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 "radeonkmsfw_R100_cp"
49 #define FIRMWARE_R200 "radeonkmsfw_R200_cp"
50 #define FIRMWARE_R300 "radeonkmsfw_R300_cp"
51 #define FIRMWARE_R420 "radeonkmsfw_R420_cp"
52 #define FIRMWARE_RS690 "radeonkmsfw_RS690_cp"
53 #define FIRMWARE_RS600 "radeonkmsfw_RS600_cp"
54 #define FIRMWARE_R520 "radeonkmsfw_R520_cp"
55
56 MODULE_FIRMWARE(FIRMWARE_R100);
57 MODULE_FIRMWARE(FIRMWARE_R200);
58 MODULE_FIRMWARE(FIRMWARE_R300);
59 MODULE_FIRMWARE(FIRMWARE_R420);
60 MODULE_FIRMWARE(FIRMWARE_RS690);
61 MODULE_FIRMWARE(FIRMWARE_RS600);
62 MODULE_FIRMWARE(FIRMWARE_R520);
63
64 #include "r100_track.h"
65
66 /* This files gather functions specifics to:
67 * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
68 * and others in some cases.
69 */
70
r100_is_in_vblank(struct radeon_device * rdev,int crtc)71 static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
72 {
73 if (crtc == 0) {
74 if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
75 return true;
76 else
77 return false;
78 } else {
79 if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
80 return true;
81 else
82 return false;
83 }
84 }
85
r100_is_counter_moving(struct radeon_device * rdev,int crtc)86 static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
87 {
88 u32 vline1, vline2;
89
90 if (crtc == 0) {
91 vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
92 vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
93 } else {
94 vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
95 vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
96 }
97 if (vline1 != vline2)
98 return true;
99 else
100 return false;
101 }
102
103 /**
104 * r100_wait_for_vblank - vblank wait asic callback.
105 *
106 * @rdev: radeon_device pointer
107 * @crtc: crtc to wait for vblank on
108 *
109 * Wait for vblank on the requested crtc (r1xx-r4xx).
110 */
r100_wait_for_vblank(struct radeon_device * rdev,int crtc)111 void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
112 {
113 unsigned i = 0;
114
115 if (crtc >= rdev->num_crtc)
116 return;
117
118 if (crtc == 0) {
119 if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
120 return;
121 } else {
122 if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
123 return;
124 }
125
126 /* depending on when we hit vblank, we may be close to active; if so,
127 * wait for another frame.
128 */
129 while (r100_is_in_vblank(rdev, crtc)) {
130 if (i++ % 100 == 0) {
131 if (!r100_is_counter_moving(rdev, crtc))
132 break;
133 }
134 }
135
136 while (!r100_is_in_vblank(rdev, crtc)) {
137 if (i++ % 100 == 0) {
138 if (!r100_is_counter_moving(rdev, crtc))
139 break;
140 }
141 }
142 }
143
144 /**
145 * r100_page_flip - pageflip callback.
146 *
147 * @rdev: radeon_device pointer
148 * @crtc_id: crtc to cleanup pageflip on
149 * @crtc_base: new address of the crtc (GPU MC address)
150 *
151 * Does the actual pageflip (r1xx-r4xx).
152 * During vblank we take the crtc lock and wait for the update_pending
153 * bit to go high, when it does, we release the lock, and allow the
154 * double buffered update to take place.
155 */
r100_page_flip(struct radeon_device * rdev,int crtc_id,u64 crtc_base,bool async)156 void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base, bool async)
157 {
158 struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
159 u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
160 int i;
161
162 /* Lock the graphics update lock */
163 /* update the scanout addresses */
164 WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
165
166 /* Wait for update_pending to go high. */
167 for (i = 0; i < rdev->usec_timeout; i++) {
168 if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
169 break;
170 udelay(1);
171 }
172 DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
173
174 /* Unlock the lock, so double-buffering can take place inside vblank */
175 tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
176 WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
177
178 }
179
180 /**
181 * r100_page_flip_pending - check if page flip is still pending
182 *
183 * @rdev: radeon_device pointer
184 * @crtc_id: crtc to check
185 *
186 * Check if the last pagefilp is still pending (r1xx-r4xx).
187 * Returns the current update pending status.
188 */
r100_page_flip_pending(struct radeon_device * rdev,int crtc_id)189 bool r100_page_flip_pending(struct radeon_device *rdev, int crtc_id)
190 {
191 struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
192
193 /* Return current update_pending status: */
194 return !!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) &
195 RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET);
196 }
197
198 /**
199 * r100_pm_get_dynpm_state - look up dynpm power state callback.
200 *
201 * @rdev: radeon_device pointer
202 *
203 * Look up the optimal power state based on the
204 * current state of the GPU (r1xx-r5xx).
205 * Used for dynpm only.
206 */
r100_pm_get_dynpm_state(struct radeon_device * rdev)207 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
208 {
209 int i;
210 rdev->pm.dynpm_can_upclock = true;
211 rdev->pm.dynpm_can_downclock = true;
212
213 switch (rdev->pm.dynpm_planned_action) {
214 case DYNPM_ACTION_MINIMUM:
215 rdev->pm.requested_power_state_index = 0;
216 rdev->pm.dynpm_can_downclock = false;
217 break;
218 case DYNPM_ACTION_DOWNCLOCK:
219 if (rdev->pm.current_power_state_index == 0) {
220 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
221 rdev->pm.dynpm_can_downclock = false;
222 } else {
223 if (rdev->pm.active_crtc_count > 1) {
224 for (i = 0; i < rdev->pm.num_power_states; i++) {
225 if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
226 continue;
227 else if (i >= rdev->pm.current_power_state_index) {
228 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
229 break;
230 } else {
231 rdev->pm.requested_power_state_index = i;
232 break;
233 }
234 }
235 } else
236 rdev->pm.requested_power_state_index =
237 rdev->pm.current_power_state_index - 1;
238 }
239 /* don't use the power state if crtcs are active and no display flag is set */
240 if ((rdev->pm.active_crtc_count > 0) &&
241 (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
242 RADEON_PM_MODE_NO_DISPLAY)) {
243 rdev->pm.requested_power_state_index++;
244 }
245 break;
246 case DYNPM_ACTION_UPCLOCK:
247 if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
248 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
249 rdev->pm.dynpm_can_upclock = false;
250 } else {
251 if (rdev->pm.active_crtc_count > 1) {
252 for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
253 if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
254 continue;
255 else if (i <= rdev->pm.current_power_state_index) {
256 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
257 break;
258 } else {
259 rdev->pm.requested_power_state_index = i;
260 break;
261 }
262 }
263 } else
264 rdev->pm.requested_power_state_index =
265 rdev->pm.current_power_state_index + 1;
266 }
267 break;
268 case DYNPM_ACTION_DEFAULT:
269 rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
270 rdev->pm.dynpm_can_upclock = false;
271 break;
272 case DYNPM_ACTION_NONE:
273 default:
274 DRM_ERROR("Requested mode for not defined action\n");
275 return;
276 }
277 /* only one clock mode per power state */
278 rdev->pm.requested_clock_mode_index = 0;
279
280 DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
281 rdev->pm.power_state[rdev->pm.requested_power_state_index].
282 clock_info[rdev->pm.requested_clock_mode_index].sclk,
283 rdev->pm.power_state[rdev->pm.requested_power_state_index].
284 clock_info[rdev->pm.requested_clock_mode_index].mclk,
285 rdev->pm.power_state[rdev->pm.requested_power_state_index].
286 pcie_lanes);
287 }
288
289 /**
290 * r100_pm_init_profile - Initialize power profiles callback.
291 *
292 * @rdev: radeon_device pointer
293 *
294 * Initialize the power states used in profile mode
295 * (r1xx-r3xx).
296 * Used for profile mode only.
297 */
r100_pm_init_profile(struct radeon_device * rdev)298 void r100_pm_init_profile(struct radeon_device *rdev)
299 {
300 /* default */
301 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
302 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
303 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
304 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
305 /* low sh */
306 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
307 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
308 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
309 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
310 /* mid sh */
311 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
312 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
313 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
314 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
315 /* high sh */
316 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
317 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
318 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
319 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
320 /* low mh */
321 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
322 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
323 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
324 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
325 /* mid mh */
326 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
327 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
328 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
329 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
330 /* high mh */
331 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
332 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
333 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
334 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
335 }
336
337 /**
338 * r100_pm_misc - set additional pm hw parameters callback.
339 *
340 * @rdev: radeon_device pointer
341 *
342 * Set non-clock parameters associated with a power state
343 * (voltage, pcie lanes, etc.) (r1xx-r4xx).
344 */
r100_pm_misc(struct radeon_device * rdev)345 void r100_pm_misc(struct radeon_device *rdev)
346 {
347 int requested_index = rdev->pm.requested_power_state_index;
348 struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
349 struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
350 u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
351
352 if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
353 if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
354 tmp = RREG32(voltage->gpio.reg);
355 if (voltage->active_high)
356 tmp |= voltage->gpio.mask;
357 else
358 tmp &= ~(voltage->gpio.mask);
359 WREG32(voltage->gpio.reg, tmp);
360 if (voltage->delay)
361 udelay(voltage->delay);
362 } else {
363 tmp = RREG32(voltage->gpio.reg);
364 if (voltage->active_high)
365 tmp &= ~voltage->gpio.mask;
366 else
367 tmp |= voltage->gpio.mask;
368 WREG32(voltage->gpio.reg, tmp);
369 if (voltage->delay)
370 udelay(voltage->delay);
371 }
372 }
373
374 sclk_cntl = RREG32_PLL(SCLK_CNTL);
375 sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
376 sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
377 sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
378 sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
379 if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
380 sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
381 if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
382 sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
383 else
384 sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
385 if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
386 sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
387 else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
388 sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
389 } else
390 sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
391
392 if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
393 sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
394 if (voltage->delay) {
395 sclk_more_cntl |= VOLTAGE_DROP_SYNC;
396 switch (voltage->delay) {
397 case 33:
398 sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
399 break;
400 case 66:
401 sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
402 break;
403 case 99:
404 sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
405 break;
406 case 132:
407 sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
408 break;
409 }
410 } else
411 sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
412 } else
413 sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
414
415 if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
416 sclk_cntl &= ~FORCE_HDP;
417 else
418 sclk_cntl |= FORCE_HDP;
419
420 WREG32_PLL(SCLK_CNTL, sclk_cntl);
421 WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
422 WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
423
424 /* set pcie lanes */
425 if ((rdev->flags & RADEON_IS_PCIE) &&
426 !(rdev->flags & RADEON_IS_IGP) &&
427 rdev->asic->pm.set_pcie_lanes &&
428 (ps->pcie_lanes !=
429 rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
430 radeon_set_pcie_lanes(rdev,
431 ps->pcie_lanes);
432 DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
433 }
434 }
435
436 /**
437 * r100_pm_prepare - pre-power state change callback.
438 *
439 * @rdev: radeon_device pointer
440 *
441 * Prepare for a power state change (r1xx-r4xx).
442 */
r100_pm_prepare(struct radeon_device * rdev)443 void r100_pm_prepare(struct radeon_device *rdev)
444 {
445 struct drm_device *ddev = rdev->ddev;
446 struct drm_crtc *crtc;
447 struct radeon_crtc *radeon_crtc;
448 u32 tmp;
449
450 /* disable any active CRTCs */
451 list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
452 radeon_crtc = to_radeon_crtc(crtc);
453 if (radeon_crtc->enabled) {
454 if (radeon_crtc->crtc_id) {
455 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
456 tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
457 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
458 } else {
459 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
460 tmp |= RADEON_CRTC_DISP_REQ_EN_B;
461 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
462 }
463 }
464 }
465 }
466
467 /**
468 * r100_pm_finish - post-power state change callback.
469 *
470 * @rdev: radeon_device pointer
471 *
472 * Clean up after a power state change (r1xx-r4xx).
473 */
r100_pm_finish(struct radeon_device * rdev)474 void r100_pm_finish(struct radeon_device *rdev)
475 {
476 struct drm_device *ddev = rdev->ddev;
477 struct drm_crtc *crtc;
478 struct radeon_crtc *radeon_crtc;
479 u32 tmp;
480
481 /* enable any active CRTCs */
482 list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
483 radeon_crtc = to_radeon_crtc(crtc);
484 if (radeon_crtc->enabled) {
485 if (radeon_crtc->crtc_id) {
486 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
487 tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
488 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
489 } else {
490 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
491 tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
492 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
493 }
494 }
495 }
496 }
497
498 /**
499 * r100_gui_idle - gui idle callback.
500 *
501 * @rdev: radeon_device pointer
502 *
503 * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
504 * Returns true if idle, false if not.
505 */
r100_gui_idle(struct radeon_device * rdev)506 bool r100_gui_idle(struct radeon_device *rdev)
507 {
508 if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
509 return false;
510 else
511 return true;
512 }
513
514 /* hpd for digital panel detect/disconnect */
515 /**
516 * r100_hpd_sense - hpd sense callback.
517 *
518 * @rdev: radeon_device pointer
519 * @hpd: hpd (hotplug detect) pin
520 *
521 * Checks if a digital monitor is connected (r1xx-r4xx).
522 * Returns true if connected, false if not connected.
523 */
r100_hpd_sense(struct radeon_device * rdev,enum radeon_hpd_id hpd)524 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
525 {
526 bool connected = false;
527
528 switch (hpd) {
529 case RADEON_HPD_1:
530 if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
531 connected = true;
532 break;
533 case RADEON_HPD_2:
534 if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
535 connected = true;
536 break;
537 default:
538 break;
539 }
540 return connected;
541 }
542
543 /**
544 * r100_hpd_set_polarity - hpd set polarity callback.
545 *
546 * @rdev: radeon_device pointer
547 * @hpd: hpd (hotplug detect) pin
548 *
549 * Set the polarity of the hpd pin (r1xx-r4xx).
550 */
r100_hpd_set_polarity(struct radeon_device * rdev,enum radeon_hpd_id hpd)551 void r100_hpd_set_polarity(struct radeon_device *rdev,
552 enum radeon_hpd_id hpd)
553 {
554 u32 tmp;
555 bool connected = r100_hpd_sense(rdev, hpd);
556
557 switch (hpd) {
558 case RADEON_HPD_1:
559 tmp = RREG32(RADEON_FP_GEN_CNTL);
560 if (connected)
561 tmp &= ~RADEON_FP_DETECT_INT_POL;
562 else
563 tmp |= RADEON_FP_DETECT_INT_POL;
564 WREG32(RADEON_FP_GEN_CNTL, tmp);
565 break;
566 case RADEON_HPD_2:
567 tmp = RREG32(RADEON_FP2_GEN_CNTL);
568 if (connected)
569 tmp &= ~RADEON_FP2_DETECT_INT_POL;
570 else
571 tmp |= RADEON_FP2_DETECT_INT_POL;
572 WREG32(RADEON_FP2_GEN_CNTL, tmp);
573 break;
574 default:
575 break;
576 }
577 }
578
579 /**
580 * r100_hpd_init - hpd setup callback.
581 *
582 * @rdev: radeon_device pointer
583 *
584 * Setup the hpd pins used by the card (r1xx-r4xx).
585 * Set the polarity, and enable the hpd interrupts.
586 */
r100_hpd_init(struct radeon_device * rdev)587 void r100_hpd_init(struct radeon_device *rdev)
588 {
589 struct drm_device *dev = rdev->ddev;
590 struct drm_connector *connector;
591 unsigned enable = 0;
592
593 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
594 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
595 if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
596 enable |= 1 << radeon_connector->hpd.hpd;
597 radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
598 }
599 radeon_irq_kms_enable_hpd(rdev, enable);
600 }
601
602 /**
603 * r100_hpd_fini - hpd tear down callback.
604 *
605 * @rdev: radeon_device pointer
606 *
607 * Tear down the hpd pins used by the card (r1xx-r4xx).
608 * Disable the hpd interrupts.
