drivers/gpu/drm/radeon/r100.c

   1 /*
   2  * Copyright 2008 Advanced Micro Devices, Inc.
   3  * Copyright 2008 Red Hat Inc.
   4  * Copyright 2009 Jerome Glisse.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  22  * OTHER DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors: Dave Airlie
  25  *          Alex Deucher
  26  *          Jerome Glisse
  27  */
  28 #include <linux/seq_file.h>
  29 #include <linux/slab.h>
  30 #include <drm/drmP.h>
  31 #include <drm/radeon_drm.h>
  32 #include "radeon_reg.h"
  33 #include "radeon.h"
  34 #include "radeon_asic.h"
  35 #include "r100d.h"
  36 #include "rs100d.h"
  37 #include "rv200d.h"
  38 #include "rv250d.h"
  39 #include "atom.h"
  40
  41 #include <linux/firmware.h>
  42 #include <linux/module.h>
  43
  44 #include "r100_reg_safe.h"
  45 #include "rn50_reg_safe.h"
  46
  47 /* Firmware Names */
  48 #define FIRMWARE_R100           "/*(DEBLOBBED)*/"
  49 #define FIRMWARE_R200           "/*(DEBLOBBED)*/"
  50 #define FIRMWARE_R300           "/*(DEBLOBBED)*/"
  51 #define FIRMWARE_R420           "/*(DEBLOBBED)*/"
  52 #define FIRMWARE_RS690          "/*(DEBLOBBED)*/"
  53 #define FIRMWARE_RS600          "/*(DEBLOBBED)*/"
  54 #define FIRMWARE_R520           "/*(DEBLOBBED)*/"
  55
  56 /*(DEBLOBBED)*/
  57
  58 #include "r100_track.h"
  59
  60 /* This files gather functions specifics to:
  61  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
  62  * and others in some cases.
  63  */
  64
  65 static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
  66 {
  67         if (crtc == 0) {
  68                 if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
  69                         return true;
  70                 else
  71                         return false;
  72         } else {
  73                 if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
  74                         return true;
  75                 else
  76                         return false;
  77         }
  78 }
  79
  80 static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
  81 {
  82         u32 vline1, vline2;
  83
  84         if (crtc == 0) {
  85                 vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
  86                 vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
  87         } else {
  88                 vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
  89                 vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
  90         }
  91         if (vline1 != vline2)
  92                 return true;
  93         else
  94                 return false;
  95 }
  96
  97 /**
  98  * r100_wait_for_vblank - vblank wait asic callback.
  99  *
 100  * @rdev: radeon_device pointer
 101  * @crtc: crtc to wait for vblank on
 102  *
 103  * Wait for vblank on the requested crtc (r1xx-r4xx).
 104  */
 105 void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
 106 {
 107         unsigned i = 0;
 108
 109         if (crtc >= rdev->num_crtc)
 110                 return;
 111
 112         if (crtc == 0) {
 113                 if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
 114                         return;
 115         } else {
 116                 if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
 117                         return;
 118         }
 119
 120         /* depending on when we hit vblank, we may be close to active; if so,
 121          * wait for another frame.
 122          */
 123         while (r100_is_in_vblank(rdev, crtc)) {
 124                 if (i++ % 100 == 0) {
 125                         if (!r100_is_counter_moving(rdev, crtc))
 126                                 break;
 127                 }
 128         }
 129
 130         while (!r100_is_in_vblank(rdev, crtc)) {
 131                 if (i++ % 100 == 0) {
 132                         if (!r100_is_counter_moving(rdev, crtc))
 133                                 break;
 134                 }
 135         }
 136 }
 137
 138 /**
 139  * r100_pre_page_flip - pre-pageflip callback.
 140  *
 141  * @rdev: radeon_device pointer
 142  * @crtc: crtc to prepare for pageflip on
 143  *
 144  * Pre-pageflip callback (r1xx-r4xx).
 145  * Enables the pageflip irq (vblank irq).
 146  */
 147 void r100_pre_page_flip(struct radeon_device *rdev, int crtc)
 148 {
 149         /* enable the pflip int */
 150         radeon_irq_kms_pflip_irq_get(rdev, crtc);
 151 }
 152
 153 /**
 154  * r100_post_page_flip - pos-pageflip callback.
 155  *
 156  * @rdev: radeon_device pointer
 157  * @crtc: crtc to cleanup pageflip on
 158  *
 159  * Post-pageflip callback (r1xx-r4xx).
 160  * Disables the pageflip irq (vblank irq).
 161  */
 162 void r100_post_page_flip(struct radeon_device *rdev, int crtc)
 163 {
 164         /* disable the pflip int */
 165         radeon_irq_kms_pflip_irq_put(rdev, crtc);
 166 }
 167
 168 /**
 169  * r100_page_flip - pageflip callback.
 170  *
 171  * @rdev: radeon_device pointer
 172  * @crtc_id: crtc to cleanup pageflip on
 173  * @crtc_base: new address of the crtc (GPU MC address)
 174  *
 175  * Does the actual pageflip (r1xx-r4xx).
 176  * During vblank we take the crtc lock and wait for the update_pending
 177  * bit to go high, when it does, we release the lock, and allow the
 178  * double buffered update to take place.
 179  * Returns the current update pending status.
 180  */
 181 u32 r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
 182 {
 183         struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
 184         u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
 185         int i;
 186
 187         /* Lock the graphics update lock */
 188         /* update the scanout addresses */
 189         WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 190
 191         /* Wait for update_pending to go high. */
 192         for (i = 0; i < rdev->usec_timeout; i++) {
 193                 if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
 194                         break;
 195                 udelay(1);
 196         }
 197         DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
 198
 199         /* Unlock the lock, so double-buffering can take place inside vblank */
 200         tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
 201         WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 202
 203         /* Return current update_pending status: */
 204         return RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET;
 205 }
 206
 207 /**
 208  * r100_pm_get_dynpm_state - look up dynpm power state callback.
 209  *
 210  * @rdev: radeon_device pointer
 211  *
 212  * Look up the optimal power state based on the
 213  * current state of the GPU (r1xx-r5xx).
 214  * Used for dynpm only.
 215  */
 216 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
 217 {
 218         int i;
 219         rdev->pm.dynpm_can_upclock = true;
 220         rdev->pm.dynpm_can_downclock = true;
 221
 222         switch (rdev->pm.dynpm_planned_action) {
 223         case DYNPM_ACTION_MINIMUM:
 224                 rdev->pm.requested_power_state_index = 0;
 225                 rdev->pm.dynpm_can_downclock = false;
 226                 break;
 227         case DYNPM_ACTION_DOWNCLOCK:
 228                 if (rdev->pm.current_power_state_index == 0) {
 229                         rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 230                         rdev->pm.dynpm_can_downclock = false;
 231                 } else {
 232                         if (rdev->pm.active_crtc_count > 1) {
 233                                 for (i = 0; i < rdev->pm.num_power_states; i++) {
 234                                         if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 235                                                 continue;
 236                                         else if (i >= rdev->pm.current_power_state_index) {
 237                                                 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 238                                                 break;
 239                                         } else {
 240                                                 rdev->pm.requested_power_state_index = i;
 241                                                 break;
 242                                         }
 243                                 }
 244                         } else
 245                                 rdev->pm.requested_power_state_index =
 246                                         rdev->pm.current_power_state_index - 1;
 247                 }
 248                 /* don't use the power state if crtcs are active and no display flag is set */
 249                 if ((rdev->pm.active_crtc_count > 0) &&
 250                     (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
 251                      RADEON_PM_MODE_NO_DISPLAY)) {
 252                         rdev->pm.requested_power_state_index++;
 253                 }
 254                 break;
 255         case DYNPM_ACTION_UPCLOCK:
 256                 if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
 257                         rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 258                         rdev->pm.dynpm_can_upclock = false;
 259                 } else {
 260                         if (rdev->pm.active_crtc_count > 1) {
 261                                 for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
 262                                         if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 263                                                 continue;
 264                                         else if (i <= rdev->pm.current_power_state_index) {
 265                                                 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 266                                                 break;
 267                                         } else {
 268                                                 rdev->pm.requested_power_state_index = i;
 269                                                 break;
 270                                         }
 271                                 }
 272                         } else
 273                                 rdev->pm.requested_power_state_index =
 274                                         rdev->pm.current_power_state_index + 1;
 275                 }
 276                 break;
 277         case DYNPM_ACTION_DEFAULT:
 278                 rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
 279                 rdev->pm.dynpm_can_upclock = false;
 280                 break;
 281         case DYNPM_ACTION_NONE:
 282         default:
 283                 DRM_ERROR("Requested mode for not defined action\n");
 284                 return;
 285         }
 286         /* only one clock mode per power state */
 287         rdev->pm.requested_clock_mode_index = 0;
 288
 289         DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
 290                   rdev->pm.power_state[rdev->pm.requested_power_state_index].
 291                   clock_info[rdev->pm.requested_clock_mode_index].sclk,
 292                   rdev->pm.power_state[rdev->pm.requested_power_state_index].
 293                   clock_info[rdev->pm.requested_clock_mode_index].mclk,
 294                   rdev->pm.power_state[rdev->pm.requested_power_state_index].
 295                   pcie_lanes);
 296 }
 297
 298 /**
 299  * r100_pm_init_profile - Initialize power profiles callback.
 300  *
 301  * @rdev: radeon_device pointer
 302  *
 303  * Initialize the power states used in profile mode
 304  * (r1xx-r3xx).
 305  * Used for profile mode only.
 306  */
 307 void r100_pm_init_profile(struct radeon_device *rdev)
 308 {
 309         /* default */
 310         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
 311         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 312         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
 313         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
 314         /* low sh */
 315         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
 316         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
 317         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
 318         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
 319         /* mid sh */
 320         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
 321         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
 322         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
 323         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
 324         /* high sh */
 325         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
 326         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 327         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
 328         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
 329         /* low mh */
 330         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
 331         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 332         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
 333         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
 334         /* mid mh */
 335         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
 336         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 337         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
 338         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
 339         /* high mh */
 340         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
 341         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 342         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
 343         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
 344 }
 345
 346 /**
 347  * r100_pm_misc - set additional pm hw parameters callback.
 348  *
 349  * @rdev: radeon_device pointer
 350  *
 351  * Set non-clock parameters associated with a power state
 352  * (voltage, pcie lanes, etc.) (r1xx-r4xx).
 353  */
 354 void r100_pm_misc(struct radeon_device *rdev)
 355 {
 356         int requested_index = rdev->pm.requested_power_state_index;
 357         struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
 358         struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
 359         u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
 360
 361         if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
 362                 if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
 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                 } else {
 372                         tmp = RREG32(voltage->gpio.reg);
 373                         if (voltage->active_high)
 374                                 tmp &= ~voltage->gpio.mask;
 375                         else
 376                                 tmp |= voltage->gpio.mask;
 377                         WREG32(voltage->gpio.reg, tmp);
 378                         if (voltage->delay)
 379                                 udelay(voltage->delay);
 380                 }
 381         }
 382
 383         sclk_cntl = RREG32_PLL(SCLK_CNTL);
 384         sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
 385         sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
 386         sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
 387         sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
 388         if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
 389                 sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
 390                 if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
 391                         sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
 392                 else
 393                         sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
 394                 if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
 395                         sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
 396                 else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
 397                         sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
 398         } else
 399                 sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
 400
 401         if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
 402                 sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
 403                 if (voltage->delay) {
 404                         sclk_more_cntl |= VOLTAGE_DROP_SYNC;
 405                         switch (voltage->delay) {
 406                         case 33:
 407                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
 408                                 break;
 409                         case 66:
 410                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
 411                                 break;
 412                         case 99:
 413                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
 414                                 break;
 415                         case 132:
 416                                 sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
 417                                 break;
 418                         }
 419                 } else
 420                         sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
 421         } else
 422                 sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
 423
 424         if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
 425                 sclk_cntl &= ~FORCE_HDP;
 426         else
 427                 sclk_cntl |= FORCE_HDP;
 428
 429         WREG32_PLL(SCLK_CNTL, sclk_cntl);
 430         WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
 431         WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
 432
 433         /* set pcie lanes */
 434         if ((rdev->flags & RADEON_IS_PCIE) &&
 435             !(rdev->flags & RADEON_IS_IGP) &&
 436             rdev->asic->pm.set_pcie_lanes &&
 437             (ps->pcie_lanes !=
 438              rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
 439                 radeon_set_pcie_lanes(rdev,
 440                                       ps->pcie_lanes);
 441                 DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
 442         }
 443 }
 444
 445 /**
 446  * r100_pm_prepare - pre-power state change callback.
 447  *
 448  * @rdev: radeon_device pointer
 449  *
 450  * Prepare for a power state change (r1xx-r4xx).
 451  */
 452 void r100_pm_prepare(struct radeon_device *rdev)
 453 {
 454         struct drm_device *ddev = rdev->ddev;
 455         struct drm_crtc *crtc;
 456         struct radeon_crtc *radeon_crtc;
 457         u32 tmp;
 458
 459         /* disable any active CRTCs */
 460         list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 461                 radeon_crtc = to_radeon_crtc(crtc);
 462                 if (radeon_crtc->enabled) {
 463                         if (radeon_crtc->crtc_id) {
 464                                 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 465                                 tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
 466                                 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 467                         } else {
 468                                 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 469                                 tmp |= RADEON_CRTC_DISP_REQ_EN_B;
 470                                 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 471                         }
 472                 }
 473         }
 474 }
 475
 476 /**
 477  * r100_pm_finish - post-power state change callback.
 478  *
 479  * @rdev: radeon_device pointer
 480  *
 481  * Clean up after a power state change (r1xx-r4xx).
 482  */
 483 void r100_pm_finish(struct radeon_device *rdev)
 484 {
 485         struct drm_device *ddev = rdev->ddev;
 486         struct drm_crtc *crtc;
 487         struct radeon_crtc *radeon_crtc;
 488         u32 tmp;
 489
 490         /* enable any active CRTCs */
 491         list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 492                 radeon_crtc = to_radeon_crtc(crtc);
 493                 if (radeon_crtc->enabled) {
 494                         if (radeon_crtc->crtc_id) {
 495                                 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 496                                 tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
 497                                 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 498                         } else {
 499                                 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 500                                 tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
 501                                 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 502                         }
 503                 }
 504         }
 505 }
 506
 507 /**
 508  * r100_gui_idle - gui idle callback.
 509  *
 510  * @rdev: radeon_device pointer
 511  *
 512  * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
 513  * Returns true if idle, false if not.
 514  */
 515 bool r100_gui_idle(struct radeon_device *rdev)
 516 {
 517         if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
 518                 return false;
 519         else
 520                 return true;
 521 }
 522
 523 /* hpd for digital panel detect/disconnect */
 524 /**
 525  * r100_hpd_sense - hpd sense callback.
 526  *
 527  * @rdev: radeon_device pointer
 528  * @hpd: hpd (hotplug detect) pin
 529  *
 530  * Checks if a digital monitor is connected (r1xx-r4xx).
 531  * Returns true if connected, false if not connected.
 532  */
 533 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
 534 {
 535         bool connected = false;
 536
 537         switch (hpd) {
 538         case RADEON_HPD_1:
 539                 if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
 540                         connected = true;
 541                 break;
 542         case RADEON_HPD_2:
 543                 if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
 544                         connected = true;
 545                 break;
 546         default:
 547                 break;
 548         }
 549         return connected;
 550 }
 551
 552 /**
 553  * r100_hpd_set_polarity - hpd set polarity callback.
 554  *
 555  * @rdev: radeon_device pointer
 556  * @hpd: hpd (hotplug detect) pin
 557  *
 558  * Set the polarity of the hpd pin (r1xx-r4xx).
 559  */
 560 void r100_hpd_set_polarity(struct radeon_device *rdev,
 561                            enum radeon_hpd_id hpd)
 562 {
 563         u32 tmp;
 564         bool connected = r100_hpd_sense(rdev, hpd);
 565
 566         switch (hpd) {
 567         case RADEON_HPD_1:
 568                 tmp = RREG32(RADEON_FP_GEN_CNTL);
 569                 if (connected)
 570                         tmp &= ~RADEON_FP_DETECT_INT_POL;
 571                 else
 572                         tmp |= RADEON_FP_DETECT_INT_POL;
 573                 WREG32(RADEON_FP_GEN_CNTL, tmp);
 574                 break;
 575         case RADEON_HPD_2:
 576                 tmp = RREG32(RADEON_FP2_GEN_CNTL);
 577                 if (connected)
 578                         tmp &= ~RADEON_FP2_DETECT_INT_POL;
 579                 else
 580                         tmp |= RADEON_FP2_DETECT_INT_POL;
 581                 WREG32(RADEON_FP2_GEN_CNTL, tmp);
 582                 break;
 583         default:
 584                 break;
 585         }
 586 }
 587
 588 /**
 589  * r100_hpd_init - hpd setup callback.
