Linux Audio

Check our new training course

Loading...
v4.6
   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
  56MODULE_FIRMWARE(FIRMWARE_R100);
  57MODULE_FIRMWARE(FIRMWARE_R200);
  58MODULE_FIRMWARE(FIRMWARE_R300);
  59MODULE_FIRMWARE(FIRMWARE_R420);
  60MODULE_FIRMWARE(FIRMWARE_RS690);
  61MODULE_FIRMWARE(FIRMWARE_RS600);
  62MODULE_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
  71static 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
  86static 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 */
 111void 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 */
 156void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
 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 */
 189bool 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 */
 207void 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 */
 298void 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 */
 345void 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 */
 443void 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 */
 474void 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 */
 506bool 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 */
 524bool 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 */
 551void 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 */
 587void 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		enable |= 1 << radeon_connector->hpd.hpd;
 
 
 
 
 
 
 
 
 
 596		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
 597	}
 598	radeon_irq_kms_enable_hpd(rdev, enable);
 
 599}
 600
 601/**
 602 * r100_hpd_fini - hpd tear down callback.
 603 *
 604 * @rdev: radeon_device pointer
 605 *
 606 * Tear down the hpd pins used by the card (r1xx-r4xx).
 607 * Disable the hpd interrupts.
 608 */
 609void r100_hpd_fini(struct radeon_device *rdev)
 610{
 611	struct drm_device *dev = rdev->ddev;
 612	struct drm_connector *connector;
 613	unsigned disable = 0;
 614
 615	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 616		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 617		disable |= 1 << radeon_connector->hpd.hpd;
 
 
 
 
 
 
 
 
 
 618	}
 619	radeon_irq_kms_disable_hpd(rdev, disable);
 620}
 621
 622/*
 623 * PCI GART
 624 */
 625void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
 626{
 627	/* TODO: can we do somethings here ? */
 628	/* It seems hw only cache one entry so we should discard this
 629	 * entry otherwise if first GPU GART read hit this entry it
 630	 * could end up in wrong address. */
 631}
 632
 633int r100_pci_gart_init(struct radeon_device *rdev)
 634{
 635	int r;
 636
 637	if (rdev->gart.ptr) {
 638		WARN(1, "R100 PCI GART already initialized\n");
 639		return 0;
 640	}
 641	/* Initialize common gart structure */
 642	r = radeon_gart_init(rdev);
 643	if (r)
 644		return r;
 645	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 646	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
 647	rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
 648	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
 649	return radeon_gart_table_ram_alloc(rdev);
 650}
 651
 
 
 
 
 
 
 
 
 
 652int r100_pci_gart_enable(struct radeon_device *rdev)
 653{
 654	uint32_t tmp;
 655
 
 656	/* discard memory request outside of configured range */
 657	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 658	WREG32(RADEON_AIC_CNTL, tmp);
 659	/* set address range for PCI address translate */
 660	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
 661	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
 662	/* set PCI GART page-table base address */
 663	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
 664	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
 665	WREG32(RADEON_AIC_CNTL, tmp);
 666	r100_pci_gart_tlb_flush(rdev);
 667	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
 668		 (unsigned)(rdev->mc.gtt_size >> 20),
 669		 (unsigned long long)rdev->gart.table_addr);
 670	rdev->gart.ready = true;
 671	return 0;
 672}
 673
 674void r100_pci_gart_disable(struct radeon_device *rdev)
 675{
 676	uint32_t tmp;
 677
 678	/* discard memory request outside of configured range */
 679	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 680	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
 681	WREG32(RADEON_AIC_LO_ADDR, 0);
 682	WREG32(RADEON_AIC_HI_ADDR, 0);
 683}
 684
 685uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
 686{
 687	return addr;
 688}
 689
 690void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
 691			    uint64_t entry)
 692{
 693	u32 *gtt = rdev->gart.ptr;
 694	gtt[i] = cpu_to_le32(lower_32_bits(entry));
 
 
 
 
 
 695}
 696
 697void r100_pci_gart_fini(struct radeon_device *rdev)
 698{
 699	radeon_gart_fini(rdev);
 700	r100_pci_gart_disable(rdev);
 701	radeon_gart_table_ram_free(rdev);
 702}
 703
 704int r100_irq_set(struct radeon_device *rdev)
 705{
 706	uint32_t tmp = 0;
 707
 708	if (!rdev->irq.installed) {
 709		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
 710		WREG32(R_000040_GEN_INT_CNTL, 0);
 711		return -EINVAL;
 712	}
 713	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
 714		tmp |= RADEON_SW_INT_ENABLE;
 715	}
 
 
 
 716	if (rdev->irq.crtc_vblank_int[0] ||
 717	    atomic_read(&rdev->irq.pflip[0])) {
 718		tmp |= RADEON_CRTC_VBLANK_MASK;
 719	}
 720	if (rdev->irq.crtc_vblank_int[1] ||
 721	    atomic_read(&rdev->irq.pflip[1])) {
 722		tmp |= RADEON_CRTC2_VBLANK_MASK;
 723	}
 724	if (rdev->irq.hpd[0]) {
 725		tmp |= RADEON_FP_DETECT_MASK;
 726	}
 727	if (rdev->irq.hpd[1]) {
 728		tmp |= RADEON_FP2_DETECT_MASK;
 729	}
 730	WREG32(RADEON_GEN_INT_CNTL, tmp);
 731
 732	/* read back to post the write */
 733	RREG32(RADEON_GEN_INT_CNTL);
 734
 735	return 0;
 736}
 737
 738void r100_irq_disable(struct radeon_device *rdev)
 739{
 740	u32 tmp;
 741
 742	WREG32(R_000040_GEN_INT_CNTL, 0);
 743	/* Wait and acknowledge irq */
 744	mdelay(1);
 745	tmp = RREG32(R_000044_GEN_INT_STATUS);
 746	WREG32(R_000044_GEN_INT_STATUS, tmp);
 747}
 748
 749static uint32_t r100_irq_ack(struct radeon_device *rdev)
 750{
 751	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
 752	uint32_t irq_mask = RADEON_SW_INT_TEST |
 753		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
 754		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
 755
 
 
 
 
 
 
 756	if (irqs) {
 757		WREG32(RADEON_GEN_INT_STATUS, irqs);
 758	}
 759	return irqs & irq_mask;
 760}
 761
 762int r100_irq_process(struct radeon_device *rdev)
 763{
 764	uint32_t status, msi_rearm;
 765	bool queue_hotplug = false;
 766
 
 
 
 767	status = r100_irq_ack(rdev);
 768	if (!status) {
 769		return IRQ_NONE;
 770	}
 771	if (rdev->shutdown) {
 772		return IRQ_NONE;
 773	}
 774	while (status) {
 775		/* SW interrupt */
 776		if (status & RADEON_SW_INT_TEST) {
 777			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
 778		}
 
 
 
 
 
 
 779		/* Vertical blank interrupts */
 780		if (status & RADEON_CRTC_VBLANK_STAT) {
 781			if (rdev->irq.crtc_vblank_int[0]) {
 782				drm_handle_vblank(rdev->ddev, 0);
 783				rdev->pm.vblank_sync = true;
 784				wake_up(&rdev->irq.vblank_queue);
 785			}
 786			if (atomic_read(&rdev->irq.pflip[0]))
 787				radeon_crtc_handle_vblank(rdev, 0);
 788		}
 789		if (status & RADEON_CRTC2_VBLANK_STAT) {
 790			if (rdev->irq.crtc_vblank_int[1]) {
 791				drm_handle_vblank(rdev->ddev, 1);
 792				rdev->pm.vblank_sync = true;
 793				wake_up(&rdev->irq.vblank_queue);
 794			}
 795			if (atomic_read(&rdev->irq.pflip[1]))
 796				radeon_crtc_handle_vblank(rdev, 1);
 797		}
 798		if (status & RADEON_FP_DETECT_STAT) {
 799			queue_hotplug = true;
 800			DRM_DEBUG("HPD1\n");
 801		}
 802		if (status & RADEON_FP2_DETECT_STAT) {
 803			queue_hotplug = true;
 804			DRM_DEBUG("HPD2\n");
 805		}
 806		status = r100_irq_ack(rdev);
 807	}
 
 
 808	if (queue_hotplug)
 809		schedule_delayed_work(&rdev->hotplug_work, 0);
 810	if (rdev->msi_enabled) {
 811		switch (rdev->family) {
 812		case CHIP_RS400:
 813		case CHIP_RS480:
 814			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
 815			WREG32(RADEON_AIC_CNTL, msi_rearm);
 816			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
 817			break;
 818		default:
 819			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
 820			break;
 821		}
 822	}
 823	return IRQ_HANDLED;
 824}
 825
 826u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
 827{
 828	if (crtc == 0)
 829		return RREG32(RADEON_CRTC_CRNT_FRAME);
 830	else
 831		return RREG32(RADEON_CRTC2_CRNT_FRAME);
 832}
 833
 834/**
 835 * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
 836 * rdev: radeon device structure
 837 * ring: ring buffer struct for emitting packets
 838 */
 839static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
 840{
 841	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 842	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
 843				RADEON_HDP_READ_BUFFER_INVALIDATE);
 844	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 845	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
 846}
 847
 848/* Who ever call radeon_fence_emit should call ring_lock and ask
 849 * for enough space (today caller are ib schedule and buffer move) */
 850void r100_fence_ring_emit(struct radeon_device *rdev,
 851			  struct radeon_fence *fence)
 852{
 853	struct radeon_ring *ring = &rdev->ring[fence->ring];
 854
 855	/* We have to make sure that caches are flushed before
 856	 * CPU might read something from VRAM. */
 857	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
 858	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
 859	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
 860	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
 861	/* Wait until IDLE & CLEAN */
 862	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 863	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
 864	r100_ring_hdp_flush(rdev, ring);
 
 
 
 
 865	/* Emit fence sequence & fire IRQ */
 866	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
 867	radeon_ring_write(ring, fence->seq);
 868	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
 869	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
 870}
 871
 872bool r100_semaphore_ring_emit(struct radeon_device *rdev,
 873			      struct radeon_ring *ring,
 874			      struct radeon_semaphore *semaphore,
 875			      bool emit_wait)
 876{
 877	/* Unused on older asics, since we don't have semaphores or multiple rings */
 878	BUG();
 879	return false;
 880}
 881
 882struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
 883				    uint64_t src_offset,
 884				    uint64_t dst_offset,
 885				    unsigned num_gpu_pages,
 886				    struct reservation_object *resv)
 887{
 888	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 889	struct radeon_fence *fence;
 890	uint32_t cur_pages;
 891	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
 892	uint32_t pitch;
 893	uint32_t stride_pixels;
 894	unsigned ndw;
 895	int num_loops;
 896	int r = 0;
 897
 898	/* radeon limited to 16k stride */
 899	stride_bytes &= 0x3fff;
 900	/* radeon pitch is /64 */
 901	pitch = stride_bytes / 64;
 902	stride_pixels = stride_bytes / 4;
 903	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
 904
 905	/* Ask for enough room for blit + flush + fence */
 906	ndw = 64 + (10 * num_loops);
 907	r = radeon_ring_lock(rdev, ring, ndw);
 908	if (r) {
 909		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 910		return ERR_PTR(-EINVAL);
 911	}
 912	while (num_gpu_pages > 0) {
 913		cur_pages = num_gpu_pages;
 914		if (cur_pages > 8191) {
 915			cur_pages = 8191;
 916		}
 917		num_gpu_pages -= cur_pages;
 918
 919		/* pages are in Y direction - height
 920		   page width in X direction - width */
 921		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
 922		radeon_ring_write(ring,
 923				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 924				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 925				  RADEON_GMC_SRC_CLIPPING |
 926				  RADEON_GMC_DST_CLIPPING |
 927				  RADEON_GMC_BRUSH_NONE |
 928				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 929				  RADEON_GMC_SRC_DATATYPE_COLOR |
 930				  RADEON_ROP3_S |
 931				  RADEON_DP_SRC_SOURCE_MEMORY |
 932				  RADEON_GMC_CLR_CMP_CNTL_DIS |
 933				  RADEON_GMC_WR_MSK_DIS);
 934		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
 935		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
 936		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 937		radeon_ring_write(ring, 0);
 938		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 939		radeon_ring_write(ring, num_gpu_pages);
 940		radeon_ring_write(ring, num_gpu_pages);
 941		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
 942	}
 943	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 944	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
 945	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 946	radeon_ring_write(ring,
 947			  RADEON_WAIT_2D_IDLECLEAN |
 948			  RADEON_WAIT_HOST_IDLECLEAN |
 949			  RADEON_WAIT_DMA_GUI_IDLE);
 950	r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
 951	if (r) {
 952		radeon_ring_unlock_undo(rdev, ring);
 953		return ERR_PTR(r);
 954	}
 955	radeon_ring_unlock_commit(rdev, ring, false);
 956	return fence;
 957}
 958
 959static int r100_cp_wait_for_idle(struct radeon_device *rdev)
 960{
 961	unsigned i;
 962	u32 tmp;
 963
 964	for (i = 0; i < rdev->usec_timeout; i++) {
 965		tmp = RREG32(R_000E40_RBBM_STATUS);
 966		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
 967			return 0;
 968		}
 969		udelay(1);
 970	}
 971	return -1;
 972}
 973
 974void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
 975{
 976	int r;
 977
 978	r = radeon_ring_lock(rdev, ring, 2);
 979	if (r) {
 980		return;
 981	}
 982	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
 983	radeon_ring_write(ring,
 984			  RADEON_ISYNC_ANY2D_IDLE3D |
 985			  RADEON_ISYNC_ANY3D_IDLE2D |
 986			  RADEON_ISYNC_WAIT_IDLEGUI |
 987			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 988	radeon_ring_unlock_commit(rdev, ring, false);
 989}
 990
 991
 992/* Load the microcode for the CP */
 993static int r100_cp_init_microcode(struct radeon_device *rdev)
 994{
 