609 */
r100_hpd_fini(struct radeon_device * rdev)610 void r100_hpd_fini(struct radeon_device *rdev)
611 {
612 struct drm_device *dev = rdev->ddev;
613 struct drm_connector *connector;
614 unsigned disable = 0;
615
616 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
617 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
618 if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
619 disable |= 1 << radeon_connector->hpd.hpd;
620 }
621 radeon_irq_kms_disable_hpd(rdev, disable);
622 }
623
624 /*
625 * PCI GART
626 */
r100_pci_gart_tlb_flush(struct radeon_device * rdev)627 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
628 {
629 /* TODO: can we do somethings here ? */
630 /* It seems hw only cache one entry so we should discard this
631 * entry otherwise if first GPU GART read hit this entry it
632 * could end up in wrong address. */
633 }
634
r100_pci_gart_init(struct radeon_device * rdev)635 int r100_pci_gart_init(struct radeon_device *rdev)
636 {
637 int r;
638
639 if (rdev->gart.ptr) {
640 WARN(1, "R100 PCI GART already initialized\n");
641 return 0;
642 }
643 /* Initialize common gart structure */
644 r = radeon_gart_init(rdev);
645 if (r)
646 return r;
647 rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
648 rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
649 rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
650 rdev->asic->gart.set_page = &r100_pci_gart_set_page;
651 return radeon_gart_table_ram_alloc(rdev);
652 }
653
r100_pci_gart_enable(struct radeon_device * rdev)654 int r100_pci_gart_enable(struct radeon_device *rdev)
655 {
656 uint32_t tmp;
657
658 /* discard memory request outside of configured range */
659 tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
660 WREG32(RADEON_AIC_CNTL, tmp);
661 /* set address range for PCI address translate */
662 WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
663 WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
664 /* set PCI GART page-table base address */
665 WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
666 tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
667 WREG32(RADEON_AIC_CNTL, tmp);
668 r100_pci_gart_tlb_flush(rdev);
669 DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
670 (unsigned)(rdev->mc.gtt_size >> 20),
671 (unsigned long long)rdev->gart.table_addr);
672 rdev->gart.ready = true;
673 return 0;
674 }
675
r100_pci_gart_disable(struct radeon_device * rdev)676 void r100_pci_gart_disable(struct radeon_device *rdev)
677 {
678 uint32_t tmp;
679
680 /* discard memory request outside of configured range */
681 tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
682 WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
683 WREG32(RADEON_AIC_LO_ADDR, 0);
684 WREG32(RADEON_AIC_HI_ADDR, 0);
685 }
686
r100_pci_gart_get_page_entry(uint64_t addr,uint32_t flags)687 uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
688 {
689 return addr;
690 }
691
r100_pci_gart_set_page(struct radeon_device * rdev,unsigned i,uint64_t entry)692 void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
693 uint64_t entry)
694 {
695 u32 *gtt = rdev->gart.ptr;
696 gtt[i] = cpu_to_le32(lower_32_bits(entry));
697 }
698
r100_pci_gart_fini(struct radeon_device * rdev)699 void r100_pci_gart_fini(struct radeon_device *rdev)
700 {
701 radeon_gart_fini(rdev);
702 r100_pci_gart_disable(rdev);
703 radeon_gart_table_ram_free(rdev);
704 }
705
r100_irq_set(struct radeon_device * rdev)706 int r100_irq_set(struct radeon_device *rdev)
707 {
708 uint32_t tmp = 0;
709
710 if (!rdev->irq.installed) {
711 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
712 WREG32(R_000040_GEN_INT_CNTL, 0);
713 return -EINVAL;
714 }
715 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
716 tmp |= RADEON_SW_INT_ENABLE;
717 }
718 if (rdev->irq.crtc_vblank_int[0] ||
719 atomic_read(&rdev->irq.pflip[0])) {
720 tmp |= RADEON_CRTC_VBLANK_MASK;
721 }
722 if (rdev->irq.crtc_vblank_int[1] ||
723 atomic_read(&rdev->irq.pflip[1])) {
724 tmp |= RADEON_CRTC2_VBLANK_MASK;
725 }
726 if (rdev->irq.hpd[0]) {
727 tmp |= RADEON_FP_DETECT_MASK;
728 }
729 if (rdev->irq.hpd[1]) {
730 tmp |= RADEON_FP2_DETECT_MASK;
731 }
732 WREG32(RADEON_GEN_INT_CNTL, tmp);
733
734 /* read back to post the write */
735 RREG32(RADEON_GEN_INT_CNTL);
736
737 return 0;
738 }
739
r100_irq_disable(struct radeon_device * rdev)740 void r100_irq_disable(struct radeon_device *rdev)
741 {
742 u32 tmp;
743
744 WREG32(R_000040_GEN_INT_CNTL, 0);
745 /* Wait and acknowledge irq */
746 mdelay(1);
747 tmp = RREG32(R_000044_GEN_INT_STATUS);
748 WREG32(R_000044_GEN_INT_STATUS, tmp);
749 }
750
r100_irq_ack(struct radeon_device * rdev)751 static uint32_t r100_irq_ack(struct radeon_device *rdev)
752 {
753 uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
754 uint32_t irq_mask = RADEON_SW_INT_TEST |
755 RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
756 RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
757
758 if (irqs) {
759 WREG32(RADEON_GEN_INT_STATUS, irqs);
760 }
761 return irqs & irq_mask;
762 }
763
r100_irq_process(struct radeon_device * rdev)764 irqreturn_t r100_irq_process(struct radeon_device *rdev)
765 {
766 uint32_t status, msi_rearm;
767 bool queue_hotplug = false;
768
769 status = r100_irq_ack(rdev);
770 if (!status) {
771 return IRQ_NONE;
772 }
773 if (rdev->shutdown) {
774 return IRQ_NONE;
775 }
776 while (status) {
777 /* SW interrupt */
778 if (status & RADEON_SW_INT_TEST) {
779 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
780 }
781 /* Vertical blank interrupts */
782 if (status & RADEON_CRTC_VBLANK_STAT) {
783 if (rdev->irq.crtc_vblank_int[0]) {
784 drm_handle_vblank(rdev->ddev, 0);
785 rdev->pm.vblank_sync = true;
786 wake_up(&rdev->irq.vblank_queue);
787 }
788 if (atomic_read(&rdev->irq.pflip[0]))
789 radeon_crtc_handle_vblank(rdev, 0);
790 }
791 if (status & RADEON_CRTC2_VBLANK_STAT) {
792 if (rdev->irq.crtc_vblank_int[1]) {
793 drm_handle_vblank(rdev->ddev, 1);
794 rdev->pm.vblank_sync = true;
795 wake_up(&rdev->irq.vblank_queue);
796 }
797 if (atomic_read(&rdev->irq.pflip[1]))
798 radeon_crtc_handle_vblank(rdev, 1);
799 }
800 if (status & RADEON_FP_DETECT_STAT) {
801 queue_hotplug = true;
802 DRM_DEBUG("HPD1\n");
803 }
804 if (status & RADEON_FP2_DETECT_STAT) {
805 queue_hotplug = true;
806 DRM_DEBUG("HPD2\n");
807 }
808 status = r100_irq_ack(rdev);
809 }
810 if (queue_hotplug)
811 schedule_delayed_work(&rdev->hotplug_work, 0);
812 if (rdev->msi_enabled) {
813 switch (rdev->family) {
814 case CHIP_RS400:
815 case CHIP_RS480:
816 msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
817 WREG32(RADEON_AIC_CNTL, msi_rearm);
818 WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
819 break;
820 default:
821 WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
822 break;
823 }
824 }
825 return IRQ_HANDLED;
826 }
827
r100_get_vblank_counter(struct radeon_device * rdev,int crtc)828 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
829 {
830 if (crtc == 0)
831 return RREG32(RADEON_CRTC_CRNT_FRAME);
832 else
833 return RREG32(RADEON_CRTC2_CRNT_FRAME);
834 }
835
836 /**
837 * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
838 * rdev: radeon device structure
839 * ring: ring buffer struct for emitting packets
840 */
r100_ring_hdp_flush(struct radeon_device * rdev,struct radeon_ring * ring)841 static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
842 {
843 radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
844 radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
845 RADEON_HDP_READ_BUFFER_INVALIDATE);
846 radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
847 radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
848 }
849
850 /* Who ever call radeon_fence_emit should call ring_lock and ask
851 * for enough space (today caller are ib schedule and buffer move) */
r100_fence_ring_emit(struct radeon_device * rdev,struct radeon_fence * fence)852 void r100_fence_ring_emit(struct radeon_device *rdev,
853 struct radeon_fence *fence)
854 {
855 struct radeon_ring *ring = &rdev->ring[fence->ring];
856
857 /* We have to make sure that caches are flushed before
858 * CPU might read something from VRAM. */
859 radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
860 radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
861 radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
862 radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
863 /* Wait until IDLE & CLEAN */
864 radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
865 radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
866 r100_ring_hdp_flush(rdev, ring);
867 /* Emit fence sequence & fire IRQ */
868 radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
869 radeon_ring_write(ring, fence->seq);
870 radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
871 radeon_ring_write(ring, RADEON_SW_INT_FIRE);
872 }
873
r100_semaphore_ring_emit(struct radeon_device * rdev,struct radeon_ring * ring,struct radeon_semaphore * semaphore,bool emit_wait)874 bool r100_semaphore_ring_emit(struct radeon_device *rdev,
875 struct radeon_ring *ring,
876 struct radeon_semaphore *semaphore,
877 bool emit_wait)
878 {
879 /* Unused on older asics, since we don't have semaphores or multiple rings */
880 BUG();
881 return false;
882 }
883
r100_copy_blit(struct radeon_device * rdev,uint64_t src_offset,uint64_t dst_offset,unsigned num_gpu_pages,struct reservation_object * resv)884 struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
885 uint64_t src_offset,
886 uint64_t dst_offset,
887 unsigned num_gpu_pages,
888 struct reservation_object *resv)
889 {
890 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
891 struct radeon_fence *fence;
892 uint32_t cur_pages;
893 uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
894 uint32_t pitch;
895 uint32_t stride_pixels;
896 unsigned ndw;
897 int num_loops;
898 int r = 0;
899
900 /* radeon limited to 16k stride */
901 stride_bytes &= 0x3fff;
902 /* radeon pitch is /64 */
903 pitch = stride_bytes / 64;
904 stride_pixels = stride_bytes / 4;
905 num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
906
907 /* Ask for enough room for blit + flush + fence */
908 ndw = 64 + (10 * num_loops);
909 r = radeon_ring_lock(rdev, ring, ndw);
910 if (r) {
911 DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
912 return ERR_PTR(-EINVAL);
913 }
914 while (num_gpu_pages > 0) {
915 cur_pages = num_gpu_pages;
916 if (cur_pages > 8191) {
917 cur_pages = 8191;
918 }
919 num_gpu_pages -= cur_pages;
920
921 /* pages are in Y direction - height
922 page width in X direction - width */
923 radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
924 radeon_ring_write(ring,
925 RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
926 RADEON_GMC_DST_PITCH_OFFSET_CNTL |
927 RADEON_GMC_SRC_CLIPPING |
928 RADEON_GMC_DST_CLIPPING |
929 RADEON_GMC_BRUSH_NONE |
930 (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
931 RADEON_GMC_SRC_DATATYPE_COLOR |
932 RADEON_ROP3_S |
933 RADEON_DP_SRC_SOURCE_MEMORY |
934 RADEON_GMC_CLR_CMP_CNTL_DIS |
935 RADEON_GMC_WR_MSK_DIS);
936 radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
937 radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
938 radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
939 radeon_ring_write(ring, 0);
940 radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
941 radeon_ring_write(ring, num_gpu_pages);
942 radeon_ring_write(ring, num_gpu_pages);
943 radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
944 }
945 radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
946 radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
947 radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
948 radeon_ring_write(ring,
949 RADEON_WAIT_2D_IDLECLEAN |
950 RADEON_WAIT_HOST_IDLECLEAN |
951 RADEON_WAIT_DMA_GUI_IDLE);
952 r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
953 if (r) {
954 radeon_ring_unlock_undo(rdev, ring);
955 return ERR_PTR(r);
956 }
957 radeon_ring_unlock_commit(rdev, ring, false);
958 return fence;
959 }
960
r100_cp_wait_for_idle(struct radeon_device * rdev)961 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
962 {
963 unsigned i;
964 u32 tmp;
965
966 for (i = 0; i < rdev->usec_timeout; i++) {
967 tmp = RREG32(R_000E40_RBBM_STATUS);
968 if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
969 return 0;
970 }
971 udelay(1);
972 }
973 return -1;
974 }
975
r100_ring_start(struct radeon_device * rdev,struct radeon_ring * ring)976 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
977 {
978 int r;
979
980 r = radeon_ring_lock(rdev, ring, 2);
981 if (r) {
982 return;
983 }
984 radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
985 radeon_ring_write(ring,
986 RADEON_ISYNC_ANY2D_IDLE3D |
987 RADEON_ISYNC_ANY3D_IDLE2D |
988 RADEON_ISYNC_WAIT_IDLEGUI |
989 RADEON_ISYNC_CPSCRATCH_IDLEGUI);
990 radeon_ring_unlock_commit(rdev, ring, false);
991 }
992
993
994 /* Load the microcode for the CP */
r100_cp_init_microcode(struct radeon_device * rdev)995 static int r100_cp_init_microcode(struct radeon_device *rdev)
996 {
997 const char *fw_name = NULL;
998 int err;
999
1000 DRM_DEBUG_KMS("\n");
1001
1002 if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
1003 (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
1004 (rdev->family == CHIP_RS200)) {
1005 DRM_INFO("Loading R100 Microcode\n");
1006 fw_name = FIRMWARE_R100;
1007 } else if ((rdev->family == CHIP_R200) ||
1008 (rdev->family == CHIP_RV250) ||
1009 (rdev->family == CHIP_RV280) ||
1010 (rdev->family == CHIP_RS300)) {
1011 DRM_INFO("Loading R200 Microcode\n");
1012 fw_name = FIRMWARE_R200;
1013 } else if ((rdev->family == CHIP_R300) ||
1014 (rdev->family == CHIP_R350) ||
1015 (rdev->family == CHIP_RV350) ||
1016 (rdev->family == CHIP_RV380) ||
1017 (rdev->family == CHIP_RS400) ||
1018 (rdev->family == CHIP_RS480)) {
1019 DRM_INFO("Loading R300 Microcode\n");
1020 fw_name = FIRMWARE_R300;
1021 } else if ((rdev->family == CHIP_R420) ||
1022 (rdev->family == CHIP_R423) ||
1023 (rdev->family == CHIP_RV410)) {
1024 DRM_INFO("Loading R400 Microcode\n");
1025 fw_name = FIRMWARE_R420;
1026 } else if ((rdev->family == CHIP_RS690) ||
1027 (rdev->family == CHIP_RS740)) {
1028 DRM_INFO("Loading RS690/RS740 Microcode\n");
1029 fw_name = FIRMWARE_RS690;
1030 } else if (rdev->family == CHIP_RS600) {
1031 DRM_INFO("Loading RS600 Microcode\n");
1032 fw_name = FIRMWARE_RS600;
1033 } else if ((rdev->family == CHIP_RV515) ||
1034 (rdev->family == CHIP_R520) ||
1035 (rdev->family == CHIP_RV530) ||
1036 (rdev->family == CHIP_R580) ||
1037 (rdev->family == CHIP_RV560) ||
1038 (rdev->family == CHIP_RV570)) {
1039 DRM_INFO("Loading R500 Microcode\n");
1040 fw_name = FIRMWARE_R520;
1041 }
1042
1043 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1044 if (err) {
1045 printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1046 fw_name);
1047 } else if (rdev->me_fw->datasize % 8) {
1048 printk(KERN_ERR
1049 "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1050 rdev->me_fw->datasize, fw_name);
1051 err = -EINVAL;
1052 release_firmware(rdev->me_fw);
1053 rdev->me_fw = NULL;
1054 }
1055 return err;
1056 }
1057
r100_gfx_get_rptr(struct radeon_device * rdev,struct radeon_ring * ring)1058 u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1059 struct radeon_ring *ring)
1060 {
1061 u32 rptr;
1062
1063 if (rdev->wb.enabled)
1064 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1065 else
1066 rptr = RREG32(RADEON_CP_RB_RPTR);
1067
1068 return rptr;
1069 }
1070
r100_gfx_get_wptr(struct radeon_device * rdev,struct radeon_ring * ring)1071 u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1072 struct radeon_ring *ring)
1073 {
1074 return RREG32(RADEON_CP_RB_WPTR);
1075 }
1076
r100_gfx_set_wptr(struct radeon_device * rdev,struct radeon_ring * ring)1077 void r100_gfx_set_wptr(struct radeon_device *rdev,
1078 struct radeon_ring *ring)
1079 {
1080 WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1081 (void)RREG32(RADEON_CP_RB_WPTR);
1082 }
1083
1084 /**
1085 * r100_cp_fini_microcode - drop the firmware image reference
1086 *
1087 * @rdev: radeon_device pointer
1088 *
1089 * Drop the me firmware image reference.
1090 * Called at driver shutdown.