 590  *
 591  * @rdev: radeon_device pointer
 592  *
 593  * Setup the hpd pins used by the card (r1xx-r4xx).
 594  * Set the polarity, and enable the hpd interrupts.
 595  */
 596 void r100_hpd_init(struct radeon_device *rdev)
 597 {
 598         struct drm_device *dev = rdev->ddev;
 599         struct drm_connector *connector;
 600         unsigned enable = 0;
 601
 602         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 603                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 604                 enable |= 1 << radeon_connector->hpd.hpd;
 605                 radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
 606         }
 607         radeon_irq_kms_enable_hpd(rdev, enable);
 608 }
 609
 610 /**
 611  * r100_hpd_fini - hpd tear down callback.
 612  *
 613  * @rdev: radeon_device pointer
 614  *
 615  * Tear down the hpd pins used by the card (r1xx-r4xx).
 616  * Disable the hpd interrupts.
 617  */
 618 void r100_hpd_fini(struct radeon_device *rdev)
 619 {
 620         struct drm_device *dev = rdev->ddev;
 621         struct drm_connector *connector;
 622         unsigned disable = 0;
 623
 624         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 625                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 626                 disable |= 1 << radeon_connector->hpd.hpd;
 627         }
 628         radeon_irq_kms_disable_hpd(rdev, disable);
 629 }
 630
 631 /*
 632  * PCI GART
 633  */
 634 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
 635 {
 636         /* TODO: can we do somethings here ? */
 637         /* It seems hw only cache one entry so we should discard this
 638          * entry otherwise if first GPU GART read hit this entry it
 639          * could end up in wrong address. */
 640 }
 641
 642 int r100_pci_gart_init(struct radeon_device *rdev)
 643 {
 644         int r;
 645
 646         if (rdev->gart.ptr) {
 647                 WARN(1, "R100 PCI GART already initialized\n");
 648                 return 0;
 649         }
 650         /* Initialize common gart structure */
 651         r = radeon_gart_init(rdev);
 652         if (r)
 653                 return r;
 654         rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 655         rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
 656         rdev->asic->gart.set_page = &r100_pci_gart_set_page;
 657         return radeon_gart_table_ram_alloc(rdev);
 658 }
 659
 660 int r100_pci_gart_enable(struct radeon_device *rdev)
 661 {
 662         uint32_t tmp;
 663
 664         radeon_gart_restore(rdev);
 665         /* discard memory request outside of configured range */
 666         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 667         WREG32(RADEON_AIC_CNTL, tmp);
 668         /* set address range for PCI address translate */
 669         WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
 670         WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
 671         /* set PCI GART page-table base address */
 672         WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
 673         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
 674         WREG32(RADEON_AIC_CNTL, tmp);
 675         r100_pci_gart_tlb_flush(rdev);
 676         DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
 677                  (unsigned)(rdev->mc.gtt_size >> 20),
 678                  (unsigned long long)rdev->gart.table_addr);
 679         rdev->gart.ready = true;
 680         return 0;
 681 }
 682
 683 void r100_pci_gart_disable(struct radeon_device *rdev)
 684 {
 685         uint32_t tmp;
 686
 687         /* discard memory request outside of configured range */
 688         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 689         WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
 690         WREG32(RADEON_AIC_LO_ADDR, 0);
 691         WREG32(RADEON_AIC_HI_ADDR, 0);
 692 }
 693
 694 int r100_pci_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
 695 {
 696         u32 *gtt = rdev->gart.ptr;
 697
 698         if (i < 0 || i > rdev->gart.num_gpu_pages) {
 699                 return -EINVAL;
 700         }
 701         gtt[i] = cpu_to_le32(lower_32_bits(addr));
 702         return 0;
 703 }
 704
 705 void r100_pci_gart_fini(struct radeon_device *rdev)
 706 {
 707         radeon_gart_fini(rdev);
 708         r100_pci_gart_disable(rdev);
 709         radeon_gart_table_ram_free(rdev);
 710 }
 711
 712 int r100_irq_set(struct radeon_device *rdev)
 713 {
 714         uint32_t tmp = 0;
 715
 716         if (!rdev->irq.installed) {
 717                 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
 718                 WREG32(R_000040_GEN_INT_CNTL, 0);
 719                 return -EINVAL;
 720         }
 721         if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
 722                 tmp |= RADEON_SW_INT_ENABLE;
 723         }
 724         if (rdev->irq.crtc_vblank_int[0] ||
 725             atomic_read(&rdev->irq.pflip[0])) {
 726                 tmp |= RADEON_CRTC_VBLANK_MASK;
 727         }
 728         if (rdev->irq.crtc_vblank_int[1] ||
 729             atomic_read(&rdev->irq.pflip[1])) {
 730                 tmp |= RADEON_CRTC2_VBLANK_MASK;
 731         }
 732         if (rdev->irq.hpd[0]) {
 733                 tmp |= RADEON_FP_DETECT_MASK;
 734         }
 735         if (rdev->irq.hpd[1]) {
 736                 tmp |= RADEON_FP2_DETECT_MASK;
 737         }
 738         WREG32(RADEON_GEN_INT_CNTL, tmp);
 739
 740         /* read back to post the write */
 741         RREG32(RADEON_GEN_INT_CNTL);
 742
 743         return 0;
 744 }
 745
 746 void r100_irq_disable(struct radeon_device *rdev)
 747 {
 748         u32 tmp;
 749
 750         WREG32(R_000040_GEN_INT_CNTL, 0);
 751         /* Wait and acknowledge irq */
 752         mdelay(1);
 753         tmp = RREG32(R_000044_GEN_INT_STATUS);
 754         WREG32(R_000044_GEN_INT_STATUS, tmp);
 755 }
 756
 757 static uint32_t r100_irq_ack(struct radeon_device *rdev)
 758 {
 759         uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
 760         uint32_t irq_mask = RADEON_SW_INT_TEST |
 761                 RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
 762                 RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
 763
 764         if (irqs) {
 765                 WREG32(RADEON_GEN_INT_STATUS, irqs);
 766         }
 767         return irqs & irq_mask;
 768 }
 769
 770 int r100_irq_process(struct radeon_device *rdev)
 771 {
 772         uint32_t status, msi_rearm;
 773         bool queue_hotplug = false;
 774
 775         status = r100_irq_ack(rdev);
 776         if (!status) {
 777                 return IRQ_NONE;
 778         }
 779         if (rdev->shutdown) {
 780                 return IRQ_NONE;
 781         }
 782         while (status) {
 783                 /* SW interrupt */
 784                 if (status & RADEON_SW_INT_TEST) {
 785                         radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
 786                 }
 787                 /* Vertical blank interrupts */
 788                 if (status & RADEON_CRTC_VBLANK_STAT) {
 789                         if (rdev->irq.crtc_vblank_int[0]) {
 790                                 drm_handle_vblank(rdev->ddev, 0);
 791                                 rdev->pm.vblank_sync = true;
 792                                 wake_up(&rdev->irq.vblank_queue);
 793                         }
 794                         if (atomic_read(&rdev->irq.pflip[0]))
 795                                 radeon_crtc_handle_flip(rdev, 0);
 796                 }
 797                 if (status & RADEON_CRTC2_VBLANK_STAT) {
 798                         if (rdev->irq.crtc_vblank_int[1]) {
 799                                 drm_handle_vblank(rdev->ddev, 1);
 800                                 rdev->pm.vblank_sync = true;
 801                                 wake_up(&rdev->irq.vblank_queue);
 802                         }
 803                         if (atomic_read(&rdev->irq.pflip[1]))
 804                                 radeon_crtc_handle_flip(rdev, 1);
 805                 }
 806                 if (status & RADEON_FP_DETECT_STAT) {
 807                         queue_hotplug = true;
 808                         DRM_DEBUG("HPD1\n");
 809                 }
 810                 if (status & RADEON_FP2_DETECT_STAT) {
 811                         queue_hotplug = true;
 812                         DRM_DEBUG("HPD2\n");
 813                 }
 814                 status = r100_irq_ack(rdev);
 815         }
 816         if (queue_hotplug)
 817                 schedule_work(&rdev->hotplug_work);
 818         if (rdev->msi_enabled) {
 819                 switch (rdev->family) {
 820                 case CHIP_RS400:
 821                 case CHIP_RS480:
 822                         msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
 823                         WREG32(RADEON_AIC_CNTL, msi_rearm);
 824                         WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
 825                         break;
 826                 default:
 827                         WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
 828                         break;
 829                 }
 830         }
 831         return IRQ_HANDLED;
 832 }
 833
 834 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
 835 {
 836         if (crtc == 0)
 837                 return RREG32(RADEON_CRTC_CRNT_FRAME);
 838         else
 839                 return RREG32(RADEON_CRTC2_CRNT_FRAME);
 840 }
 841
 842 /* Who ever call radeon_fence_emit should call ring_lock and ask
 843  * for enough space (today caller are ib schedule and buffer move) */
 844 void r100_fence_ring_emit(struct radeon_device *rdev,
 845                           struct radeon_fence *fence)
 846 {
 847         struct radeon_ring *ring = &rdev->ring[fence->ring];
 848
 849         /* We have to make sure that caches are flushed before
 850          * CPU might read something from VRAM. */
 851         radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
 852         radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
 853         radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
 854         radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
 855         /* Wait until IDLE & CLEAN */
 856         radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 857         radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
 858         radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 859         radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
 860                                 RADEON_HDP_READ_BUFFER_INVALIDATE);
 861         radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 862         radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
 863         /* Emit fence sequence & fire IRQ */
 864         radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
 865         radeon_ring_write(ring, fence->seq);
 866         radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
 867         radeon_ring_write(ring, RADEON_SW_INT_FIRE);
 868 }
 869
 870 bool r100_semaphore_ring_emit(struct radeon_device *rdev,
 871                               struct radeon_ring *ring,
 872                               struct radeon_semaphore *semaphore,
 873                               bool emit_wait)
 874 {
 875         /* Unused on older asics, since we don't have semaphores or multiple rings */
 876         BUG();
 877         return false;
 878 }
 879
 880 int r100_copy_blit(struct radeon_device *rdev,
 881                    uint64_t src_offset,
 882                    uint64_t dst_offset,
 883                    unsigned num_gpu_pages,
 884                    struct radeon_fence **fence)
 885 {
 886         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 887         uint32_t cur_pages;
 888         uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
 889         uint32_t pitch;
 890         uint32_t stride_pixels;
 891         unsigned ndw;
 892         int num_loops;
 893         int r = 0;
 894
 895         /* radeon limited to 16k stride */
 896         stride_bytes &= 0x3fff;
 897         /* radeon pitch is /64 */
 898         pitch = stride_bytes / 64;
 899         stride_pixels = stride_bytes / 4;
 900         num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
 901
 902         /* Ask for enough room for blit + flush + fence */
 903         ndw = 64 + (10 * num_loops);
 904         r = radeon_ring_lock(rdev, ring, ndw);
 905         if (r) {
 906                 DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 907                 return -EINVAL;
 908         }
 909         while (num_gpu_pages > 0) {
 910                 cur_pages = num_gpu_pages;
 911                 if (cur_pages > 8191) {
 912                         cur_pages = 8191;
 913                 }
 914                 num_gpu_pages -= cur_pages;
 915
 916                 /* pages are in Y direction - height
 917                    page width in X direction - width */
 918                 radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
 919                 radeon_ring_write(ring,
 920                                   RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 921                                   RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 922                                   RADEON_GMC_SRC_CLIPPING |
 923                                   RADEON_GMC_DST_CLIPPING |
 924                                   RADEON_GMC_BRUSH_NONE |
 925                                   (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 926                                   RADEON_GMC_SRC_DATATYPE_COLOR |
 927                                   RADEON_ROP3_S |
 928                                   RADEON_DP_SRC_SOURCE_MEMORY |
 929                                   RADEON_GMC_CLR_CMP_CNTL_DIS |
 930                                   RADEON_GMC_WR_MSK_DIS);
 931                 radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
 932                 radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
 933                 radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 934                 radeon_ring_write(ring, 0);
 935                 radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 936                 radeon_ring_write(ring, num_gpu_pages);
 937                 radeon_ring_write(ring, num_gpu_pages);
 938                 radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
 939         }
 940         radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 941         radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
 942         radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 943         radeon_ring_write(ring,
 944                           RADEON_WAIT_2D_IDLECLEAN |
 945                           RADEON_WAIT_HOST_IDLECLEAN |
 946                           RADEON_WAIT_DMA_GUI_IDLE);
 947         if (fence) {
 948                 r = radeon_fence_emit(rdev, fence, RADEON_RING_TYPE_GFX_INDEX);
 949         }
 950         radeon_ring_unlock_commit(rdev, ring);
 951         return r;
 952 }
 953
 954 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
 955 {
 956         unsigned i;
 957         u32 tmp;
 958
 959         for (i = 0; i < rdev->usec_timeout; i++) {
 960                 tmp = RREG32(R_000E40_RBBM_STATUS);
 961                 if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
 962                         return 0;
 963                 }
 964                 udelay(1);
 965         }
 966         return -1;
 967 }
 968
 969 void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
 970 {
 971         int r;
 972
 973         r = radeon_ring_lock(rdev, ring, 2);
 974         if (r) {
 975                 return;
 976         }
 977         radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
 978         radeon_ring_write(ring,
 979                           RADEON_ISYNC_ANY2D_IDLE3D |
 980                           RADEON_ISYNC_ANY3D_IDLE2D |
 981                           RADEON_ISYNC_WAIT_IDLEGUI |
 982                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 983         radeon_ring_unlock_commit(rdev, ring);
 984 }
 985
 986
 987 /* Load the microcode for the CP */
 988 static int r100_cp_init_microcode(struct radeon_device *rdev)
 989 {
 990         const char *fw_name = NULL;
 991         int err;
 992
 993         DRM_DEBUG_KMS("\n");
 994
 995         if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
 996             (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
 997             (rdev->family == CHIP_RS200)) {
 998                 DRM_INFO("Loading R100 Microcode\n");
 999                 fw_name = FIRMWARE_R100;
1000         } else if ((rdev->family == CHIP_R200) ||
1001                    (rdev->family == CHIP_RV250) ||
1002                    (rdev->family == CHIP_RV280) ||
1003                    (rdev->family == CHIP_RS300)) {
1004                 DRM_INFO("Loading R200 Microcode\n");
1005                 fw_name = FIRMWARE_R200;
1006         } else if ((rdev->family == CHIP_R300) ||
1007                    (rdev->family == CHIP_R350) ||
1008                    (rdev->family == CHIP_RV350) ||
1009                    (rdev->family == CHIP_RV380) ||
1010                    (rdev->family == CHIP_RS400) ||
1011                    (rdev->family == CHIP_RS480)) {
1012                 DRM_INFO("Loading R300 Microcode\n");
1013                 fw_name = FIRMWARE_R300;
1014         } else if ((rdev->family == CHIP_R420) ||
1015                    (rdev->family == CHIP_R423) ||
1016                    (rdev->family == CHIP_RV410)) {
1017                 DRM_INFO("Loading R400 Microcode\n");
1018                 fw_name = FIRMWARE_R420;
1019         } else if ((rdev->family == CHIP_RS690) ||
1020                    (rdev->family == CHIP_RS740)) {
1021                 DRM_INFO("Loading RS690/RS740 Microcode\n");
1022                 fw_name = FIRMWARE_RS690;
1023         } else if (rdev->family == CHIP_RS600) {
1024                 DRM_INFO("Loading RS600 Microcode\n");
1025                 fw_name = FIRMWARE_RS600;
1026         } else if ((rdev->family == CHIP_RV515) ||
1027                    (rdev->family == CHIP_R520) ||
1028                    (rdev->family == CHIP_RV530) ||
1029                    (rdev->family == CHIP_R580) ||
1030                    (rdev->family == CHIP_RV560) ||
1031                    (rdev->family == CHIP_RV570)) {
1032                 DRM_INFO("Loading R500 Microcode\n");
1033                 fw_name = FIRMWARE_R520;
1034         }
1035
1036         err = reject_firmware(&rdev->me_fw, fw_name, rdev->dev);
1037         if (err) {
1038                 printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1039                        fw_name);
1040         } else if (rdev->me_fw->size % 8) {
1041                 printk(KERN_ERR
1042                        "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1043                        rdev->me_fw->size, fw_name);
1044                 err = -EINVAL;
1045                 release_firmware(rdev->me_fw);
1046                 rdev->me_fw = NULL;
1047         }
1048         return err;
1049 }
1050
1051 u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1052                       struct radeon_ring *ring)
1053 {
1054         u32 rptr;
1055
1056         if (rdev->wb.enabled)
1057                 rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1058         else
1059                 rptr = RREG32(RADEON_CP_RB_RPTR);
1060
1061         return rptr;
1062 }
1063
1064 u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1065                       struct radeon_ring *ring)
1066 {
1067         u32 wptr;
1068
1069         wptr = RREG32(RADEON_CP_RB_WPTR);
1070
1071         return wptr;
1072 }
1073
1074 void r100_gfx_set_wptr(struct radeon_device *rdev,
1075                        struct radeon_ring *ring)
1076 {
1077         WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1078         (void)RREG32(RADEON_CP_RB_WPTR);
1079 }
1080
1081 static void r100_cp_load_microcode(struct radeon_device *rdev)
1082 {
1083         const __be32 *fw_data;
1084         int i, size;
1085
1086         if (r100_gui_wait_for_idle(rdev)) {
1087                 printk(KERN_WARNING "Failed to wait GUI idle while "
1088                        "programming pipes. Bad things might happen.\n");
1089         }
1090
1091         if (rdev->me_fw) {
1092                 size = rdev->me_fw->size / 4;
1093                 fw_data = (const __be32 *)&rdev->me_fw->data[0];
1094                 WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1095                 for (i = 0; i < size; i += 2) {
1096                         WREG32(RADEON_CP_ME_RAM_DATAH,
1097                                be32_to_cpup(&fw_data[i]));
1098                         WREG32(RADEON_CP_ME_RAM_DATAL,
1099                                be32_to_cpup(&fw_data[i + 1]));
1100                 }
1101         }
1102 }
1103
1104 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1105 {
1106         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1107         unsigned rb_bufsz;
1108         unsigned rb_blksz;
1109         unsigned max_fetch;
1110         unsigned pre_write_timer;
1111         unsigned pre_write_limit;
1112         unsigned indirect2_start;
1113         unsigned indirect1_start;
1114         uint32_t tmp;
1115         int r;
1116
1117         if (r100_debugfs_cp_init(rdev)) {
1118                 DRM_ERROR("Failed to register debugfs file for CP !\n");
1119         }
1120         if (!rdev->me_fw) {
1121                 r = r100_cp_init_microcode(rdev);
1122                 if (r) {
1123                         DRM_ERROR("Failed to load firmware!\n");
1124                         return r;
1125                 }
1126         }
1127
1128         /* Align ring size */
1129         rb_bufsz = order_base_2(ring_size / 8);
1130         ring_size = (1 << (rb_bufsz + 1)) * 4;
1131         r100_cp_load_microcode(rdev);
1132         r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1133                              RADEON_CP_PACKET2);
1134         if (r) {
1135                 return r;
1136         }
1137         /* Each time the cp read 1024 bytes (16 dword/quadword) update
1138          * the rptr copy in system ram */
1139         rb_blksz = 9;
1140         /* cp will read 128bytes at a time (4 dwords) */
1141         max_fetch = 1;
1142         ring->align_mask = 16 - 1;
1143         /* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1144         pre_write_timer = 64;
1145         /* Force CP_RB_WPTR write if written more than one time before the
1146          * delay expire
1147          */
1148         pre_write_limit = 0;
1149         /* Setup the cp cache like this (cache size is 96 dwords) :
1150          *      RING            0  to 15
1151          *      INDIRECT1       16 to 79
1152          *      INDIRECT2       80 to 95
1153          * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1154          *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1155          *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1156          * Idea being that most of the gpu cmd will be through indirect1 buffer
1157          * so it gets the bigger cache.
