid this
3997  * problem.
3998  */
3999 void r100_restore_sanity(struct radeon_device *rdev)
4000 {
4001         u32 tmp;
4002
4003         tmp = RREG32(RADEON_CP_CSQ_CNTL);
4004         if (tmp) {
4005                 WREG32(RADEON_CP_CSQ_CNTL, 0);
4006         }
4007         tmp = RREG32(RADEON_CP_RB_CNTL);
4008         if (tmp) {
4009                 WREG32(RADEON_CP_RB_CNTL, 0);
4010         }
4011         tmp = RREG32(RADEON_SCRATCH_UMSK);
4012         if (tmp) {
4013                 WREG32(RADEON_SCRATCH_UMSK, 0);
4014         }
4015 }
4016
4017 int r100_init(struct radeon_device *rdev)
4018 {
4019         int r;
4020
4021         /* Register debugfs file specific to this group of asics */
4022         r100_debugfs(rdev);
4023         /* Disable VGA */
4024         r100_vga_render_disable(rdev);
4025         /* Initialize scratch registers */
4026         radeon_scratch_init(rdev);
4027         /* Initialize surface registers */
4028         radeon_surface_init(rdev);
4029         /* sanity check some register to avoid hangs like after kexec */
4030         r100_restore_sanity(rdev);
4031         /* TODO: disable VGA need to use VGA request */
4032         /* BIOS*/
4033         if (!radeon_get_bios(rdev)) {
4034                 if (ASIC_IS_AVIVO(rdev))
4035                         return -EINVAL;
4036         }
4037         if (rdev->is_atom_bios) {
4038                 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4039                 return -EINVAL;
4040         } else {
4041                 r = radeon_combios_init(rdev);
4042                 if (r)
4043                         return r;
4044         }
4045         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
4046         if (radeon_asic_reset(rdev)) {
4047                 dev_warn(rdev->dev,
4048                         "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4049                         RREG32(R_000E40_RBBM_STATUS),
4050                         RREG32(R_0007C0_CP_STAT));
4051         }
4052         /* check if cards are posted or not */
4053         if (radeon_boot_test_post_card(rdev) == false)
4054                 return -EINVAL;
4055         /* Set asic errata */
4056         r100_errata(rdev);
4057         /* Initialize clocks */
4058         radeon_get_clock_info(rdev->ddev);
4059         /* initialize AGP */
4060         if (rdev->flags & RADEON_IS_AGP) {
4061                 r = radeon_agp_init(rdev);
4062                 if (r) {
4063                         radeon_agp_disable(rdev);
4064                 }
4065         }
4066         /* initialize VRAM */
4067         r100_mc_init(rdev);
4068         /* Fence driver */
4069         r = radeon_fence_driver_init(rdev);
4070         if (r)
4071                 return r;
4072         /* Memory manager */
4073         r = radeon_bo_init(rdev);
4074         if (r)
4075                 return r;
4076         if (rdev->flags & RADEON_IS_PCI) {
4077                 r = r100_pci_gart_init(rdev);
4078                 if (r)
4079                         return r;
4080         }
4081         r100_set_safe_registers(rdev);
4082
4083         /* Initialize power management */
4084         radeon_pm_init(rdev);
4085
4086         rdev->accel_working = true;
4087         r = r100_startup(rdev);
4088         if (r) {
4089                 /* Somethings want wront with the accel init stop accel */
4090                 dev_err(rdev->dev, "Disabling GPU acceleration\n");
4091                 r100_cp_fini(rdev);
4092                 radeon_wb_fini(rdev);
4093                 radeon_ib_pool_fini(rdev);
4094                 radeon_irq_kms_fini(rdev);
4095                 if (rdev->flags & RADEON_IS_PCI)
4096                         r100_pci_gart_fini(rdev);
4097                 rdev->accel_working = false;
4098         }
4099         return 0;
4100 }
4101
4102 uint32_t r100_mm_rreg_slow(struct radeon_device *rdev, uint32_t reg)
4103 {
4104         unsigned long flags;
4105         uint32_t ret;
4106
4107         spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4108         writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4109         ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4110         spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4111         return ret;
4112 }
4113
4114 void r100_mm_wreg_slow(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4115 {
4116         unsigned long flags;
4117
4118         spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4119         writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4120         writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4121         spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4122 }
4123
4124 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4125 {
4126         if (reg < rdev->rio_mem_size)
4127                 return ioread32(rdev->rio_mem + reg);
4128         else {
4129                 iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4130                 return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4131         }
4132 }
4133
4134 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4135 {
4136         if (reg < rdev->rio_mem_size)
4137                 iowrite32(v, rdev->rio_mem + reg);
4138         else {
4139                 iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4140                 iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4141         }
4142 }