1091 */
r100_cp_fini_microcode(struct radeon_device * rdev)1092 static void r100_cp_fini_microcode (struct radeon_device *rdev)
1093 {
1094 release_firmware(rdev->me_fw);
1095 rdev->me_fw = NULL;
1096 }
1097
r100_cp_load_microcode(struct radeon_device * rdev)1098 static void r100_cp_load_microcode(struct radeon_device *rdev)
1099 {
1100 const __be32 *fw_data;
1101 int i, size;
1102
1103 if (r100_gui_wait_for_idle(rdev)) {
1104 pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1105 }
1106
1107 if (rdev->me_fw) {
1108 size = rdev->me_fw->datasize / 4;
1109 fw_data = (const __be32 *)&rdev->me_fw->data[0];
1110 WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1111 for (i = 0; i < size; i += 2) {
1112 WREG32(RADEON_CP_ME_RAM_DATAH,
1113 be32_to_cpup(&fw_data[i]));
1114 WREG32(RADEON_CP_ME_RAM_DATAL,
1115 be32_to_cpup(&fw_data[i + 1]));
1116 }
1117 }
1118 }
1119
r100_cp_init(struct radeon_device * rdev,unsigned ring_size)1120 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1121 {
1122 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1123 unsigned rb_bufsz;
1124 unsigned rb_blksz;
1125 unsigned max_fetch;
1126 unsigned pre_write_timer;
1127 unsigned pre_write_limit;
1128 unsigned indirect2_start;
1129 unsigned indirect1_start;
1130 uint32_t tmp;
1131 int r;
1132
1133 if (r100_debugfs_cp_init(rdev)) {
1134 DRM_ERROR("Failed to register debugfs file for CP !\n");
1135 }
1136 if (!rdev->me_fw) {
1137 r = r100_cp_init_microcode(rdev);
1138 if (r) {
1139 DRM_ERROR("Failed to load firmware!\n");
1140 return r;
1141 }
1142 }
1143
1144 /* Align ring size */
1145 rb_bufsz = order_base_2(ring_size / 8);
1146 ring_size = (1 << (rb_bufsz + 1)) * 4;
1147 r100_cp_load_microcode(rdev);
1148 r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1149 RADEON_CP_PACKET2);
1150 if (r) {
1151 return r;
1152 }
1153 /* Each time the cp read 1024 bytes (16 dword/quadword) update
1154 * the rptr copy in system ram */
1155 rb_blksz = 9;
1156 /* cp will read 128bytes at a time (4 dwords) */
1157 max_fetch = 1;
1158 ring->align_mask = 16 - 1;
1159 /* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1160 pre_write_timer = 64;
1161 /* Force CP_RB_WPTR write if written more than one time before the
1162 * delay expire
1163 */
1164 pre_write_limit = 0;
1165 /* Setup the cp cache like this (cache size is 96 dwords) :
1166 * RING 0 to 15
1167 * INDIRECT1 16 to 79
1168 * INDIRECT2 80 to 95
1169 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1170 * indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1171 * indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1172 * Idea being that most of the gpu cmd will be through indirect1 buffer
1173 * so it gets the bigger cache.
1174 */
1175 indirect2_start = 80;
1176 indirect1_start = 16;
1177 /* cp setup */
1178 WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1179 tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1180 REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1181 REG_SET(RADEON_MAX_FETCH, max_fetch));
1182 #ifdef __BIG_ENDIAN
1183 tmp |= RADEON_BUF_SWAP_32BIT;
1184 #endif
1185 WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1186
1187 /* Set ring address */
1188 DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1189 WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1190 /* Force read & write ptr to 0 */
1191 WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1192 WREG32(RADEON_CP_RB_RPTR_WR, 0);
1193 ring->wptr = 0;
1194 WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1195
1196 /* set the wb address whether it's enabled or not */
1197 WREG32(R_00070C_CP_RB_RPTR_ADDR,
1198 S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1199 WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1200
1201 if (rdev->wb.enabled)
1202 WREG32(R_000770_SCRATCH_UMSK, 0xff);
1203 else {
1204 tmp |= RADEON_RB_NO_UPDATE;
1205 WREG32(R_000770_SCRATCH_UMSK, 0);
1206 }
1207
1208 WREG32(RADEON_CP_RB_CNTL, tmp);
1209 udelay(10);
1210 /* Set cp mode to bus mastering & enable cp*/
1211 WREG32(RADEON_CP_CSQ_MODE,
1212 REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1213 REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1214 WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1215 WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1216 WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1217
1218 /* at this point everything should be setup correctly to enable master */
1219 pci_enable_busmaster(rdev->dev->bsddev);
1220
1221 radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1222 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1223 if (r) {
1224 DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1225 return r;
1226 }
1227 ring->ready = true;
1228 radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1229
1230 if (!ring->rptr_save_reg /* not resuming from suspend */
1231 && radeon_ring_supports_scratch_reg(rdev, ring)) {
1232 r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1233 if (r) {
1234 DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1235 ring->rptr_save_reg = 0;
1236 }
1237 }
1238 return 0;
1239 }
1240
r100_cp_fini(struct radeon_device * rdev)1241 void r100_cp_fini(struct radeon_device *rdev)
1242 {
1243 if (r100_cp_wait_for_idle(rdev)) {
1244 DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1245 }
1246 /* Disable ring */
1247 r100_cp_disable(rdev);
1248 radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1249 radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1250 DRM_INFO("radeon: cp finalized\n");
1251 }
1252
r100_cp_disable(struct radeon_device * rdev)1253 void r100_cp_disable(struct radeon_device *rdev)
1254 {
1255 /* Disable ring */
1256 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1257 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1258 WREG32(RADEON_CP_CSQ_MODE, 0);
1259 WREG32(RADEON_CP_CSQ_CNTL, 0);
1260 WREG32(R_000770_SCRATCH_UMSK, 0);
1261 if (r100_gui_wait_for_idle(rdev)) {
1262 pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1263 }
1264 }
1265
1266 /*
1267 * CS functions
1268 */
r100_reloc_pitch_offset(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,unsigned idx,unsigned reg)1269 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1270 struct radeon_cs_packet *pkt,
1271 unsigned idx,
1272 unsigned reg)
1273 {
1274 int r;
1275 u32 tile_flags = 0;
1276 u32 tmp;
1277 struct radeon_bo_list *reloc;
1278 u32 value;
1279
1280 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1281 if (r) {
1282 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1283 idx, reg);
1284 radeon_cs_dump_packet(p, pkt);
1285 return r;
1286 }
1287
1288 value = radeon_get_ib_value(p, idx);
1289 tmp = value & 0x003fffff;
1290 tmp += (((u32)reloc->gpu_offset) >> 10);
1291
1292 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1293 if (reloc->tiling_flags & RADEON_TILING_MACRO)
1294 tile_flags |= RADEON_DST_TILE_MACRO;
1295 if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1296 if (reg == RADEON_SRC_PITCH_OFFSET) {
1297 DRM_ERROR("Cannot src blit from microtiled surface\n");
1298 radeon_cs_dump_packet(p, pkt);
1299 return -EINVAL;
1300 }
1301 tile_flags |= RADEON_DST_TILE_MICRO;
1302 }
1303
1304 tmp |= tile_flags;
1305 p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1306 } else
1307 p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1308 return 0;
1309 }
1310
r100_packet3_load_vbpntr(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,int idx)1311 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1312 struct radeon_cs_packet *pkt,
1313 int idx)
1314 {
1315 unsigned c, i;
1316 struct radeon_bo_list *reloc;
1317 struct r100_cs_track *track;
1318 int r = 0;
1319 volatile uint32_t *ib;
1320 u32 idx_value;
1321
1322 ib = p->ib.ptr;
1323 track = (struct r100_cs_track *)p->track;
1324 c = radeon_get_ib_value(p, idx++) & 0x1F;
1325 if (c > 16) {
1326 DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1327 pkt->opcode);
1328 radeon_cs_dump_packet(p, pkt);
1329 return -EINVAL;
1330 }
1331 track->num_arrays = c;
1332 for (i = 0; i < (c - 1); i+=2, idx+=3) {
1333 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1334 if (r) {
1335 DRM_ERROR("No reloc for packet3 %d\n",
1336 pkt->opcode);
1337 radeon_cs_dump_packet(p, pkt);
1338 return r;
1339 }
1340 idx_value = radeon_get_ib_value(p, idx);
1341 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1342
1343 track->arrays[i + 0].esize = idx_value >> 8;
1344 track->arrays[i + 0].robj = reloc->robj;
1345 track->arrays[i + 0].esize &= 0x7F;
1346 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1347 if (r) {
1348 DRM_ERROR("No reloc for packet3 %d\n",
1349 pkt->opcode);
1350 radeon_cs_dump_packet(p, pkt);
1351 return r;
1352 }
1353 ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
1354 track->arrays[i + 1].robj = reloc->robj;
1355 track->arrays[i + 1].esize = idx_value >> 24;
1356 track->arrays[i + 1].esize &= 0x7F;
1357 }
1358 if (c & 1) {
1359 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1360 if (r) {
1361 DRM_ERROR("No reloc for packet3 %d\n",
1362 pkt->opcode);
1363 radeon_cs_dump_packet(p, pkt);
1364 return r;
1365 }
1366 idx_value = radeon_get_ib_value(p, idx);
1367 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1368 track->arrays[i + 0].robj = reloc->robj;
1369 track->arrays[i + 0].esize = idx_value >> 8;
1370 track->arrays[i + 0].esize &= 0x7F;
1371 }
1372 return r;
1373 }
1374
r100_cs_parse_packet0(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,const unsigned * auth,unsigned n,radeon_packet0_check_t check)1375 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1376 struct radeon_cs_packet *pkt,
1377 const unsigned *auth, unsigned n,
1378 radeon_packet0_check_t check)
1379 {
1380 unsigned reg;
1381 unsigned i, j, m;
1382 unsigned idx;
1383 int r;
1384
1385 idx = pkt->idx + 1;
1386 reg = pkt->reg;
1387 /* Check that register fall into register range
1388 * determined by the number of entry (n) in the
1389 * safe register bitmap.
1390 */
1391 if (pkt->one_reg_wr) {
1392 if ((reg >> 7) > n) {
1393 return -EINVAL;
1394 }
1395 } else {
1396 if (((reg + (pkt->count << 2)) >> 7) > n) {
1397 return -EINVAL;
1398 }
1399 }
1400 for (i = 0; i <= pkt->count; i++, idx++) {
1401 j = (reg >> 7);
1402 m = 1 << ((reg >> 2) & 31);
1403 if (auth[j] & m) {
1404 r = check(p, pkt, idx, reg);
1405 if (r) {
1406 return r;
1407 }
1408 }
1409 if (pkt->one_reg_wr) {
1410 if (!(auth[j] & m)) {
1411 break;
1412 }
1413 } else {
1414 reg += 4;
1415 }
1416 }
1417 return 0;
1418 }
1419
1420 /**
1421 * r100_cs_packet_next_vline() - parse userspace VLINE packet
1422 * @parser: parser structure holding parsing context.
1423 *
1424 * Userspace sends a special sequence for VLINE waits.
1425 * PACKET0 - VLINE_START_END + value
1426 * PACKET0 - WAIT_UNTIL +_value
1427 * RELOC (P3) - crtc_id in reloc.
1428 *
1429 * This function parses this and relocates the VLINE START END
1430 * and WAIT UNTIL packets to the correct crtc.
1431 * It also detects a switched off crtc and nulls out the
1432 * wait in that case.
1433 */
r100_cs_packet_parse_vline(struct radeon_cs_parser * p)1434 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1435 {
1436 struct drm_crtc *crtc;
1437 struct radeon_crtc *radeon_crtc;
1438 struct radeon_cs_packet p3reloc, waitreloc;
1439 int crtc_id;
1440 int r;
1441 uint32_t header, h_idx, reg;
1442 volatile uint32_t *ib;
1443
1444 ib = p->ib.ptr;
1445
1446 /* parse the wait until */
1447 r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1448 if (r)
1449 return r;
1450
1451 /* check its a wait until and only 1 count */
1452 if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1453 waitreloc.count != 0) {
1454 DRM_ERROR("vline wait had illegal wait until segment\n");
1455 return -EINVAL;
1456 }
1457
1458 if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1459 DRM_ERROR("vline wait had illegal wait until\n");
1460 return -EINVAL;
1461 }
1462
1463 /* jump over the NOP */
1464 r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1465 if (r)
1466 return r;
1467
1468 h_idx = p->idx - 2;
1469 p->idx += waitreloc.count + 2;
1470 p->idx += p3reloc.count + 2;
1471
1472 header = radeon_get_ib_value(p, h_idx);
1473 crtc_id = radeon_get_ib_value(p, h_idx + 5);
1474 reg = R100_CP_PACKET0_GET_REG(header);
1475 crtc = drm_crtc_find(p->rdev->ddev, p->filp, crtc_id);
1476 if (!crtc) {
1477 DRM_ERROR("cannot find crtc %d\n", crtc_id);
1478 return -ENOENT;
1479 }
1480 radeon_crtc = to_radeon_crtc(crtc);
1481 crtc_id = radeon_crtc->crtc_id;
1482
1483 if (!crtc->enabled) {
1484 /* if the CRTC isn't enabled - we need to nop out the wait until */
1485 ib[h_idx + 2] = PACKET2(0);
1486 ib[h_idx + 3] = PACKET2(0);
1487 } else if (crtc_id == 1) {
1488 switch (reg) {
1489 case AVIVO_D1MODE_VLINE_START_END:
1490 header &= ~R300_CP_PACKET0_REG_MASK;
1491 header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1492 break;
1493 case RADEON_CRTC_GUI_TRIG_VLINE:
1494 header &= ~R300_CP_PACKET0_REG_MASK;
1495 header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1496 break;
1497 default:
1498 DRM_ERROR("unknown crtc reloc\n");
1499 return -EINVAL;
1500 }
1501 ib[h_idx] = header;
1502 ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1503 }
1504
1505 return 0;
1506 }
1507
r100_get_vtx_size(uint32_t vtx_fmt)1508 static int r100_get_vtx_size(uint32_t vtx_fmt)
1509 {
1510 int vtx_size;
1511 vtx_size = 2;
1512 /* ordered according to bits in spec */
1513 if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1514 vtx_size++;
1515 if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1516 vtx_size += 3;
1517 if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1518 vtx_size++;
1519 if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1520 vtx_size++;
1521 if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1522 vtx_size += 3;
1523 if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1524 vtx_size++;
1525 if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1526 vtx_size++;
1527 if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1528 vtx_size += 2;
1529 if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1530 vtx_size += 2;
1531 if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1532 vtx_size++;
1533 if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1534 vtx_size += 2;
1535 if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1536 vtx_size++;
1537 if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1538 vtx_size += 2;
1539 if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1540 vtx_size++;
1541 if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1542 vtx_size++;
1543 /* blend weight */
1544 if (vtx_fmt & (0x7 << 15))
1545 vtx_size += (vtx_fmt >> 15) & 0x7;
1546 if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1547 vtx_size += 3;
1548 if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1549 vtx_size += 2;
1550 if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1551 vtx_size++;
1552 if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1553 vtx_size++;
1554 if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1555 vtx_size++;
1556 if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1557 vtx_size++;
1558 return vtx_size;
1559 }
1560
r100_packet0_check(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,unsigned idx,unsigned reg)1561 static int r100_packet0_check(struct radeon_cs_parser *p,
1562 struct radeon_cs_packet *pkt,
1563 unsigned idx, unsigned reg)
1564 {
1565 struct radeon_bo_list *reloc;
1566 struct r100_cs_track *track;
1567 volatile uint32_t *ib;
1568 uint32_t tmp;
1569 int r;
1570 int i, face;
1571 u32 tile_flags = 0;
1572 u32 idx_value;
1573
1574 ib = p->ib.ptr;
1575 track = (struct r100_cs_track *)p->track;
1576
1577 idx_value = radeon_get_ib_value(p, idx);
1578
1579 switch (reg) {
1580 case RADEON_CRTC_GUI_TRIG_VLINE:
1581 r = r100_cs_packet_parse_vline(p);
1582 if (r) {
1583 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1584 idx, reg);
1585 radeon_cs_dump_packet(p, pkt);
1586 return r;
1587 }
1588 break;
1589 /* FIXME: only allow PACKET3 blit? easier to check for out of
1590 * range access */
1591 case RADEON_DST_PITCH_OFFSET:
1592 case RADEON_SRC_PITCH_OFFSET:
1593 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1594 if (r)
1595 return r;
1596 break;
1597 case RADEON_RB3D_DEPTHOFFSET:
1598 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1599 if (r) {