1158          */
1159         indirect2_start = 80;
1160         indirect1_start = 16;
1161         /* cp setup */
1162         WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1163         tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1164                REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1165                REG_SET(RADEON_MAX_FETCH, max_fetch));
1166 #ifdef __BIG_ENDIAN
1167         tmp |= RADEON_BUF_SWAP_32BIT;
1168 #endif
1169         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1170
1171         /* Set ring address */
1172         DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1173         WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1174         /* Force read & write ptr to 0 */
1175         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1176         WREG32(RADEON_CP_RB_RPTR_WR, 0);
1177         ring->wptr = 0;
1178         WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1179
1180         /* set the wb address whether it's enabled or not */
1181         WREG32(R_00070C_CP_RB_RPTR_ADDR,
1182                 S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1183         WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1184
1185         if (rdev->wb.enabled)
1186                 WREG32(R_000770_SCRATCH_UMSK, 0xff);
1187         else {
1188                 tmp |= RADEON_RB_NO_UPDATE;
1189                 WREG32(R_000770_SCRATCH_UMSK, 0);
1190         }
1191
1192         WREG32(RADEON_CP_RB_CNTL, tmp);
1193         udelay(10);
1194         ring->rptr = RREG32(RADEON_CP_RB_RPTR);
1195         /* Set cp mode to bus mastering & enable cp*/
1196         WREG32(RADEON_CP_CSQ_MODE,
1197                REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1198                REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1199         WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1200         WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1201         WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1202
1203         /* at this point everything should be setup correctly to enable master */
1204         pci_set_master(rdev->pdev);
1205
1206         radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1207         r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1208         if (r) {
1209                 DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1210                 return r;
1211         }
1212         ring->ready = true;
1213         radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1214
1215         if (!ring->rptr_save_reg /* not resuming from suspend */
1216             && radeon_ring_supports_scratch_reg(rdev, ring)) {
1217                 r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1218                 if (r) {
1219                         DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1220                         ring->rptr_save_reg = 0;
1221                 }
1222         }
1223         return 0;
1224 }
1225
1226 void r100_cp_fini(struct radeon_device *rdev)
1227 {
1228         if (r100_cp_wait_for_idle(rdev)) {
1229                 DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1230         }
1231         /* Disable ring */
1232         r100_cp_disable(rdev);
1233         radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1234         radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1235         DRM_INFO("radeon: cp finalized\n");
1236 }
1237
1238 void r100_cp_disable(struct radeon_device *rdev)
1239 {
1240         /* Disable ring */
1241         radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1242         rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1243         WREG32(RADEON_CP_CSQ_MODE, 0);
1244         WREG32(RADEON_CP_CSQ_CNTL, 0);
1245         WREG32(R_000770_SCRATCH_UMSK, 0);
1246         if (r100_gui_wait_for_idle(rdev)) {
1247                 printk(KERN_WARNING "Failed to wait GUI idle while "
1248                        "programming pipes. Bad things might happen.\n");
1249         }
1250 }
1251
1252 /*
1253  * CS functions
1254  */
1255 int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1256                             struct radeon_cs_packet *pkt,
1257                             unsigned idx,
1258                             unsigned reg)
1259 {
1260         int r;
1261         u32 tile_flags = 0;
1262         u32 tmp;
1263         struct radeon_cs_reloc *reloc;
1264         u32 value;
1265
1266         r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1267         if (r) {
1268                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1269                           idx, reg);
1270                 radeon_cs_dump_packet(p, pkt);
1271                 return r;
1272         }
1273
1274         value = radeon_get_ib_value(p, idx);
1275         tmp = value & 0x003fffff;
1276         tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
1277
1278         if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1279                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1280                         tile_flags |= RADEON_DST_TILE_MACRO;
1281                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
1282                         if (reg == RADEON_SRC_PITCH_OFFSET) {
1283                                 DRM_ERROR("Cannot src blit from microtiled surface\n");
1284                                 radeon_cs_dump_packet(p, pkt);
1285                                 return -EINVAL;
1286                         }
1287                         tile_flags |= RADEON_DST_TILE_MICRO;
1288                 }
1289
1290                 tmp |= tile_flags;
1291                 p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1292         } else
1293                 p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1294         return 0;
1295 }
1296
1297 int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1298                              struct radeon_cs_packet *pkt,
1299                              int idx)
1300 {
1301         unsigned c, i;
1302         struct radeon_cs_reloc *reloc;
1303         struct r100_cs_track *track;
1304         int r = 0;
1305         volatile uint32_t *ib;
1306         u32 idx_value;
1307
1308         ib = p->ib.ptr;
1309         track = (struct r100_cs_track *)p->track;
1310         c = radeon_get_ib_value(p, idx++) & 0x1F;
1311         if (c > 16) {
1312             DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1313                       pkt->opcode);
1314             radeon_cs_dump_packet(p, pkt);
1315             return -EINVAL;
1316         }
1317         track->num_arrays = c;
1318         for (i = 0; i < (c - 1); i+=2, idx+=3) {
1319                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1320                 if (r) {
1321                         DRM_ERROR("No reloc for packet3 %d\n",
1322                                   pkt->opcode);
1323                         radeon_cs_dump_packet(p, pkt);
1324                         return r;
1325                 }
1326                 idx_value = radeon_get_ib_value(p, idx);
1327                 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1328
1329                 track->arrays[i + 0].esize = idx_value >> 8;
1330                 track->arrays[i + 0].robj = reloc->robj;
1331                 track->arrays[i + 0].esize &= 0x7F;
1332                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1333                 if (r) {
1334                         DRM_ERROR("No reloc for packet3 %d\n",
1335                                   pkt->opcode);
1336                         radeon_cs_dump_packet(p, pkt);
1337                         return r;
1338                 }
1339                 ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->lobj.gpu_offset);
1340                 track->arrays[i + 1].robj = reloc->robj;
1341                 track->arrays[i + 1].esize = idx_value >> 24;
1342                 track->arrays[i + 1].esize &= 0x7F;
1343         }
1344         if (c & 1) {
1345                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1346                 if (r) {
1347                         DRM_ERROR("No reloc for packet3 %d\n",
1348                                           pkt->opcode);
1349                         radeon_cs_dump_packet(p, pkt);
1350                         return r;
1351                 }
1352                 idx_value = radeon_get_ib_value(p, idx);
1353                 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1354                 track->arrays[i + 0].robj = reloc->robj;
1355                 track->arrays[i + 0].esize = idx_value >> 8;
1356                 track->arrays[i + 0].esize &= 0x7F;
1357         }
1358         return r;
1359 }
1360
1361 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1362                           struct radeon_cs_packet *pkt,
1363                           const unsigned *auth, unsigned n,
1364                           radeon_packet0_check_t check)
1365 {
1366         unsigned reg;
1367         unsigned i, j, m;
1368         unsigned idx;
1369         int r;
1370
1371         idx = pkt->idx + 1;
1372         reg = pkt->reg;
1373         /* Check that register fall into register range
1374          * determined by the number of entry (n) in the
1375          * safe register bitmap.
1376          */
1377         if (pkt->one_reg_wr) {
1378                 if ((reg >> 7) > n) {
1379                         return -EINVAL;
1380                 }
1381         } else {
1382                 if (((reg + (pkt->count << 2)) >> 7) > n) {
1383                         return -EINVAL;
1384                 }
1385         }
1386         for (i = 0; i <= pkt->count; i++, idx++) {
1387                 j = (reg >> 7);
1388                 m = 1 << ((reg >> 2) & 31);
1389                 if (auth[j] & m) {
1390                         r = check(p, pkt, idx, reg);
1391                         if (r) {
1392                                 return r;
1393                         }
1394                 }
1395                 if (pkt->one_reg_wr) {
1396                         if (!(auth[j] & m)) {
1397                                 break;
1398                         }
1399                 } else {
1400                         reg += 4;
1401                 }
1402         }
1403         return 0;
1404 }
1405
1406 /**
1407  * r100_cs_packet_next_vline() - parse userspace VLINE packet
1408  * @parser:             parser structure holding parsing context.
1409  *
1410  * Userspace sends a special sequence for VLINE waits.
1411  * PACKET0 - VLINE_START_END + value
1412  * PACKET0 - WAIT_UNTIL +_value
1413  * RELOC (P3) - crtc_id in reloc.
1414  *
1415  * This function parses this and relocates the VLINE START END
1416  * and WAIT UNTIL packets to the correct crtc.
1417  * It also detects a switched off crtc and nulls out the
1418  * wait in that case.