 995	const char *fw_name = NULL;
 996	int err;
 997
 998	DRM_DEBUG_KMS("\n");
 999
 
 
 
 
 
 
1000	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
1001	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
1002	    (rdev->family == CHIP_RS200)) {
1003		DRM_INFO("Loading R100 Microcode\n");
1004		fw_name = FIRMWARE_R100;
1005	} else if ((rdev->family == CHIP_R200) ||
1006		   (rdev->family == CHIP_RV250) ||
1007		   (rdev->family == CHIP_RV280) ||
1008		   (rdev->family == CHIP_RS300)) {
1009		DRM_INFO("Loading R200 Microcode\n");
1010		fw_name = FIRMWARE_R200;
1011	} else if ((rdev->family == CHIP_R300) ||
1012		   (rdev->family == CHIP_R350) ||
1013		   (rdev->family == CHIP_RV350) ||
1014		   (rdev->family == CHIP_RV380) ||
1015		   (rdev->family == CHIP_RS400) ||
1016		   (rdev->family == CHIP_RS480)) {
1017		DRM_INFO("Loading R300 Microcode\n");
1018		fw_name = FIRMWARE_R300;
1019	} else if ((rdev->family == CHIP_R420) ||
1020		   (rdev->family == CHIP_R423) ||
1021		   (rdev->family == CHIP_RV410)) {
1022		DRM_INFO("Loading R400 Microcode\n");
1023		fw_name = FIRMWARE_R420;
1024	} else if ((rdev->family == CHIP_RS690) ||
1025		   (rdev->family == CHIP_RS740)) {
1026		DRM_INFO("Loading RS690/RS740 Microcode\n");
1027		fw_name = FIRMWARE_RS690;
1028	} else if (rdev->family == CHIP_RS600) {
1029		DRM_INFO("Loading RS600 Microcode\n");
1030		fw_name = FIRMWARE_RS600;
1031	} else if ((rdev->family == CHIP_RV515) ||
1032		   (rdev->family == CHIP_R520) ||
1033		   (rdev->family == CHIP_RV530) ||
1034		   (rdev->family == CHIP_R580) ||
1035		   (rdev->family == CHIP_RV560) ||
1036		   (rdev->family == CHIP_RV570)) {
1037		DRM_INFO("Loading R500 Microcode\n");
1038		fw_name = FIRMWARE_R520;
1039	}
1040
1041	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
 
1042	if (err) {
1043		printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1044		       fw_name);
1045	} else if (rdev->me_fw->size % 8) {
1046		printk(KERN_ERR
1047		       "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1048		       rdev->me_fw->size, fw_name);
1049		err = -EINVAL;
1050		release_firmware(rdev->me_fw);
1051		rdev->me_fw = NULL;
1052	}
1053	return err;
1054}
1055
1056u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1057		      struct radeon_ring *ring)
1058{
1059	u32 rptr;
1060
1061	if (rdev->wb.enabled)
1062		rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1063	else
1064		rptr = RREG32(RADEON_CP_RB_RPTR);
1065
1066	return rptr;
1067}
1068
1069u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1070		      struct radeon_ring *ring)
1071{
1072	u32 wptr;
1073
1074	wptr = RREG32(RADEON_CP_RB_WPTR);
1075
1076	return wptr;
1077}
1078
1079void r100_gfx_set_wptr(struct radeon_device *rdev,
1080		       struct radeon_ring *ring)
1081{
1082	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1083	(void)RREG32(RADEON_CP_RB_WPTR);
1084}
1085
1086static void r100_cp_load_microcode(struct radeon_device *rdev)
1087{
1088	const __be32 *fw_data;
1089	int i, size;
1090
1091	if (r100_gui_wait_for_idle(rdev)) {
1092		printk(KERN_WARNING "Failed to wait GUI idle while "
1093		       "programming pipes. Bad things might happen.\n");
1094	}
1095
1096	if (rdev->me_fw) {
1097		size = rdev->me_fw->size / 4;
1098		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1099		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1100		for (i = 0; i < size; i += 2) {
1101			WREG32(RADEON_CP_ME_RAM_DATAH,
1102			       be32_to_cpup(&fw_data[i]));
1103			WREG32(RADEON_CP_ME_RAM_DATAL,
1104			       be32_to_cpup(&fw_data[i + 1]));
1105		}
1106	}
1107}
1108
1109int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1110{
1111	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1112	unsigned rb_bufsz;
1113	unsigned rb_blksz;
1114	unsigned max_fetch;
1115	unsigned pre_write_timer;
1116	unsigned pre_write_limit;
1117	unsigned indirect2_start;
1118	unsigned indirect1_start;
1119	uint32_t tmp;
1120	int r;
1121
1122	if (r100_debugfs_cp_init(rdev)) {
1123		DRM_ERROR("Failed to register debugfs file for CP !\n");
1124	}
1125	if (!rdev->me_fw) {
1126		r = r100_cp_init_microcode(rdev);
1127		if (r) {
1128			DRM_ERROR("Failed to load firmware!\n");
1129			return r;
1130		}
1131	}
1132
1133	/* Align ring size */
1134	rb_bufsz = order_base_2(ring_size / 8);
1135	ring_size = (1 << (rb_bufsz + 1)) * 4;
1136	r100_cp_load_microcode(rdev);
1137	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1138			     RADEON_CP_PACKET2);
 
1139	if (r) {
1140		return r;
1141	}
1142	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1143	 * the rptr copy in system ram */
1144	rb_blksz = 9;
1145	/* cp will read 128bytes at a time (4 dwords) */
1146	max_fetch = 1;
1147	ring->align_mask = 16 - 1;
1148	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1149	pre_write_timer = 64;
1150	/* Force CP_RB_WPTR write if written more than one time before the
1151	 * delay expire
1152	 */
1153	pre_write_limit = 0;
1154	/* Setup the cp cache like this (cache size is 96 dwords) :
1155	 *	RING		0  to 15
1156	 *	INDIRECT1	16 to 79
1157	 *	INDIRECT2	80 to 95
1158	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1159	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1160	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1161	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1162	 * so it gets the bigger cache.
1163	 */
1164	indirect2_start = 80;
1165	indirect1_start = 16;
1166	/* cp setup */
1167	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1168	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1169	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1170	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1171#ifdef __BIG_ENDIAN
1172	tmp |= RADEON_BUF_SWAP_32BIT;
1173#endif
1174	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1175
1176	/* Set ring address */
1177	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1178	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1179	/* Force read & write ptr to 0 */
1180	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1181	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1182	ring->wptr = 0;
1183	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1184
1185	/* set the wb address whether it's enabled or not */
1186	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1187		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1188	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1189
1190	if (rdev->wb.enabled)
1191		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1192	else {
1193		tmp |= RADEON_RB_NO_UPDATE;
1194		WREG32(R_000770_SCRATCH_UMSK, 0);
1195	}
1196
1197	WREG32(RADEON_CP_RB_CNTL, tmp);
1198	udelay(10);
 
1199	/* Set cp mode to bus mastering & enable cp*/
1200	WREG32(RADEON_CP_CSQ_MODE,
1201	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1202	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1203	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1204	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1205	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1206
1207	/* at this point everything should be setup correctly to enable master */
1208	pci_set_master(rdev->pdev);
1209
1210	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1211	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1212	if (r) {
1213		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1214		return r;
1215	}
1216	ring->ready = true;
1217	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1218
1219	if (!ring->rptr_save_reg /* not resuming from suspend */
1220	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1221		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1222		if (r) {
1223			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1224			ring->rptr_save_reg = 0;
1225		}
1226	}
1227	return 0;
1228}
1229
1230void r100_cp_fini(struct radeon_device *rdev)
1231{
1232	if (r100_cp_wait_for_idle(rdev)) {
1233		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1234	}
1235	/* Disable ring */
1236	r100_cp_disable(rdev);
1237	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1238	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1239	DRM_INFO("radeon: cp finalized\n");
1240}
1241
1242void r100_cp_disable(struct radeon_device *rdev)
1243{
1244	/* Disable ring */
1245	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1246	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1247	WREG32(RADEON_CP_CSQ_MODE, 0);
1248	WREG32(RADEON_CP_CSQ_CNTL, 0);
1249	WREG32(R_000770_SCRATCH_UMSK, 0);
1250	if (r100_gui_wait_for_idle(rdev)) {
1251		printk(KERN_WARNING "Failed to wait GUI idle while "
1252		       "programming pipes. Bad things might happen.\n");
1253	}
1254}
1255
1256/*
1257 * CS functions
1258 */
1259int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1260			    struct radeon_cs_packet *pkt,
1261			    unsigned idx,
1262			    unsigned reg)
1263{
1264	int r;
1265	u32 tile_flags = 0;
1266	u32 tmp;
1267	struct radeon_bo_list *reloc;
1268	u32 value;
1269
1270	r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1271	if (r) {
1272		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1273			  idx, reg);
1274		radeon_cs_dump_packet(p, pkt);
1275		return r;
1276	}
1277
1278	value = radeon_get_ib_value(p, idx);
1279	tmp = value & 0x003fffff;
1280	tmp += (((u32)reloc->gpu_offset) >> 10);
1281
1282	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1283		if (reloc->tiling_flags & RADEON_TILING_MACRO)
1284			tile_flags |= RADEON_DST_TILE_MACRO;
1285		if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1286			if (reg == RADEON_SRC_PITCH_OFFSET) {
1287				DRM_ERROR("Cannot src blit from microtiled surface\n");
1288				radeon_cs_dump_packet(p, pkt);
1289				return -EINVAL;
1290			}
1291			tile_flags |= RADEON_DST_TILE_MICRO;
1292		}
1293
1294		tmp |= tile_flags;
1295		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1296	} else
1297		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1298	return 0;
1299}
1300
1301int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1302			     struct radeon_cs_packet *pkt,
1303			     int idx)
1304{
1305	unsigned c, i;
1306	struct radeon_bo_list *reloc;
1307	struct r100_cs_track *track;
1308	int r = 0;
1309	volatile uint32_t *ib;
1310	u32 idx_value;
1311
1312	ib = p->ib.ptr;
1313	track = (struct r100_cs_track *)p->track;
1314	c = radeon_get_ib_value(p, idx++) & 0x1F;
1315	if (c > 16) {
1316	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1317		      pkt->opcode);
1318	    radeon_cs_dump_packet(p, pkt);
1319	    return -EINVAL;
1320	}
1321	track->num_arrays = c;
1322	for (i = 0; i < (c - 1); i+=2, idx+=3) {
1323		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1324		if (r) {
1325			DRM_ERROR("No reloc for packet3 %d\n",
1326				  pkt->opcode);
1327			radeon_cs_dump_packet(p, pkt);
1328			return r;
1329		}
1330		idx_value = radeon_get_ib_value(p, idx);
1331		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1332
1333		track->arrays[i + 0].esize = idx_value >> 8;
1334		track->arrays[i + 0].robj = reloc->robj;
1335		track->arrays[i + 0].esize &= 0x7F;
1336		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1337		if (r) {
1338			DRM_ERROR("No reloc for packet3 %d\n",
1339				  pkt->opcode);
1340			radeon_cs_dump_packet(p, pkt);
1341			return r;
1342		}
1343		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
1344		track->arrays[i + 1].robj = reloc->robj;
1345		track->arrays[i + 1].esize = idx_value >> 24;
1346		track->arrays[i + 1].esize &= 0x7F;
1347	}
1348	if (c & 1) {
1349		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1350		if (r) {
1351			DRM_ERROR("No reloc for packet3 %d\n",
1352					  pkt->opcode);
1353			radeon_cs_dump_packet(p, pkt);
1354			return r;
1355		}
1356		idx_value = radeon_get_ib_value(p, idx);
1357		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1358		track->arrays[i + 0].robj = reloc->robj;
1359		track->arrays[i + 0].esize = idx_value >> 8;
1360		track->arrays[i + 0].esize &= 0x7F;
1361	}
1362	return r;
1363}
1364
1365int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1366			  struct radeon_cs_packet *pkt,
1367			  const unsigned *auth, unsigned n,
1368			  radeon_packet0_check_t check)
1369{
1370	unsigned reg;
1371	unsigned i, j, m;
1372	unsigned idx;
1373	int r;
1374
1375	idx = pkt->idx + 1;
1376	reg = pkt->reg;
1377	/* Check that register fall into register range
1378	 * determined by the number of entry (n) in the
1379	 * safe register bitmap.
1380	 */
1381	if (pkt->one_reg_wr) {
1382		if ((reg >> 7) > n) {
1383			return -EINVAL;
1384		}
1385	} else {
1386		if (((reg + (pkt->count << 2)) >> 7) > n) {
1387			return -EINVAL;
1388		}
1389	}
1390	for (i = 0; i <= pkt->count; i++, idx++) {
1391		j = (reg >> 7);
1392		m = 1 << ((reg >> 2) & 31);
1393		if (auth[j] & m) {
1394			r = check(p, pkt, idx, reg);
1395			if (r) {
1396				return r;
1397			}
1398		}
1399		if (pkt->one_reg_wr) {
1400			if (!(auth[j] & m)) {
1401				break;
1402			}
1403		} else {
1404			reg += 4;
1405		}
1406	}
1407	return 0;
1408}
1409
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1410/**
1411 * r100_cs_packet_next_vline() - parse userspace VLINE packet
1412 * @parser:		parser structure holding parsing context.
1413 *
1414 * Userspace sends a special sequence for VLINE waits.
1415 * PACKET0 - VLINE_START_END + value
1416 * PACKET0 - WAIT_UNTIL +_value
1417 * RELOC (P3) - crtc_id in reloc.
1418 *
1419 * This function parses this and relocates the VLINE START END
1420 * and WAIT UNTIL packets to the correct crtc.
1421 * It also detects a switched off crtc and nulls out the
1422 * wait in that case.
1423 */
1424int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1425{
 