1600 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1601 idx, reg);
1602 radeon_cs_dump_packet(p, pkt);
1603 return r;
1604 }
1605 track->zb.robj = reloc->robj;
1606 track->zb.offset = idx_value;
1607 track->zb_dirty = true;
1608 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1609 break;
1610 case RADEON_RB3D_COLOROFFSET:
1611 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1612 if (r) {
1613 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1614 idx, reg);
1615 radeon_cs_dump_packet(p, pkt);
1616 return r;
1617 }
1618 track->cb[0].robj = reloc->robj;
1619 track->cb[0].offset = idx_value;
1620 track->cb_dirty = true;
1621 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1622 break;
1623 case RADEON_PP_TXOFFSET_0:
1624 case RADEON_PP_TXOFFSET_1:
1625 case RADEON_PP_TXOFFSET_2:
1626 i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1627 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1628 if (r) {
1629 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1630 idx, reg);
1631 radeon_cs_dump_packet(p, pkt);
1632 return r;
1633 }
1634 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1635 if (reloc->tiling_flags & RADEON_TILING_MACRO)
1636 tile_flags |= RADEON_TXO_MACRO_TILE;
1637 if (reloc->tiling_flags & RADEON_TILING_MICRO)
1638 tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1639
1640 tmp = idx_value & ~(0x7 << 2);
1641 tmp |= tile_flags;
1642 ib[idx] = tmp + ((u32)reloc->gpu_offset);
1643 } else
1644 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1645 track->textures[i].robj = reloc->robj;
1646 track->tex_dirty = true;
1647 break;
1648 case RADEON_PP_CUBIC_OFFSET_T0_0:
1649 case RADEON_PP_CUBIC_OFFSET_T0_1:
1650 case RADEON_PP_CUBIC_OFFSET_T0_2:
1651 case RADEON_PP_CUBIC_OFFSET_T0_3:
1652 case RADEON_PP_CUBIC_OFFSET_T0_4:
1653 i = (reg - RADEON_PP_CUBIC_OFFSET_T0_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[0].cube_info[i].offset = idx_value;
1662 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1663 track->textures[0].cube_info[i].robj = reloc->robj;
1664 track->tex_dirty = true;
1665 break;
1666 case RADEON_PP_CUBIC_OFFSET_T1_0:
1667 case RADEON_PP_CUBIC_OFFSET_T1_1:
1668 case RADEON_PP_CUBIC_OFFSET_T1_2:
1669 case RADEON_PP_CUBIC_OFFSET_T1_3:
1670 case RADEON_PP_CUBIC_OFFSET_T1_4:
1671 i = (reg - RADEON_PP_CUBIC_OFFSET_T1_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[1].cube_info[i].offset = idx_value;
1680 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1681 track->textures[1].cube_info[i].robj = reloc->robj;
1682 track->tex_dirty = true;
1683 break;
1684 case RADEON_PP_CUBIC_OFFSET_T2_0:
1685 case RADEON_PP_CUBIC_OFFSET_T2_1:
1686 case RADEON_PP_CUBIC_OFFSET_T2_2:
1687 case RADEON_PP_CUBIC_OFFSET_T2_3:
1688 case RADEON_PP_CUBIC_OFFSET_T2_4:
1689 i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
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 track->textures[2].cube_info[i].offset = idx_value;
1698 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1699 track->textures[2].cube_info[i].robj = reloc->robj;
1700 track->tex_dirty = true;
1701 break;
1702 case RADEON_RE_WIDTH_HEIGHT:
1703 track->maxy = ((idx_value >> 16) & 0x7FF);
1704 track->cb_dirty = true;
1705 track->zb_dirty = true;
1706 break;
1707 case RADEON_RB3D_COLORPITCH:
1708 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1709 if (r) {
1710 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1711 idx, reg);
1712 radeon_cs_dump_packet(p, pkt);
1713 return r;
1714 }
1715 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1716 if (reloc->tiling_flags & RADEON_TILING_MACRO)
1717 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1718 if (reloc->tiling_flags & RADEON_TILING_MICRO)
1719 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1720
1721 tmp = idx_value & ~(0x7 << 16);
1722 tmp |= tile_flags;
1723 ib[idx] = tmp;
1724 } else
1725 ib[idx] = idx_value;
1726
1727 track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1728 track->cb_dirty = true;
1729 break;
1730 case RADEON_RB3D_DEPTHPITCH:
1731 track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1732 track->zb_dirty = true;
1733 break;
1734 case RADEON_RB3D_CNTL:
1735 switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1736 case 7:
1737 case 8:
1738 case 9:
1739 case 11:
1740 case 12:
1741 track->cb[0].cpp = 1;
1742 break;
1743 case 3:
1744 case 4:
1745 case 15:
1746 track->cb[0].cpp = 2;
1747 break;
1748 case 6:
1749 track->cb[0].cpp = 4;
1750 break;
1751 default:
1752 DRM_ERROR("Invalid color buffer format (%d) !\n",
1753 ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1754 return -EINVAL;
1755 }
1756 track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1757 track->cb_dirty = true;
1758 track->zb_dirty = true;
1759 break;
1760 case RADEON_RB3D_ZSTENCILCNTL:
1761 switch (idx_value & 0xf) {
1762 case 0:
1763 track->zb.cpp = 2;
1764 break;
1765 case 2:
1766 case 3:
1767 case 4:
1768 case 5:
1769 case 9:
1770 case 11:
1771 track->zb.cpp = 4;
1772 break;
1773 default:
1774 break;
1775 }
1776 track->zb_dirty = true;
1777 break;
1778 case RADEON_RB3D_ZPASS_ADDR:
1779 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1780 if (r) {
1781 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1782 idx, reg);
1783 radeon_cs_dump_packet(p, pkt);
1784 return r;
1785 }
1786 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1787 break;
1788 case RADEON_PP_CNTL:
1789 {
1790 uint32_t temp = idx_value >> 4;
1791 for (i = 0; i < track->num_texture; i++)
1792 track->textures[i].enabled = !!(temp & (1 << i));
1793 track->tex_dirty = true;
1794 }
1795 break;
1796 case RADEON_SE_VF_CNTL:
1797 track->vap_vf_cntl = idx_value;
1798 break;
1799 case RADEON_SE_VTX_FMT:
1800 track->vtx_size = r100_get_vtx_size(idx_value);
1801 break;
1802 case RADEON_PP_TEX_SIZE_0:
1803 case RADEON_PP_TEX_SIZE_1:
1804 case RADEON_PP_TEX_SIZE_2:
1805 i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1806 track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1807 track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1808 track->tex_dirty = true;
1809 break;
1810 case RADEON_PP_TEX_PITCH_0:
1811 case RADEON_PP_TEX_PITCH_1:
1812 case RADEON_PP_TEX_PITCH_2:
1813 i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1814 track->textures[i].pitch = idx_value + 32;
1815 track->tex_dirty = true;
1816 break;
1817 case RADEON_PP_TXFILTER_0:
1818 case RADEON_PP_TXFILTER_1:
1819 case RADEON_PP_TXFILTER_2:
1820 i = (reg - RADEON_PP_TXFILTER_0) / 24;
1821 track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1822 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1823 tmp = (idx_value >> 23) & 0x7;
1824 if (tmp == 2 || tmp == 6)
1825 track->textures[i].roundup_w = false;
1826 tmp = (idx_value >> 27) & 0x7;
1827 if (tmp == 2 || tmp == 6)
1828 track->textures[i].roundup_h = false;
1829 track->tex_dirty = true;
1830 break;
1831 case RADEON_PP_TXFORMAT_0:
1832 case RADEON_PP_TXFORMAT_1:
1833 case RADEON_PP_TXFORMAT_2:
1834 i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1835 if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1836 track->textures[i].use_pitch = 1;
1837 } else {
1838 track->textures[i].use_pitch = 0;
1839 track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1840 track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1841 }
1842 if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1843 track->textures[i].tex_coord_type = 2;
1844 switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1845 case RADEON_TXFORMAT_I8:
1846 case RADEON_TXFORMAT_RGB332:
1847 case RADEON_TXFORMAT_Y8:
1848 track->textures[i].cpp = 1;
1849 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1850 break;
1851 case RADEON_TXFORMAT_AI88:
1852 case RADEON_TXFORMAT_ARGB1555:
1853 case RADEON_TXFORMAT_RGB565:
1854 case RADEON_TXFORMAT_ARGB4444:
1855 case RADEON_TXFORMAT_VYUY422:
1856 case RADEON_TXFORMAT_YVYU422:
1857 case RADEON_TXFORMAT_SHADOW16:
1858 case RADEON_TXFORMAT_LDUDV655:
1859 case RADEON_TXFORMAT_DUDV88:
1860 track->textures[i].cpp = 2;
1861 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1862 break;
1863 case RADEON_TXFORMAT_ARGB8888:
1864 case RADEON_TXFORMAT_RGBA8888:
1865 case RADEON_TXFORMAT_SHADOW32:
1866 case RADEON_TXFORMAT_LDUDUV8888:
1867 track->textures[i].cpp = 4;
1868 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1869 break;
1870 case RADEON_TXFORMAT_DXT1:
1871 track->textures[i].cpp = 1;
1872 track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1873 break;
1874 case RADEON_TXFORMAT_DXT23:
1875 case RADEON_TXFORMAT_DXT45:
1876 track->textures[i].cpp = 1;
1877 track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1878 break;
1879 }
1880 track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1881 track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1882 track->tex_dirty = true;
1883 break;
1884 case RADEON_PP_CUBIC_FACES_0:
1885 case RADEON_PP_CUBIC_FACES_1:
1886 case RADEON_PP_CUBIC_FACES_2:
1887 tmp = idx_value;
1888 i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1889 for (face = 0; face < 4; face++) {
1890 track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1891 track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1892 }
1893 track->tex_dirty = true;
1894 break;
1895 default:
1896 pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
1897 return -EINVAL;
1898 }
1899 return 0;
1900 }
1901
r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt,struct radeon_bo * robj)1902 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1903 struct radeon_cs_packet *pkt,
1904 struct radeon_bo *robj)
1905 {
1906 unsigned idx;
1907 u32 value;
1908 idx = pkt->idx + 1;
1909 value = radeon_get_ib_value(p, idx + 2);
1910 if ((value + 1) > radeon_bo_size(robj)) {
1911 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1912 "(need %u have %lu) !\n",
1913 value + 1,
1914 radeon_bo_size(robj));
1915 return -EINVAL;
1916 }
1917 return 0;
1918 }
1919
r100_packet3_check(struct radeon_cs_parser * p,struct radeon_cs_packet * pkt)1920 static int r100_packet3_check(struct radeon_cs_parser *p,
1921 struct radeon_cs_packet *pkt)
1922 {
1923 struct radeon_bo_list *reloc;
1924 struct r100_cs_track *track;
1925 unsigned idx;
1926 volatile uint32_t *ib;
1927 int r;
1928
1929 ib = p->ib.ptr;
1930 idx = pkt->idx + 1;
1931 track = (struct r100_cs_track *)p->track;
1932 switch (pkt->opcode) {
1933 case PACKET3_3D_LOAD_VBPNTR:
1934 r = r100_packet3_load_vbpntr(p, pkt, idx);
1935 if (r)
1936 return r;
1937 break;
1938 case PACKET3_INDX_BUFFER:
1939 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1940 if (r) {
1941 DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1942 radeon_cs_dump_packet(p, pkt);
1943 return r;
1944 }
1945 ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1946 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1947 if (r) {
1948 return r;
1949 }
1950 break;
1951 case 0x23:
1952 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1953 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1954 if (r) {
1955 DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1956 radeon_cs_dump_packet(p, pkt);
1957 return r;
1958 }
1959 ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1960 track->num_arrays = 1;
1961 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1962
1963 track->arrays[0].robj = reloc->robj;
1964 track->arrays[0].esize = track->vtx_size;
1965
1966 track->max_indx = radeon_get_ib_value(p, idx+1);
1967
1968 track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1969 track->immd_dwords = pkt->count - 1;
1970 r = r100_cs_track_check(p->rdev, track);
1971 if (r)
1972 return r;
1973 break;
1974 case PACKET3_3D_DRAW_IMMD:
1975 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1976 DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1977 return -EINVAL;
1978 }
1979 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1980 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1981 track->immd_dwords = pkt->count - 1;
1982 r = r100_cs_track_check(p->rdev, track);
1983 if (r)
1984 return r;
1985 break;
1986 /* triggers drawing using in-packet vertex data */
1987 case PACKET3_3D_DRAW_IMMD_2:
1988 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1989 DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1990 return -EINVAL;
1991 }
1992 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1993 track->immd_dwords = pkt->count;
1994 r = r100_cs_track_check(p->rdev, track);
1995 if (r)
1996 return r;
1997 break;
1998 /* triggers drawing using in-packet vertex data */
1999 case PACKET3_3D_DRAW_VBUF_2:
2000 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
2001 r = r100_cs_track_check(p->rdev, track);
2002 if (r)
2003 return r;
2004 break;
2005 /* triggers drawing of vertex buffers setup elsewhere */
2006 case PACKET3_3D_DRAW_INDX_2:
2007 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
2008 r = r100_cs_track_check(p->rdev, track);
2009 if (r)
2010 return r;
2011 break;
2012 /* triggers drawing using indices to vertex buffer */
2013 case PACKET3_3D_DRAW_VBUF:
2014 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2015 r = r100_cs_track_check(p->rdev, track);
2016 if (r)
2017 return r;
2018 break;
2019 /* triggers drawing of vertex buffers setup elsewhere */
2020 case PACKET3_3D_DRAW_INDX:
2021 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2022 r = r100_cs_track_check(p->rdev, track);
2023 if (r)
2024 return r;
2025 break;
2026 /* triggers drawing using indices to vertex buffer */
2027 case PACKET3_3D_CLEAR_HIZ:
2028 case PACKET3_3D_CLEAR_ZMASK:
2029 if (p->rdev->hyperz_filp != p->filp)
2030 return -EINVAL;
2031 break;
2032 case PACKET3_NOP:
2033 break;
2034 default:
2035 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2036 return -EINVAL;
2037 }
2038 return 0;
2039 }
2040
r100_cs_parse(struct radeon_cs_parser * p)2041 int r100_cs_parse(struct radeon_cs_parser *p)
2042 {
2043 struct radeon_cs_packet pkt;
2044 struct r100_cs_track *track;
2045 int r;
2046
2047 track = kzalloc(sizeof(*track), GFP_KERNEL);
2048 if (!track)
2049 return -ENOMEM;
2050 r100_cs_track_clear(p->rdev, track);
2051 p->track = track;
2052 do {
2053 r = radeon_cs_packet_parse(p, &pkt, p->idx);
2054 if (r) {
2055 return r;
2056 }
2057 p->idx += pkt.count + 2;
2058 switch (pkt.type) {
2059 case RADEON_PACKET_TYPE0:
2060 if (p->rdev->family >= CHIP_R200)
2061 r = r100_cs_parse_packet0(p, &pkt,
2062 p->rdev->config.r100.reg_safe_bm,
2063 p->rdev->config.r100.reg_safe_bm_size,
2064 &r200_packet0_check);
2065 else
2066 r = r100_cs_parse_packet0(p, &pkt,
2067 p->rdev->config.r100.reg_safe_bm,
2068 p->rdev->config.r100.reg_safe_bm_size,
2069 &r100_packet0_check);
2070 break;
2071 case RADEON_PACKET_TYPE2:
2072 break;
2073 case RADEON_PACKET_TYPE3:
2074 r = r100_packet3_check(p, &pkt);
2075 break;
2076 default:
2077 DRM_ERROR("Unknown packet type %d !\n",
2078 pkt.type);
2079 return -EINVAL;
2080 }
2081 if (r)
2082 return r;
2083 } while (p->idx < p->chunk_ib->length_dw);
2084 return 0;
2085 }
2086
r100_cs_track_texture_print(struct r100_cs_track_texture * t)2087 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2088 {
2089 DRM_ERROR("pitch %d\n", t->pitch);
2090 DRM_ERROR("use_pitch %d\n", t->use_pitch);
2091 DRM_ERROR("width %d\n", t->width);
2092 DRM_ERROR("width_11 %d\n", t->width_11);
2093 DRM_ERROR("height %d\n", t->height);
2094 DRM_ERROR("height_11 %d\n", t->height_11);
2095 DRM_ERROR("num levels %d\n", t->num_levels);
2096 DRM_ERROR("depth %d\n", t->txdepth);
2097 DRM_ERROR("bpp %d\n", t->cpp);
2098 DRM_ERROR("coordinate type %d\n", t->tex_coord_type);
2099 DRM_ERROR("width round to power of 2 %d\n", t->roundup_w);
2100 DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2101 DRM_ERROR("compress format %d\n", t->compress_format);
2102 }
2103
r100_track_compress_size(int compress_format,int w,int h)2104 static int r100_track_compress_size(int compress_format, int w, int h)
2105 {
2106 int block_width, block_height, block_bytes;
2107 int wblocks, hblocks;
2108 int min_wblocks;
2109 int sz;
2110
2111 block_width = 4;
2112 block_height = 4;
2113
2114 switch (compress_format) {
2115 case R100_TRACK_COMP_DXT1:
2116 block_bytes = 8;
2117 min_wblocks = 4;
2118 break;
2119 default:
2120 case R100_TRACK_COMP_DXT35:
2121 block_bytes = 16;
2122 min_wblocks = 2;
2123 break;
2124 }
2125
2126 hblocks = (h + block_height - 1) / block_height;