1419  */
1420 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1421 {
1422         struct drm_mode_object *obj;
1423         struct drm_crtc *crtc;
1424         struct radeon_crtc *radeon_crtc;
1425         struct radeon_cs_packet p3reloc, waitreloc;
1426         int crtc_id;
1427         int r;
1428         uint32_t header, h_idx, reg;
1429         volatile uint32_t *ib;
1430
1431         ib = p->ib.ptr;
1432
1433         /* parse the wait until */
1434         r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1435         if (r)
1436                 return r;
1437
1438         /* check its a wait until and only 1 count */
1439         if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1440             waitreloc.count != 0) {
1441                 DRM_ERROR("vline wait had illegal wait until segment\n");
1442                 return -EINVAL;
1443         }
1444
1445         if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1446                 DRM_ERROR("vline wait had illegal wait until\n");
1447                 return -EINVAL;
1448         }
1449
1450         /* jump over the NOP */
1451         r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1452         if (r)
1453                 return r;
1454
1455         h_idx = p->idx - 2;
1456         p->idx += waitreloc.count + 2;
1457         p->idx += p3reloc.count + 2;
1458
1459         header = radeon_get_ib_value(p, h_idx);
1460         crtc_id = radeon_get_ib_value(p, h_idx + 5);
1461         reg = R100_CP_PACKET0_GET_REG(header);
1462         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
1463         if (!obj) {
1464                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
1465                 return -ENOENT;
1466         }
1467         crtc = obj_to_crtc(obj);
1468         radeon_crtc = to_radeon_crtc(crtc);
1469         crtc_id = radeon_crtc->crtc_id;
1470
1471         if (!crtc->enabled) {
1472                 /* if the CRTC isn't enabled - we need to nop out the wait until */
1473                 ib[h_idx + 2] = PACKET2(0);
1474                 ib[h_idx + 3] = PACKET2(0);
1475         } else if (crtc_id == 1) {
1476                 switch (reg) {
1477                 case AVIVO_D1MODE_VLINE_START_END:
1478                         header &= ~R300_CP_PACKET0_REG_MASK;
1479                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1480                         break;
1481                 case RADEON_CRTC_GUI_TRIG_VLINE:
1482                         header &= ~R300_CP_PACKET0_REG_MASK;
1483                         header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1484                         break;
1485                 default:
1486                         DRM_ERROR("unknown crtc reloc\n");
1487                         return -EINVAL;
1488                 }
1489                 ib[h_idx] = header;
1490                 ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1491         }
1492
1493         return 0;
1494 }
1495
1496 static int r100_get_vtx_size(uint32_t vtx_fmt)
1497 {
1498         int vtx_size;
1499         vtx_size = 2;
1500         /* ordered according to bits in spec */
1501         if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1502                 vtx_size++;
1503         if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1504                 vtx_size += 3;
1505         if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1506                 vtx_size++;
1507         if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1508                 vtx_size++;
1509         if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1510                 vtx_size += 3;
1511         if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1512                 vtx_size++;
1513         if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1514                 vtx_size++;
1515         if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1516                 vtx_size += 2;
1517         if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1518                 vtx_size += 2;
1519         if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1520                 vtx_size++;
1521         if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1522                 vtx_size += 2;
1523         if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1524                 vtx_size++;
1525         if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1526                 vtx_size += 2;
1527         if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1528                 vtx_size++;
1529         if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1530                 vtx_size++;
1531         /* blend weight */
1532         if (vtx_fmt & (0x7 << 15))
1533                 vtx_size += (vtx_fmt >> 15) & 0x7;
1534         if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1535                 vtx_size += 3;
1536         if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1537                 vtx_size += 2;
1538         if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1539                 vtx_size++;
1540         if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1541                 vtx_size++;
1542         if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1543                 vtx_size++;
1544         if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1545                 vtx_size++;
1546         return vtx_size;
1547 }
1548
1549 static int r100_packet0_check(struct radeon_cs_parser *p,
1550                               struct radeon_cs_packet *pkt,
1551                               unsigned idx, unsigned reg)
1552 {
1553         struct radeon_cs_reloc *reloc;
1554         struct r100_cs_track *track;
1555         volatile uint32_t *ib;
1556         uint32_t tmp;
1557         int r;
1558         int i, face;
1559         u32 tile_flags = 0;
1560         u32 idx_value;
1561
1562         ib = p->ib.ptr;
1563         track = (struct r100_cs_track *)p->track;
1564
1565         idx_value = radeon_get_ib_value(p, idx);
1566
1567         switch (reg) {
1568         case RADEON_CRTC_GUI_TRIG_VLINE:
1569                 r = r100_cs_packet_parse_vline(p);
1570                 if (r) {
1571                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1572                                   idx, reg);
1573                         radeon_cs_dump_packet(p, pkt);
1574                         return r;
1575                 }
1576                 break;
1577                 /* FIXME: only allow PACKET3 blit? easier to check for out of
1578                  * range access */
1579         case RADEON_DST_PITCH_OFFSET:
1580         case RADEON_SRC_PITCH_OFFSET:
1581                 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1582                 if (r)
1583                         return r;
1584                 break;
1585         case RADEON_RB3D_DEPTHOFFSET:
1586                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1587                 if (r) {
1588                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1589                                   idx, reg);
1590                         radeon_cs_dump_packet(p, pkt);
1591                         return r;
1592                 }
1593                 track->zb.robj = reloc->robj;
1594                 track->zb.offset = idx_value;
1595                 track->zb_dirty = true;
1596                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1597                 break;
1598         case RADEON_RB3D_COLOROFFSET:
1599                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1600                 if (r) {
1601                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1602                                   idx, reg);
1603                         radeon_cs_dump_packet(p, pkt);
1604                         return r;
1605                 }
1606                 track->cb[0].robj = reloc->robj;
1607                 track->cb[0].offset = idx_value;
1608                 track->cb_dirty = true;
1609                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1610                 break;
1611         case RADEON_PP_TXOFFSET_0:
1612         case RADEON_PP_TXOFFSET_1:
1613         case RADEON_PP_TXOFFSET_2:
1614                 i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1615                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1616                 if (r) {
1617                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1618                                   idx, reg);
1619                         radeon_cs_dump_packet(p, pkt);
1620                         return r;
1621                 }
1622                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1623                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1624                                 tile_flags |= RADEON_TXO_MACRO_TILE;
1625                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1626                                 tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1627
1628                         tmp = idx_value & ~(0x7 << 2);
1629                         tmp |= tile_flags;
1630                         ib[idx] = tmp + ((u32)reloc->lobj.gpu_offset);
1631                 } else
1632                         ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1633                 track->textures[i].robj = reloc->robj;
1634                 track->tex_dirty = true;
1635                 break;
1636         case RADEON_PP_CUBIC_OFFSET_T0_0:
1637         case RADEON_PP_CUBIC_OFFSET_T0_1:
1638         case RADEON_PP_CUBIC_OFFSET_T0_2:
1639         case RADEON_PP_CUBIC_OFFSET_T0_3:
1640         case RADEON_PP_CUBIC_OFFSET_T0_4:
1641                 i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1642                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1643                 if (r) {
1644                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1645                                   idx, reg);
1646                         radeon_cs_dump_packet(p, pkt);
1647                         return r;
1648                 }
1649                 track->textures[0].cube_info[i].offset = idx_value;
1650                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1651                 track->textures[0].cube_info[i].robj = reloc->robj;
1652                 track->tex_dirty = true;
1653                 break;
1654         case RADEON_PP_CUBIC_OFFSET_T1_0:
1655         case RADEON_PP_CUBIC_OFFSET_T1_1:
1656         case RADEON_PP_CUBIC_OFFSET_T1_2:
1657         case RADEON_PP_CUBIC_OFFSET_T1_3:
1658         case RADEON_PP_CUBIC_OFFSET_T1_4:
1659                 i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1660                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1661                 if (r) {
1662                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1663                                   idx, reg);
1664                         radeon_cs_dump_packet(p, pkt);
1665                         return r;
1666                 }
1667                 track->textures[1].cube_info[i].offset = idx_value;
1668                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1669                 track->textures[1].cube_info[i].robj = reloc->robj;
1670                 track->tex_dirty = true;
1671                 break;
1672         case RADEON_PP_CUBIC_OFFSET_T2_0:
1673         case RADEON_PP_CUBIC_OFFSET_T2_1:
1674         case RADEON_PP_CUBIC_OFFSET_T2_2:
1675         case RADEON_PP_CUBIC_OFFSET_T2_3:
1676         case RADEON_PP_CUBIC_OFFSET_T2_4:
1677                 i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1678                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1679                 if (r) {
1680                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1681                                   idx, reg);
1682                         radeon_cs_dump_packet(p, pkt);
1683                         return r;
1684                 }
1685                 track->textures[2].cube_info[i].offset = idx_value;
1686                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1687                 track->textures[2].cube_info[i].robj = reloc->robj;
1688                 track->tex_dirty = true;
1689                 break;
1690         case RADEON_RE_WIDTH_HEIGHT:
1691                 track->maxy = ((idx_value >> 16) & 0x7FF);
1692                 track->cb_dirty = true;
1693                 track->zb_dirty = true;
1694                 break;
1695         case RADEON_RB3D_COLORPITCH:
1696                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1697                 if (r) {
1698                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1699                                   idx, reg);
1700                         radeon_cs_dump_packet(p, pkt);
1701                         return r;
1702                 }
1703                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1704                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1705                                 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1706                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1707                                 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1708
1709                         tmp = idx_value & ~(0x7 << 16);
1710                         tmp |= tile_flags;
1711                         ib[idx] = tmp;
1712                 } else
1713                         ib[idx] = idx_value;
1714
1715                 track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1716                 track->cb_dirty = true;
1717                 break;
1718         case RADEON_RB3D_DEPTHPITCH:
1719                 track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1720                 track->zb_dirty = true;
1721                 break;
1722         case RADEON_RB3D_CNTL:
1723                 switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1724                 case 7:
1725                 case 8:
1726                 case 9:
1727                 case 11:
1728                 case 12:
1729                         track->cb[0].cpp = 1;
1730                         break;
1731                 case 3:
1732                 case 4:
1733                 case 15:
1734                         track->cb[0].cpp = 2;
1735                         break;
1736                 case 6:
1737                         track->cb[0].cpp = 4;
1738                         break;
1739                 default:
1740                         DRM_ERROR("Invalid color buffer format (%d) !\n",
1741                                   ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1742                         return -EINVAL;
1743                 }
1744                 track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1745                 track->cb_dirty = true;
1746                 track->zb_dirty = true;
1747                 break;
1748         case RADEON_RB3D_ZSTENCILCNTL:
1749                 switch (idx_value & 0xf) {
1750                 case 0:
1751                         track->zb.cpp = 2;
1752                         break;
1753                 case 2:
1754                 case 3:
1755                 case 4:
1756                 case 5:
1757                 case 9:
1758                 case 11:
1759                         track->zb.cpp = 4;
1760                         break;
1761                 default:
1762                         break;
1763                 }
1764                 track->zb_dirty = true;
1765                 break;
1766         case RADEON_RB3D_ZPASS_ADDR:
1767                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1768                 if (r) {
1769                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1770                                   idx, reg);
1771                         radeon_cs_dump_packet(p, pkt);
1772                         return r;
1773                 }
1774                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1775                 break;
1776         case RADEON_PP_CNTL:
1777                 {
1778                         uint32_t temp = idx_value >> 4;
1779                         for (i = 0; i < track->num_texture; i++)
1780                                 track->textures[i].enabled = !!(temp & (1 << i));
1781                         track->tex_dirty = true;
1782                 }
1783                 break;
1784         case RADEON_SE_VF_CNTL:
1785                 track->vap_vf_cntl = idx_value;
1786                 break;
1787         case RADEON_SE_VTX_FMT:
1788                 track->vtx_size = r100_get_vtx_size(idx_value);
1789                 break;
1790         case RADEON_PP_TEX_SIZE_0:
1791         case RADEON_PP_TEX_SIZE_1:
1792         case RADEON_PP_TEX_SIZE_2:
1793                 i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1794                 track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1795                 track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1796                 track->tex_dirty = true;
1797                 break;
1798         case RADEON_PP_TEX_PITCH_0:
1799         case RADEON_PP_TEX_PITCH_1:
1800         case RADEON_PP_TEX_PITCH_2:
1801                 i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1802                 track->textures[i].pitch = idx_value + 32;
1803                 track->tex_dirty = true;
1804                 break;
1805         case RADEON_PP_TXFILTER_0:
1806         case RADEON_PP_TXFILTER_1:
1807         case RADEON_PP_TXFILTER_2:
1808                 i = (reg - RADEON_PP_TXFILTER_0) / 24;
1809                 track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1810                                                  >> RADEON_MAX_MIP_LEVEL_SHIFT);
1811                 tmp = (idx_value >> 23) & 0x7;
1812                 if (tmp == 2 || tmp == 6)
1813                         track->textures[i].roundup_w = false;
1814                 tmp = (idx_value >> 27) & 0x7;
1815                 if (tmp == 2 || tmp == 6)
1816                         track->textures[i].roundup_h = false;
1817                 track->tex_dirty = true;
1818                 break;
1819         case RADEON_PP_TXFORMAT_0:
1820         case RADEON_PP_TXFORMAT_1:
1821         case RADEON_PP_TXFORMAT_2:
1822                 i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1823                 if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1824                         track->textures[i].use_pitch = 1;
1825                 } else {
1826                         track->textures[i].use_pitch = 0;
1827                         track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1828                         track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1829                 }
1830                 if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1831                         track->textures[i].tex_coord_type = 2;
1832                 switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1833                 case RADEON_TXFORMAT_I8:
1834                 case RADEON_TXFORMAT_RGB332:
1835                 case RADEON_TXFORMAT_Y8:
1836                         track->textures[i].cpp = 1;
1837                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1838                         break;
1839                 case RADEON_TXFORMAT_AI88:
1840                 case RADEON_TXFORMAT_ARGB1555:
1841                 case RADEON_TXFORMAT_RGB565:
1842                 case RADEON_TXFORMAT_ARGB4444:
1843                 case RADEON_TXFORMAT_VYUY422:
1844                 case RADEON_TXFORMAT_YVYU422:
1845                 case RADEON_TXFORMAT_SHADOW16:
1846                 case RADEON_TXFORMAT_LDUDV655:
1847                 case RADEON_TXFORMAT_DUDV88:
1848                         track->textures[i].cpp = 2;
1849                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1850                         break;
1851                 case RADEON_TXFORMAT_ARGB8888:
1852                 case RADEON_TXFORMAT_RGBA8888:
1853                 case RADEON_TXFORMAT_SHADOW32:
1854                 case RADEON_TXFORMAT_LDUDUV8888:
1855                         track->textures[i].cpp = 4;
1856                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1857                         break;
1858                 case RADEON_TXFORMAT_DXT1:
1859                         track->textures[i].cpp = 1;
1860                         track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1861                         break;
1862                 case RADEON_TXFORMAT_DXT23:
1863                 case RADEON_TXFORMAT_DXT45:
1864                         track->textures[i].cpp = 1;
1865                         track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1866                         break;
1867                 }
1868                 track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1869                 track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1870                 track->tex_dirty = true;
1871                 break;
1872         case RADEON_PP_CUBIC_FACES_0:
1873         case RADEON_PP_CUBIC_FACES_1:
1874         case RADEON_PP_CUBIC_FACES_2:
1875                 tmp = idx_value;
1876                 i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1877                 for (face = 0; face < 4; face++) {
1878                         track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1879                         track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1880                 }
1881                 track->tex_dirty = true;
1882                 break;
1883         default:
1884                 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1885                        reg, idx);
1886                 return -EINVAL;
1887         }
1888         return 0;
1889 }
1890
1891 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1892                                          struct radeon_cs_packet *pkt,
1893                                          struct radeon_bo *robj)
1894 {
1895         unsigned idx;
1896         u32 value;
1897         idx = pkt->idx + 1;
1898         value = radeon_get_ib_value(p, idx + 2);
1899         if ((value + 1) > radeon_bo_size(robj)) {
1900                 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1901                           "(need %u have %lu) !\n",
1902                           value + 1,
1903                           radeon_bo_size(robj));
1904                 return -EINVAL;
1905         }
1906         return 0;
1907 }
1908
1909 static int r100_packet3_check(struct radeon_cs_parser *p,
1910                               struct radeon_cs_packet *pkt)
1911 {
1912         struct radeon_cs_reloc *reloc;
1913         struct r100_cs_track *track;
1914         unsigned idx;
1915         volatile uint32_t *ib;
1916         int r;
1917
1918         ib = p->ib.ptr;
1919         idx = pkt->idx + 1;
1920         track = (struct r100_cs_track *)p->track;
1921         switch (pkt->opcode) {
1922         case PACKET3_3D_LOAD_VBPNTR:
1923                 r = r100_packet3_load_vbpntr(p, pkt, idx);
1924                 if (r)
1925                         return r;
1926                 break;
1927         case PACKET3_INDX_BUFFER:
1928                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1929                 if (r) {
1930                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1931                         radeon_cs_dump_packet(p, pkt);
1932                         return r;
1933                 }
1934                 ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->lobj.gpu_offset);
1935                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1936                 if (r) {
1937                         return r;
1938                 }
1939                 break;
1940         case 0x23:
1941                 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1942                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1943                 if (r) {
1944                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1945                         radeon_cs_dump_packet(p, pkt);
1946                         return r;
1947                 }
1948                 ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->lobj.gpu_offset);
1949                 track->num_arrays = 1;
1950                 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1951
1952                 track->arrays[0].robj = reloc->robj;
1953                 track->arrays[0].esize = track->vtx_size;
1954
1955                 track->max_indx = radeon_get_ib_value(p, idx+1);
1956
1957                 track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1958                 track->immd_dwords = pkt->count - 1;
1959                 r = r100_cs_track_check(p->rdev, track);
1960                 if (r)
1961                         return r;
1962                 break;
1963         case PACKET3_3D_DRAW_IMMD:
1964                 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1965                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1966                         return -EINVAL;
1967                 }
1968                 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1969                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1970                 track->immd_dwords = pkt->count - 1;
1971                 r = r100_cs_track_check(p->rdev, track);
1972                 if (r)
1973                         return r;
1974                 break;
1975                 /* triggers drawing using in-packet vertex data */
1976         case PACKET3_3D_DRAW_IMMD_2:
1977                 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1978                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1979                         return -EINVAL;
1980                 }
1981                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1982                 track->immd_dwords = pkt->count;
1983                 r = r100_cs_track_check(p->rdev, track);
1984                 if (r)
1985                         return r;
1986                 break;
1987                 /* triggers drawing using in-packet vertex data */
1988         case PACKET3_3D_DRAW_VBUF_2:
1989                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1990                 r = r100_cs_track_check(p->rdev, track);
1991                 if (r)
1992                         return r;
1993                 break;
1994                 /* triggers drawing of vertex buffers setup elsewhere */
1995         case PACKET3_3D_DRAW_INDX_2:
1996                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1997                 r = r100_cs_track_check(p->rdev, track);
1998                 if (r)
1999                         return r;
2000                 break;
2001                 /* triggers drawing using indices to vertex buffer */
2002         case PACKET3_3D_DRAW_VBUF:
2003                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2004                 r = r100_cs_track_check(p->rdev, track);
2005                 if (r)
2006                         return r;
2007                 break;
2008                 /* triggers drawing of vertex buffers setup elsewhere */