1426	struct drm_crtc *crtc;
1427	struct radeon_crtc *radeon_crtc;
1428	struct radeon_cs_packet p3reloc, waitreloc;
1429	int crtc_id;
1430	int r;
1431	uint32_t header, h_idx, reg;
1432	volatile uint32_t *ib;
1433
1434	ib = p->ib.ptr;
1435
1436	/* parse the wait until */
1437	r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1438	if (r)
1439		return r;
1440
1441	/* check its a wait until and only 1 count */
1442	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1443	    waitreloc.count != 0) {
1444		DRM_ERROR("vline wait had illegal wait until segment\n");
1445		return -EINVAL;
1446	}
1447
1448	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1449		DRM_ERROR("vline wait had illegal wait until\n");
1450		return -EINVAL;
1451	}
1452
1453	/* jump over the NOP */
1454	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1455	if (r)
1456		return r;
1457
1458	h_idx = p->idx - 2;
1459	p->idx += waitreloc.count + 2;
1460	p->idx += p3reloc.count + 2;
1461
1462	header = radeon_get_ib_value(p, h_idx);
1463	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1464	reg = R100_CP_PACKET0_GET_REG(header);
1465	crtc = drm_crtc_find(p->rdev->ddev, crtc_id);
1466	if (!crtc) {
1467		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1468		return -ENOENT;
1469	}
 