2127 wblocks = (w + block_width - 1) / block_width;
2128 if (wblocks < min_wblocks)
2129 wblocks = min_wblocks;
2130 sz = wblocks * hblocks * block_bytes;
2131 return sz;
2132 }
2133
r100_cs_track_cube(struct radeon_device * rdev,struct r100_cs_track * track,unsigned idx)2134 static int r100_cs_track_cube(struct radeon_device *rdev,
2135 struct r100_cs_track *track, unsigned idx)
2136 {
2137 unsigned face, w, h;
2138 struct radeon_bo *cube_robj;
2139 unsigned long size;
2140 unsigned compress_format = track->textures[idx].compress_format;
2141
2142 for (face = 0; face < 5; face++) {
2143 cube_robj = track->textures[idx].cube_info[face].robj;
2144 w = track->textures[idx].cube_info[face].width;
2145 h = track->textures[idx].cube_info[face].height;
2146
2147 if (compress_format) {
2148 size = r100_track_compress_size(compress_format, w, h);
2149 } else
2150 size = w * h;
2151 size *= track->textures[idx].cpp;
2152
2153 size += track->textures[idx].cube_info[face].offset;
2154
2155 if (size > radeon_bo_size(cube_robj)) {
2156 DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2157 size, radeon_bo_size(cube_robj));
2158 r100_cs_track_texture_print(&track->textures[idx]);
2159 return -1;
2160 }
2161 }
2162 return 0;
2163 }
2164
r100_cs_track_texture_check(struct radeon_device * rdev,struct r100_cs_track * track)2165 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2166 struct r100_cs_track *track)
2167 {
2168 struct radeon_bo *robj;
2169 unsigned long size;
2170 unsigned u, i, w, h, d;
2171 int ret;
2172
2173 for (u = 0; u < track->num_texture; u++) {
2174 if (!track->textures[u].enabled)
2175 continue;
2176 if (track->textures[u].lookup_disable)
2177 continue;
2178 robj = track->textures[u].robj;
2179 if (robj == NULL) {
2180 DRM_ERROR("No texture bound to unit %u\n", u);
2181 return -EINVAL;
2182 }
2183 size = 0;
2184 for (i = 0; i <= track->textures[u].num_levels; i++) {
2185 if (track->textures[u].use_pitch) {
2186 if (rdev->family < CHIP_R300)
2187 w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2188 else
2189 w = track->textures[u].pitch / (1 << i);
2190 } else {
2191 w = track->textures[u].width;
2192 if (rdev->family >= CHIP_RV515)
2193 w |= track->textures[u].width_11;
2194 w = w / (1 << i);
2195 if (track->textures[u].roundup_w)
2196 w = roundup_pow_of_two(w);
2197 }
2198 h = track->textures[u].height;
2199 if (rdev->family >= CHIP_RV515)
2200 h |= track->textures[u].height_11;
2201 h = h / (1 << i);
2202 if (track->textures[u].roundup_h)
2203 h = roundup_pow_of_two(h);
2204 if (track->textures[u].tex_coord_type == 1) {
2205 d = (1 << track->textures[u].txdepth) / (1 << i);
2206 if (!d)
2207 d = 1;
2208 } else {
2209 d = 1;
2210 }
2211 if (track->textures[u].compress_format) {
2212
2213 size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2214 /* compressed textures are block based */
2215 } else
2216 size += w * h * d;
2217 }
2218 size *= track->textures[u].cpp;
2219
2220 switch (track->textures[u].tex_coord_type) {
2221 case 0:
2222 case 1:
2223 break;
2224 case 2:
2225 if (track->separate_cube) {
2226 ret = r100_cs_track_cube(rdev, track, u);
2227 if (ret)
2228 return ret;
2229 } else
2230 size *= 6;
2231 break;
2232 default:
2233 DRM_ERROR("Invalid texture coordinate type %u for unit "
2234 "%u\n", track->textures[u].tex_coord_type, u);
2235 return -EINVAL;
2236 }
2237 if (size > radeon_bo_size(robj)) {
2238 DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2239 "%lu\n", u, size, radeon_bo_size(robj));
2240 r100_cs_track_texture_print(&track->textures[u]);
2241 return -EINVAL;
2242 }
2243 }
2244 return 0;
2245 }
2246
r100_cs_track_check(struct radeon_device * rdev,struct r100_cs_track * track)2247 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2248 {
2249 unsigned i;
2250 unsigned long size;
2251 unsigned prim_walk;
2252 unsigned nverts;
2253 unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2254
2255 if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2256 !track->blend_read_enable)
2257 num_cb = 0;
2258
2259 for (i = 0; i < num_cb; i++) {
2260 if (track->cb[i].robj == NULL) {
2261 DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2262 return -EINVAL;
2263 }
2264 size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2265 size += track->cb[i].offset;
2266 if (size > radeon_bo_size(track->cb[i].robj)) {
2267 DRM_ERROR("[drm] Buffer too small for color buffer %d "
2268 "(need %lu have %lu) !\n", i, size,
2269 radeon_bo_size(track->cb[i].robj));
2270 DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2271 i, track->cb[i].pitch, track->cb[i].cpp,
2272 track->cb[i].offset, track->maxy);
2273 return -EINVAL;
2274 }
2275 }
2276 track->cb_dirty = false;
2277
2278 if (track->zb_dirty && track->z_enabled) {
2279 if (track->zb.robj == NULL) {
2280 DRM_ERROR("[drm] No buffer for z buffer !\n");
2281 return -EINVAL;
2282 }
2283 size = track->zb.pitch * track->zb.cpp * track->maxy;
2284 size += track->zb.offset;
2285 if (size > radeon_bo_size(track->zb.robj)) {
2286 DRM_ERROR("[drm] Buffer too small for z buffer "
2287 "(need %lu have %lu) !\n", size,
2288 radeon_bo_size(track->zb.robj));
2289 DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2290 track->zb.pitch, track->zb.cpp,
2291 track->zb.offset, track->maxy);
2292 return -EINVAL;
2293 }
2294 }
2295 track->zb_dirty = false;
2296
2297 if (track->aa_dirty && track->aaresolve) {
2298 if (track->aa.robj == NULL) {
2299 DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2300 return -EINVAL;
2301 }
2302 /* I believe the format comes from colorbuffer0. */
2303 size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2304 size += track->aa.offset;
2305 if (size > radeon_bo_size(track->aa.robj)) {
2306 DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2307 "(need %lu have %lu) !\n", i, size,
2308 radeon_bo_size(track->aa.robj));
2309 DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2310 i, track->aa.pitch, track->cb[0].cpp,
2311 track->aa.offset, track->maxy);
2312 return -EINVAL;
2313 }
2314 }
2315 track->aa_dirty = false;
2316
2317 prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2318 if (track->vap_vf_cntl & (1 << 14)) {
2319 nverts = track->vap_alt_nverts;
2320 } else {
2321 nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2322 }
2323 switch (prim_walk) {
2324 case 1:
2325 for (i = 0; i < track->num_arrays; i++) {
2326 size = track->arrays[i].esize * track->max_indx * 4;
2327 if (track->arrays[i].robj == NULL) {
2328 DRM_ERROR("(PW %u) Vertex array %u no buffer "
2329 "bound\n", prim_walk, i);
2330 return -EINVAL;
2331 }
2332 if (size > radeon_bo_size(track->arrays[i].robj)) {
2333 dev_err(rdev->dev, "(PW %u) Vertex array %u "
2334 "need %lu dwords have %lu dwords\n",
2335 prim_walk, i, size >> 2,
2336 radeon_bo_size(track->arrays[i].robj)
2337 >> 2);
2338 DRM_ERROR("Max indices %u\n", track->max_indx);
2339 return -EINVAL;
2340 }
2341 }
2342 break;
2343 case 2:
2344 for (i = 0; i < track->num_arrays; i++) {
2345 size = track->arrays[i].esize * (nverts - 1) * 4;
2346 if (track->arrays[i].robj == NULL) {
2347 DRM_ERROR("(PW %u) Vertex array %u no buffer "
2348 "bound\n", prim_walk, i);
2349 return -EINVAL;
2350 }
2351 if (size > radeon_bo_size(track->arrays[i].robj)) {
2352 dev_err(rdev->dev, "(PW %u) Vertex array %u "
2353 "need %lu dwords have %lu dwords\n",
2354 prim_walk, i, size >> 2,
2355 radeon_bo_size(track->arrays[i].robj)
2356 >> 2);
2357 return -EINVAL;
2358 }
2359 }
2360 break;
2361 case 3:
2362 size = track->vtx_size * nverts;
2363 if (size != track->immd_dwords) {
2364 DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2365 track->immd_dwords, size);
2366 DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2367 nverts, track->vtx_size);
2368 return -EINVAL;
2369 }
2370 break;
2371 default:
2372 DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2373 prim_walk);
2374 return -EINVAL;
2375 }
2376
2377 if (track->tex_dirty) {
2378 track->tex_dirty = false;
2379 return r100_cs_track_texture_check(rdev, track);
2380 }
2381 return 0;
2382 }
2383
r100_cs_track_clear(struct radeon_device * rdev,struct r100_cs_track * track)2384 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2385 {
2386 unsigned i, face;
2387
2388 track->cb_dirty = true;
2389 track->zb_dirty = true;
2390 track->tex_dirty = true;
2391 track->aa_dirty = true;
2392
2393 if (rdev->family < CHIP_R300) {
2394 track->num_cb = 1;
2395 if (rdev->family <= CHIP_RS200)
2396 track->num_texture = 3;
2397 else
2398 track->num_texture = 6;
2399 track->maxy = 2048;
2400 track->separate_cube = 1;
2401 } else {
2402 track->num_cb = 4;
2403 track->num_texture = 16;
2404 track->maxy = 4096;
2405 track->separate_cube = 0;
2406 track->aaresolve = false;
2407 track->aa.robj = NULL;
2408 }
2409
2410 for (i = 0; i < track->num_cb; i++) {
2411 track->cb[i].robj = NULL;
2412 track->cb[i].pitch = 8192;
2413 track->cb[i].cpp = 16;
2414 track->cb[i].offset = 0;
2415 }
2416 track->z_enabled = true;
2417 track->zb.robj = NULL;
2418 track->zb.pitch = 8192;
2419 track->zb.cpp = 4;
2420 track->zb.offset = 0;
2421 track->vtx_size = 0x7F;
2422 track->immd_dwords = 0xFFFFFFFFUL;
2423 track->num_arrays = 11;
2424 track->max_indx = 0x00FFFFFFUL;
2425 for (i = 0; i < track->num_arrays; i++) {
2426 track->arrays[i].robj = NULL;
2427 track->arrays[i].esize = 0x7F;
2428 }
2429 for (i = 0; i < track->num_texture; i++) {
2430 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2431 track->textures[i].pitch = 16536;
2432 track->textures[i].width = 16536;
2433 track->textures[i].height = 16536;
2434 track->textures[i].width_11 = 1 << 11;
2435 track->textures[i].height_11 = 1 << 11;
2436 track->textures[i].num_levels = 12;
2437 if (rdev->family <= CHIP_RS200) {
2438 track->textures[i].tex_coord_type = 0;
2439 track->textures[i].txdepth = 0;
2440 } else {
2441 track->textures[i].txdepth = 16;
2442 track->textures[i].tex_coord_type = 1;
2443 }
2444 track->textures[i].cpp = 64;
2445 track->textures[i].robj = NULL;
2446 /* CS IB emission code makes sure texture unit are disabled */
2447 track->textures[i].enabled = false;
2448 track->textures[i].lookup_disable = false;
2449 track->textures[i].roundup_w = true;
2450 track->textures[i].roundup_h = true;
2451 if (track->separate_cube)
2452 for (face = 0; face < 5; face++) {
2453 track->textures[i].cube_info[face].robj = NULL;
2454 track->textures[i].cube_info[face].width = 16536;
2455 track->textures[i].cube_info[face].height = 16536;
2456 track->textures[i].cube_info[face].offset = 0;
2457 }
2458 }
2459 }
2460
2461 /*
2462 * Global GPU functions
2463 */
r100_errata(struct radeon_device * rdev)2464 static void r100_errata(struct radeon_device *rdev)
2465 {
2466 rdev->pll_errata = 0;
2467
2468 if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2469 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2470 }
2471
2472 if (rdev->family == CHIP_RV100 ||
2473 rdev->family == CHIP_RS100 ||
2474 rdev->family == CHIP_RS200) {
2475 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2476 }
2477 }
2478
r100_rbbm_fifo_wait_for_entry(struct radeon_device * rdev,unsigned n)2479 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2480 {
2481 unsigned i;
2482 uint32_t tmp;
2483
2484 for (i = 0; i < rdev->usec_timeout; i++) {
2485 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2486 if (tmp >= n) {
2487 return 0;
2488 }
2489 DRM_UDELAY(1);
2490 }
2491 return -1;
2492 }
2493
r100_gui_wait_for_idle(struct radeon_device * rdev)2494 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2495 {
2496 unsigned i;
2497 uint32_t tmp;
2498
2499 if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2500 pr_warn("radeon: wait for empty RBBM fifo failed! Bad things might happen.\n");
2501 }
2502 for (i = 0; i < rdev->usec_timeout; i++) {
2503 tmp = RREG32(RADEON_RBBM_STATUS);
2504 if (!(tmp & RADEON_RBBM_ACTIVE)) {
2505 return 0;
2506 }
2507 DRM_UDELAY(1);
2508 }
2509 return -1;
2510 }
2511
r100_mc_wait_for_idle(struct radeon_device * rdev)2512 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2513 {
2514 unsigned i;
2515 uint32_t tmp;
2516
2517 for (i = 0; i < rdev->usec_timeout; i++) {
2518 /* read MC_STATUS */
2519 tmp = RREG32(RADEON_MC_STATUS);
2520 if (tmp & RADEON_MC_IDLE) {
2521 return 0;
2522 }
2523 DRM_UDELAY(1);
2524 }
2525 return -1;
2526 }
2527
r100_gpu_is_lockup(struct radeon_device * rdev,struct radeon_ring * ring)2528 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2529 {
2530 u32 rbbm_status;
2531
2532 rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2533 if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2534 radeon_ring_lockup_update(rdev, ring);
2535 return false;
2536 }
2537 return radeon_ring_test_lockup(rdev, ring);
2538 }
2539
2540 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
r100_enable_bm(struct radeon_device * rdev)2541 void r100_enable_bm(struct radeon_device *rdev)
2542 {
2543 uint32_t tmp;
2544 /* Enable bus mastering */
2545 tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2546 WREG32(RADEON_BUS_CNTL, tmp);
2547 }
2548
r100_bm_disable(struct radeon_device * rdev)2549 void r100_bm_disable(struct radeon_device *rdev)
2550 {
2551 u32 tmp;
2552
2553 /* disable bus mastering */
2554 tmp = RREG32(R_000030_BUS_CNTL);
2555 WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2556 mdelay(1);
2557 WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2558 mdelay(1);
2559 WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2560 tmp = RREG32(RADEON_BUS_CNTL);
2561 mdelay(1);
2562 pci_disable_busmaster(rdev->dev->bsddev);
2563 mdelay(1);
2564 }
2565
r100_asic_reset(struct radeon_device * rdev,bool hard)2566 int r100_asic_reset(struct radeon_device *rdev, bool hard)
2567 {
2568 struct r100_mc_save save;
2569 u32 status, tmp;
2570 int ret = 0;
2571
2572 status = RREG32(R_000E40_RBBM_STATUS);
2573 if (!G_000E40_GUI_ACTIVE(status)) {
2574 return 0;
2575 }
2576 r100_mc_stop(rdev, &save);
2577 status = RREG32(R_000E40_RBBM_STATUS);
2578 dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2579 /* stop CP */
2580 WREG32(RADEON_CP_CSQ_CNTL, 0);
2581 tmp = RREG32(RADEON_CP_RB_CNTL);
2582 WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2583 WREG32(RADEON_CP_RB_RPTR_WR, 0);
2584 WREG32(RADEON_CP_RB_WPTR, 0);
2585 WREG32(RADEON_CP_RB_CNTL, tmp);
2586 /* save PCI state */
2587 pci_save_state(device_get_parent(rdev->dev->bsddev));
2588 /* disable bus mastering */
2589 r100_bm_disable(rdev);
2590 WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2591 S_0000F0_SOFT_RESET_RE(1) |
2592 S_0000F0_SOFT_RESET_PP(1) |
2593 S_0000F0_SOFT_RESET_RB(1));
2594 RREG32(R_0000F0_RBBM_SOFT_RESET);
2595 mdelay(500);
2596 WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2597 mdelay(1);
2598 status = RREG32(R_000E40_RBBM_STATUS);
2599 dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2600 /* reset CP */
2601 WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2602 RREG32(R_0000F0_RBBM_SOFT_RESET);
2603 mdelay(500);
2604 WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2605 mdelay(1);
2606 status = RREG32(R_000E40_RBBM_STATUS);
2607 dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2608 /* restore PCI & busmastering */
2609 pci_restore_state(device_get_parent(rdev->dev->bsddev));
2610 r100_enable_bm(rdev);
2611 /* Check if GPU is idle */
2612 if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2613 G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2614 dev_err(rdev->dev, "failed to reset GPU\n");
2615 ret = -1;
2616 } else
2617 dev_info(rdev->dev, "GPU reset succeed\n");
2618 r100_mc_resume(rdev, &save);
2619 return ret;
2620 }
2621
r100_set_common_regs(struct radeon_device * rdev)2622 void r100_set_common_regs(struct radeon_device *rdev)
2623 {
2624 struct drm_device *dev = rdev->ddev;
2625 bool force_dac2 = false;
2626 u32 tmp;
2627
2628 /* set these so they don't interfere with anything */
2629 WREG32(RADEON_OV0_SCALE_CNTL, 0);
2630 WREG32(RADEON_SUBPIC_CNTL, 0);
2631 WREG32(RADEON_VIPH_CONTROL, 0);
2632 WREG32(RADEON_I2C_CNTL_1, 0);
2633 WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2634 WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2635 WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2636
2637 /* always set up dac2 on rn50 and some rv100 as lots
2638 * of servers seem to wire it up to a VGA port but
2639 * don't report it in the bios connector
2640 * table.