2009         case PACKET3_3D_DRAW_INDX:
2010                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2011                 r = r100_cs_track_check(p->rdev, track);
2012                 if (r)
2013                         return r;
2014                 break;
2015                 /* triggers drawing using indices to vertex buffer */
2016         case PACKET3_3D_CLEAR_HIZ:
2017         case PACKET3_3D_CLEAR_ZMASK:
2018                 if (p->rdev->hyperz_filp != p->filp)
2019                         return -EINVAL;
2020                 break;
2021         case PACKET3_NOP:
2022                 break;
2023         default:
2024                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2025                 return -EINVAL;
2026         }
2027         return 0;
2028 }
2029
2030 int r100_cs_parse(struct radeon_cs_parser *p)
2031 {
2032         struct radeon_cs_packet pkt;
2033         struct r100_cs_track *track;
2034         int r;
2035
2036         track = kzalloc(sizeof(*track), GFP_KERNEL);
2037         if (!track)
2038                 return -ENOMEM;
2039         r100_cs_track_clear(p->rdev, track);
2040         p->track = track;
2041         do {
2042                 r = radeon_cs_packet_parse(p, &pkt, p->idx);
2043                 if (r) {
2044                         return r;
2045                 }
2046                 p->idx += pkt.count + 2;
2047                 switch (pkt.type) {
2048                 case RADEON_PACKET_TYPE0:
2049                         if (p->rdev->family >= CHIP_R200)
2050                                 r = r100_cs_parse_packet0(p, &pkt,
2051                                         p->rdev->config.r100.reg_safe_bm,
2052                                         p->rdev->config.r100.reg_safe_bm_size,
2053                                         &r200_packet0_check);
2054                         else
2055                                 r = r100_cs_parse_packet0(p, &pkt,
2056                                         p->rdev->config.r100.reg_safe_bm,
2057                                         p->rdev->config.r100.reg_safe_bm_size,
2058                                         &r100_packet0_check);
2059                         break;
2060                 case RADEON_PACKET_TYPE2:
2061                         break;
2062                 case RADEON_PACKET_TYPE3:
2063                         r = r100_packet3_check(p, &pkt);
2064                         break;
2065                 default:
2066                         DRM_ERROR("Unknown packet type %d !\n",
2067                                   pkt.type);
2068                         return -EINVAL;
2069                 }
2070                 if (r)
2071                         return r;
2072         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
2073         return 0;
2074 }
2075
2076 static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2077 {
2078         DRM_ERROR("pitch                      %d\n", t->pitch);
2079         DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2080         DRM_ERROR("width                      %d\n", t->width);
2081         DRM_ERROR("width_11                   %d\n", t->width_11);
2082         DRM_ERROR("height                     %d\n", t->height);
2083         DRM_ERROR("height_11                  %d\n", t->height_11);
2084         DRM_ERROR("num levels                 %d\n", t->num_levels);
2085         DRM_ERROR("depth                      %d\n", t->txdepth);
2086         DRM_ERROR("bpp                        %d\n", t->cpp);
2087         DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2088         DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2089         DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2090         DRM_ERROR("compress format            %d\n", t->compress_format);
2091 }
2092
2093 static int r100_track_compress_size(int compress_format, int w, int h)
2094 {
2095         int block_width, block_height, block_bytes;
2096         int wblocks, hblocks;
2097         int min_wblocks;
2098         int sz;
2099
2100         block_width = 4;
2101         block_height = 4;
2102
2103         switch (compress_format) {
2104         case R100_TRACK_COMP_DXT1:
2105                 block_bytes = 8;
2106                 min_wblocks = 4;
2107                 break;
2108         default:
2109         case R100_TRACK_COMP_DXT35:
2110                 block_bytes = 16;
2111                 min_wblocks = 2;
2112                 break;
2113         }
2114
2115         hblocks = (h + block_height - 1) / block_height;
2116         wblocks = (w + block_width - 1) / block_width;
2117         if (wblocks < min_wblocks)
2118                 wblocks = min_wblocks;
2119         sz = wblocks * hblocks * block_bytes;
2120         return sz;
2121 }
2122
2123 static int r100_cs_track_cube(struct radeon_device *rdev,
2124                               struct r100_cs_track *track, unsigned idx)
2125 {
2126         unsigned face, w, h;
2127         struct radeon_bo *cube_robj;
2128         unsigned long size;
2129         unsigned compress_format = track->textures[idx].compress_format;
2130
2131         for (face = 0; face < 5; face++) {
2132                 cube_robj = track->textures[idx].cube_info[face].robj;
2133                 w = track->textures[idx].cube_info[face].width;
2134                 h = track->textures[idx].cube_info[face].height;
2135
2136                 if (compress_format) {
2137                         size = r100_track_compress_size(compress_format, w, h);
2138                 } else
2139                         size = w * h;
2140                 size *= track->textures[idx].cpp;
2141
2142                 size += track->textures[idx].cube_info[face].offset;
2143
2144                 if (size > radeon_bo_size(cube_robj)) {
2145                         DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2146                                   size, radeon_bo_size(cube_robj));
2147                         r100_cs_track_texture_print(&track->textures[idx]);
2148                         return -1;
2149                 }
2150         }
2151         return 0;
2152 }
2153
2154 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2155                                        struct r100_cs_track *track)
2156 {
2157         struct radeon_bo *robj;
2158         unsigned long size;
2159         unsigned u, i, w, h, d;
2160         int ret;
2161
2162         for (u = 0; u < track->num_texture; u++) {
2163                 if (!track->textures[u].enabled)
2164                         continue;
2165                 if (track->textures[u].lookup_disable)
2166                         continue;
2167                 robj = track->textures[u].robj;
2168                 if (robj == NULL) {
2169                         DRM_ERROR("No texture bound to unit %u\n", u);
2170                         return -EINVAL;
2171                 }
2172                 size = 0;
2173                 for (i = 0; i <= track->textures[u].num_levels; i++) {
2174                         if (track->textures[u].use_pitch) {
2175                                 if (rdev->family < CHIP_R300)
2176                                         w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2177                                 else
2178                                         w = track->textures[u].pitch / (1 << i);
2179                         } else {
2180                                 w = track->textures[u].width;
2181                                 if (rdev->family >= CHIP_RV515)
2182                                         w |= track->textures[u].width_11;
2183                                 w = w / (1 << i);
2184                                 if (track->textures[u].roundup_w)
2185                                         w = roundup_pow_of_two(w);
2186                         }
2187                         h = track->textures[u].height;
2188                         if (rdev->family >= CHIP_RV515)
2189                                 h |= track->textures[u].height_11;
2190                         h = h / (1 << i);
2191                         if (track->textures[u].roundup_h)
2192                                 h = roundup_pow_of_two(h);
2193                         if (track->textures[u].tex_coord_type == 1) {
2194                                 d = (1 << track->textures[u].txdepth) / (1 << i);
2195                                 if (!d)
2196                                         d = 1;
2197                         } else {
2198                                 d = 1;
2199                         }
2200                         if (track->textures[u].compress_format) {
2201
2202                                 size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2203                                 /* compressed textures are block based */
2204                         } else
2205                                 size += w * h * d;
2206                 }
2207                 size *= track->textures[u].cpp;
2208
2209                 switch (track->textures[u].tex_coord_type) {
2210                 case 0:
2211                 case 1:
2212                         break;
2213                 case 2:
2214                         if (track->separate_cube) {
2215                                 ret = r100_cs_track_cube(rdev, track, u);
2216                                 if (ret)
2217                                         return ret;
2218                         } else
2219                                 size *= 6;
2220                         break;
2221                 default:
2222                         DRM_ERROR("Invalid texture coordinate type %u for unit "
2223                                   "%u\n", track->textures[u].tex_coord_type, u);
2224                         return -EINVAL;
2225                 }
2226                 if (size > radeon_bo_size(robj)) {
2227                         DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2228                                   "%lu\n", u, size, radeon_bo_size(robj));
2229                         r100_cs_track_texture_print(&track->textures[u]);
2230                         return -EINVAL;
2231                 }
2232         }
2233         return 0;
2234 }
2235
2236 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2237 {
2238         unsigned i;
2239         unsigned long size;
2240         unsigned prim_walk;
2241         unsigned nverts;
2242         unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2243
2244         if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2245             !track->blend_read_enable)
2246                 num_cb = 0;
2247
2248         for (i = 0; i < num_cb; i++) {
2249                 if (track->cb[i].robj == NULL) {
2250                         DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2251                         return -EINVAL;
2252                 }
2253                 size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2254                 size += track->cb[i].offset;
2255                 if (size > radeon_bo_size(track->cb[i].robj)) {
2256                         DRM_ERROR("[drm] Buffer too small for color buffer %d "
2257                                   "(need %lu have %lu) !\n", i, size,
2258                                   radeon_bo_size(track->cb[i].robj));
2259                         DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2260                                   i, track->cb[i].pitch, track->cb[i].cpp,
2261                                   track->cb[i].offset, track->maxy);
2262                         return -EINVAL;
2263                 }
2264         }
2265         track->cb_dirty = false;
2266
2267         if (track->zb_dirty && track->z_enabled) {
2268                 if (track->zb.robj == NULL) {
2269                         DRM_ERROR("[drm] No buffer for z buffer !\n");
2270                         return -EINVAL;
2271                 }
2272                 size = track->zb.pitch * track->zb.cpp * track->maxy;
2273                 size += track->zb.offset;
2274                 if (size > radeon_bo_size(track->zb.robj)) {
2275                         DRM_ERROR("[drm] Buffer too small for z buffer "
2276                                   "(need %lu have %lu) !\n", size,
2277                                   radeon_bo_size(track->zb.robj));
2278                         DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2279                                   track->zb.pitch, track->zb.cpp,
2280                                   track->zb.offset, track->maxy);
2281                         return -EINVAL;
2282                 }
2283         }
2284         track->zb_dirty = false;
2285
2286         if (track->aa_dirty && track->aaresolve) {
2287                 if (track->aa.robj == NULL) {
2288                         DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2289                         return -EINVAL;
2290                 }
2291                 /* I believe the format comes from colorbuffer0. */
2292                 size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2293                 size += track->aa.offset;
2294                 if (size > radeon_bo_size(track->aa.robj)) {
2295                         DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2296                                   "(need %lu have %lu) !\n", i, size,
2297                                   radeon_bo_size(track->aa.robj));
2298                         DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2299                                   i, track->aa.pitch, track->cb[0].cpp,
2300                                   track->aa.offset, track->maxy);
2301                         return -EINVAL;
2302                 }
2303         }
2304         track->aa_dirty = false;
2305
2306         prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2307         if (track->vap_vf_cntl & (1 << 14)) {
2308                 nverts = track->vap_alt_nverts;
2309         } else {
2310                 nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2311         }
2312         switch (prim_walk) {
2313         case 1:
2314                 for (i = 0; i < track->num_arrays; i++) {
2315                         size = track->arrays[i].esize * track->max_indx * 4;
2316                         if (track->arrays[i].robj == NULL) {
2317                                 DRM_ERROR("(PW %u) Vertex array %u no buffer "
2318                                           "bound\n", prim_walk, i);
2319                                 return -EINVAL;
2320                         }
2321                         if (size > radeon_bo_size(track->arrays[i].robj)) {
2322                                 dev_err(rdev->dev, "(PW %u) Vertex array %u "
2323                                         "need %lu dwords have %lu dwords\n",
2324                                         prim_walk, i, size >> 2,
2325                                         radeon_bo_size(track->arrays[i].robj)
2326                                         >> 2);
2327                                 DRM_ERROR("Max indices %u\n", track->max_indx);
2328                                 return -EINVAL;
2329                         }
2330                 }
2331                 break;
2332         case 2:
2333                 for (i = 0; i < track->num_arrays; i++) {
2334                         size = track->arrays[i].esize * (nverts - 1) * 4;
2335                         if (track->arrays[i].robj == NULL) {
2336                                 DRM_ERROR("(PW %u) Vertex array %u no buffer "
2337                                           "bound\n", prim_walk, i);
2338                                 return -EINVAL;
2339                         }
2340                         if (size > radeon_bo_size(track->arrays[i].robj)) {
2341                                 dev_err(rdev->dev, "(PW %u) Vertex array %u "
2342                                         "need %lu dwords have %lu dwords\n",
2343                                         prim_walk, i, size >> 2,
2344                                         radeon_bo_size(track->arrays[i].robj)
2345                                         >> 2);
2346                                 return -EINVAL;
2347                         }
2348                 }
2349                 break;
2350         case 3:
2351                 size = track->vtx_size * nverts;
2352                 if (size != track->immd_dwords) {
2353                         DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2354                                   track->immd_dwords, size);
2355                         DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2356                                   nverts, track->vtx_size);
2357                         return -EINVAL;
2358                 }
2359                 break;
2360         default:
2361                 DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2362                           prim_walk);
2363                 return -EINVAL;
2364         }
2365
2366         if (track->tex_dirty) {
2367                 track->tex_dirty = false;
2368                 return r100_cs_track_texture_check(rdev, track);
2369         }
2370         return 0;
2371 }
2372
2373 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2374 {
2375         unsigned i, face;
2376
2377         track->cb_dirty = true;
2378         track->zb_dirty = true;
2379         track->tex_dirty = true;
2380         track->aa_dirty = true;
2381
2382         if (rdev->family < CHIP_R300) {
2383                 track->num_cb = 1;
2384                 if (rdev->family <= CHIP_RS200)
2385                         track->num_texture = 3;
2386                 else
2387                         track->num_texture = 6;
2388                 track->maxy = 2048;
2389                 track->separate_cube = 1;
2390         } else {
2391                 track->num_cb = 4;
2392                 track->num_texture = 16;
2393                 track->maxy = 4096;
2394                 track->separate_cube = 0;
2395                 track->aaresolve = false;
2396                 track->aa.robj = NULL;
2397         }
2398
2399         for (i = 0; i < track->num_cb; i++) {
2400                 track->cb[i].robj = NULL;
2401                 track->cb[i].pitch = 8192;
2402                 track->cb[i].cpp = 16;
2403                 track->cb[i].offset = 0;
2404         }
2405         track->z_enabled = true;
2406         track->zb.robj = NULL;
2407         track->zb.pitch = 8192;
2408         track->zb.cpp = 4;
2409         track->zb.offset = 0;
2410         track->vtx_size = 0x7F;
2411         track->immd_dwords = 0xFFFFFFFFUL;
2412         track->num_arrays = 11;
2413         track->max_indx = 0x00FFFFFFUL;
2414         for (i = 0; i < track->num_arrays; i++) {
2415                 track->arrays[i].robj = NULL;
2416                 track->arrays[i].esize = 0x7F;
2417         }
2418         for (i = 0; i < track->num_texture; i++) {
2419                 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2420                 track->textures[i].pitch = 16536;
2421                 track->textures[i].width = 16536;
2422                 track->textures[i].height = 16536;
2423                 track->textures[i].width_11 = 1 << 11;
2424                 track->textures[i].height_11 = 1 << 11;
2425                 track->textures[i].num_levels = 12;
2426                 if (rdev->family <= CHIP_RS200) {
2427                         track->textures[i].tex_coord_type = 0;
2428                         track->textures[i].txdepth = 0;
2429                 } else {
2430                         track->textures[i].txdepth = 16;
2431                         track->textures[i].tex_coord_type = 1;
2432                 }
2433                 track->textures[i].cpp = 64;
2434                 track->textures[i].robj = NULL;
2435                 /* CS IB emission code makes sure texture unit are disabled */
2436                 track->textures[i].enabled = false;
2437                 track->textures[i].lookup_disable = false;
2438                 track->textures[i].roundup_w = true;
2439                 track->textures[i].roundup_h = true;
2440                 if (track->separate_cube)
2441                         for (face = 0; face < 5; face++) {
2442                                 track->textures[i].cube_info[face].robj = NULL;
2443                                 track->textures[i].cube_info[face].width = 16536;
2444                                 track->textures[i].cube_info[face].height = 16536;
2445                                 track->textures[i].cube_info[face].offset = 0;
2446                         }
2447         }
2448 }
2449
2450 /*
2451  * Global GPU functions
2452  */
2453 static void r100_errata(struct radeon_device *rdev)
2454 {
2455         rdev->pll_errata = 0;
2456
2457         if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2458                 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2459         }
2460
2461         if (rdev->family == CHIP_RV100 ||
2462             rdev->family == CHIP_RS100 ||
2463             rdev->family == CHIP_RS200) {
2464                 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2465         }
2466 }
2467
2468 static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2469 {
2470         unsigned i;
2471         uint32_t tmp;
2472
2473         for (i = 0; i < rdev->usec_timeout; i++) {
2474                 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2475                 if (tmp >= n) {
2476                         return 0;
2477                 }
2478                 DRM_UDELAY(1);
2479         }
2480         return -1;
2481 }
2482
2483 int r100_gui_wait_for_idle(struct radeon_device *rdev)
2484 {
2485         unsigned i;
2486         uint32_t tmp;
2487
2488         if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2489                 printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
2490                        " Bad things might happen.\n");
2491         }
2492         for (i = 0; i < rdev->usec_timeout; i++) {
2493                 tmp = RREG32(RADEON_RBBM_STATUS);
2494                 if (!(tmp & RADEON_RBBM_ACTIVE)) {
2495                         return 0;
2496                 }
2497                 DRM_UDELAY(1);
2498         }
2499         return -1;
2500 }
2501
2502 int r100_mc_wait_for_idle(struct radeon_device *rdev)
2503 {
2504         unsigned i;
2505         uint32_t tmp;
2506
2507         for (i = 0; i < rdev->usec_timeout; i++) {
2508                 /* read MC_STATUS */
2509                 tmp = RREG32(RADEON_MC_STATUS);
2510                 if (tmp & RADEON_MC_IDLE) {
2511                         return 0;
2512                 }
2513                 DRM_UDELAY(1);
2514         }
2515         return -1;
2516 }
2517
2518 bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2519 {
2520         u32 rbbm_status;
2521
2522         rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2523         if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2524                 radeon_ring_lockup_update(ring);
2525                 return false;
2526         }
2527         /* force CP activities */
2528         radeon_ring_force_activity(rdev, ring);
2529         return radeon_ring_test_lockup(rdev, ring);
2530 }
2531
2532 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2533 void r100_enable_bm(struct radeon_device *rdev)
2534 {
2535         uint32_t tmp;
2536         /* Enable bus mastering */
2537         tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2538         WREG32(RADEON_BUS_CNTL, tmp);
2539 }
2540
2541 void r100_bm_disable(struct radeon_device *rdev)
2542 {
2543         u32 tmp;
2544
2545         /* disable bus mastering */
2546         tmp = RREG32(R_000030_BUS_CNTL);
2547         WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2548         mdelay(1);
2549         WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2550         mdelay(1);
2551         WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2552         tmp = RREG32(RADEON_BUS_CNTL);
2553         mdelay(1);
2554         pci_clear_master(rdev->pdev);
2555         mdelay(1);
2556 }
2557
2558 int r100_asic_reset(struct radeon_device *rdev)
2559 {
2560         struct r100_mc_save save;
2561         u32 status, tmp;
2562         int ret = 0;
2563
2564         status = RREG32(R_000E40_RBBM_STATUS);
2565         if (!G_000E40_GUI_ACTIVE(status)) {
2566                 return 0;
2567         }
2568         r100_mc_stop(rdev, &save);
2569         status = RREG32(R_000E40_RBBM_STATUS);
2570         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2571         /* stop CP */
2572         WREG32(RADEON_CP_CSQ_CNTL, 0);
2573         tmp = RREG32(RADEON_CP_RB_CNTL);
2574         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2575         WREG32(RADEON_CP_RB_RPTR_WR, 0);
2576         WREG32(RADEON_CP_RB_WPTR, 0);
2577         WREG32(RADEON_CP_RB_CNTL, tmp);
2578         /* save PCI state */
2579         pci_save_state(rdev->pdev);
2580         /* disable bus mastering */
2581         r100_bm_disable(rdev);
2582         WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2583                                         S_0000F0_SOFT_RESET_RE(1) |
2584                                         S_0000F0_SOFT_RESET_PP(1) |
2585                                         S_0000F0_SOFT_RESET_RB(1));
2586         RREG32(R_0000F0_RBBM_SOFT_RESET);
2587         mdelay(500);
2588         WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2589         mdelay(1);
2590         status = RREG32(R_000E40_RBBM_STATUS);
2591         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2592         /* reset CP */
2593         WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2594         RREG32(R_0000F0_RBBM_SOFT_RESET);
2595         mdelay(500);
2596         WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2597         mdelay(1);
2598         status = RREG32(R_000E40_RBBM_STATUS);
2599         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2600         /* restore PCI & busmastering */
2601         pci_restore_state(rdev->pdev);
2602         r100_enable_bm(rdev);
2603         /* Check if GPU is idle */
2604         if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2605                 G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2606                 dev_err(rdev->dev, "failed to reset GPU\n");
2607                 ret = -1;
2608         } else
2609                 dev_info(rdev->dev, "GPU reset succeed\n");
2610         r100_mc_resume(rdev, &save);
2611         return ret;
2612 }
2613
2614 void r100_set_common_regs(struct radeon_device *rdev)
2615 {
2616         struct drm_device *dev = rdev->ddev;
2617         bool force_dac2 = false;
2618         u32 tmp;
2619
2620         /* set these so they don't interfere with anything */
2621         WREG32(RADEON_OV0_SCALE_CNTL, 0);
2622         WREG32(RADEON_SUBPIC_CNTL, 0);
2623         WREG32(RADEON_VIPH_CONTROL, 0);
2624         WREG32(RADEON_I2C_CNTL_1, 0);
2625         WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2626         WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2627         WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2628
2629         /* always set up dac2 on rn50 and some rv100 as lots
2630          * of servers seem to wire it up to a VGA port but
2631          * don't report it in the bios connector
2632          * table.