1470	radeon_crtc = to_radeon_crtc(crtc);
1471	crtc_id = radeon_crtc->crtc_id;
1472
1473	if (!crtc->enabled) {
1474		/* if the CRTC isn't enabled - we need to nop out the wait until */
1475		ib[h_idx + 2] = PACKET2(0);
1476		ib[h_idx + 3] = PACKET2(0);
1477	} else if (crtc_id == 1) {
1478		switch (reg) {
1479		case AVIVO_D1MODE_VLINE_START_END:
1480			header &= ~R300_CP_PACKET0_REG_MASK;
1481			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1482			break;
1483		case RADEON_CRTC_GUI_TRIG_VLINE:
1484			header &= ~R300_CP_PACKET0_REG_MASK;
1485			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1486			break;
1487		default:
1488			DRM_ERROR("unknown crtc reloc\n");
1489			return -EINVAL;
1490		}
1491		ib[h_idx] = header;
1492		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1493	}
1494
1495	return 0;
1496}
1497
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1498static int r100_get_vtx_size(uint32_t vtx_fmt)
1499{
1500	int vtx_size;
1501	vtx_size = 2;
1502	/* ordered according to bits in spec */
1503	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1504		vtx_size++;
1505	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1506		vtx_size += 3;
1507	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1508		vtx_size++;
1509	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1510		vtx_size++;
1511	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1512		vtx_size += 3;
1513	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1514		vtx_size++;
1515	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1516		vtx_size++;
1517	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1518		vtx_size += 2;
1519	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1520		vtx_size += 2;
1521	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1522		vtx_size++;
1523	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1524		vtx_size += 2;
1525	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1526		vtx_size++;
1527	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1528		vtx_size += 2;
1529	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1530		vtx_size++;
1531	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1532		vtx_size++;
1533	/* blend weight */
1534	if (vtx_fmt & (0x7 << 15))
1535		vtx_size += (vtx_fmt >> 15) & 0x7;
1536	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1537		vtx_size += 3;
1538	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1539		vtx_size += 2;
1540	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1541		vtx_size++;
1542	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1543		vtx_size++;
1544	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1545		vtx_size++;
1546	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1547		vtx_size++;
1548	return vtx_size;
1549}
1550
1551static int r100_packet0_check(struct radeon_cs_parser *p,
1552			      struct radeon_cs_packet *pkt,
1553			      unsigned idx, unsigned reg)
1554{
1555	struct radeon_bo_list *reloc;
1556	struct r100_cs_track *track;
1557	volatile uint32_t *ib;
1558	uint32_t tmp;
1559	int r;
1560	int i, face;
1561	u32 tile_flags = 0;
1562	u32 idx_value;
1563
1564	ib = p->ib.ptr;
1565	track = (struct r100_cs_track *)p->track;
1566
1567	idx_value = radeon_get_ib_value(p, idx);
1568
1569	switch (reg) {
1570	case RADEON_CRTC_GUI_TRIG_VLINE:
1571		r = r100_cs_packet_parse_vline(p);
1572		if (r) {
1573			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1574				  idx, reg);
1575			radeon_cs_dump_packet(p, pkt);
1576			return r;
1577		}
1578		break;
1579		/* FIXME: only allow PACKET3 blit? easier to check for out of
1580		 * range access */
1581	case RADEON_DST_PITCH_OFFSET:
1582	case RADEON_SRC_PITCH_OFFSET:
1583		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1584		if (r)
1585			return r;
1586		break;
1587	case RADEON_RB3D_DEPTHOFFSET:
1588		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1589		if (r) {
1590			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1591				  idx, reg);
1592			radeon_cs_dump_packet(p, pkt);
1593			return r;
1594		}
1595		track->zb.robj = reloc->robj;
1596		track->zb.offset = idx_value;
1597		track->zb_dirty = true;
1598		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1599		break;
1600	case RADEON_RB3D_COLOROFFSET:
1601		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1602		if (r) {
1603			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1604				  idx, reg);
1605			radeon_cs_dump_packet(p, pkt);
1606			return r;
1607		}
1608		track->cb[0].robj = reloc->robj;
1609		track->cb[0].offset = idx_value;
1610		track->cb_dirty = true;
1611		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1612		break;
1613	case RADEON_PP_TXOFFSET_0:
1614	case RADEON_PP_TXOFFSET_1:
1615	case RADEON_PP_TXOFFSET_2:
1616		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1617		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1618		if (r) {
1619			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1620				  idx, reg);
1621			radeon_cs_dump_packet(p, pkt);
1622			return r;
1623		}
1624		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1625			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1626				tile_flags |= RADEON_TXO_MACRO_TILE;
1627			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1628				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1629
1630			tmp = idx_value & ~(0x7 << 2);
1631			tmp |= tile_flags;
1632			ib[idx] = tmp + ((u32)reloc->gpu_offset);
1633		} else
1634			ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1635		track->textures[i].robj = reloc->robj;
1636		track->tex_dirty = true;
1637		break;
1638	case RADEON_PP_CUBIC_OFFSET_T0_0:
1639	case RADEON_PP_CUBIC_OFFSET_T0_1:
1640	case RADEON_PP_CUBIC_OFFSET_T0_2:
1641	case RADEON_PP_CUBIC_OFFSET_T0_3:
1642	case RADEON_PP_CUBIC_OFFSET_T0_4:
1643		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1644		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1645		if (r) {
1646			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1647				  idx, reg);
1648			radeon_cs_dump_packet(p, pkt);
1649			return r;
1650		}
1651		track->textures[0].cube_info[i].offset = idx_value;
1652		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1653		track->textures[0].cube_info[i].robj = reloc->robj;
1654		track->tex_dirty = true;
1655		break;
1656	case RADEON_PP_CUBIC_OFFSET_T1_0:
1657	case RADEON_PP_CUBIC_OFFSET_T1_1:
1658	case RADEON_PP_CUBIC_OFFSET_T1_2:
1659	case RADEON_PP_CUBIC_OFFSET_T1_3:
1660	case RADEON_PP_CUBIC_OFFSET_T1_4:
1661		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1662		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1663		if (r) {
1664			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1665				  idx, reg);
1666			radeon_cs_dump_packet(p, pkt);
1667			return r;
1668		}
1669		track->textures[1].cube_info[i].offset = idx_value;
1670		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1671		track->textures[1].cube_info[i].robj = reloc->robj;
1672		track->tex_dirty = true;
1673		break;
1674	case RADEON_PP_CUBIC_OFFSET_T2_0:
1675	case RADEON_PP_CUBIC_OFFSET_T2_1:
1676	case RADEON_PP_CUBIC_OFFSET_T2_2:
1677	case RADEON_PP_CUBIC_OFFSET_T2_3:
1678	case RADEON_PP_CUBIC_OFFSET_T2_4:
1679		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1680		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1681		if (r) {
1682			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1683				  idx, reg);
1684			radeon_cs_dump_packet(p, pkt);
1685			return r;
1686		}
1687		track->textures[2].cube_info[i].offset = idx_value;
1688		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1689		track->textures[2].cube_info[i].robj = reloc->robj;
1690		track->tex_dirty = true;
1691		break;
1692	case RADEON_RE_WIDTH_HEIGHT:
1693		track->maxy = ((idx_value >> 16) & 0x7FF);
1694		track->cb_dirty = true;
1695		track->zb_dirty = true;
1696		break;
1697	case RADEON_RB3D_COLORPITCH:
1698		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1699		if (r) {
1700			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1701				  idx, reg);
1702			radeon_cs_dump_packet(p, pkt);
1703			return r;
1704		}
1705		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1706			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1707				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1708			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1709				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1710
1711			tmp = idx_value & ~(0x7 << 16);
1712			tmp |= tile_flags;
1713			ib[idx] = tmp;
1714		} else
1715			ib[idx] = idx_value;
1716
1717		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1718		track->cb_dirty = true;
1719		break;
1720	case RADEON_RB3D_DEPTHPITCH:
1721		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1722		track->zb_dirty = true;
1723		break;
1724	case RADEON_RB3D_CNTL:
1725		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1726		case 7:
1727		case 8:
1728		case 9:
1729		case 11:
1730		case 12:
1731			track->cb[0].cpp = 1;
1732			break;
1733		case 3:
1734		case 4:
1735		case 15:
1736			track->cb[0].cpp = 2;
1737			break;
1738		case 6:
1739			track->cb[0].cpp = 4;
1740			break;
1741		default:
1742			DRM_ERROR("Invalid color buffer format (%d) !\n",
1743				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1744			return -EINVAL;
1745		}
1746		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1747		track->cb_dirty = true;
1748		track->zb_dirty = true;
1749		break;
1750	case RADEON_RB3D_ZSTENCILCNTL:
1751		switch (idx_value & 0xf) {
1752		case 0:
1753			track->zb.cpp = 2;
1754			break;
1755		case 2:
1756		case 3:
1757		case 4:
1758		case 5:
1759		case 9:
1760		case 11:
1761			track->zb.cpp = 4;
1762			break;
1763		default:
1764			break;
1765		}
1766		track->zb_dirty = true;
1767		break;
1768	case RADEON_RB3D_ZPASS_ADDR:
1769		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1770		if (r) {
1771			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1772				  idx, reg);
1773			radeon_cs_dump_packet(p, pkt);
1774			return r;
1775		}
1776		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1777		break;
1778	case RADEON_PP_CNTL:
1779		{
1780			uint32_t temp = idx_value >> 4;
1781			for (i = 0; i < track->num_texture; i++)
1782				track->textures[i].enabled = !!(temp & (1 << i));
1783			track->tex_dirty = true;
1784		}
1785		break;
1786	case RADEON_SE_VF_CNTL:
1787		track->vap_vf_cntl = idx_value;
1788		break;
1789	case RADEON_SE_VTX_FMT:
1790		track->vtx_size = r100_get_vtx_size(idx_value);
1791		break;
1792	case RADEON_PP_TEX_SIZE_0:
1793	case RADEON_PP_TEX_SIZE_1:
1794	case RADEON_PP_TEX_SIZE_2:
1795		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1796		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1797		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1798		track->tex_dirty = true;
1799		break;
1800	case RADEON_PP_TEX_PITCH_0:
1801	case RADEON_PP_TEX_PITCH_1:
1802	case RADEON_PP_TEX_PITCH_2:
1803		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1804		track->textures[i].pitch = idx_value + 32;
1805		track->tex_dirty = true;
1806		break;
1807	case RADEON_PP_TXFILTER_0:
1808	case RADEON_PP_TXFILTER_1:
1809	case RADEON_PP_TXFILTER_2:
1810		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1811		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1812						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1813		tmp = (idx_value >> 23) & 0x7;
1814		if (tmp == 2 || tmp == 6)
1815			track->textures[i].roundup_w = false;
1816		tmp = (idx_value >> 27) & 0x7;
1817		if (tmp == 2 || tmp == 6)
1818			track->textures[i].roundup_h = false;
1819		track->tex_dirty = true;
1820		break;
1821	case RADEON_PP_TXFORMAT_0:
1822	case RADEON_PP_TXFORMAT_1:
1823	case RADEON_PP_TXFORMAT_2:
1824		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1825		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1826			track->textures[i].use_pitch = 1;
1827		} else {
1828			track->textures[i].use_pitch = 0;
1829			track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1830			track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1831		}
1832		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1833			track->textures[i].tex_coord_type = 2;
1834		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1835		case RADEON_TXFORMAT_I8:
1836		case RADEON_TXFORMAT_RGB332:
1837		case RADEON_TXFORMAT_Y8:
1838			track->textures[i].cpp = 1;
1839			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1840			break;
1841		case RADEON_TXFORMAT_AI88:
1842		case RADEON_TXFORMAT_ARGB1555:
1843		case RADEON_TXFORMAT_RGB565:
1844		case RADEON_TXFORMAT_ARGB4444:
1845		case RADEON_TXFORMAT_VYUY422:
1846		case RADEON_TXFORMAT_YVYU422:
1847		case RADEON_TXFORMAT_SHADOW16:
1848		case RADEON_TXFORMAT_LDUDV655:
1849		case RADEON_TXFORMAT_DUDV88:
1850			track->textures[i].cpp = 2;
1851			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1852			break;
1853		case RADEON_TXFORMAT_ARGB8888:
1854		case RADEON_TXFORMAT_RGBA8888:
1855		case RADEON_TXFORMAT_SHADOW32:
1856		case RADEON_TXFORMAT_LDUDUV8888:
1857			track->textures[i].cpp = 4;
1858			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1859			break;
1860		case RADEON_TXFORMAT_DXT1:
1861			track->textures[i].cpp = 1;
1862			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1863			break;
1864		case RADEON_TXFORMAT_DXT23:
1865		case RADEON_TXFORMAT_DXT45:
1866			track->textures[i].cpp = 1;
1867			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1868			break;
1869		}
1870		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1871		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1872		track->tex_dirty = true;
1873		break;
1874	case RADEON_PP_CUBIC_FACES_0:
1875	case RADEON_PP_CUBIC_FACES_1:
1876	case RADEON_PP_CUBIC_FACES_2:
1877		tmp = idx_value;
1878		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1879		for (face = 0; face < 4; face++) {
1880			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1881			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1882		}
1883		track->tex_dirty = true;
1884		break;
1885	default:
1886		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1887		       reg, idx);
1888		return -EINVAL;
1889	}
1890	return 0;
1891}
1892
1893int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1894					 struct radeon_cs_packet *pkt,
1895					 struct radeon_bo *robj)
1896{
1897	unsigned idx;
1898	u32 value;
1899	idx = pkt->idx + 1;
1900	value = radeon_get_ib_value(p, idx + 2);
1901	if ((value + 1) > radeon_bo_size(robj)) {
1902		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1903			  "(need %u have %lu) !\n",
1904			  value + 1,
1905			  radeon_bo_size(robj));
1906		return -EINVAL;
1907	}
1908	return 0;
1909}
1910
1911static int r100_packet3_check(struct radeon_cs_parser *p,