2641 */
2642 switch (dev->pdev->device) {
2643 /* RN50 */
2644 case 0x515e:
2645 case 0x5969:
2646 force_dac2 = true;
2647 break;
2648 /* RV100*/
2649 case 0x5159:
2650 case 0x515a:
2651 /* DELL triple head servers */
2652 if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2653 ((dev->pdev->subsystem_device == 0x016c) ||
2654 (dev->pdev->subsystem_device == 0x016d) ||
2655 (dev->pdev->subsystem_device == 0x016e) ||
2656 (dev->pdev->subsystem_device == 0x016f) ||
2657 (dev->pdev->subsystem_device == 0x0170) ||
2658 (dev->pdev->subsystem_device == 0x017d) ||
2659 (dev->pdev->subsystem_device == 0x017e) ||
2660 (dev->pdev->subsystem_device == 0x0183) ||
2661 (dev->pdev->subsystem_device == 0x018a) ||
2662 (dev->pdev->subsystem_device == 0x019a)))
2663 force_dac2 = true;
2664 break;
2665 }
2666
2667 if (force_dac2) {
2668 u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2669 u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2670 u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2671
2672 /* For CRT on DAC2, don't turn it on if BIOS didn't
2673 enable it, even it's detected.
2674 */
2675
2676 /* force it to crtc0 */
2677 dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2678 dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2679 disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2680
2681 /* set up the TV DAC */
2682 tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2683 RADEON_TV_DAC_STD_MASK |
2684 RADEON_TV_DAC_RDACPD |
2685 RADEON_TV_DAC_GDACPD |
2686 RADEON_TV_DAC_BDACPD |
2687 RADEON_TV_DAC_BGADJ_MASK |
2688 RADEON_TV_DAC_DACADJ_MASK);
2689 tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2690 RADEON_TV_DAC_NHOLD |
2691 RADEON_TV_DAC_STD_PS2 |
2692 (0x58 << 16));
2693
2694 WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2695 WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2696 WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2697 }
2698
2699 /* switch PM block to ACPI mode */
2700 tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2701 tmp &= ~RADEON_PM_MODE_SEL;
2702 WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2703
2704 }
2705
2706 /*
2707 * VRAM info
2708 */
r100_vram_get_type(struct radeon_device * rdev)2709 static void r100_vram_get_type(struct radeon_device *rdev)
2710 {
2711 uint32_t tmp;
2712
2713 rdev->mc.vram_is_ddr = false;
2714 if (rdev->flags & RADEON_IS_IGP)
2715 rdev->mc.vram_is_ddr = true;
2716 else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2717 rdev->mc.vram_is_ddr = true;
2718 if ((rdev->family == CHIP_RV100) ||
2719 (rdev->family == CHIP_RS100) ||
2720 (rdev->family == CHIP_RS200)) {
2721 tmp = RREG32(RADEON_MEM_CNTL);
2722 if (tmp & RV100_HALF_MODE) {
2723 rdev->mc.vram_width = 32;
2724 } else {
2725 rdev->mc.vram_width = 64;
2726 }
2727 if (rdev->flags & RADEON_SINGLE_CRTC) {
2728 rdev->mc.vram_width /= 4;
2729 rdev->mc.vram_is_ddr = true;
2730 }
2731 } else if (rdev->family <= CHIP_RV280) {
2732 tmp = RREG32(RADEON_MEM_CNTL);
2733 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2734 rdev->mc.vram_width = 128;
2735 } else {
2736 rdev->mc.vram_width = 64;
2737 }
2738 } else {
2739 /* newer IGPs */
2740 rdev->mc.vram_width = 128;
2741 }
2742 }
2743
r100_get_accessible_vram(struct radeon_device * rdev)2744 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2745 {
2746 u32 aper_size;
2747 u8 byte;
2748
2749 aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2750
2751 /* Set HDP_APER_CNTL only on cards that are known not to be broken,
2752 * that is has the 2nd generation multifunction PCI interface
2753 */
2754 if (rdev->family == CHIP_RV280 ||
2755 rdev->family >= CHIP_RV350) {
2756 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2757 ~RADEON_HDP_APER_CNTL);
2758 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2759 return aper_size * 2;
2760 }
2761
2762 /* Older cards have all sorts of funny issues to deal with. First
2763 * check if it's a multifunction card by reading the PCI config
2764 * header type... Limit those to one aperture size
2765 */
2766 pci_read_config_byte(rdev->pdev, 0xe, &byte);
2767 if (byte & 0x80) {
2768 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2769 DRM_INFO("Limiting VRAM to one aperture\n");
2770 return aper_size;
2771 }
2772
2773 /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2774 * have set it up. We don't write this as it's broken on some ASICs but
2775 * we expect the BIOS to have done the right thing (might be too optimistic...)
2776 */
2777 if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2778 return aper_size * 2;
2779 return aper_size;
2780 }
2781
r100_vram_init_sizes(struct radeon_device * rdev)2782 void r100_vram_init_sizes(struct radeon_device *rdev)
2783 {
2784 u64 config_aper_size;
2785
2786 /* work out accessible VRAM */
2787 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2788 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2789 rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2790 /* FIXME we don't use the second aperture yet when we could use it */
2791 if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2792 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2793 config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2794 if (rdev->flags & RADEON_IS_IGP) {
2795 uint32_t tom;
2796 /* read NB_TOM to get the amount of ram stolen for the GPU */
2797 tom = RREG32(RADEON_NB_TOM);
2798 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2799 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2800 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2801 } else {
2802 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2803 /* Some production boards of m6 will report 0
2804 * if it's 8 MB
2805 */
2806 if (rdev->mc.real_vram_size == 0) {
2807 rdev->mc.real_vram_size = 8192 * 1024;
2808 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2809 }
2810 /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2811 * Novell bug 204882 + along with lots of ubuntu ones
2812 */
2813 if (rdev->mc.aper_size > config_aper_size)
2814 config_aper_size = rdev->mc.aper_size;
2815
2816 if (config_aper_size > rdev->mc.real_vram_size)
2817 rdev->mc.mc_vram_size = config_aper_size;
2818 else
2819 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2820 }
2821 }
2822
r100_vga_set_state(struct radeon_device * rdev,bool state)2823 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2824 {
2825 uint32_t temp;
2826
2827 temp = RREG32(RADEON_CONFIG_CNTL);
2828 if (state == false) {
2829 temp &= ~RADEON_CFG_VGA_RAM_EN;
2830 temp |= RADEON_CFG_VGA_IO_DIS;
2831 } else {
2832 temp &= ~RADEON_CFG_VGA_IO_DIS;
2833 }
2834 WREG32(RADEON_CONFIG_CNTL, temp);
2835 }
2836
r100_mc_init(struct radeon_device * rdev)2837 static void r100_mc_init(struct radeon_device *rdev)
2838 {
2839 u64 base;
2840
2841 r100_vram_get_type(rdev);
2842 r100_vram_init_sizes(rdev);
2843 base = rdev->mc.aper_base;
2844 if (rdev->flags & RADEON_IS_IGP)
2845 base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2846 radeon_vram_location(rdev, &rdev->mc, base);
2847 rdev->mc.gtt_base_align = 0;
2848 if (!(rdev->flags & RADEON_IS_AGP))
2849 radeon_gtt_location(rdev, &rdev->mc);
2850 radeon_update_bandwidth_info(rdev);
2851 }
2852
2853
2854 /*
2855 * Indirect registers accessor
2856 */
r100_pll_errata_after_index(struct radeon_device * rdev)2857 void r100_pll_errata_after_index(struct radeon_device *rdev)
2858 {
2859 if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2860 (void)RREG32(RADEON_CLOCK_CNTL_DATA);
2861 (void)RREG32(RADEON_CRTC_GEN_CNTL);
2862 }
2863 }
2864
r100_pll_errata_after_data(struct radeon_device * rdev)2865 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2866 {
2867 /* This workarounds is necessary on RV100, RS100 and RS200 chips
2868 * or the chip could hang on a subsequent access
2869 */
2870 if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2871 mdelay(5);
2872 }
2873
2874 /* This function is required to workaround a hardware bug in some (all?)
2875 * revisions of the R300. This workaround should be called after every
2876 * CLOCK_CNTL_INDEX register access. If not, register reads afterward
2877 * may not be correct.
2878 */
2879 if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2880 uint32_t save, tmp;
2881
2882 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2883 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2884 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2885 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2886 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2887 }
2888 }
2889
r100_pll_rreg(struct radeon_device * rdev,uint32_t reg)2890 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2891 {
2892 unsigned long flags;
2893 uint32_t data;
2894
2895 spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2896 WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2897 r100_pll_errata_after_index(rdev);
2898 data = RREG32(RADEON_CLOCK_CNTL_DATA);
2899 r100_pll_errata_after_data(rdev);
2900 spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2901 return data;
2902 }
2903
r100_pll_wreg(struct radeon_device * rdev,uint32_t reg,uint32_t v)2904 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2905 {
2906 unsigned long flags;
2907
2908 spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2909 WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2910 r100_pll_errata_after_index(rdev);
2911 WREG32(RADEON_CLOCK_CNTL_DATA, v);
2912 r100_pll_errata_after_data(rdev);
2913 spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2914 }
2915
r100_set_safe_registers(struct radeon_device * rdev)2916 static void r100_set_safe_registers(struct radeon_device *rdev)
2917 {
2918 if (ASIC_IS_RN50(rdev)) {
2919 rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2920 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2921 } else if (rdev->family < CHIP_R200) {
2922 rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2923 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2924 } else {
2925 r200_set_safe_registers(rdev);
2926 }
2927 }
2928
2929 /*
2930 * Debugfs info
2931 */
2932 #if defined(CONFIG_DEBUG_FS)
r100_debugfs_rbbm_info(struct seq_file * m,void * data)2933 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2934 {
2935 struct drm_info_node *node = (struct drm_info_node *) m->private;
2936 struct drm_device *dev = node->minor->dev;
2937 struct radeon_device *rdev = dev->dev_private;
2938 uint32_t reg, value;
2939 unsigned i;
2940
2941 seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2942 seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2943 seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2944 for (i = 0; i < 64; i++) {
2945 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2946 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2947 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2948 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2949 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2950 }
2951 return 0;
2952 }
2953
r100_debugfs_cp_ring_info(struct seq_file * m,void * data)2954 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2955 {
2956 struct drm_info_node *node = (struct drm_info_node *) m->private;
2957 struct drm_device *dev = node->minor->dev;
2958 struct radeon_device *rdev = dev->dev_private;
2959 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2960 uint32_t rdp, wdp;
2961 unsigned count, i, j;
2962
2963 radeon_ring_free_size(rdev, ring);
2964 rdp = RREG32(RADEON_CP_RB_RPTR);
2965 wdp = RREG32(RADEON_CP_RB_WPTR);
2966 count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2967 seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2968 seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2969 seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2970 seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2971 seq_printf(m, "%u dwords in ring\n", count);
2972 if (ring->ready) {
2973 for (j = 0; j <= count; j++) {
2974 i = (rdp + j) & ring->ptr_mask;
2975 seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2976 }
2977 }
2978 return 0;
2979 }
2980
2981
r100_debugfs_cp_csq_fifo(struct seq_file * m,void * data)2982 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2983 {
2984 struct drm_info_node *node = (struct drm_info_node *) m->private;
2985 struct drm_device *dev = node->minor->dev;
2986 struct radeon_device *rdev = dev->dev_private;
2987 uint32_t csq_stat, csq2_stat, tmp;
2988 unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2989 unsigned i;
2990
2991 seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2992 seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2993 csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2994 csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2995 r_rptr = (csq_stat >> 0) & 0x3ff;
2996 r_wptr = (csq_stat >> 10) & 0x3ff;
2997 ib1_rptr = (csq_stat >> 20) & 0x3ff;
2998 ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2999 ib2_rptr = (csq2_stat >> 10) & 0x3ff;
3000 ib2_wptr = (csq2_stat >> 20) & 0x3ff;
3001 seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
3002 seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
3003 seq_printf(m, "Ring rptr %u\n", r_rptr);
3004 seq_printf(m, "Ring wptr %u\n", r_wptr);
3005 seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
3006 seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
3007 seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
3008 seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
3009 /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
3010 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
3011 seq_printf(m, "Ring fifo:\n");
3012 for (i = 0; i < 256; i++) {
3013 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3014 tmp = RREG32(RADEON_CP_CSQ_DATA);
3015 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3016 }
3017 seq_printf(m, "Indirect1 fifo:\n");
3018 for (i = 256; i <= 512; i++) {
3019 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3020 tmp = RREG32(RADEON_CP_CSQ_DATA);
3021 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3022 }
3023 seq_printf(m, "Indirect2 fifo:\n");
3024 for (i = 640; i < ib1_wptr; i++) {
3025 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3026 tmp = RREG32(RADEON_CP_CSQ_DATA);
3027 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3028 }
3029 return 0;
3030 }
3031
r100_debugfs_mc_info(struct seq_file * m,void * data)3032 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
3033 {
3034 struct drm_info_node *node = (struct drm_info_node *) m->private;
3035 struct drm_device *dev = node->minor->dev;
3036 struct radeon_device *rdev = dev->dev_private;
3037 uint32_t tmp;
3038
3039 tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3040 seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3041 tmp = RREG32(RADEON_MC_FB_LOCATION);
3042 seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3043 tmp = RREG32(RADEON_BUS_CNTL);
3044 seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3045 tmp = RREG32(RADEON_MC_AGP_LOCATION);
3046 seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3047 tmp = RREG32(RADEON_AGP_BASE);
3048 seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3049 tmp = RREG32(RADEON_HOST_PATH_CNTL);
3050 seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3051 tmp = RREG32(0x01D0);
3052 seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3053 tmp = RREG32(RADEON_AIC_LO_ADDR);
3054 seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3055 tmp = RREG32(RADEON_AIC_HI_ADDR);
3056 seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3057 tmp = RREG32(0x01E4);
3058 seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3059 return 0;
3060 }
3061
3062 static struct drm_info_list r100_debugfs_rbbm_list[] = {
3063 {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3064 };
3065
3066 static struct drm_info_list r100_debugfs_cp_list[] = {
3067 {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3068 {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3069 };
3070
3071 static struct drm_info_list r100_debugfs_mc_info_list[] = {
3072 {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3073 };
3074 #endif
3075
r100_debugfs_rbbm_init(struct radeon_device * rdev)3076 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3077 {