2633          */
2634         switch (dev->pdev->device) {
2635                 /* RN50 */
2636         case 0x515e:
2637         case 0x5969:
2638                 force_dac2 = true;
2639                 break;
2640                 /* RV100*/
2641         case 0x5159:
2642         case 0x515a:
2643                 /* DELL triple head servers */
2644                 if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2645                     ((dev->pdev->subsystem_device == 0x016c) ||
2646                      (dev->pdev->subsystem_device == 0x016d) ||
2647                      (dev->pdev->subsystem_device == 0x016e) ||
2648                      (dev->pdev->subsystem_device == 0x016f) ||
2649                      (dev->pdev->subsystem_device == 0x0170) ||
2650                      (dev->pdev->subsystem_device == 0x017d) ||
2651                      (dev->pdev->subsystem_device == 0x017e) ||
2652                      (dev->pdev->subsystem_device == 0x0183) ||
2653                      (dev->pdev->subsystem_device == 0x018a) ||
2654                      (dev->pdev->subsystem_device == 0x019a)))
2655                         force_dac2 = true;
2656                 break;
2657         }
2658
2659         if (force_dac2) {
2660                 u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2661                 u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2662                 u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2663
2664                 /* For CRT on DAC2, don't turn it on if BIOS didn't
2665                    enable it, even it's detected.
2666                 */
2667
2668                 /* force it to crtc0 */
2669                 dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2670                 dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2671                 disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2672
2673                 /* set up the TV DAC */
2674                 tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2675                                  RADEON_TV_DAC_STD_MASK |
2676                                  RADEON_TV_DAC_RDACPD |
2677                                  RADEON_TV_DAC_GDACPD |
2678                                  RADEON_TV_DAC_BDACPD |
2679                                  RADEON_TV_DAC_BGADJ_MASK |
2680                                  RADEON_TV_DAC_DACADJ_MASK);
2681                 tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2682                                 RADEON_TV_DAC_NHOLD |
2683                                 RADEON_TV_DAC_STD_PS2 |
2684                                 (0x58 << 16));
2685
2686                 WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2687                 WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2688                 WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2689         }
2690
2691         /* switch PM block to ACPI mode */
2692         tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2693         tmp &= ~RADEON_PM_MODE_SEL;
2694         WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2695
2696 }
2697
2698 /*
2699  * VRAM info
2700  */
2701 static void r100_vram_get_type(struct radeon_device *rdev)
2702 {
2703         uint32_t tmp;
2704
2705         rdev->mc.vram_is_ddr = false;
2706         if (rdev->flags & RADEON_IS_IGP)
2707                 rdev->mc.vram_is_ddr = true;
2708         else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2709                 rdev->mc.vram_is_ddr = true;
2710         if ((rdev->family == CHIP_RV100) ||
2711             (rdev->family == CHIP_RS100) ||
2712             (rdev->family == CHIP_RS200)) {
2713                 tmp = RREG32(RADEON_MEM_CNTL);
2714                 if (tmp & RV100_HALF_MODE) {
2715                         rdev->mc.vram_width = 32;
2716                 } else {
2717                         rdev->mc.vram_width = 64;
2718                 }
2719                 if (rdev->flags & RADEON_SINGLE_CRTC) {
2720                         rdev->mc.vram_width /= 4;
2721                         rdev->mc.vram_is_ddr = true;
2722                 }
2723         } else if (rdev->family <= CHIP_RV280) {
2724                 tmp = RREG32(RADEON_MEM_CNTL);
2725                 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2726                         rdev->mc.vram_width = 128;
2727                 } else {
2728                         rdev->mc.vram_width = 64;
2729                 }
2730         } else {
2731                 /* newer IGPs */
2732                 rdev->mc.vram_width = 128;
2733         }
2734 }
2735
2736 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2737 {
2738         u32 aper_size;
2739         u8 byte;
2740
2741         aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2742
2743         /* Set HDP_APER_CNTL only on cards that are known not to be broken,
2744          * that is has the 2nd generation multifunction PCI interface
2745          */
2746         if (rdev->family == CHIP_RV280 ||
2747             rdev->family >= CHIP_RV350) {
2748                 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2749                        ~RADEON_HDP_APER_CNTL);
2750                 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2751                 return aper_size * 2;
2752         }
2753
2754         /* Older cards have all sorts of funny issues to deal with. First
2755          * check if it's a multifunction card by reading the PCI config
2756          * header type... Limit those to one aperture size
2757          */
2758         pci_read_config_byte(rdev->pdev, 0xe, &byte);
2759         if (byte & 0x80) {
2760                 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2761                 DRM_INFO("Limiting VRAM to one aperture\n");
2762                 return aper_size;
2763         }
2764
2765         /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2766          * have set it up. We don't write this as it's broken on some ASICs but
2767          * we expect the BIOS to have done the right thing (might be too optimistic...)
2768          */
2769         if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2770                 return aper_size * 2;
2771         return aper_size;
2772 }
2773
2774 void r100_vram_init_sizes(struct radeon_device *rdev)
2775 {
2776         u64 config_aper_size;
2777
2778         /* work out accessible VRAM */
2779         rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2780         rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2781         rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2782         /* FIXME we don't use the second aperture yet when we could use it */
2783         if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2784                 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2785         config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2786         if (rdev->flags & RADEON_IS_IGP) {
2787                 uint32_t tom;
2788                 /* read NB_TOM to get the amount of ram stolen for the GPU */
2789                 tom = RREG32(RADEON_NB_TOM);
2790                 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2791                 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2792                 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2793         } else {
2794                 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2795                 /* Some production boards of m6 will report 0
2796                  * if it's 8 MB
2797                  */
2798                 if (rdev->mc.real_vram_size == 0) {
2799                         rdev->mc.real_vram_size = 8192 * 1024;
2800                         WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2801                 }
2802                 /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2803                  * Novell bug 204882 + along with lots of ubuntu ones
2804                  */
2805                 if (rdev->mc.aper_size > config_aper_size)
2806                         config_aper_size = rdev->mc.aper_size;
2807
2808                 if (config_aper_size > rdev->mc.real_vram_size)
2809                         rdev->mc.mc_vram_size = config_aper_size;
2810                 else
2811                         rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2812         }
2813 }
2814
2815 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2816 {
2817         uint32_t temp;
2818
2819         temp = RREG32(RADEON_CONFIG_CNTL);
2820         if (state == false) {
2821                 temp &= ~RADEON_CFG_VGA_RAM_EN;
2822                 temp |= RADEON_CFG_VGA_IO_DIS;
2823         } else {
2824                 temp &= ~RADEON_CFG_VGA_IO_DIS;
2825         }
2826         WREG32(RADEON_CONFIG_CNTL, temp);
2827 }
2828
2829 static void r100_mc_init(struct radeon_device *rdev)
2830 {
2831         u64 base;
2832
2833         r100_vram_get_type(rdev);
2834         r100_vram_init_sizes(rdev);
2835         base = rdev->mc.aper_base;
2836         if (rdev->flags & RADEON_IS_IGP)
2837                 base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2838         radeon_vram_location(rdev, &rdev->mc, base);
2839         rdev->mc.gtt_base_align = 0;
2840         if (!(rdev->flags & RADEON_IS_AGP))
2841                 radeon_gtt_location(rdev, &rdev->mc);
2842         radeon_update_bandwidth_info(rdev);
2843 }
2844
2845
2846 /*
2847  * Indirect registers accessor
2848  */
2849 void r100_pll_errata_after_index(struct radeon_device *rdev)
2850 {
2851         if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2852                 (void)RREG32(RADEON_CLOCK_CNTL_DATA);
2853                 (void)RREG32(RADEON_CRTC_GEN_CNTL);
2854         }
2855 }
2856
2857 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2858 {
2859         /* This workarounds is necessary on RV100, RS100 and RS200 chips
2860          * or the chip could hang on a subsequent access
2861          */
2862         if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2863                 mdelay(5);
2864         }
2865
2866         /* This function is required to workaround a hardware bug in some (all?)
2867          * revisions of the R300.  This workaround should be called after every
2868          * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2869          * may not be correct.
2870          */
2871         if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2872                 uint32_t save, tmp;
2873
2874                 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2875                 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2876                 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2877                 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2878                 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2879         }
2880 }
2881
2882 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2883 {
2884         unsigned long flags;
2885         uint32_t data;
2886
2887         spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2888         WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2889         r100_pll_errata_after_index(rdev);
2890         data = RREG32(RADEON_CLOCK_CNTL_DATA);
2891         r100_pll_errata_after_data(rdev);
2892         spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2893         return data;
2894 }
2895
2896 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2897 {
2898         unsigned long flags;
2899
2900         spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2901         WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2902         r100_pll_errata_after_index(rdev);
2903         WREG32(RADEON_CLOCK_CNTL_DATA, v);
2904         r100_pll_errata_after_data(rdev);
2905         spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2906 }
2907
2908 static void r100_set_safe_registers(struct radeon_device *rdev)
2909 {
2910         if (ASIC_IS_RN50(rdev)) {
2911                 rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2912                 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2913         } else if (rdev->family < CHIP_R200) {
2914                 rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2915                 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2916         } else {
2917                 r200_set_safe_registers(rdev);
2918         }
2919 }
2920
2921 /*
2922  * Debugfs info
2923  */
2924 #if defined(CONFIG_DEBUG_FS)
2925 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2926 {
2927         struct drm_info_node *node = (struct drm_info_node *) m->private;
2928         struct drm_device *dev = node->minor->dev;
2929         struct radeon_device *rdev = dev->dev_private;
2930         uint32_t reg, value;
2931         unsigned i;
2932
2933         seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2934         seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2935         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2936         for (i = 0; i < 64; i++) {
2937                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2938                 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2939                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2940                 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2941                 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2942         }
2943         return 0;
2944 }
2945
2946 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2947 {
2948         struct drm_info_node *node = (struct drm_info_node *) m->private;
2949         struct drm_device *dev = node->minor->dev;
2950         struct radeon_device *rdev = dev->dev_private;
2951         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2952         uint32_t rdp, wdp;
2953         unsigned count, i, j;
2954
2955         radeon_ring_free_size(rdev, ring);
2956         rdp = RREG32(RADEON_CP_RB_RPTR);
2957         wdp = RREG32(RADEON_CP_RB_WPTR);
2958         count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2959         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2960         seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2961         seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2962         seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2963         seq_printf(m, "%u dwords in ring\n", count);
2964         if (ring->ready) {
2965                 for (j = 0; j <= count; j++) {
2966                         i = (rdp + j) & ring->ptr_mask;
2967                         seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2968                 }
2969         }
2970         return 0;
2971 }
2972
2973
2974 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2975 {
2976         struct drm_info_node *node = (struct drm_info_node *) m->private;
2977         struct drm_device *dev = node->minor->dev;
2978         struct radeon_device *rdev = dev->dev_private;
2979         uint32_t csq_stat, csq2_stat, tmp;
2980         unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2981         unsigned i;
2982
2983         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2984         seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2985         csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2986         csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2987         r_rptr = (csq_stat >> 0) & 0x3ff;
2988         r_wptr = (csq_stat >> 10) & 0x3ff;
2989         ib1_rptr = (csq_stat >> 20) & 0x3ff;
2990         ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2991         ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2992         ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2993         seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2994         seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2995         seq_printf(m, "Ring rptr %u\n", r_rptr);
2996         seq_printf(m, "Ring wptr %u\n", r_wptr);
2997         seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2998         seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2999         seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
3000         seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
3001         /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
3002          * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
3003         seq_printf(m, "Ring fifo:\n");
3004         for (i = 0; i < 256; i++) {
3005                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3006                 tmp = RREG32(RADEON_CP_CSQ_DATA);
3007                 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3008         }
3009         seq_printf(m, "Indirect1 fifo:\n");
3010         for (i = 256; i <= 512; i++) {
3011                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3012                 tmp = RREG32(RADEON_CP_CSQ_DATA);
3013                 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3014         }
3015         seq_printf(m, "Indirect2 fifo:\n");
3016         for (i = 640; i < ib1_wptr; i++) {
3017                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3018                 tmp = RREG32(RADEON_CP_CSQ_DATA);
3019                 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3020         }
3021         return 0;
3022 }
3023
3024 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
3025 {
3026         struct drm_info_node *node = (struct drm_info_node *) m->private;
3027         struct drm_device *dev = node->minor->dev;
3028         struct radeon_device *rdev = dev->dev_private;
3029         uint32_t tmp;
3030
3031         tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3032         seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3033         tmp = RREG32(RADEON_MC_FB_LOCATION);
3034         seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3035         tmp = RREG32(RADEON_BUS_CNTL);
3036         seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3037         tmp = RREG32(RADEON_MC_AGP_LOCATION);
3038         seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3039         tmp = RREG32(RADEON_AGP_BASE);
3040         seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3041         tmp = RREG32(RADEON_HOST_PATH_CNTL);
3042         seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3043         tmp = RREG32(0x01D0);
3044         seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3045         tmp = RREG32(RADEON_AIC_LO_ADDR);
3046         seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3047         tmp = RREG32(RADEON_AIC_HI_ADDR);
3048         seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3049         tmp = RREG32(0x01E4);
3050         seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3051         return 0;
3052 }
3053
3054 static struct drm_info_list r100_debugfs_rbbm_list[] = {
3055         {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3056 };
3057
3058 static struct drm_info_list r100_debugfs_cp_list[] = {
3059         {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3060         {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3061 };
3062
3063 static struct drm_info_list r100_debugfs_mc_info_list[] = {
3064         {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3065 };
3066 #endif
3067
3068 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3069 {
3070 #if defined(CONFIG_DEBUG_FS)
3071         return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3072 #else
3073         return 0;
3074 #endif
3075 }
3076
3077 int r100_debugfs_cp_init(struct radeon_device *rdev)
3078 {
3079 #if defined(CONFIG_DEBUG_FS)
3080         return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3081 #else
3082         return 0;
3083 #endif
3084 }
3085
3086 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3087 {
3088 #if defined(CONFIG_DEBUG_FS)
3089         return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3090 #else
3091         return 0;
3092 #endif
3093 }
3094
3095 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3096                          uint32_t tiling_flags, uint32_t pitch,
3097                          uint32_t offset, uint32_t obj_size)
3098 {
3099         int surf_index = reg * 16;
3100         int flags = 0;
3101
3102         if (rdev->family <= CHIP_RS200) {
3103                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3104                                  == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3105                         flags |= RADEON_SURF_TILE_COLOR_BOTH;