1912			      struct radeon_cs_packet *pkt)
1913{
1914	struct radeon_bo_list *reloc;
1915	struct r100_cs_track *track;
1916	unsigned idx;
1917	volatile uint32_t *ib;
1918	int r;
1919
1920	ib = p->ib.ptr;
1921	idx = pkt->idx + 1;
1922	track = (struct r100_cs_track *)p->track;
1923	switch (pkt->opcode) {
1924	case PACKET3_3D_LOAD_VBPNTR:
1925		r = r100_packet3_load_vbpntr(p, pkt, idx);
1926		if (r)
1927			return r;
1928		break;
1929	case PACKET3_INDX_BUFFER:
1930		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1931		if (r) {
1932			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1933			radeon_cs_dump_packet(p, pkt);
1934			return r;
1935		}
1936		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1937		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1938		if (r) {
1939			return r;
1940		}
1941		break;
1942	case 0x23:
1943		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1944		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1945		if (r) {
1946			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1947			radeon_cs_dump_packet(p, pkt);
1948			return r;
1949		}
1950		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1951		track->num_arrays = 1;
1952		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1953
1954		track->arrays[0].robj = reloc->robj;
1955		track->arrays[0].esize = track->vtx_size;
1956
1957		track->max_indx = radeon_get_ib_value(p, idx+1);
1958
1959		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1960		track->immd_dwords = pkt->count - 1;
1961		r = r100_cs_track_check(p->rdev, track);
1962		if (r)
1963			return r;
1964		break;
1965	case PACKET3_3D_DRAW_IMMD:
1966		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1967			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1968			return -EINVAL;
1969		}
1970		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1971		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1972		track->immd_dwords = pkt->count - 1;
1973		r = r100_cs_track_check(p->rdev, track);
1974		if (r)
1975			return r;
1976		break;
1977		/* triggers drawing using in-packet vertex data */
1978	case PACKET3_3D_DRAW_IMMD_2:
1979		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1980			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1981			return -EINVAL;
1982		}
1983		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1984		track->immd_dwords = pkt->count;
1985		r = r100_cs_track_check(p->rdev, track);
1986		if (r)
1987			return r;
1988		break;
1989		/* triggers drawing using in-packet vertex data */
1990	case PACKET3_3D_DRAW_VBUF_2:
1991		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1992		r = r100_cs_track_check(p->rdev, track);
1993		if (r)
1994			return r;
1995		break;
1996		/* triggers drawing of vertex buffers setup elsewhere */
1997	case PACKET3_3D_DRAW_INDX_2:
1998		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1999		r = r100_cs_track_check(p->rdev, track);
2000		if (r)
2001			return r;
2002		break;
2003		/* triggers drawing using indices to vertex buffer */
2004	case PACKET3_3D_DRAW_VBUF:
2005		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2006		r = r100_cs_track_check(p->rdev, track);
2007		if (r)
2008			return r;
2009		break;
2010		/* triggers drawing of vertex buffers setup elsewhere */
2011	case PACKET3_3D_DRAW_INDX:
2012		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2013		r = r100_cs_track_check(p->rdev, track);
2014		if (r)
2015			return r;
2016		break;
2017		/* triggers drawing using indices to vertex buffer */
2018	case PACKET3_3D_CLEAR_HIZ:
2019	case PACKET3_3D_CLEAR_ZMASK:
2020		if (p->rdev->hyperz_filp != p->filp)
2021			return -EINVAL;
2022		break;
2023	case PACKET3_NOP:
2024		break;
2025	default:
2026		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2027		return -EINVAL;
2028	}
2029	return 0;
2030}
2031
2032int r100_cs_parse(struct radeon_cs_parser *p)
2033{
2034	struct radeon_cs_packet pkt;
2035	struct r100_cs_track *track;
2036	int r;
2037
2038	track = kzalloc(sizeof(*track), GFP_KERNEL);
2039	if (!track)
2040		return -ENOMEM;
2041	r100_cs_track_clear(p->rdev, track);
2042	p->track = track;
2043	do {
2044		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2045		if (r) {
2046			return r;
2047		}
2048		p->idx += pkt.count + 2;
2049		switch (pkt.type) {
2050		case RADEON_PACKET_TYPE0:
2051			if (p->rdev->family >= CHIP_R200)
2052				r = r100_cs_parse_packet0(p, &pkt,
2053					p->rdev->config.r100.reg_safe_bm,
2054					p->rdev->config.r100.reg_safe_bm_size,
2055					&r200_packet0_check);
2056			else
2057				r = r100_cs_parse_packet0(p, &pkt,
2058					p->rdev->config.r100.reg_safe_bm,
2059					p->rdev->config.r100.reg_safe_bm_size,
2060					&r100_packet0_check);
2061			break;
2062		case RADEON_PACKET_TYPE2:
2063			break;
2064		case RADEON_PACKET_TYPE3:
2065			r = r100_packet3_check(p, &pkt);
2066			break;
2067		default:
2068			DRM_ERROR("Unknown packet type %d !\n",
2069				  pkt.type);
2070			return -EINVAL;
2071		}
2072		if (r)
2073			return r;
2074	} while (p->idx < p->chunk_ib->length_dw);
2075	return 0;
2076}
2077
2078static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2079{
2080	DRM_ERROR("pitch                      %d\n", t->pitch);
2081	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2082	DRM_ERROR("width                      %d\n", t->width);
2083	DRM_ERROR("width_11                   %d\n", t->width_11);
2084	DRM_ERROR("height                     %d\n", t->height);
2085	DRM_ERROR("height_11                  %d\n", t->height_11);
2086	DRM_ERROR("num levels                 %d\n", t->num_levels);
2087	DRM_ERROR("depth                      %d\n", t->txdepth);
2088	DRM_ERROR("bpp                        %d\n", t->cpp);
2089	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2090	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2091	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2092	DRM_ERROR("compress format            %d\n", t->compress_format);
2093}
2094
2095static int r100_track_compress_size(int compress_format, int w, int h)
2096{
2097	int block_width, block_height, block_bytes;
2098	int wblocks, hblocks;
2099	int min_wblocks;
2100	int sz;
2101
2102	block_width = 4;
2103	block_height = 4;
2104
2105	switch (compress_format) {
2106	case R100_TRACK_COMP_DXT1:
2107		block_bytes = 8;
2108		min_wblocks = 4;
2109		break;
2110	default:
2111	case R100_TRACK_COMP_DXT35:
2112		block_bytes = 16;
2113		min_wblocks = 2;
2114		break;
2115	}
2116
2117	hblocks = (h + block_height - 1) / block_height;
2118	wblocks = (w + block_width - 1) / block_width;
2119	if (wblocks < min_wblocks)
2120		wblocks = min_wblocks;
2121	sz = wblocks * hblocks * block_bytes;
2122	return sz;
2123}
2124
2125static int r100_cs_track_cube(struct radeon_device *rdev,
2126			      struct r100_cs_track *track, unsigned idx)
2127{
2128	unsigned face, w, h;
2129	struct radeon_bo *cube_robj;
2130	unsigned long size;
2131	unsigned compress_format = track->textures[idx].compress_format;
2132
2133	for (face = 0; face < 5; face++) {
2134		cube_robj = track->textures[idx].cube_info[face].robj;
2135		w = track->textures[idx].cube_info[face].width;
2136		h = track->textures[idx].cube_info[face].height;
2137
2138		if (compress_format) {
2139			size = r100_track_compress_size(compress_format, w, h);
2140		} else
2141			size = w * h;
2142		size *= track->textures[idx].cpp;
2143
2144		size += track->textures[idx].cube_info[face].offset;
2145
2146		if (size > radeon_bo_size(cube_robj)) {
2147			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2148				  size, radeon_bo_size(cube_robj));
2149			r100_cs_track_texture_print(&track->textures[idx]);
2150			return -1;
2151		}
2152	}
2153	return 0;
2154}
2155
2156static int r100_cs_track_texture_check(struct radeon_device *rdev,
2157				       struct r100_cs_track *track)
2158{
2159	struct radeon_bo *robj;
2160	unsigned long size;
2161	unsigned u, i, w, h, d;
2162	int ret;
2163
2164	for (u = 0; u < track->num_texture; u++) {
2165		if (!track->textures[u].enabled)
2166			continue;
2167		if (track->textures[u].lookup_disable)
2168			continue;
2169		robj = track->textures[u].robj;
2170		if (robj == NULL) {
2171			DRM_ERROR("No texture bound to unit %u\n", u);
2172			return -EINVAL;
2173		}
2174		size = 0;
2175		for (i = 0; i <= track->textures[u].num_levels; i++) {
2176			if (track->textures[u].use_pitch) {
2177				if (rdev->family < CHIP_R300)
2178					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2179				else
2180					w = track->textures[u].pitch / (1 << i);
2181			} else {
2182				w = track->textures[u].width;
2183				if (rdev->family >= CHIP_RV515)
2184					w |= track->textures[u].width_11;
2185				w = w / (1 << i);
2186				if (track->textures[u].roundup_w)
2187					w = roundup_pow_of_two(w);
2188			}
2189			h = track->textures[u].height;
2190			if (rdev->family >= CHIP_RV515)
2191				h |= track->textures[u].height_11;
2192			h = h / (1 << i);
2193			if (track->textures[u].roundup_h)
2194				h = roundup_pow_of_two(h);
2195			if (track->textures[u].tex_coord_type == 1) {
2196				d = (1 << track->textures[u].txdepth) / (1 << i);
2197				if (!d)
2198					d = 1;
2199			} else {
2200				d = 1;
2201			}
2202			if (track->textures[u].compress_format) {
2203
2204				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2205				/* compressed textures are block based */
2206			} else
2207				size += w * h * d;
2208		}
2209		size *= track->textures[u].cpp;
2210
2211		switch (track->textures[u].tex_coord_type) {
2212		case 0:
2213		case 1:
2214			break;
2215		case 2:
2216			if (track->separate_cube) {
2217				ret = r100_cs_track_cube(rdev, track, u);
2218				if (ret)
2219					return ret;
2220			} else
2221				size *= 6;
2222			break;
2223		default:
2224			DRM_ERROR("Invalid texture coordinate type %u for unit "
2225				  "%u\n", track->textures[u].tex_coord_type, u);
2226			return -EINVAL;
2227		}
2228		if (size > radeon_bo_size(robj)) {
2229			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2230				  "%lu\n", u, size, radeon_bo_size(robj));
2231			r100_cs_track_texture_print(&track->textures[u]);
2232			return -EINVAL;
2233		}
2234	}
2235	return 0;
2236}
2237
2238int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2239{
2240	unsigned i;
2241	unsigned long size;
2242	unsigned prim_walk;
2243	unsigned nverts;
2244	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2245
2246	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2247	    !track->blend_read_enable)
2248		num_cb = 0;
2249
2250	for (i = 0; i < num_cb; i++) {
2251		if (track->cb[i].robj == NULL) {
2252			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2253			return -EINVAL;
2254		}
2255		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2256		size += track->cb[i].offset;
2257		if (size > radeon_bo_size(track->cb[i].robj)) {
2258			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2259				  "(need %lu have %lu) !\n", i, size,
2260				  radeon_bo_size(track->cb[i].robj));
2261			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2262				  i, track->cb[i].pitch, track->cb[i].cpp,
2263				  track->cb[i].offset, track->maxy);
2264			return -EINVAL;
2265		}
2266	}
2267	track->cb_dirty = false;
2268
2269	if (track->zb_dirty && track->z_enabled) {
2270		if (track->zb.robj == NULL) {
2271			DRM_ERROR("[drm] No buffer for z buffer !\n");
2272			return -EINVAL;
2273		}
2274		size = track->zb.pitch * track->zb.cpp * track->maxy;
2275		size += track->zb.offset;
2276		if (size > radeon_bo_size(track->zb.robj)) {
2277			DRM_ERROR("[drm] Buffer too small for z buffer "
2278				  "(need %lu have %lu) !\n", size,
2279				  radeon_bo_size(track->zb.robj));
2280			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2281				  track->zb.pitch, track->zb.cpp,
2282				  track->zb.offset, track->maxy);
2283			return -EINVAL;
2284		}
2285	}
2286	track->zb_dirty = false;
2287
2288	if (track->aa_dirty && track->aaresolve) {
2289		if (track->aa.robj == NULL) {
2290			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2291			return -EINVAL;
2292		}
2293		/* I believe the format comes from colorbuffer0. */
2294		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2295		size += track->aa.offset;
2296		if (size > radeon_bo_size(track->aa.robj)) {
2297			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2298				  "(need %lu have %lu) !\n", i, size,
2299				  radeon_bo_size(track->aa.robj));
2300			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2301				  i, track->aa.pitch, track->cb[0].cpp,
2302				  track->aa.offset, track->maxy);
2303			return -EINVAL;
2304		}
2305	}
2306	track->aa_dirty = false;
2307
2308	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2309	if (track->vap_vf_cntl & (1 << 14)) {
2310		nverts = track->vap_alt_nverts;
2311	} else {
2312		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2313	}
2314	switch (prim_walk) {
2315	case 1:
2316		for (i = 0; i < track->num_arrays; i++) {
2317			size = track->arrays[i].esize * track->max_indx * 4;
2318			if (track->arrays[i].robj == NULL) {
2319				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2320					  "bound\n", prim_walk, i);
2321				return -EINVAL;
2322			}
2323			if (size > radeon_bo_size(track->arrays[i].robj)) {
2324				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2325					"need %lu dwords have %lu dwords\n",
2326					prim_walk, i, size >> 2,
2327					radeon_bo_size(track->arrays[i].robj)
2328					>> 2);
2329				DRM_ERROR("Max indices %u\n", track->max_indx);
2330				return -EINVAL;
2331			}
2332		}
2333		break;
2334	case 2:
2335		for (i = 0; i < track->num_arrays; i++) {
2336			size = track->arrays[i].esize * (nverts - 1) * 4;
2337			if (track->arrays[i].robj == NULL) {
2338				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2339					  "bound\n", prim_walk, i);
2340				return -EINVAL;
2341			}
2342			if (size > radeon_bo_size(track->arrays[i].robj)) {
2343				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2344					"need %lu dwords have %lu dwords\n",
2345					prim_walk, i, size >> 2,
2346					radeon_bo_size(track->arrays[i].robj)
2347					>> 2);
2348				return -EINVAL;
2349			}
2350		}
2351		break;
2352	case 3:
2353		size = track->vtx_size * nverts;
2354		if (size != track->immd_dwords) {
2355			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2356				  track->immd_dwords, size);
2357			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2358				  nverts, track->vtx_size);
2359			return -EINVAL;
2360		}
2361		break;
2362	default:
2363		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2364			  prim_walk);
2365		return -EINVAL;
2366	}
2367
2368	if (track->tex_dirty) {
2369		track->tex_dirty = false;
2370		return r100_cs_track_texture_check(rdev, track);
 