3078 #if defined(CONFIG_DEBUG_FS)
3079 return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3080 #else
3081 return 0;
3082 #endif
3083 }
3084
r100_debugfs_cp_init(struct radeon_device * rdev)3085 int r100_debugfs_cp_init(struct radeon_device *rdev)
3086 {
3087 #if defined(CONFIG_DEBUG_FS)
3088 return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3089 #else
3090 return 0;
3091 #endif
3092 }
3093
r100_debugfs_mc_info_init(struct radeon_device * rdev)3094 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3095 {
3096 #if defined(CONFIG_DEBUG_FS)
3097 return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3098 #else
3099 return 0;
3100 #endif
3101 }
3102
r100_set_surface_reg(struct radeon_device * rdev,int reg,uint32_t tiling_flags,uint32_t pitch,uint32_t offset,uint32_t obj_size)3103 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3104 uint32_t tiling_flags, uint32_t pitch,
3105 uint32_t offset, uint32_t obj_size)
3106 {
3107 int surf_index = reg * 16;
3108 int flags = 0;
3109
3110 if (rdev->family <= CHIP_RS200) {
3111 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3112 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3113 flags |= RADEON_SURF_TILE_COLOR_BOTH;
3114 if (tiling_flags & RADEON_TILING_MACRO)
3115 flags |= RADEON_SURF_TILE_COLOR_MACRO;
3116 /* setting pitch to 0 disables tiling */
3117 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3118 == 0)
3119 pitch = 0;
3120 } else if (rdev->family <= CHIP_RV280) {
3121 if (tiling_flags & (RADEON_TILING_MACRO))
3122 flags |= R200_SURF_TILE_COLOR_MACRO;
3123 if (tiling_flags & RADEON_TILING_MICRO)
3124 flags |= R200_SURF_TILE_COLOR_MICRO;
3125 } else {
3126 if (tiling_flags & RADEON_TILING_MACRO)
3127 flags |= R300_SURF_TILE_MACRO;
3128 if (tiling_flags & RADEON_TILING_MICRO)
3129 flags |= R300_SURF_TILE_MICRO;
3130 }
3131
3132 if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3133 flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3134 if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3135 flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3136
3137 /* r100/r200 divide by 16 */
3138 if (rdev->family < CHIP_R300)
3139 flags |= pitch / 16;
3140 else
3141 flags |= pitch / 8;
3142
3143
3144 DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3145 WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3146 WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3147 WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3148 return 0;
3149 }
3150
r100_clear_surface_reg(struct radeon_device * rdev,int reg)3151 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3152 {
3153 int surf_index = reg * 16;
3154 WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3155 }
3156
r100_bandwidth_update(struct radeon_device * rdev)3157 void r100_bandwidth_update(struct radeon_device *rdev)
3158 {
3159 fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3160 fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3161 fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
3162 fixed20_12 crit_point_ff = {0};
3163 uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3164 fixed20_12 memtcas_ff[8] = {
3165 dfixed_init(1),
3166 dfixed_init(2),
3167 dfixed_init(3),
3168 dfixed_init(0),
3169 dfixed_init_half(1),
3170 dfixed_init_half(2),
3171 dfixed_init(0),
3172 };
3173 fixed20_12 memtcas_rs480_ff[8] = {
3174 dfixed_init(0),
3175 dfixed_init(1),
3176 dfixed_init(2),
3177 dfixed_init(3),
3178 dfixed_init(0),
3179 dfixed_init_half(1),
3180 dfixed_init_half(2),
3181 dfixed_init_half(3),
3182 };
3183 fixed20_12 memtcas2_ff[8] = {
3184 dfixed_init(0),
3185 dfixed_init(1),
3186 dfixed_init(2),
3187 dfixed_init(3),
3188 dfixed_init(4),
3189 dfixed_init(5),
3190 dfixed_init(6),
3191 dfixed_init(7),
3192 };
3193 fixed20_12 memtrbs[8] = {
3194 dfixed_init(1),
3195 dfixed_init_half(1),
3196 dfixed_init(2),
3197 dfixed_init_half(2),
3198 dfixed_init(3),
3199 dfixed_init_half(3),
3200 dfixed_init(4),
3201 dfixed_init_half(4)
3202 };
3203 fixed20_12 memtrbs_r4xx[8] = {
3204 dfixed_init(4),
3205 dfixed_init(5),
3206 dfixed_init(6),
3207 dfixed_init(7),
3208 dfixed_init(8),
3209 dfixed_init(9),
3210 dfixed_init(10),
3211 dfixed_init(11)
3212 };
3213 fixed20_12 min_mem_eff;
3214 fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3215 fixed20_12 cur_latency_mclk, cur_latency_sclk;
3216 fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate = {0},
3217 disp_drain_rate2, read_return_rate;
3218 fixed20_12 time_disp1_drop_priority;
3219 int c;
3220 int cur_size = 16; /* in octawords */
3221 int critical_point = 0, critical_point2;
3222 /* uint32_t read_return_rate, time_disp1_drop_priority; */
3223 int stop_req, max_stop_req;
3224 struct drm_display_mode *mode1 = NULL;
3225 struct drm_display_mode *mode2 = NULL;
3226 uint32_t pixel_bytes1 = 0;
3227 uint32_t pixel_bytes2 = 0;
3228
3229 /* Guess line buffer size to be 8192 pixels */
3230 u32 lb_size = 8192;
3231
3232 if (!rdev->mode_info.mode_config_initialized)
3233 return;
3234
3235 radeon_update_display_priority(rdev);
3236
3237 if (rdev->mode_info.crtcs[0]->base.enabled) {
3238 const struct drm_framebuffer *fb =
3239 rdev->mode_info.crtcs[0]->base.primary->fb;
3240
3241 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3242 pixel_bytes1 = fb->format->cpp[0];
3243 }
3244 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3245 if (rdev->mode_info.crtcs[1]->base.enabled) {
3246 const struct drm_framebuffer *fb =
3247 rdev->mode_info.crtcs[1]->base.primary->fb;
3248
3249 mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3250 pixel_bytes2 = fb->format->cpp[0];
3251 }
3252 }
3253
3254 min_mem_eff.full = dfixed_const_8(0);
3255 /* get modes */
3256 if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3257 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3258 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3259 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3260 /* check crtc enables */
3261 if (mode2)
3262 mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3263 if (mode1)
3264 mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3265 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3266 }
3267
3268 /*
3269 * determine is there is enough bw for current mode
3270 */
3271 sclk_ff = rdev->pm.sclk;
3272 mclk_ff = rdev->pm.mclk;
3273
3274 temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3275 temp_ff.full = dfixed_const(temp);
3276 mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3277
3278 pix_clk.full = 0;
3279 pix_clk2.full = 0;
3280 peak_disp_bw.full = 0;
3281 if (mode1) {
3282 temp_ff.full = dfixed_const(1000);
3283 pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3284 pix_clk.full = dfixed_div(pix_clk, temp_ff);
3285 temp_ff.full = dfixed_const(pixel_bytes1);
3286 peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3287 }
3288 if (mode2) {
3289 temp_ff.full = dfixed_const(1000);
3290 pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3291 pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3292 temp_ff.full = dfixed_const(pixel_bytes2);
3293 peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3294 }
3295
3296 mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3297 if (peak_disp_bw.full >= mem_bw.full) {
3298 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3299 "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3300 }
3301
3302 /* Get values from the EXT_MEM_CNTL register...converting its contents. */
3303 temp = RREG32(RADEON_MEM_TIMING_CNTL);
3304 if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3305 mem_trcd = ((temp >> 2) & 0x3) + 1;
3306 mem_trp = ((temp & 0x3)) + 1;
3307 mem_tras = ((temp & 0x70) >> 4) + 1;
3308 } else if (rdev->family == CHIP_R300 ||
3309 rdev->family == CHIP_R350) { /* r300, r350 */
3310 mem_trcd = (temp & 0x7) + 1;
3311 mem_trp = ((temp >> 8) & 0x7) + 1;
3312 mem_tras = ((temp >> 11) & 0xf) + 4;
3313 } else if (rdev->family == CHIP_RV350 ||
3314 rdev->family == CHIP_RV380) {
3315 /* rv3x0 */
3316 mem_trcd = (temp & 0x7) + 3;
3317 mem_trp = ((temp >> 8) & 0x7) + 3;
3318 mem_tras = ((temp >> 11) & 0xf) + 6;
3319 } else if (rdev->family == CHIP_R420 ||
3320 rdev->family == CHIP_R423 ||
3321 rdev->family == CHIP_RV410) {
3322 /* r4xx */
3323 mem_trcd = (temp & 0xf) + 3;
3324 if (mem_trcd > 15)
3325 mem_trcd = 15;
3326 mem_trp = ((temp >> 8) & 0xf) + 3;
3327 if (mem_trp > 15)
3328 mem_trp = 15;
3329 mem_tras = ((temp >> 12) & 0x1f) + 6;
3330 if (mem_tras > 31)
3331 mem_tras = 31;
3332 } else { /* RV200, R200 */
3333 mem_trcd = (temp & 0x7) + 1;
3334 mem_trp = ((temp >> 8) & 0x7) + 1;
3335 mem_tras = ((temp >> 12) & 0xf) + 4;
3336 }
3337 /* convert to FF */
3338 trcd_ff.full = dfixed_const(mem_trcd);
3339 trp_ff.full = dfixed_const(mem_trp);
3340 tras_ff.full = dfixed_const(mem_tras);
3341
3342 /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3343 temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3344 data = (temp & (7 << 20)) >> 20;
3345 if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3346 if (rdev->family == CHIP_RS480) /* don't think rs400 */
3347 tcas_ff = memtcas_rs480_ff[data];
3348 else
3349 tcas_ff = memtcas_ff[data];
3350 } else
3351 tcas_ff = memtcas2_ff[data];
3352
3353 if (rdev->family == CHIP_RS400 ||
3354 rdev->family == CHIP_RS480) {
3355 /* extra cas latency stored in bits 23-25 0-4 clocks */
3356 data = (temp >> 23) & 0x7;
3357 if (data < 5)
3358 tcas_ff.full += dfixed_const(data);
3359 }
3360
3361 if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3362 /* on the R300, Tcas is included in Trbs.
3363 */
3364 temp = RREG32(RADEON_MEM_CNTL);
3365 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3366 if (data == 1) {
3367 if (R300_MEM_USE_CD_CH_ONLY & temp) {
3368 temp = RREG32(R300_MC_IND_INDEX);
3369 temp &= ~R300_MC_IND_ADDR_MASK;
3370 temp |= R300_MC_READ_CNTL_CD_mcind;
3371 WREG32(R300_MC_IND_INDEX, temp);
3372 temp = RREG32(R300_MC_IND_DATA);
3373 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3374 } else {
3375 temp = RREG32(R300_MC_READ_CNTL_AB);
3376 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3377 }
3378 } else {
3379 temp = RREG32(R300_MC_READ_CNTL_AB);
3380 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3381 }
3382 if (rdev->family == CHIP_RV410 ||
3383 rdev->family == CHIP_R420 ||
3384 rdev->family == CHIP_R423)
3385 trbs_ff = memtrbs_r4xx[data];
3386 else
3387 trbs_ff = memtrbs[data];
3388 tcas_ff.full += trbs_ff.full;
3389 }
3390
3391 sclk_eff_ff.full = sclk_ff.full;
3392
3393 if (rdev->flags & RADEON_IS_AGP) {
3394 fixed20_12 agpmode_ff;
3395 agpmode_ff.full = dfixed_const(radeon_agpmode);
3396 temp_ff.full = dfixed_const_666(16);
3397 sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3398 }
3399 /* TODO PCIE lanes may affect this - agpmode == 16?? */
3400
3401 if (ASIC_IS_R300(rdev)) {
3402 sclk_delay_ff.full = dfixed_const(250);
3403 } else {
3404 if ((rdev->family == CHIP_RV100) ||
3405 rdev->flags & RADEON_IS_IGP) {
3406 if (rdev->mc.vram_is_ddr)
3407 sclk_delay_ff.full = dfixed_const(41);
3408 else
3409 sclk_delay_ff.full = dfixed_const(33);
3410 } else {
3411 if (rdev->mc.vram_width == 128)
3412 sclk_delay_ff.full = dfixed_const(57);
3413 else
3414 sclk_delay_ff.full = dfixed_const(41);
3415 }
3416 }
3417
3418 mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3419
3420 if (rdev->mc.vram_is_ddr) {
3421 if (rdev->mc.vram_width == 32) {
3422 k1.full = dfixed_const(40);
3423 c = 3;
3424 } else {
3425 k1.full = dfixed_const(20);
3426 c = 1;
3427 }
3428 } else {
3429 k1.full = dfixed_const(40);
3430 c = 3;
3431 }
3432
3433 temp_ff.full = dfixed_const(2);
3434 mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3435 temp_ff.full = dfixed_const(c);
3436 mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3437 temp_ff.full = dfixed_const(4);
3438 mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3439 mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3440 mc_latency_mclk.full += k1.full;
3441
3442 mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3443 mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3444
3445 /*
3446 HW cursor time assuming worst case of full size colour cursor.
3447 */
3448 temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3449 temp_ff.full += trcd_ff.full;
3450 if (temp_ff.full < tras_ff.full)
3451 temp_ff.full = tras_ff.full;
3452 cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3453
3454 temp_ff.full = dfixed_const(cur_size);
3455 cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3456 /*
3457 Find the total latency for the display data.
3458 */
3459 disp_latency_overhead.full = dfixed_const(8);
3460 disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3461 mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3462 mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3463
3464 if (mc_latency_mclk.full > mc_latency_sclk.full)
3465 disp_latency.full = mc_latency_mclk.full;
3466 else
3467 disp_latency.full = mc_latency_sclk.full;
3468
3469 /* setup Max GRPH_STOP_REQ default value */
3470 if (ASIC_IS_RV100(rdev))
3471 max_stop_req = 0x5c;
3472 else
3473 max_stop_req = 0x7c;
3474
3475 if (mode1) {
3476 /* CRTC1
3477 Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3478 GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3479 */
3480 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3481
3482 if (stop_req > max_stop_req)
3483 stop_req = max_stop_req;
3484
3485 /*
3486 Find the drain rate of the display buffer.
3487 */
3488 temp_ff.full = dfixed_const((16/pixel_bytes1));
3489 disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3490
3491 /*
3492 Find the critical point of the display buffer.
3493 */
3494 crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3495 crit_point_ff.full += dfixed_const_half(0);
3496
3497 critical_point = dfixed_trunc(crit_point_ff);
3498
3499 if (rdev->disp_priority == 2) {
3500 critical_point = 0;
3501 }
3502
3503 /*
3504 The critical point should never be above max_stop_req-4. Setting
3505 GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3506 */
3507 if (max_stop_req - critical_point < 4)
3508 critical_point = 0;
3509
3510 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3511 /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3512 critical_point = 0x10;
3513 }
3514
3515 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3516 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3517 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3518 temp &= ~(RADEON_GRPH_START_REQ_MASK);
3519 if ((rdev->family == CHIP_R350) &&
3520 (stop_req > 0x15)) {
3521 stop_req -= 0x10;
3522 }
3523 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3524 temp |= RADEON_GRPH_BUFFER_SIZE;
3525 temp &= ~(RADEON_GRPH_CRITICAL_CNTL |
3526 RADEON_GRPH_CRITICAL_AT_SOF |
3527 RADEON_GRPH_STOP_CNTL);
3528 /*
3529 Write the result into the register.
3530 */
3531 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3532 (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3533
3534 #if 0
3535 if ((rdev->family == CHIP_RS400) ||
3536 (rdev->family == CHIP_RS480)) {
3537 /* attempt to program RS400 disp regs correctly ??? */
3538 temp = RREG32(RS400_DISP1_REG_CNTL);
3539 temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3540 RS400_DISP1_STOP_REQ_LEVEL_MASK);
3541 WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3542 (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3543 (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3544 temp = RREG32(RS400_DMIF_MEM_CNTL1);
3545 temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3546 RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3547 WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3548 (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3549 (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3550 }
3551 #endif
3552
3553 DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3554 /* (unsigned int)info->SavedReg->grph_buffer_cntl, */
3555 (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3556 }
3557
3558 if (mode2) {
3559 u32 grph2_cntl;
3560 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3561
3562 if (stop_req > max_stop_req)
3563 stop_req = max_stop_req;