3106                 if (tiling_flags & RADEON_TILING_MACRO)
3107                         flags |= RADEON_SURF_TILE_COLOR_MACRO;
3108                 /* setting pitch to 0 disables tiling */
3109                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3110                                 == 0)
3111                         pitch = 0;
3112         } else if (rdev->family <= CHIP_RV280) {
3113                 if (tiling_flags & (RADEON_TILING_MACRO))
3114                         flags |= R200_SURF_TILE_COLOR_MACRO;
3115                 if (tiling_flags & RADEON_TILING_MICRO)
3116                         flags |= R200_SURF_TILE_COLOR_MICRO;
3117         } else {
3118                 if (tiling_flags & RADEON_TILING_MACRO)
3119                         flags |= R300_SURF_TILE_MACRO;
3120                 if (tiling_flags & RADEON_TILING_MICRO)
3121                         flags |= R300_SURF_TILE_MICRO;
3122         }
3123
3124         if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3125                 flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3126         if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3127                 flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3128
3129         /* r100/r200 divide by 16 */
3130         if (rdev->family < CHIP_R300)
3131                 flags |= pitch / 16;
3132         else
3133                 flags |= pitch / 8;
3134
3135
3136         DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3137         WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3138         WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3139         WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3140         return 0;
3141 }
3142
3143 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3144 {
3145         int surf_index = reg * 16;
3146         WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3147 }
3148
3149 void r100_bandwidth_update(struct radeon_device *rdev)
3150 {
3151         fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3152         fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3153         fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
3154         uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3155         fixed20_12 memtcas_ff[8] = {
3156                 dfixed_init(1),
3157                 dfixed_init(2),
3158                 dfixed_init(3),
3159                 dfixed_init(0),
3160                 dfixed_init_half(1),
3161                 dfixed_init_half(2),
3162                 dfixed_init(0),
3163         };
3164         fixed20_12 memtcas_rs480_ff[8] = {
3165                 dfixed_init(0),
3166                 dfixed_init(1),
3167                 dfixed_init(2),
3168                 dfixed_init(3),
3169                 dfixed_init(0),
3170                 dfixed_init_half(1),
3171                 dfixed_init_half(2),
3172                 dfixed_init_half(3),
3173         };
3174         fixed20_12 memtcas2_ff[8] = {
3175                 dfixed_init(0),
3176                 dfixed_init(1),
3177                 dfixed_init(2),
3178                 dfixed_init(3),
3179                 dfixed_init(4),
3180                 dfixed_init(5),
3181                 dfixed_init(6),
3182                 dfixed_init(7),
3183         };
3184         fixed20_12 memtrbs[8] = {
3185                 dfixed_init(1),
3186                 dfixed_init_half(1),
3187                 dfixed_init(2),
3188                 dfixed_init_half(2),
3189                 dfixed_init(3),
3190                 dfixed_init_half(3),
3191                 dfixed_init(4),
3192                 dfixed_init_half(4)
3193         };
3194         fixed20_12 memtrbs_r4xx[8] = {
3195                 dfixed_init(4),
3196                 dfixed_init(5),
3197                 dfixed_init(6),
3198                 dfixed_init(7),
3199                 dfixed_init(8),
3200                 dfixed_init(9),
3201                 dfixed_init(10),
3202                 dfixed_init(11)
3203         };
3204         fixed20_12 min_mem_eff;
3205         fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3206         fixed20_12 cur_latency_mclk, cur_latency_sclk;
3207         fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
3208                 disp_drain_rate2, read_return_rate;
3209         fixed20_12 time_disp1_drop_priority;
3210         int c;
3211         int cur_size = 16;       /* in octawords */
3212         int critical_point = 0, critical_point2;
3213 /*      uint32_t read_return_rate, time_disp1_drop_priority; */
3214         int stop_req, max_stop_req;
3215         struct drm_display_mode *mode1 = NULL;
3216         struct drm_display_mode *mode2 = NULL;
3217         uint32_t pixel_bytes1 = 0;
3218         uint32_t pixel_bytes2 = 0;
3219
3220         if (!rdev->mode_info.mode_config_initialized)
3221                 return;
3222
3223         radeon_update_display_priority(rdev);
3224
3225         if (rdev->mode_info.crtcs[0]->base.enabled) {
3226                 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3227                 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
3228         }
3229         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3230                 if (rdev->mode_info.crtcs[1]->base.enabled) {
3231                         mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3232                         pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
3233                 }
3234         }
3235
3236         min_mem_eff.full = dfixed_const_8(0);
3237         /* get modes */
3238         if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3239                 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3240                 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3241                 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3242                 /* check crtc enables */
3243                 if (mode2)
3244                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3245                 if (mode1)
3246                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3247                 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3248         }
3249
3250         /*
3251          * determine is there is enough bw for current mode
3252          */
3253         sclk_ff = rdev->pm.sclk;
3254         mclk_ff = rdev->pm.mclk;
3255
3256         temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3257         temp_ff.full = dfixed_const(temp);
3258         mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3259
3260         pix_clk.full = 0;
3261         pix_clk2.full = 0;
3262         peak_disp_bw.full = 0;
3263         if (mode1) {
3264                 temp_ff.full = dfixed_const(1000);
3265                 pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3266                 pix_clk.full = dfixed_div(pix_clk, temp_ff);
3267                 temp_ff.full = dfixed_const(pixel_bytes1);
3268                 peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3269         }
3270         if (mode2) {
3271                 temp_ff.full = dfixed_const(1000);
3272                 pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3273                 pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3274                 temp_ff.full = dfixed_const(pixel_bytes2);
3275                 peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3276         }
3277
3278         mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3279         if (peak_disp_bw.full >= mem_bw.full) {
3280                 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3281                           "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3282         }
3283
3284         /*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3285         temp = RREG32(RADEON_MEM_TIMING_CNTL);
3286         if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3287                 mem_trcd = ((temp >> 2) & 0x3) + 1;
3288                 mem_trp  = ((temp & 0x3)) + 1;
3289                 mem_tras = ((temp & 0x70) >> 4) + 1;
3290         } else if (rdev->family == CHIP_R300 ||
3291                    rdev->family == CHIP_R350) { /* r300, r350 */
3292                 mem_trcd = (temp & 0x7) + 1;
3293                 mem_trp = ((temp >> 8) & 0x7) + 1;
3294                 mem_tras = ((temp >> 11) & 0xf) + 4;
3295         } else if (rdev->family == CHIP_RV350 ||
3296                    rdev->family <= CHIP_RV380) {
3297                 /* rv3x0 */
3298                 mem_trcd = (temp & 0x7) + 3;
3299                 mem_trp = ((temp >> 8) & 0x7) + 3;
3300                 mem_tras = ((temp >> 11) & 0xf) + 6;
3301         } else if (rdev->family == CHIP_R420 ||
3302                    rdev->family == CHIP_R423 ||
3303                    rdev->family == CHIP_RV410) {
3304                 /* r4xx */
3305                 mem_trcd = (temp & 0xf) + 3;
3306                 if (mem_trcd > 15)
3307                         mem_trcd = 15;
3308                 mem_trp = ((temp >> 8) & 0xf) + 3;
3309                 if (mem_trp > 15)
3310                         mem_trp = 15;
3311                 mem_tras = ((temp >> 12) & 0x1f) + 6;
3312                 if (mem_tras > 31)
3313                         mem_tras = 31;
3314         } else { /* RV200, R200 */
3315                 mem_trcd = (temp & 0x7) + 1;
3316                 mem_trp = ((temp >> 8) & 0x7) + 1;
3317                 mem_tras = ((temp >> 12) & 0xf) + 4;
3318         }
3319         /* convert to FF */
3320         trcd_ff.full = dfixed_const(mem_trcd);
3321         trp_ff.full = dfixed_const(mem_trp);
3322         tras_ff.full = dfixed_const(mem_tras);
3323
3324         /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3325         temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3326         data = (temp & (7 << 20)) >> 20;
3327         if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3328                 if (rdev->family == CHIP_RS480) /* don't think rs400 */
3329                         tcas_ff = memtcas_rs480_ff[data];
3330                 else
3331                         tcas_ff = memtcas_ff[data];
3332         } else
3333                 tcas_ff = memtcas2_ff[data];
3334
3335         if (rdev->family == CHIP_RS400 ||
3336             rdev->family == CHIP_RS480) {
3337                 /* extra cas latency stored in bits 23-25 0-4 clocks */
3338                 data = (temp >> 23) & 0x7;
3339                 if (data < 5)
3340                         tcas_ff.full += dfixed_const(data);
3341         }
3342
3343         if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3344                 /* on the R300, Tcas is included in Trbs.
3345                  */
3346                 temp = RREG32(RADEON_MEM_CNTL);
3347                 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3348                 if (data == 1) {
3349                         if (R300_MEM_USE_CD_CH_ONLY & temp) {
3350                                 temp = RREG32(R300_MC_IND_INDEX);
3351                                 temp &= ~R300_MC_IND_ADDR_MASK;
3352                                 temp |= R300_MC_READ_CNTL_CD_mcind;
3353                                 WREG32(R300_MC_IND_INDEX, temp);
3354                                 temp = RREG32(R300_MC_IND_DATA);
3355                                 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3356                         } else {
3357                                 temp = RREG32(R300_MC_READ_CNTL_AB);
3358                                 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3359                         }
3360                 } else {
3361                         temp = RREG32(R300_MC_READ_CNTL_AB);
3362                         data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3363                 }
3364                 if (rdev->family == CHIP_RV410 ||
3365                     rdev->family == CHIP_R420 ||
3366                     rdev->family == CHIP_R423)
3367                         trbs_ff = memtrbs_r4xx[data];
3368                 else
3369                         trbs_ff = memtrbs[data];
3370                 tcas_ff.full += trbs_ff.full;
3371         }
3372
3373         sclk_eff_ff.full = sclk_ff.full;
3374
3375         if (rdev->flags & RADEON_IS_AGP) {
3376                 fixed20_12 agpmode_ff;
3377                 agpmode_ff.full = dfixed_const(radeon_agpmode);
3378                 temp_ff.full = dfixed_const_666(16);
3379                 sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3380         }
3381         /* TODO PCIE lanes may affect this - agpmode == 16?? */
3382
3383         if (ASIC_IS_R300(rdev)) {
3384                 sclk_delay_ff.full = dfixed_const(250);
3385         } else {
3386                 if ((rdev->family == CHIP_RV100) ||
3387                     rdev->flags & RADEON_IS_IGP) {
3388                         if (rdev->mc.vram_is_ddr)
3389                                 sclk_delay_ff.full = dfixed_const(41);
3390                         else
3391                                 sclk_delay_ff.full = dfixed_const(33);
3392                 } else {
3393                         if (rdev->mc.vram_width == 128)
3394                                 sclk_delay_ff.full = dfixed_const(57);
3395                         else
3396                                 sclk_delay_ff.full = dfixed_const(41);
3397                 }
3398         }
3399
3400         mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3401
3402         if (rdev->mc.vram_is_ddr) {
3403                 if (rdev->mc.vram_width == 32) {
3404                         k1.full = dfixed_const(40);
3405                         c  = 3;
3406                 } else {
3407                         k1.full = dfixed_const(20);
3408                         c  = 1;
3409                 }
3410         } else {
3411                 k1.full = dfixed_const(40);
3412                 c  = 3;
3413         }
3414
3415         temp_ff.full = dfixed_const(2);
3416         mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3417         temp_ff.full = dfixed_const(c);
3418         mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3419         temp_ff.full = dfixed_const(4);
3420         mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3421         mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3422         mc_latency_mclk.full += k1.full;
3423
3424         mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3425         mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3426
3427         /*
3428           HW cursor time assuming worst case of full size colour cursor.
3429         */
3430         temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3431         temp_ff.full += trcd_ff.full;
3432         if (temp_ff.full < tras_ff.full)
3433                 temp_ff.full = tras_ff.full;
3434         cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3435
3436         temp_ff.full = dfixed_const(cur_size);
3437         cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3438         /*
3439           Find the total latency for the display data.
3440         */
3441         disp_latency_overhead.full = dfixed_const(8);
3442         disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3443         mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3444         mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3445
3446         if (mc_latency_mclk.full > mc_latency_sclk.full)
3447                 disp_latency.full = mc_latency_mclk.full;
3448         else
3449                 disp_latency.full = mc_latency_sclk.full;
3450
3451         /* setup Max GRPH_STOP_REQ default value */
3452         if (ASIC_IS_RV100(rdev))
3453                 max_stop_req = 0x5c;
3454         else
3455                 max_stop_req = 0x7c;
3456
3457         if (mode1) {
3458                 /*  CRTC1
3459                     Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3460                     GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3461                 */
3462                 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3463
3464                 if (stop_req > max_stop_req)
3465                         stop_req = max_stop_req;
3466
3467                 /*
3468                   Find the drain rate of the display buffer.
3469                 */
3470                 temp_ff.full = dfixed_const((16/pixel_bytes1));
3471                 disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3472
3473                 /*
3474                   Find the critical point of the display buffer.
3475                 */
3476                 crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3477                 crit_point_ff.full += dfixed_const_half(0);
3478
3479                 critical_point = dfixed_trunc(crit_point_ff);
3480
3481                 if (rdev->disp_priority == 2) {
3482                         critical_point = 0;
3483                 }
3484
3485                 /*
3486                   The critical point should never be above max_stop_req-4.  Setting
3487                   GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3488                 */
3489                 if (max_stop_req - critical_point < 4)
3490                         critical_point = 0;
3491
3492                 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3493                         /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3494                         critical_point = 0x10;
3495                 }
3496
3497                 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3498                 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3499                 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3500                 temp &= ~(RADEON_GRPH_START_REQ_MASK);
3501                 if ((rdev->family == CHIP_R350) &&
3502                     (stop_req > 0x15)) {
3503                         stop_req -= 0x10;
3504                 }
3505                 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3506                 temp |= RADEON_GRPH_BUFFER_SIZE;
3507                 temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3508                           RADEON_GRPH_CRITICAL_AT_SOF |
3509                           RADEON_GRPH_STOP_CNTL);
3510                 /*
3511                   Write the result into the register.
3512                 */
3513                 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3514                                                        (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3515
3516 #if 0
3517                 if ((rdev->family == CHIP_RS400) ||
3518                     (rdev->family == CHIP_RS480)) {
3519                         /* attempt to program RS400 disp regs correctly ??? */
3520                         temp = RREG32(RS400_DISP1_REG_CNTL);
3521                         temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3522                                   RS400_DISP1_STOP_REQ_LEVEL_MASK);
3523                         WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3524                                                        (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3525                                                        (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3526                         temp = RREG32(RS400_DMIF_MEM_CNTL1);
3527                         temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3528                                   RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3529                         WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3530                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3531                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3532                 }
3533 #endif
3534
3535                 DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3536                           /*      (unsigned int)info->SavedReg->grph_buffer_cntl, */
3537                           (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3538         }
3539
3540         if (mode2) {
3541                 u32 grph2_cntl;
3542                 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3543
3544                 if (stop_req > max_stop_req)
3545                         stop_req = max_stop_req;