2371	}
2372	return 0;
2373}
2374
2375void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
 
2376{
2377	unsigned i, face;
2378
2379	track->cb_dirty = true;
2380	track->zb_dirty = true;
2381	track->tex_dirty = true;
2382	track->aa_dirty = true;
2383
2384	if (rdev->family < CHIP_R300) {
2385		track->num_cb = 1;
2386		if (rdev->family <= CHIP_RS200)
2387			track->num_texture = 3;
2388		else
2389			track->num_texture = 6;
2390		track->maxy = 2048;
2391		track->separate_cube = 1;
2392	} else {
2393		track->num_cb = 4;
2394		track->num_texture = 16;
2395		track->maxy = 4096;
2396		track->separate_cube = 0;
2397		track->aaresolve = false;
2398		track->aa.robj = NULL;
2399	}
2400
2401	for (i = 0; i < track->num_cb; i++) {
2402		track->cb[i].robj = NULL;
2403		track->cb[i].pitch = 8192;
2404		track->cb[i].cpp = 16;
2405		track->cb[i].offset = 0;
2406	}
2407	track->z_enabled = true;
2408	track->zb.robj = NULL;
2409	track->zb.pitch = 8192;
2410	track->zb.cpp = 4;
2411	track->zb.offset = 0;
2412	track->vtx_size = 0x7F;
2413	track->immd_dwords = 0xFFFFFFFFUL;
2414	track->num_arrays = 11;
2415	track->max_indx = 0x00FFFFFFUL;
2416	for (i = 0; i < track->num_arrays; i++) {
2417		track->arrays[i].robj = NULL;
2418		track->arrays[i].esize = 0x7F;
2419	}
2420	for (i = 0; i < track->num_texture; i++) {
2421		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2422		track->textures[i].pitch = 16536;
2423		track->textures[i].width = 16536;
2424		track->textures[i].height = 16536;
2425		track->textures[i].width_11 = 1 << 11;
2426		track->textures[i].height_11 = 1 << 11;
2427		track->textures[i].num_levels = 12;
2428		if (rdev->family <= CHIP_RS200) {
2429			track->textures[i].tex_coord_type = 0;
2430			track->textures[i].txdepth = 0;
2431		} else {
2432			track->textures[i].txdepth = 16;
2433			track->textures[i].tex_coord_type = 1;
2434		}
2435		track->textures[i].cpp = 64;
2436		track->textures[i].robj = NULL;
2437		/* CS IB emission code makes sure texture unit are disabled */
2438		track->textures[i].enabled = false;
2439		track->textures[i].lookup_disable = false;
2440		track->textures[i].roundup_w = true;
2441		track->textures[i].roundup_h = true;
2442		if (track->separate_cube)
2443			for (face = 0; face < 5; face++) {
2444				track->textures[i].cube_info[face].robj = NULL;
2445				track->textures[i].cube_info[face].width = 16536;
2446				track->textures[i].cube_info[face].height = 16536;
2447				track->textures[i].cube_info[face].offset = 0;
2448			}
2449	}
2450}
2451
2452/*
2453 * Global GPU functions
2454 */
2455static void r100_errata(struct radeon_device *rdev)
2456{
2457	rdev->pll_errata = 0;
 
2458
2459	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2460		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2461	}
 
2462
2463	if (rdev->family == CHIP_RV100 ||
2464	    rdev->family == CHIP_RS100 ||
2465	    rdev->family == CHIP_RS200) {
2466		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
 
 
 
 
2467	}
2468}
2469
2470static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2471{
2472	unsigned i;
2473	uint32_t tmp;
2474
2475	for (i = 0; i < rdev->usec_timeout; i++) {
2476		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2477		if (tmp >= n) {
2478			return 0;
2479		}
2480		DRM_UDELAY(1);
2481	}
2482	return -1;
2483}
2484
2485int r100_gui_wait_for_idle(struct radeon_device *rdev)
2486{
2487	unsigned i;
2488	uint32_t tmp;
2489
2490	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2491		printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
2492		       " Bad things might happen.\n");
2493	}
2494	for (i = 0; i < rdev->usec_timeout; i++) {
2495		tmp = RREG32(RADEON_RBBM_STATUS);
2496		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2497			return 0;
2498		}
2499		DRM_UDELAY(1);
2500	}
2501	return -1;
2502}
2503
2504int r100_mc_wait_for_idle(struct radeon_device *rdev)
2505{
2506	unsigned i;
2507	uint32_t tmp;
2508
2509	for (i = 0; i < rdev->usec_timeout; i++) {
2510		/* read MC_STATUS */
2511		tmp = RREG32(RADEON_MC_STATUS);
2512		if (tmp & RADEON_MC_IDLE) {
2513			return 0;
2514		}
2515		DRM_UDELAY(1);
2516	}
2517	return -1;
2518}
2519
2520bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2521{
2522	u32 rbbm_status;
2523
2524	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2525	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2526		radeon_ring_lockup_update(rdev, ring);
2527		return false;
2528	}
 
