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