3546
3547                 /*
3548                   Find the drain rate of the display buffer.
3549                 */
3550                 temp_ff.full = dfixed_const((16/pixel_bytes2));
3551                 disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3552
3553                 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3554                 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3555                 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3556                 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3557                 if ((rdev->family == CHIP_R350) &&
3558                     (stop_req > 0x15)) {
3559                         stop_req -= 0x10;
3560                 }
3561                 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3562                 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3563                 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3564                           RADEON_GRPH_CRITICAL_AT_SOF |
3565                           RADEON_GRPH_STOP_CNTL);
3566
3567                 if ((rdev->family == CHIP_RS100) ||
3568                     (rdev->family == CHIP_RS200))
3569                         critical_point2 = 0;
3570                 else {
3571                         temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3572                         temp_ff.full = dfixed_const(temp);
3573                         temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3574                         if (sclk_ff.full < temp_ff.full)
3575                                 temp_ff.full = sclk_ff.full;
3576
3577                         read_return_rate.full = temp_ff.full;
3578
3579                         if (mode1) {
3580                                 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3581                                 time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3582                         } else {
3583                                 time_disp1_drop_priority.full = 0;
3584                         }
3585                         crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3586                         crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3587                         crit_point_ff.full += dfixed_const_half(0);
3588
3589                         critical_point2 = dfixed_trunc(crit_point_ff);
3590
3591                         if (rdev->disp_priority == 2) {
3592                                 critical_point2 = 0;
3593                         }
3594
3595                         if (max_stop_req - critical_point2 < 4)
3596                                 critical_point2 = 0;
3597
3598                 }
3599
3600                 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3601                         /* some R300 cards have problem with this set to 0 */
3602                         critical_point2 = 0x10;
3603                 }
3604
3605                 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3606                                                   (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3607
3608                 if ((rdev->family == CHIP_RS400) ||
3609                     (rdev->family == CHIP_RS480)) {
3610 #if 0
3611                         /* attempt to program RS400 disp2 regs correctly ??? */
3612                         temp = RREG32(RS400_DISP2_REQ_CNTL1);
3613                         temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3614                                   RS400_DISP2_STOP_REQ_LEVEL_MASK);
3615                         WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3616                                                        (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3617                                                        (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3618                         temp = RREG32(RS400_DISP2_REQ_CNTL2);
3619                         temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3620                                   RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3621                         WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3622                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3623                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3624 #endif
3625                         WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3626                         WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3627                         WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3628                         WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3629                 }
3630
3631                 DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3632                           (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3633         }
3634 }
3635
3636 int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3637 {
3638         uint32_t scratch;
3639         uint32_t tmp = 0;
3640         unsigned i;
3641         int r;
3642
3643         r = radeon_scratch_get(rdev, &scratch);
3644         if (r) {
3645                 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3646                 return r;
3647         }
3648         WREG32(scratch, 0xCAFEDEAD);
3649         r = radeon_ring_lock(rdev, ring, 2);
3650         if (r) {
3651                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3652                 radeon_scratch_free(rdev, scratch);
3653                 return r;
3654         }
3655         radeon_ring_write(ring, PACKET0(scratch, 0));
3656         radeon_ring_write(ring, 0xDEADBEEF);
3657         radeon_ring_unlock_commit(rdev, ring);
3658         for (i = 0; i < rdev->usec_timeout; i++) {
3659                 tmp = RREG32(scratch);
3660                 if (tmp == 0xDEADBEEF) {
3661                         break;
3662                 }
3663                 DRM_UDELAY(1);
3664         }
3665         if (i < rdev->usec_timeout) {
3666                 DRM_INFO("ring test succeeded in %d usecs\n", i);
3667         } else {
3668                 DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3669                           scratch, tmp);
3670                 r = -EINVAL;
3671         }
3672         radeon_scratch_free(rdev, scratch);
3673         return r;
3674 }
3675
3676 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3677 {
3678         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3679
3680         if (ring->rptr_save_reg) {
3681                 u32 next_rptr = ring->wptr + 2 + 3;
3682                 radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3683                 radeon_ring_write(ring, next_rptr);
3684         }
3685
3686         radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3687         radeon_ring_write(ring, ib->gpu_addr);
3688         radeon_ring_write(ring, ib->length_dw);
3689 }
3690
3691 int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3692 {
3693         struct radeon_ib ib;
3694         uint32_t scratch;
3695         uint32_t tmp = 0;
3696         unsigned i;
3697         int r;
3698
3699         r = radeon_scratch_get(rdev, &scratch);
3700         if (r) {
3701                 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3702                 return r;
3703         }
3704         WREG32(scratch, 0xCAFEDEAD);
3705         r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3706         if (r) {
3707                 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3708                 goto free_scratch;
3709         }
3710         ib.ptr[0] = PACKET0(scratch, 0);
3711         ib.ptr[1] = 0xDEADBEEF;
3712         ib.ptr[2] = PACKET2(0);
3713         ib.ptr[3] = PACKET2(0);
3714         ib.ptr[4] = PACKET2(0);
3715         ib.ptr[5] = PACKET2(0);
3716         ib.ptr[6] = PACKET2(0);
3717         ib.ptr[7] = PACKET2(0);
3718         ib.length_dw = 8;
3719         r = radeon_ib_schedule(rdev, &ib, NULL);
3720         if (r) {
3721                 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3722                 goto free_ib;
3723         }
3724         r = radeon_fence_wait(ib.fence, false);
3725         if (r) {
3726                 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3727                 goto free_ib;
3728         }
3729         for (i = 0; i < rdev->usec_timeout; i++) {
3730                 tmp = RREG32(scratch);
3731                 if (tmp == 0xDEADBEEF) {
3732                         break;
3733                 }
3734                 DRM_UDELAY(1);
3735         }
3736         if (i < rdev->usec_timeout) {
3737                 DRM_INFO("ib test succeeded in %u usecs\n", i);
3738         } else {
3739                 DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3740                           scratch, tmp);
3741                 r = -EINVAL;
3742         }
3743 free_ib:
3744         radeon_ib_free(rdev, &ib);
3745 free_scratch:
3746         radeon_scratch_free(rdev, scratch);
3747         return r;
3748 }
3749
3750 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3751 {
3752         /* Shutdown CP we shouldn't need to do that but better be safe than
3753          * sorry
3754          */
3755         rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3756         WREG32(R_000740_CP_CSQ_CNTL, 0);
3757
3758         /* Save few CRTC registers */
3759         save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3760         save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3761         save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3762         save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3763         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3764                 save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3765                 save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3766         }
3767
3768         /* Disable VGA aperture access */
3769         WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3770         /* Disable cursor, overlay, crtc */
3771         WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3772         WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3773                                         S_000054_CRTC_DISPLAY_DIS(1));
3774         WREG32(R_000050_CRTC_GEN_CNTL,
3775                         (C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3776                         S_000050_CRTC_DISP_REQ_EN_B(1));
3777         WREG32(R_000420_OV0_SCALE_CNTL,
3778                 C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3779         WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3780         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3781                 WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3782                                                 S_000360_CUR2_LOCK(1));
3783                 WREG32(R_0003F8_CRTC2_GEN_CNTL,
3784                         (C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3785                         S_0003F8_CRTC2_DISPLAY_DIS(1) |
3786                         S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3787                 WREG32(R_000360_CUR2_OFFSET,
3788                         C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3789         }
3790 }
3791
3792 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3793 {
3794         /* Update base address for crtc */
3795         WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3796         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3797                 WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3798         }
3799         /* Restore CRTC registers */
3800         WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3801         WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3802         WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3803         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3804                 WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3805         }
3806 }
3807
3808 void r100_vga_render_disable(struct radeon_device *rdev)
3809 {
3810         u32 tmp;
3811
3812         tmp = RREG8(R_0003C2_GENMO_WT);
3813         WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3814 }
3815
3816 static void r100_debugfs(struct radeon_device *rdev)
3817 {
3818         int r;
3819
3820         r = r100_debugfs_mc_info_init(rdev);
3821         if (r)
3822                 dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3823 }
3824
3825 static void r100_mc_program(struct radeon_device *rdev)
3826 {
3827         struct r100_mc_save save;
3828
3829         /* Stops all mc clients */
3830         r100_mc_stop(rdev, &save);
3831         if (rdev->flags & RADEON_IS_AGP) {
3832                 WREG32(R_00014C_MC_AGP_LOCATION,
3833                         S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3834                         S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3835                 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3836                 if (rdev->family > CHIP_RV200)
3837                         WREG32(R_00015C_AGP_BASE_2,
3838                                 upper_32_bits(rdev->mc.agp_base) & 0xff);
3839         } else {
3840                 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3841                 WREG32(R_000170_AGP_BASE, 0);
3842                 if (rdev->family > CHIP_RV200)
3843                         WREG32(R_00015C_AGP_BASE_2, 0);
3844         }
3845         /* Wait for mc idle */
3846         if (r100_mc_wait_for_idle(rdev))
3847                 dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3848         /* Program MC, should be a 32bits limited address space */
3849         WREG32(R_000148_MC_FB_LOCATION,
3850                 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3851                 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3852         r100_mc_resume(rdev, &save);
3853 }
3854
3855 static void r100_clock_startup(struct radeon_device *rdev)
3856 {
3857         u32 tmp;
3858
3859         if (radeon_dynclks != -1 && radeon_dynclks)
3860                 radeon_legacy_set_clock_gating(rdev, 1);
3861         /* We need to force on some of the block */
3862         tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3863         tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3864         if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3865                 tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3866         WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3867 }
3868
3869 static int r100_startup(struct radeon_device *rdev)
3870 {
3871         int r;
3872
3873         /* set common regs */
3874         r100_set_common_regs(rdev);
3875         /* program mc */
3876         r100_mc_program(rdev);
3877         /* Resume clock */
3878         r100_clock_startup(rdev);
3879         /* Initialize GART (initialize after TTM so we can allocate
3880          * memory through TTM but finalize after TTM) */
3881         r100_enable_bm(rdev);
3882         if (rdev->flags & RADEON_IS_PCI) {
3883                 r = r100_pci_gart_enable(rdev);
3884                 if (r)
3885                         return r;
3886         }
3887
3888         /* allocate wb buffer */
3889         r = radeon_wb_init(rdev);
3890         if (r)
3891                 return r;
3892
3893         r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3894         if (r) {
3895                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3896                 return r;
3897         }
3898
3899         /* Enable IRQ */
3900         if (!rdev->irq.installed) {
3901                 r = radeon_irq_kms_init(rdev);
3902                 if (r)
3903                         return r;
3904         }
3905
3906         r100_irq_set(rdev);
3907         rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3908         /* 1M ring buffer */
3909         r = r100_cp_init(rdev, 1024 * 1024);
3910         if (r) {
3911                 dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3912                 return r;
3913         }
3914
3915         r = radeon_ib_pool_init(rdev);
3916         if (r) {
3917                 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3918                 return r;
3919         }
3920
3921         return 0;
3922 }
3923
3924 int r100_resume(struct radeon_device *rdev)
3925 {
3926         int r;
3927
3928         /* Make sur GART are not working */
3929         if (rdev->flags & RADEON_IS_PCI)
3930                 r100_pci_gart_disable(rdev);
3931         /* Resume clock before doing reset */
3932         r100_clock_startup(rdev);
3933         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
3934         if (radeon_asic_reset(rdev)) {
3935                 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3936                         RREG32(R_000E40_RBBM_STATUS),
3937                         RREG32(R_0007C0_CP_STAT));
3938         }
3939         /* post */
3940         radeon_combios_asic_init(rdev->ddev);
3941         /* Resume clock after posting */
3942         r100_clock_startup(rdev);
3943         /* Initialize surface registers */
3944         radeon_surface_init(rdev);
3945
3946         rdev->accel_working = true;
3947         r = r100_startup(rdev);
3948         if (r) {
3949                 rdev->accel_working = false;
3950         }
3951         return r;
3952 }
3953
3954 int r100_suspend(struct radeon_device *rdev)
3955 {
3956         radeon_pm_suspend(rdev);
3957         r100_cp_disable(rdev);
3958         radeon_wb_disable(rdev);
3959         r100_irq_disable(rdev);
3960         if (rdev->flags & RADEON_IS_PCI)
3961                 r100_pci_gart_disable(rdev);
3962         return 0;
3963 }
3964
3965 void r100_fini(struct radeon_device *rdev)
3966 {
3967         radeon_pm_fini(rdev);
3968         r100_cp_fini(rdev);
3969         radeon_wb_fini(rdev);
3970         radeon_ib_pool_fini(rdev);
3971         radeon_gem_fini(rdev);
3972         if (rdev->flags & RADEON_IS_PCI)
3973                 r100_pci_gart_fini(rdev);
3974         radeon_agp_fini(rdev);
3975         radeon_irq_kms_fini(rdev);
3976         radeon_fence_driver_fini(rdev);
3977         radeon_bo_fini(rdev);
3978         radeon_atombios_fini(rdev);
3979         kfree(rdev->bios);
3980         rdev->bios = NULL;
3981 }
3982
3983 /*
3984  * Due to how kexec works, it can leave the hw fully initialised when it
3985  * boots the new kernel. However doing our init sequence with the CP and
3986  * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3987  * do some quick sanity checks and restore sane values to avoid this
3988  * problem.
3989  */
3990 void r100_restore_sanity(struct radeon_device *rdev)
3991 {
3992         u32 tmp;
3993
3994         tmp = RREG32(RADEON_CP_CSQ_CNTL);
3995         if (tmp) {
3996                 WREG32(RADEON_CP_CSQ_CNTL, 0);
3997         }
3998         tmp = RREG32(RADEON_CP_RB_CNTL);
3999         if (tmp) {
4000                 WREG32(RADEON_CP_RB_CNTL, 0);
4001         }
4002         tmp = RREG32(RADEON_SCRATCH_UMSK);
4003         if (tmp) {
4004                 WREG32(RADEON_SCRATCH_UMSK, 0);
4005         }
4006 }
4007
4008 int r100_init(struct radeon_device *rdev)
4009 {
4010         int r;
4011
4012         /* Register debugfs file specific to this group of asics */
4013         r100_debugfs(rdev);
4014         /* Disable VGA */
4015         r100_vga_render_disable(rdev);
4016         /* Initialize scratch registers */
4017         radeon_scratch_init(rdev);
4018         /* Initialize surface registers */
4019         radeon_surface_init(rdev);
4020         /* sanity check some register to avoid hangs like after kexec */
4021         r100_restore_sanity(rdev);
4022         /* TODO: disable VGA need to use VGA request */
4023         /* BIOS*/
4024         if (!radeon_get_bios(rdev)) {
4025                 if (ASIC_IS_AVIVO(rdev))
4026                         return -EINVAL;
4027         }
4028         if (rdev->is_atom_bios) {
4029                 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4030                 return -EINVAL;
4031         } else {
4032                 r = radeon_combios_init(rdev);
4033                 if (r)
4034                         return r;
4035         }
4036         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
4037         if (radeon_asic_reset(rdev)) {
4038                 dev_warn(rdev->dev,
4039                         "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4040                         RREG32(R_000E40_RBBM_STATUS),
4041                         RREG32(R_0007C0_CP_STAT));
4042         }
4043         /* check if cards are posted or not */
4044         if (radeon_boot_test_post_card(rdev) == false)
4045                 return -EINVAL;
4046         /* Set asic errata */
4047         r100_errata(rdev);
4048         /* Initialize clocks */
4049         radeon_get_clock_info(rdev->ddev);
4050         /* initialize AGP */
4051         if (rdev->flags & RADEON_IS_AGP) {
4052                 r = radeon_agp_init(rdev);
4053                 if (r) {
4054                         radeon_agp_disable(rdev);
4055                 }
4056         }
4057         /* initialize VRAM */
4058         r100_mc_init(rdev);
4059         /* Fence driver */
4060         r = radeon_fence_driver_init(rdev);
4061         if (r)
4062                 return r;
4063         /* Memory manager */
4064         r = radeon_bo_init(rdev);
4065         if (r)
4066                 return r;
4067         if (rdev->flags & RADEON_IS_PCI) {
4068                 r = r100_pci_gart_init(rdev);
4069                 if (r)
4070                         return r;
4071         }
4072         r100_set_safe_registers(rdev);
4073
4074         /* Initialize power management */
4075         radeon_pm_init(rdev);
4076
4077         rdev->accel_working = true;
4078         r = r100_startup(rdev);
4079         if (r) {
4080                 /* Somethings want wront with the accel init stop accel */
4081                 dev_err(rdev->dev, "Disabling GPU acceleration\n");
4082                 r100_cp_fini(rdev);
4083                 radeon_wb_fini(rdev);
4084                 radeon_ib_pool_fini(rdev);
4085                 radeon_irq_kms_fini(rdev);
4086                 if (rdev->flags & RADEON_IS_PCI)
4087                         r100_pci_gart_fini(rdev);
4088                 rdev->accel_working = false;
4089         }
4090         return 0;
4091 }
4092
4093 uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg,
4094                       bool always_indirect)
4095 {
4096         if (reg < rdev->rmmio_size && !always_indirect)
4097                 return readl(((void __iomem *)rdev->rmmio) + reg);
4098         else {
4099                 unsigned long flags;
4100                 uint32_t ret;
4101
4102                 spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4103                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4104                 ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4105                 spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4106
4107                 return ret;
4108         }
4109 }
4110
4111 void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v,
4112                   bool always_indirect)
4113 {
4114         if (reg < rdev->rmmio_size && !always_indirect)
4115                 writel(v, ((void __iomem *)rdev->rmmio) + reg);
4116         else {
4117                 unsigned long flags;
4118
4119                 spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4120                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4121                 writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4122                 spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4123         }
4124 }
4125
4126 u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4127 {
4128         if (reg < rdev->rio_mem_size)
4129                 return ioread32(rdev->rio_mem + reg);
4130         else {
4131                 iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4132                 return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4133         }
4134 }
4135
4136 void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4137 {
4138         if (reg < rdev->rio_mem_size)
4139                 iowrite32(v, rdev->rio_mem + reg);
4140         else {
4141                 iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4142                 iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4143         }
4144 }