 
2529	return radeon_ring_test_lockup(rdev, ring);
2530}
2531
2532/* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2533void r100_enable_bm(struct radeon_device *rdev)
2534{
2535	uint32_t tmp;
2536	/* Enable bus mastering */
2537	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2538	WREG32(RADEON_BUS_CNTL, tmp);
2539}
2540
2541void r100_bm_disable(struct radeon_device *rdev)
2542{
2543	u32 tmp;
2544
2545	/* disable bus mastering */
2546	tmp = RREG32(R_000030_BUS_CNTL);
2547	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2548	mdelay(1);
2549	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2550	mdelay(1);
2551	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2552	tmp = RREG32(RADEON_BUS_CNTL);
2553	mdelay(1);
2554	pci_clear_master(rdev->pdev);
2555	mdelay(1);
2556}
2557
2558int r100_asic_reset(struct radeon_device *rdev)
2559{
2560	struct r100_mc_save save;
2561	u32 status, tmp;
2562	int ret = 0;
2563
2564	status = RREG32(R_000E40_RBBM_STATUS);
2565	if (!G_000E40_GUI_ACTIVE(status)) {
2566		return 0;
2567	}
2568	r100_mc_stop(rdev, &save);
2569	status = RREG32(R_000E40_RBBM_STATUS);
2570	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2571	/* stop CP */
2572	WREG32(RADEON_CP_CSQ_CNTL, 0);
2573	tmp = RREG32(RADEON_CP_RB_CNTL);
2574	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2575	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2576	WREG32(RADEON_CP_RB_WPTR, 0);
2577	WREG32(RADEON_CP_RB_CNTL, tmp);
2578	/* save PCI state */
2579	pci_save_state(rdev->pdev);
2580	/* disable bus mastering */
2581	r100_bm_disable(rdev);
2582	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2583					S_0000F0_SOFT_RESET_RE(1) |
2584					S_0000F0_SOFT_RESET_PP(1) |
2585					S_0000F0_SOFT_RESET_RB(1));
2586	RREG32(R_0000F0_RBBM_SOFT_RESET);
2587	mdelay(500);
2588	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2589	mdelay(1);
2590	status = RREG32(R_000E40_RBBM_STATUS);
2591	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2592	/* reset CP */
2593	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2594	RREG32(R_0000F0_RBBM_SOFT_RESET);
2595	mdelay(500);
2596	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2597	mdelay(1);
2598	status = RREG32(R_000E40_RBBM_STATUS);
2599	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2600	/* restore PCI & busmastering */
2601	pci_restore_state(rdev->pdev);
2602	r100_enable_bm(rdev);
2603	/* Check if GPU is idle */
2604	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2605		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2606		dev_err(rdev->dev, "failed to reset GPU\n");
2607		ret = -1;
2608	} else
2609		dev_info(rdev->dev, "GPU reset succeed\n");
2610	r100_mc_resume(rdev, &save);
2611	return ret;
2612}
2613
2614void r100_set_common_regs(struct radeon_device *rdev)
2615{
2616	struct drm_device *dev = rdev->ddev;
2617	bool force_dac2 = false;
2618	u32 tmp;
2619
2620	/* set these so they don't interfere with anything */
2621	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2622	WREG32(RADEON_SUBPIC_CNTL, 0);
2623	WREG32(RADEON_VIPH_CONTROL, 0);
2624	WREG32(RADEON_I2C_CNTL_1, 0);
2625	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2626	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2627	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2628
2629	/* always set up dac2 on rn50 and some rv100 as lots
2630	 * of servers seem to wire it up to a VGA port but
2631	 * don't report it in the bios connector
2632	 * table.
2633	 */
2634	switch (dev->pdev->device) {
2635		/* RN50 */
2636	case 0x515e:
2637	case 0x5969:
2638		force_dac2 = true;
2639		break;
2640		/* RV100*/
2641	case 0x5159:
2642	case 0x515a:
2643		/* DELL triple head servers */
2644		if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2645		    ((dev->pdev->subsystem_device == 0x016c) ||
2646		     (dev->pdev->subsystem_device == 0x016d) ||
2647		     (dev->pdev->subsystem_device == 0x016e) ||
2648		     (dev->pdev->subsystem_device == 0x016f) ||
2649		     (dev->pdev->subsystem_device == 0x0170) ||
2650		     (dev->pdev->subsystem_device == 0x017d) ||
2651		     (dev->pdev->subsystem_device == 0x017e) ||
2652		     (dev->pdev->subsystem_device == 0x0183) ||
2653		     (dev->pdev->subsystem_device == 0x018a) ||
2654		     (dev->pdev->subsystem_device == 0x019a)))
2655			force_dac2 = true;
2656		break;
2657	}
2658
2659	if (force_dac2) {
2660		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2661		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2662		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2663
2664		/* For CRT on DAC2, don't turn it on if BIOS didn't
2665		   enable it, even it's detected.
2666		*/
2667
2668		/* force it to crtc0 */
2669		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2670		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2671		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2672
2673		/* set up the TV DAC */
2674		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2675				 RADEON_TV_DAC_STD_MASK |
2676				 RADEON_TV_DAC_RDACPD |
2677				 RADEON_TV_DAC_GDACPD |
2678				 RADEON_TV_DAC_BDACPD |
2679				 RADEON_TV_DAC_BGADJ_MASK |
2680				 RADEON_TV_DAC_DACADJ_MASK);
2681		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2682				RADEON_TV_DAC_NHOLD |
2683				RADEON_TV_DAC_STD_PS2 |
2684				(0x58 << 16));
2685
2686		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2687		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2688		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2689	}
2690
2691	/* switch PM block to ACPI mode */
2692	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2693	tmp &= ~RADEON_PM_MODE_SEL;
2694	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2695
2696}
2697
2698/*
2699 * VRAM info
2700 */
2701static void r100_vram_get_type(struct radeon_device *rdev)
2702{
2703	uint32_t tmp;
2704
2705	rdev->mc.vram_is_ddr = false;
2706	if (rdev->flags & RADEON_IS_IGP)
2707		rdev->mc.vram_is_ddr = true;
2708	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2709		rdev->mc.vram_is_ddr = true;
2710	if ((rdev->family == CHIP_RV100) ||
2711	    (rdev->family == CHIP_RS100) ||
2712	    (rdev->family == CHIP_RS200)) {
2713		tmp = RREG32(RADEON_MEM_CNTL);
2714		if (tmp & RV100_HALF_MODE) {
2715			rdev->mc.vram_width = 32;
2716		} else {
2717			rdev->mc.vram_width = 64;
2718		}
2719		if (rdev->flags & RADEON_SINGLE_CRTC) {
2720			rdev->mc.vram_width /= 4;
2721			rdev->mc.vram_is_ddr = true;
2722		}
2723	} else if (rdev->family <= CHIP_RV280) {
2724		tmp = RREG32(RADEON_MEM_CNTL);
2725		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2726			rdev->mc.vram_width = 128;
2727		} else {
2728			rdev->mc.vram_width = 64;
2729		}
2730	} else {
2731		/* newer IGPs */
2732		rdev->mc.vram_width = 128;
2733	}
2734}
2735
2736static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2737{
2738	u32 aper_size;
2739	u8 byte;
2740
2741	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2742
2743	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2744	 * that is has the 2nd generation multifunction PCI interface
2745	 */
2746	if (rdev->family == CHIP_RV280 ||
2747	    rdev->family >= CHIP_RV350) {
2748		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2749		       ~RADEON_HDP_APER_CNTL);
2750		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2751		return aper_size * 2;
2752	}
2753
2754	/* Older cards have all sorts of funny issues to deal with. First
2755	 * check if it's a multifunction card by reading the PCI config
2756	 * header type... Limit those to one aperture size
2757	 */
2758	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2759	if (byte & 0x80) {
2760		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2761		DRM_INFO("Limiting VRAM to one aperture\n");
2762		return aper_size;
2763	}
2764
2765	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2766	 * have set it up. We don't write this as it's broken on some ASICs but
2767	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2768	 */
2769	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2770		return aper_size * 2;
2771	return aper_size;
2772}
2773
2774void r100_vram_init_sizes(struct radeon_device *rdev)
2775{
2776	u64 config_aper_size;
2777
2778	/* work out accessible VRAM */
2779	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2780	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2781	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2782	/* FIXME we don't use the second aperture yet when we could use it */
2783	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2784		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2785	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2786	if (rdev->flags & RADEON_IS_IGP) {
2787		uint32_t tom;
2788		/* read NB_TOM to get the amount of ram stolen for the GPU */
2789		tom = RREG32(RADEON_NB_TOM);
2790		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2791		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2792		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2793	} else {
2794		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2795		/* Some production boards of m6 will report 0
2796		 * if it's 8 MB
2797		 */
2798		if (rdev->mc.real_vram_size == 0) {
2799			rdev->mc.real_vram_size = 8192 * 1024;
2800			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2801		}
2802		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM - 
2803		 * Novell bug 204882 + along with lots of ubuntu ones
2804		 */
2805		if (rdev->mc.aper_size > config_aper_size)
2806			config_aper_size = rdev->mc.aper_size;
2807
2808		if (config_aper_size > rdev->mc.real_vram_size)
2809			rdev->mc.mc_vram_size = config_aper_size;
2810		else
2811			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2812	}
2813}
2814
2815void r100_vga_set_state(struct radeon_device *rdev, bool state)
2816{
2817	uint32_t temp;
2818
2819	temp = RREG32(RADEON_CONFIG_CNTL);
2820	if (state == false) {
2821		temp &= ~RADEON_CFG_VGA_RAM_EN;
2822		temp |= RADEON_CFG_VGA_IO_DIS;
2823	} else {
2824		temp &= ~RADEON_CFG_VGA_IO_DIS;
2825	}
2826	WREG32(RADEON_CONFIG_CNTL, temp);
2827}
2828
2829static void r100_mc_init(struct radeon_device *rdev)
2830{
2831	u64 base;
2832
2833	r100_vram_get_type(rdev);
2834	r100_vram_init_sizes(rdev);
2835	base = rdev->mc.aper_base;
2836	if (rdev->flags & RADEON_IS_IGP)
2837		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2838	radeon_vram_location(rdev, &rdev->mc, base);
2839	rdev->mc.gtt_base_align = 0;
2840	if (!(rdev->flags & RADEON_IS_AGP))
2841		radeon_gtt_location(rdev, &rdev->mc);
2842	radeon_update_bandwidth_info(rdev);
2843}
2844
2845
2846/*
2847 * Indirect registers accessor
2848 */
2849void r100_pll_errata_after_index(struct radeon_device *rdev)
2850{
2851	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2852		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2853		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2854	}
2855}
2856
2857static void r100_pll_errata_after_data(struct radeon_device *rdev)
2858{
2859	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2860	 * or the chip could hang on a subsequent access
2861	 */
2862	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2863		mdelay(5);
2864	}
2865
2866	/* This function is required to workaround a hardware bug in some (all?)
2867	 * revisions of the R300.  This workaround should be called after every
2868	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2869	 * may not be correct.
2870	 */
2871	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2872		uint32_t save, tmp;
2873
2874		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2875		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2876		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2877		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2878		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2879	}
2880}
2881
2882uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2883{
2884	unsigned long flags;
2885	uint32_t data;
2886
2887	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2888	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2889	r100_pll_errata_after_index(rdev);
2890	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2891	r100_pll_errata_after_data(rdev);
2892	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2893	return data;
2894}
2895
2896void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2897{
2898	unsigned long flags;
2899
2900	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2901	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2902	r100_pll_errata_after_index(rdev);
2903	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2904	r100_pll_errata_after_data(rdev);
2905	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2906}
2907
2908static void r100_set_safe_registers(struct radeon_device *rdev)
2909{
2910	if (ASIC_IS_RN50(rdev)) {
2911		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2912		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2913	} else if (rdev->family < CHIP_R200) {
2914		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2915		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2916	} else {
2917		r200_set_safe_registers(rdev);
2918	}
2919}
2920
2921/*
2922 * Debugfs info
2923 */
2924#if defined(CONFIG_DEBUG_FS)
2925static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2926{
2927	struct drm_info_node *node = (struct drm_info_node *) m->private;
2928	struct drm_device *dev = node->minor->dev;
2929	struct radeon_device *rdev = dev->dev_private;
2930	uint32_t reg, value;
2931	unsigned i;
2932
2933	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2934	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2935	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2936	for (i = 0; i < 64; i++) {
2937		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2938		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2939		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2940		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2941		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2942	}
2943	return 0;
2944}
2945
2946static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2947{
2948	struct drm_info_node *node = (struct drm_info_node *) m->private;
2949	struct drm_device *dev = node->minor->dev;
2950	struct radeon_device *rdev = dev->dev_private;
2951	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2952	uint32_t rdp, wdp;
2953	unsigned count, i, j;
2954
2955	radeon_ring_free_size(rdev, ring);
2956	rdp = RREG32(RADEON_CP_RB_RPTR);
2957	wdp = RREG32(RADEON_CP_RB_WPTR);
2958	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2959	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2960	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2961	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2962	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2963	seq_printf(m, "%u dwords in ring\n", count);
2964	if (ring->ready) {
2965		for (j = 0; j <= count; j++) {
2966			i = (rdp + j) & ring->ptr_mask;
2967			seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2968		}
2969	}
2970	return 0;
2971}
2972
2973
2974static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2975{
2976	struct drm_info_node *node = (struct drm_info_node *) m->private;
2977	struct drm_device *dev = node->minor->dev;
2978	struct radeon_device *rdev = dev->dev_private;
2979	uint32_t csq_stat, csq2_stat, tmp;
2980	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2981	unsigned i;
2982
2983	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2984	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2985	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2986	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2987	r_rptr = (csq_stat >> 0) & 0x3ff;
2988	r_wptr = (csq_stat >> 10) & 0x3ff;
2989	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2990	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2991	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2992	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2993	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2994	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2995	seq_printf(m, "Ring rptr %u\n", r_rptr);
2996	seq_printf(m, "Ring wptr %u\n", r_wptr);
2997	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2998	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2999	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
3000	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
3001	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
3002	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
3003	seq_printf(m, "Ring fifo:\n");
3004	for (i = 0; i < 256; i++) {
3005		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3006		tmp = RREG32(RADEON_CP_CSQ_DATA);
3007		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3008	}
3009	seq_printf(m, "Indirect1 fifo:\n");
3010	for (i = 256; i <= 512; i++) {
3011		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3012		tmp = RREG32(RADEON_CP_CSQ_DATA);
3013		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3014	}
3015	seq_printf(m, "Indirect2 fifo:\n");
3016	for (i = 640; i < ib1_wptr; i++) {
3017		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3018		tmp = RREG32(RADEON_CP_CSQ_DATA);
3019		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3020	}
3021	return 0;
3022}
3023
3024static int r100_debugfs_mc_info(struct seq_file *m, void *data)
3025{
3026	struct drm_info_node *node = (struct drm_info_node *) m->private;
3027	struct drm_device *dev = node->minor->dev;
3028	struct radeon_device *rdev = dev->dev_private;
3029	uint32_t tmp;
3030
3031	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3032	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3033	tmp = RREG32(RADEON_MC_FB_LOCATION);
3034	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3035	tmp = RREG32(RADEON_BUS_CNTL);
3036	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3037	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3038	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3039	tmp = RREG32(RADEON_AGP_BASE);
3040	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3041	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3042	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3043	tmp = RREG32(0x01D0);
3044	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3045	tmp = RREG32(RADEON_AIC_LO_ADDR);
3046	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3047	tmp = RREG32(RADEON_AIC_HI_ADDR);
3048	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3049	tmp = RREG32(0x01E4);
3050	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3051	return 0;
3052}
3053
3054static struct drm_info_list r100_debugfs_rbbm_list[] = {
3055	{"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3056};
3057
3058static struct drm_info_list r100_debugfs_cp_list[] = {
3059	{"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3060	{"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3061};
3062
3063static struct drm_info_list r100_debugfs_mc_info_list[] = {
3064	{"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3065};
3066#endif
3067
3068int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3069{
3070#if defined(CONFIG_DEBUG_FS)
3071	return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3072#else
3073	return 0;
3074#endif
3075}
3076
3077int r100_debugfs_cp_init(struct radeon_device *rdev)
3078{
3079#if defined(CONFIG_DEBUG_FS)
3080	return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3081#else
3082	return 0;
3083#endif
3084}
3085
3086int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3087{
3088#if defined(CONFIG_DEBUG_FS)
3089	return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3090#else
3091	return 0;
3092#endif
3093}
3094
3095int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3096			 uint32_t tiling_flags, uint32_t pitch,
3097			 uint32_t offset, uint32_t obj_size)
3098{
3099	int surf_index = reg * 16;
3100	int flags = 0;
3101
3102	if (rdev->family <= CHIP_RS200) {
3103		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3104				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3105			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3106		if (tiling_flags & RADEON_TILING_MACRO)
3107			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3108		/* setting pitch to 0 disables tiling */
3109		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3110				== 0)
3111			pitch = 0;
3112	} else if (rdev->family <= CHIP_RV280) {
3113		if (tiling_flags & (RADEON_TILING_MACRO))
3114			flags |= R200_SURF_TILE_COLOR_MACRO;
3115		if (tiling_flags & RADEON_TILING_MICRO)
3116			flags |= R200_SURF_TILE_COLOR_MICRO;
3117	} else {
3118		if (tiling_flags & RADEON_TILING_MACRO)
3119			flags |= R300_SURF_TILE_MACRO;
3120		if (tiling_flags & RADEON_TILING_MICRO)
3121			flags |= R300_SURF_TILE_MICRO;
3122	}
3123
3124	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3125		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3126	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3127		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3128
 