3564
3565 /*
3566 Find the drain rate of the display buffer.
3567 */
3568 temp_ff.full = dfixed_const((16/pixel_bytes2));
3569 disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3570
3571 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3572 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3573 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3574 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3575 if ((rdev->family == CHIP_R350) &&
3576 (stop_req > 0x15)) {
3577 stop_req -= 0x10;
3578 }
3579 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3580 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3581 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL |
3582 RADEON_GRPH_CRITICAL_AT_SOF |
3583 RADEON_GRPH_STOP_CNTL);
3584
3585 if ((rdev->family == CHIP_RS100) ||
3586 (rdev->family == CHIP_RS200))
3587 critical_point2 = 0;
3588 else {
3589 temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3590 temp_ff.full = dfixed_const(temp);
3591 temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3592 if (sclk_ff.full < temp_ff.full)
3593 temp_ff.full = sclk_ff.full;
3594
3595 read_return_rate.full = temp_ff.full;
3596
3597 if (mode1) {
3598 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3599 time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3600 } else {
3601 time_disp1_drop_priority.full = 0;
3602 }
3603 crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3604 crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3605 crit_point_ff.full += dfixed_const_half(0);
3606
3607 critical_point2 = dfixed_trunc(crit_point_ff);
3608
3609 if (rdev->disp_priority == 2) {
3610 critical_point2 = 0;
3611 }
3612
3613 if (max_stop_req - critical_point2 < 4)
3614 critical_point2 = 0;
3615
3616 }
3617
3618 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3619 /* some R300 cards have problem with this set to 0 */
3620 critical_point2 = 0x10;
3621 }
3622
3623 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3624 (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3625
3626 if ((rdev->family == CHIP_RS400) ||
3627 (rdev->family == CHIP_RS480)) {
3628 #if 0
3629 /* attempt to program RS400 disp2 regs correctly ??? */
3630 temp = RREG32(RS400_DISP2_REQ_CNTL1);
3631 temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3632 RS400_DISP2_STOP_REQ_LEVEL_MASK);
3633 WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3634 (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3635 (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3636 temp = RREG32(RS400_DISP2_REQ_CNTL2);
3637 temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3638 RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3639 WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3640 (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3641 (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3642 #endif
3643 WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3644 WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3645 WREG32(RS400_DMIF_MEM_CNTL1, 0x29CA71DC);
3646 WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3647 }
3648
3649 DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3650 (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3651 }
3652
3653 /* Save number of lines the linebuffer leads before the scanout */
3654 if (mode1)
3655 rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
3656
3657 if (mode2)
3658 rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
3659 }
3660
r100_ring_test(struct radeon_device * rdev,struct radeon_ring * ring)3661 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3662 {
3663 uint32_t scratch;
3664 uint32_t tmp = 0;
3665 unsigned i;
3666 int r;
3667
3668 r = radeon_scratch_get(rdev, &scratch);
3669 if (r) {
3670 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3671 return r;
3672 }
3673 WREG32(scratch, 0xCAFEDEAD);
3674 r = radeon_ring_lock(rdev, ring, 2);
3675 if (r) {
3676 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3677 radeon_scratch_free(rdev, scratch);
3678 return r;
3679 }
3680 radeon_ring_write(ring, PACKET0(scratch, 0));
3681 radeon_ring_write(ring, 0xDEADBEEF);
3682 radeon_ring_unlock_commit(rdev, ring, false);
3683 for (i = 0; i < rdev->usec_timeout; i++) {
3684 tmp = RREG32(scratch);
3685 if (tmp == 0xDEADBEEF) {
3686 break;
3687 }
3688 DRM_UDELAY(1);
3689 }
3690 if (i < rdev->usec_timeout) {
3691 DRM_INFO("ring test succeeded in %d usecs\n", i);
3692 } else {
3693 DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3694 scratch, tmp);
3695 r = -EINVAL;
3696 }
3697 radeon_scratch_free(rdev, scratch);
3698 return r;
3699 }
3700
r100_ring_ib_execute(struct radeon_device * rdev,struct radeon_ib * ib)3701 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3702 {
3703 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3704
3705 if (ring->rptr_save_reg) {
3706 u32 next_rptr = ring->wptr + 2 + 3;
3707 radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3708 radeon_ring_write(ring, next_rptr);
3709 }
3710
3711 radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3712 radeon_ring_write(ring, ib->gpu_addr);
3713 radeon_ring_write(ring, ib->length_dw);
3714 }
3715
r100_ib_test(struct radeon_device * rdev,struct radeon_ring * ring)3716 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3717 {
3718 struct radeon_ib ib;
3719 uint32_t scratch;
3720 uint32_t tmp = 0;
3721 unsigned i;
3722 int r;
3723
3724 r = radeon_scratch_get(rdev, &scratch);
3725 if (r) {
3726 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3727 return r;
3728 }
3729 WREG32(scratch, 0xCAFEDEAD);
3730 r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3731 if (r) {
3732 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3733 goto free_scratch;
3734 }
3735 ib.ptr[0] = PACKET0(scratch, 0);
3736 ib.ptr[1] = 0xDEADBEEF;
3737 ib.ptr[2] = PACKET2(0);
3738 ib.ptr[3] = PACKET2(0);
3739 ib.ptr[4] = PACKET2(0);
3740 ib.ptr[5] = PACKET2(0);
3741 ib.ptr[6] = PACKET2(0);
3742 ib.ptr[7] = PACKET2(0);
3743 ib.length_dw = 8;
3744 r = radeon_ib_schedule(rdev, &ib, NULL, false);
3745 if (r) {
3746 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3747 goto free_ib;
3748 }
3749 r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
3750 RADEON_USEC_IB_TEST_TIMEOUT));
3751 if (r < 0) {
3752 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3753 goto free_ib;
3754 } else if (r == 0) {
3755 DRM_ERROR("radeon: fence wait timed out.\n");
3756 r = -ETIMEDOUT;
3757 goto free_ib;
3758 }
3759 r = 0;
3760 for (i = 0; i < rdev->usec_timeout; i++) {
3761 tmp = RREG32(scratch);
3762 if (tmp == 0xDEADBEEF) {
3763 break;
3764 }
3765 DRM_UDELAY(1);
3766 }
3767 if (i < rdev->usec_timeout) {
3768 DRM_INFO("ib test succeeded in %u usecs\n", i);
3769 } else {
3770 DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3771 scratch, tmp);
3772 r = -EINVAL;
3773 }
3774 free_ib:
3775 radeon_ib_free(rdev, &ib);
3776 free_scratch:
3777 radeon_scratch_free(rdev, scratch);
3778 return r;
3779 }
3780
r100_mc_stop(struct radeon_device * rdev,struct r100_mc_save * save)3781 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3782 {
3783 /* Shutdown CP we shouldn't need to do that but better be safe than
3784 * sorry
3785 */
3786 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3787 WREG32(R_000740_CP_CSQ_CNTL, 0);
3788
3789 /* Save few CRTC registers */
3790 save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3791 save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3792 save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3793 save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3794 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3795 save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3796 save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3797 }
3798
3799 /* Disable VGA aperture access */
3800 WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3801 /* Disable cursor, overlay, crtc */
3802 WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3803 WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3804 S_000054_CRTC_DISPLAY_DIS(1));
3805 WREG32(R_000050_CRTC_GEN_CNTL,
3806 (C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3807 S_000050_CRTC_DISP_REQ_EN_B(1));
3808 WREG32(R_000420_OV0_SCALE_CNTL,
3809 C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3810 WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3811 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3812 WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3813 S_000360_CUR2_LOCK(1));
3814 WREG32(R_0003F8_CRTC2_GEN_CNTL,
3815 (C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3816 S_0003F8_CRTC2_DISPLAY_DIS(1) |
3817 S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3818 WREG32(R_000360_CUR2_OFFSET,
3819 C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3820 }
3821 }
3822
r100_mc_resume(struct radeon_device * rdev,struct r100_mc_save * save)3823 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3824 {
3825 /* Update base address for crtc */
3826 WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3827 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3828 WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3829 }
3830 /* Restore CRTC registers */
3831 WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3832 WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3833 WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3834 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3835 WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3836 }
3837 }
3838
r100_vga_render_disable(struct radeon_device * rdev)3839 void r100_vga_render_disable(struct radeon_device *rdev)
3840 {
3841 u32 tmp;
3842
3843 tmp = RREG8(R_0003C2_GENMO_WT);
3844 WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3845 }
3846
r100_debugfs(struct radeon_device * rdev)3847 static void r100_debugfs(struct radeon_device *rdev)
3848 {
3849 int r;
3850
3851 r = r100_debugfs_mc_info_init(rdev);
3852 if (r)
3853 dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3854 }
3855
r100_mc_program(struct radeon_device * rdev)3856 static void r100_mc_program(struct radeon_device *rdev)
3857 {
3858 struct r100_mc_save save;
3859
3860 /* Stops all mc clients */
3861 r100_mc_stop(rdev, &save);
3862 if (rdev->flags & RADEON_IS_AGP) {
3863 WREG32(R_00014C_MC_AGP_LOCATION,
3864 S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3865 S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3866 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3867 if (rdev->family > CHIP_RV200)
3868 WREG32(R_00015C_AGP_BASE_2,
3869 upper_32_bits(rdev->mc.agp_base) & 0xff);
3870 } else {
3871 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3872 WREG32(R_000170_AGP_BASE, 0);
3873 if (rdev->family > CHIP_RV200)
3874 WREG32(R_00015C_AGP_BASE_2, 0);
3875 }
3876 /* Wait for mc idle */
3877 if (r100_mc_wait_for_idle(rdev))
3878 dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3879 /* Program MC, should be a 32bits limited address space */
3880 WREG32(R_000148_MC_FB_LOCATION,
3881 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3882 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3883 r100_mc_resume(rdev, &save);
3884 }
3885
r100_clock_startup(struct radeon_device * rdev)3886 static void r100_clock_startup(struct radeon_device *rdev)
3887 {
3888 u32 tmp;
3889
3890 if (radeon_dynclks != -1 && radeon_dynclks)
3891 radeon_legacy_set_clock_gating(rdev, 1);
3892 /* We need to force on some of the block */
3893 tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3894 tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3895 if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3896 tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3897 WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3898 }
3899
r100_startup(struct radeon_device * rdev)3900 static int r100_startup(struct radeon_device *rdev)
3901 {
3902 int r;
3903
3904 /* set common regs */
3905 r100_set_common_regs(rdev);
3906 /* program mc */
3907 r100_mc_program(rdev);
3908 /* Resume clock */
3909 r100_clock_startup(rdev);
3910 /* Initialize GART (initialize after TTM so we can allocate
3911 * memory through TTM but finalize after TTM) */
3912 r100_enable_bm(rdev);
3913 if (rdev->flags & RADEON_IS_PCI) {
3914 r = r100_pci_gart_enable(rdev);
3915 if (r)
3916 return r;
3917 }
3918
3919 /* allocate wb buffer */
3920 r = radeon_wb_init(rdev);
3921 if (r)
3922 return r;
3923
3924 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3925 if (r) {
3926 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3927 return r;
3928 }
3929
3930 /* Enable IRQ */
3931 if (!rdev->irq.installed) {
3932 r = radeon_irq_kms_init(rdev);
3933 if (r)
3934 return r;
3935 }
3936
3937 r100_irq_set(rdev);
3938 rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3939 /* 1M ring buffer */
3940 r = r100_cp_init(rdev, 1024 * 1024);
3941 if (r) {
3942 dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3943 return r;
3944 }
3945
3946 r = radeon_ib_pool_init(rdev);
3947 if (r) {
3948 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3949 return r;
3950 }
3951
3952 return 0;
3953 }
3954
r100_resume(struct radeon_device * rdev)3955 int r100_resume(struct radeon_device *rdev)
3956 {
3957 int r;
3958
3959 /* Make sur GART are not working */
3960 if (rdev->flags & RADEON_IS_PCI)
3961 r100_pci_gart_disable(rdev);
3962 /* Resume clock before doing reset */
3963 r100_clock_startup(rdev);
3964 /* Reset gpu before posting otherwise ATOM will enter infinite loop */
3965 if (radeon_asic_reset(rdev)) {
3966 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3967 RREG32(R_000E40_RBBM_STATUS),
3968 RREG32(R_0007C0_CP_STAT));
3969 }
3970 /* post */
3971 radeon_combios_asic_init(rdev->ddev);
3972 /* Resume clock after posting */
3973 r100_clock_startup(rdev);
3974 /* Initialize surface registers */
3975 radeon_surface_init(rdev);
3976
3977 rdev->accel_working = true;
3978 r = r100_startup(rdev);
3979 if (r) {
3980 rdev->accel_working = false;
3981 }
3982 return r;
3983 }
3984
r100_suspend(struct radeon_device * rdev)3985 int r100_suspend(struct radeon_device *rdev)
3986 {
3987 radeon_pm_suspend(rdev);
3988 r100_cp_disable(rdev);
3989 radeon_wb_disable(rdev);
3990 r100_irq_disable(rdev);
3991 if (rdev->flags & RADEON_IS_PCI)
3992 r100_pci_gart_disable(rdev);
3993 return 0;
3994 }
3995
r100_fini(struct radeon_device * rdev)3996 void r100_fini(struct radeon_device *rdev)
3997 {
3998 radeon_pm_fini(rdev);
3999 r100_cp_fini(rdev);
4000 radeon_wb_fini(rdev);
4001 radeon_ib_pool_fini(rdev);
4002 radeon_gem_fini(rdev);
4003 if (rdev->flags & RADEON_IS_PCI)
4004 r100_pci_gart_fini(rdev);
4005 radeon_agp_fini(rdev);
4006 radeon_irq_kms_fini(rdev);
4007 radeon_fence_driver_fini(rdev);
4008 radeon_bo_fini(rdev);
4009 radeon_atombios_fini(rdev);
4010 r100_cp_fini_microcode(rdev);
4011 kfree(rdev->bios);
4012 rdev->bios = NULL;
4013 }
4014
4015 /*
4016 * Due to how kexec works, it can leave the hw fully initialised when it
4017 * boots the new kernel. However doing our init sequence with the CP and
4018 * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
4019 * do some quick sanity checks and restore sane values to avoid this
4020 * problem.
4021 */
r100_restore_sanity(struct radeon_device * rdev)4022 void r100_restore_sanity(struct radeon_device *rdev)
4023 {
4024 u32 tmp;
4025
4026 tmp = RREG32(RADEON_CP_CSQ_CNTL);
4027 if (tmp) {
4028 WREG32(RADEON_CP_CSQ_CNTL, 0);
4029 }
4030 tmp = RREG32(RADEON_CP_RB_CNTL);
4031 if (tmp) {
4032 WREG32(RADEON_CP_RB_CNTL, 0);
4033 }
4034 tmp = RREG32(RADEON_SCRATCH_UMSK);
4035 if (tmp) {
4036 WREG32(RADEON_SCRATCH_UMSK, 0);
4037 }
4038 }
4039
r100_init(struct radeon_device * rdev)4040 int r100_init(struct radeon_device *rdev)
4041 {
4042 int r;
4043
4044 /* Register debugfs file specific to this group of asics */
4045 r100_debugfs(rdev);
4046 /* Disable VGA */
4047 r100_vga_render_disable(rdev);
4048 /* Initialize scratch registers */
4049 radeon_scratch_init(rdev);
4050 /* Initialize surface registers */
4051 radeon_surface_init(rdev);
4052 /* sanity check some register to avoid hangs like after kexec */
4053 r100_restore_sanity(rdev);
4054 /* TODO: disable VGA need to use VGA request */
4055 /* BIOS*/
4056 if (!radeon_get_bios(rdev)) {
4057 if (ASIC_IS_AVIVO(rdev))
4058 return -EINVAL;
4059 }
4060 if (rdev->is_atom_bios) {
4061 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4062 return -EINVAL;
4063 } else {
4064 r = radeon_combios_init(rdev);
4065 if (r)
4066 return r;
4067 }
4068 /* Reset gpu before posting otherwise ATOM will enter infinite loop */
4069 if (radeon_asic_reset(rdev)) {
4070 dev_warn(rdev->dev,
4071 "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4072 RREG32(R_000E40_RBBM_STATUS),
4073 RREG32(R_0007C0_CP_STAT));
4074 }
4075 /* check if cards are posted or not */
4076 if (radeon_boot_test_post_card(rdev) == false)
4077 return -EINVAL;
4078 /* Set asic errata */
4079 r100_errata(rdev);
4080 /* Initialize clocks */
4081 radeon_get_clock_info(rdev->ddev);
4082 /* initialize AGP */
4083 if (rdev->flags & RADEON_IS_AGP) {
4084 r = radeon_agp_init(rdev);
4085 if (r) {
4086 radeon_agp_disable(rdev);
4087 }
4088 }
4089 /* initialize VRAM */
4090 r100_mc_init(rdev);
4091 /* Fence driver */
4092 r = radeon_fence_driver_init(rdev);
4093 if (r)
4094 return r;
4095 /* Memory manager */
4096 r = radeon_bo_init(rdev);
4097 if (r)
4098 return r;
4099 if (rdev->flags & RADEON_IS_PCI) {
4100 r = r100_pci_gart_init(rdev);
4101 if (r)
4102 return r;
4103 }
4104 r100_set_safe_registers(rdev);
4105
4106 /* Initialize power management */
4107 radeon_pm_init(rdev);
4108
4109 rdev->accel_working = true;
4110 r = r100_startup(rdev);
4111 if (r) {
4112 /* Somethings want wront with the accel init stop accel */
4113 dev_err(rdev->dev, "Disabling GPU acceleration\n");
4114 r100_cp_fini(rdev);
4115 radeon_wb_fini(rdev);
4116 radeon_ib_pool_fini(rdev);
4117 radeon_irq_kms_fini(rdev);
4118 if (rdev->flags & RADEON_IS_PCI)
4119 r100_pci_gart_fini(rdev);
4120 rdev->accel_working = false;
4121 }
4122 return 0;
4123 }
4124
r100_mm_rreg_slow(struct radeon_device * rdev,uint32_t reg)4125 uint32_t r100_mm_rreg_slow(struct radeon_device *rdev, uint32_t reg)
4126 {
4127 unsigned long flags;
4128 uint32_t ret;
4129
4130 spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4131 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4132 ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4133 spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4134 return ret;
4135 }
4136
r100_mm_wreg_slow(struct radeon_device * rdev,uint32_t reg,uint32_t v)4137 void r100_mm_wreg_slow(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4138 {
4139 unsigned long flags;
4140
4141 spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4142 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4143 writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4144 spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4145 }
4146
r100_io_rreg(struct radeon_device * rdev,u32 reg)4147 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4148 {
4149 if (reg < rdev->rio_mem_size)
4150 return ioread32(rdev->rio_mem + reg);
4151 else {
4152 iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4153 return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4154 }
4155 }
4156
r100_io_wreg(struct radeon_device * rdev,u32 reg,u32 v)4157 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4158 {
4159 if (reg < rdev->rio_mem_size)
4160 iowrite32(v, rdev->rio_mem + reg);
4161 else {
4162 iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4163 iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4164 }
4165 }
4166