 
 
 
 
 
 
3129	/* r100/r200 divide by 16 */
3130	if (rdev->family < CHIP_R300)
3131		flags |= pitch / 16;
3132	else
3133		flags |= pitch / 8;
3134
3135
3136	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3137	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3138	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3139	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3140	return 0;
3141}
3142
3143void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3144{
3145	int surf_index = reg * 16;
3146	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3147}
3148
3149void r100_bandwidth_update(struct radeon_device *rdev)
3150{
3151	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3152	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3153	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
3154	fixed20_12 crit_point_ff = {0};
3155	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3156	fixed20_12 memtcas_ff[8] = {
3157		dfixed_init(1),
3158		dfixed_init(2),
3159		dfixed_init(3),
3160		dfixed_init(0),
3161		dfixed_init_half(1),
3162		dfixed_init_half(2),
3163		dfixed_init(0),
3164	};
3165	fixed20_12 memtcas_rs480_ff[8] = {
3166		dfixed_init(0),
3167		dfixed_init(1),
3168		dfixed_init(2),
3169		dfixed_init(3),
3170		dfixed_init(0),
3171		dfixed_init_half(1),
3172		dfixed_init_half(2),
3173		dfixed_init_half(3),
3174	};
3175	fixed20_12 memtcas2_ff[8] = {
3176		dfixed_init(0),
3177		dfixed_init(1),
3178		dfixed_init(2),
3179		dfixed_init(3),
3180		dfixed_init(4),
3181		dfixed_init(5),
3182		dfixed_init(6),
3183		dfixed_init(7),
3184	};
3185	fixed20_12 memtrbs[8] = {
3186		dfixed_init(1),
3187		dfixed_init_half(1),
3188		dfixed_init(2),
3189		dfixed_init_half(2),
3190		dfixed_init(3),
3191		dfixed_init_half(3),
3192		dfixed_init(4),
3193		dfixed_init_half(4)
3194	};
3195	fixed20_12 memtrbs_r4xx[8] = {
3196		dfixed_init(4),
3197		dfixed_init(5),
3198		dfixed_init(6),
3199		dfixed_init(7),
3200		dfixed_init(8),
3201		dfixed_init(9),
3202		dfixed_init(10),
3203		dfixed_init(11)
3204	};
3205	fixed20_12 min_mem_eff;
3206	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3207	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3208	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate = {0},
3209		disp_drain_rate2, read_return_rate;
3210	fixed20_12 time_disp1_drop_priority;
3211	int c;
3212	int cur_size = 16;       /* in octawords */
3213	int critical_point = 0, critical_point2;
3214/* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3215	int stop_req, max_stop_req;
3216	struct drm_display_mode *mode1 = NULL;
3217	struct drm_display_mode *mode2 = NULL;
3218	uint32_t pixel_bytes1 = 0;
3219	uint32_t pixel_bytes2 = 0;
3220
3221	/* Guess line buffer size to be 8192 pixels */
3222	u32 lb_size = 8192;
3223
3224	if (!rdev->mode_info.mode_config_initialized)
3225		return;
3226
3227	radeon_update_display_priority(rdev);
3228
3229	if (rdev->mode_info.crtcs[0]->base.enabled) {
3230		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3231		pixel_bytes1 = rdev->mode_info.crtcs[0]->base.primary->fb->bits_per_pixel / 8;
3232	}
3233	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3234		if (rdev->mode_info.crtcs[1]->base.enabled) {
3235			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3236			pixel_bytes2 = rdev->mode_info.crtcs[1]->base.primary->fb->bits_per_pixel / 8;
3237		}
3238	}
3239
3240	min_mem_eff.full = dfixed_const_8(0);
3241	/* get modes */
3242	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3243		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3244		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3245		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3246		/* check crtc enables */
3247		if (mode2)
3248			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3249		if (mode1)
3250			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3251		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3252	}
3253
3254	/*
3255	 * determine is there is enough bw for current mode
3256	 */
3257	sclk_ff = rdev->pm.sclk;
3258	mclk_ff = rdev->pm.mclk;
3259
3260	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3261	temp_ff.full = dfixed_const(temp);
3262	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3263
3264	pix_clk.full = 0;
3265	pix_clk2.full = 0;
3266	peak_disp_bw.full = 0;
3267	if (mode1) {
3268		temp_ff.full = dfixed_const(1000);
3269		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3270		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3271		temp_ff.full = dfixed_const(pixel_bytes1);
3272		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3273	}
3274	if (mode2) {
3275		temp_ff.full = dfixed_const(1000);
3276		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3277		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3278		temp_ff.full = dfixed_const(pixel_bytes2);
3279		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3280	}
3281
3282	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3283	if (peak_disp_bw.full >= mem_bw.full) {
3284		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3285			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3286	}
3287
3288	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3289	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3290	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3291		mem_trcd = ((temp >> 2) & 0x3) + 1;
3292		mem_trp  = ((temp & 0x3)) + 1;
3293		mem_tras = ((temp & 0x70) >> 4) + 1;
3294	} else if (rdev->family == CHIP_R300 ||
3295		   rdev->family == CHIP_R350) { /* r300, r350 */
3296		mem_trcd = (temp & 0x7) + 1;
3297		mem_trp = ((temp >> 8) & 0x7) + 1;
3298		mem_tras = ((temp >> 11) & 0xf) + 4;
3299	} else if (rdev->family == CHIP_RV350 ||
3300		   rdev->family <= CHIP_RV380) {
3301		/* rv3x0 */
3302		mem_trcd = (temp & 0x7) + 3;
3303		mem_trp = ((temp >> 8) & 0x7) + 3;
3304		mem_tras = ((temp >> 11) & 0xf) + 6;
3305	} else if (rdev->family == CHIP_R420 ||
3306		   rdev->family == CHIP_R423 ||
3307		   rdev->family == CHIP_RV410) {
3308		/* r4xx */
3309		mem_trcd = (temp & 0xf) + 3;
3310		if (mem_trcd > 15)
3311			mem_trcd = 15;
3312		mem_trp = ((temp >> 8) & 0xf) + 3;
3313		if (mem_trp > 15)
3314			mem_trp = 15;
3315		mem_tras = ((temp >> 12) & 0x1f) + 6;
3316		if (mem_tras > 31)
3317			mem_tras = 31;
3318	} else { /* RV200, R200 */
3319		mem_trcd = (temp & 0x7) + 1;
3320		mem_trp = ((temp >> 8) & 0x7) + 1;
3321		mem_tras = ((temp >> 12) & 0xf) + 4;
3322	}
3323	/* convert to FF */
3324	trcd_ff.full = dfixed_const(mem_trcd);
3325	trp_ff.full = dfixed_const(mem_trp);
3326	tras_ff.full = dfixed_const(mem_tras);
3327
3328	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3329	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3330	data = (temp & (7 << 20)) >> 20;
3331	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3332		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3333			tcas_ff = memtcas_rs480_ff[data];
3334		else
3335			tcas_ff = memtcas_ff[data];
3336	} else
3337		tcas_ff = memtcas2_ff[data];
3338
3339	if (rdev->family == CHIP_RS400 ||
3340	    rdev->family == CHIP_RS480) {
3341		/* extra cas latency stored in bits 23-25 0-4 clocks */
3342		data = (temp >> 23) & 0x7;
3343		if (data < 5)
3344			tcas_ff.full += dfixed_const(data);
3345	}
3346
3347	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3348		/* on the R300, Tcas is included in Trbs.
3349		 */
3350		temp = RREG32(RADEON_MEM_CNTL);
3351		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3352		if (data == 1) {
3353			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3354				temp = RREG32(R300_MC_IND_INDEX);
3355				temp &= ~R300_MC_IND_ADDR_MASK;
3356				temp |= R300_MC_READ_CNTL_CD_mcind;
3357				WREG32(R300_MC_IND_INDEX, temp);
3358				temp = RREG32(R300_MC_IND_DATA);
3359				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3360			} else {
3361				temp = RREG32(R300_MC_READ_CNTL_AB);
3362				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3363			}
3364		} else {
3365			temp = RREG32(R300_MC_READ_CNTL_AB);
3366			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3367		}
3368		if (rdev->family == CHIP_RV410 ||
3369		    rdev->family == CHIP_R420 ||
3370		    rdev->family == CHIP_R423)
3371			trbs_ff = memtrbs_r4xx[data];
3372		else
3373			trbs_ff = memtrbs[data];
3374		tcas_ff.full += trbs_ff.full;
3375	}
3376
3377	sclk_eff_ff.full = sclk_ff.full;
3378
3379	if (rdev->flags & RADEON_IS_AGP) {
3380		fixed20_12 agpmode_ff;
3381		agpmode_ff.full = dfixed_const(radeon_agpmode);
3382		temp_ff.full = dfixed_const_666(16);
3383		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3384	}
3385	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3386
3387	if (ASIC_IS_R300(rdev)) {
3388		sclk_delay_ff.full = dfixed_const(250);
3389	} else {
3390		if ((rdev->family == CHIP_RV100) ||
3391		    rdev->flags & RADEON_IS_IGP) {
3392			if (rdev->mc.vram_is_ddr)
3393				sclk_delay_ff.full = dfixed_const(41);
3394			else
3395				sclk_delay_ff.full = dfixed_const(33);
3396		} else {
3397			if (rdev->mc.vram_width == 128)
3398				sclk_delay_ff.full = dfixed_const(57);
3399			else
3400				sclk_delay_ff.full = dfixed_const(41);
3401		}
3402	}
3403
3404	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3405
3406	if (rdev->mc.vram_is_ddr) {
3407		if (rdev->mc.vram_width == 32) {
3408			k1.full = dfixed_const(40);
3409			c  = 3;
3410		} else {
3411			k1.full = dfixed_const(20);
3412			c  = 1;
3413		}
3414	} else {
3415		k1.full = dfixed_const(40);
3416		c  = 3;
3417	}
3418
3419	temp_ff.full = dfixed_const(2);
3420	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3421	temp_ff.full = dfixed_const(c);
3422	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3423	temp_ff.full = dfixed_const(4);
3424	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3425	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3426	mc_latency_mclk.full += k1.full;
3427
3428	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3429	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3430
3431	/*
3432	  HW cursor time assuming worst case of full size colour cursor.
3433	*/
3434	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3435	temp_ff.full += trcd_ff.full;
3436	if (temp_ff.full < tras_ff.full)
3437		temp_ff.full = tras_ff.full;
3438	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3439
3440	temp_ff.full = dfixed_const(cur_size);
3441	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3442	/*
3443	  Find the total latency for the display data.
3444	*/
3445	disp_latency_overhead.full = dfixed_const(8);
3446	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3447	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3448	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3449
3450	if (mc_latency_mclk.full > mc_latency_sclk.full)
3451		disp_latency.full = mc_latency_mclk.full;
3452	else
3453		disp_latency.full = mc_latency_sclk.full;
3454
3455	/* setup Max GRPH_STOP_REQ default value */
3456	if (ASIC_IS_RV100(rdev))
3457		max_stop_req = 0x5c;
3458	else
3459		max_stop_req = 0x7c;
3460
3461	if (mode1) {
3462		/*  CRTC1
3463		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3464		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3465		*/
3466		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3467
3468		if (stop_req > max_stop_req)
3469			stop_req = max_stop_req;
3470
3471		/*
3472		  Find the drain rate of the display buffer.
3473		*/
3474		temp_ff.full = dfixed_const((16/pixel_bytes1));
3475		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3476
3477		/*
3478		  Find the critical point of the display buffer.
3479		*/
3480		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3481		crit_point_ff.full += dfixed_const_half(0);
3482
3483		critical_point = dfixed_trunc(crit_point_ff);
3484
3485		if (rdev->disp_priority == 2) {
3486			critical_point = 0;
3487		}
3488
3489		/*
3490		  The critical point should never be above max_stop_req-4.  Setting
3491		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3492		*/
3493		if (max_stop_req - critical_point < 4)
3494			critical_point = 0;
3495
3496		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3497			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3498			critical_point = 0x10;
3499		}
3500
3501		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3502		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3503		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3504		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3505		if ((rdev->family == CHIP_R350) &&
3506		    (stop_req > 0x15)) {
3507			stop_req -= 0x10;
3508		}
3509		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3510		temp |= RADEON_GRPH_BUFFER_SIZE;
3511		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3512			  RADEON_GRPH_CRITICAL_AT_SOF |
3513			  RADEON_GRPH_STOP_CNTL);
3514		/*
3515		  Write the result into the register.
3516		*/
3517		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3518						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3519
3520#if 0
3521		if ((rdev->family == CHIP_RS400) ||
3522		    (rdev->family == CHIP_RS480)) {
3523			/* attempt to program RS400 disp regs correctly ??? */
3524			temp = RREG32(RS400_DISP1_REG_CNTL);
3525			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3526				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3527			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3528						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3529						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3530			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3531			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3532				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3533			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3534						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3535						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3536		}
3537#endif
3538
3539		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3540			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3541			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3542	}
3543
3544	if (mode2) {
3545		u32 grph2_cntl;
3546		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3547
3548		if (stop_req > max_stop_req)
3549			stop_req = max_stop_req;
3550
3551		/*
3552		  Find the drain rate of the display buffer.
3553		*/
3554		temp_ff.full = dfixed_const((16/pixel_bytes2));
3555		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3556
3557		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3558		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3559		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3560		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3561		if ((rdev->family == CHIP_R350) &&
3562		    (stop_req > 0x15)) {
3563			stop_req -= 0x10;
3564		}
3565		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3566		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3567		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3568			  RADEON_GRPH_CRITICAL_AT_SOF |
3569			  RADEON_GRPH_STOP_CNTL);
3570
3571		if ((rdev->family == CHIP_RS100) ||
3572		    (rdev->family == CHIP_RS200))
3573			critical_point2 = 0;
3574		else {
3575			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3576			temp_ff.full = dfixed_const(temp);
3577			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3578			if (sclk_ff.full < temp_ff.full)
3579				temp_ff.full = sclk_ff.full;
3580
3581			read_return_rate.full = temp_ff.full;
3582
3583			if (mode1) {
3584				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3585				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3586			} else {
3587				time_disp1_drop_priority.full = 0;
3588			}
3589			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3590			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3591			crit_point_ff.full += dfixed_const_half(0);
3592
3593			critical_point2 = dfixed_trunc(crit_point_ff);
3594
3595			if (rdev->disp_priority == 2) {
3596				critical_point2 = 0;
3597			}
3598
3599			if (max_stop_req - critical_point2 < 4)
3600				critical_point2 = 0;
3601
3602		}
3603
3604		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3605			/* some R300 cards have problem with this set to 0 */
3606			critical_point2 = 0x10;
3607		}
3608
3609		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3610						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3611
3612		if ((rdev->family == CHIP_RS400) ||
3613		    (rdev->family == CHIP_RS480)) {
3614#if 0
3615			/* attempt to program RS400 disp2 regs correctly ??? */
3616			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3617			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3618				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3619			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3620						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3621						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3622			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3623			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3624				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3625			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3626						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3627						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3628#endif
3629			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3630			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3631			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3632			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3633		}
3634
3635		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3636			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3637	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3638
3639	/* Save number of lines the linebuffer leads before the scanout */
3640	if (mode1)
3641	    rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3642
3643	if (mode2)
3644	    rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3645}
3646
3647int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3648{
3649	uint32_t scratch;
3650	uint32_t tmp = 0;
3651	unsigned i;
3652	int r;
3653
3654	r = radeon_scratch_get(rdev, &scratch);
3655	if (r) {
3656		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3657		return r;
3658	}
3659	WREG32(scratch, 0xCAFEDEAD);
3660	r = radeon_ring_lock(rdev, ring, 2);
3661	if (r) {
3662		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3663		radeon_scratch_free(rdev, scratch);
3664		return r;
3665	}
3666	radeon_ring_write(ring, PACKET0(scratch, 0));
3667	radeon_ring_write(ring, 0xDEADBEEF);
3668	radeon_ring_unlock_commit(rdev, ring, false);
3669	for (i = 0; i < rdev->usec_timeout; i++) {
3670		tmp = RREG32(scratch);
3671		if (tmp == 0xDEADBEEF) {
3672			break;
3673		}
3674		DRM_UDELAY(1);
3675	}
3676	if (i < rdev->usec_timeout) {
3677		DRM_INFO("ring test succeeded in %d usecs\n", i);
3678	} else {
3679		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3680			  scratch, tmp);
3681		r = -EINVAL;
3682	}
3683	radeon_scratch_free(rdev, scratch);
3684	return r;
3685}
3686
3687void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3688{
3689	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3690
3691	if (ring->rptr_save_reg) {
3692		u32 next_rptr = ring->wptr + 2 + 3;
3693		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3694		radeon_ring_write(ring, next_rptr);
3695	}
3696
3697	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3698	radeon_ring_write(ring, ib->gpu_addr);
3699	radeon_ring_write(ring, ib->length_dw);
3700}
3701
3702int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3703{
3704	struct radeon_ib ib;
3705	uint32_t scratch;
3706	uint32_t tmp = 0;
3707	unsigned i;
3708	int r;
3709
3710	r = radeon_scratch_get(rdev, &scratch);
3711	if (r) {
3712		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3713		return r;
3714	}
3715	WREG32(scratch, 0xCAFEDEAD);
3716	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3717	if (r) {
3718		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3719		goto free_scratch;
3720	}
3721	ib.ptr[0] = PACKET0(scratch, 0);
3722	ib.ptr[1] = 0xDEADBEEF;
3723	ib.ptr[2] = PACKET2(0);
3724	ib.ptr[3] = PACKET2(0);
3725	ib.ptr[4] = PACKET2(0);
3726	ib.ptr[5] = PACKET2(0);
3727	ib.ptr[6] = PACKET2(0);
3728	ib.ptr[7] = PACKET2(0);
3729	ib.length_dw = 8;
3730	r = radeon_ib_schedule(rdev, &ib, NULL, false);
3731	if (r) {
3732		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3733		goto free_ib;
 
3734	}
3735	r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
3736		RADEON_USEC_IB_TEST_TIMEOUT));
3737	if (r < 0) {
3738		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3739		goto free_ib;
3740	} else if (r == 0) {
3741		DRM_ERROR("radeon: fence wait timed out.\n");
3742		r = -ETIMEDOUT;
3743		goto free_ib;
3744	}
3745	r = 0;
3746	for (i = 0; i < rdev->usec_timeout; i++) {
3747		tmp = RREG32(scratch);
3748		if (tmp == 0xDEADBEEF) {
3749			break;
3750		}
3751		DRM_UDELAY(1);
3752	}
3753	if (i < rdev->usec_timeout) {
3754		DRM_INFO("ib test succeeded in %u usecs\n", i);
3755	} else {
3756		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3757			  scratch, tmp);
3758		r = -EINVAL;
3759	}
3760free_ib:
3761	radeon_ib_free(rdev, &ib);
3762free_scratch:
3763	radeon_scratch_free(rdev, scratch);
 
3764	return r;
3765}
3766
 
 
 
 
 
 
3767void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3768{
3769	/* Shutdown CP we shouldn't need to do that but better be safe than
3770	 * sorry
3771	 */
3772	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3773	WREG32(R_000740_CP_CSQ_CNTL, 0);
3774
3775	/* Save few CRTC registers */
3776	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3777	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3778	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3779	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3780	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3781		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3782		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3783	}
3784
3785	/* Disable VGA aperture access */
3786	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3787	/* Disable cursor, overlay, crtc */
3788	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3789	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3790					S_000054_CRTC_DISPLAY_DIS(1));
3791	WREG32(R_000050_CRTC_GEN_CNTL,
3792			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3793			S_000050_CRTC_DISP_REQ_EN_B(1));
3794	WREG32(R_000420_OV0_SCALE_CNTL,
3795		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3796	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3797	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3798		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3799						S_000360_CUR2_LOCK(1));
3800		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3801			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3802			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3803			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3804		WREG32(R_000360_CUR2_OFFSET,
3805			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3806	}
3807}
3808
3809void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3810{
3811	/* Update base address for crtc */
3812	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3813	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3814		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3815	}
3816	/* Restore CRTC registers */
3817	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3818	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3819	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3820	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3821		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3822	}
3823}
3824
3825void r100_vga_render_disable(struct radeon_device *rdev)
3826{
3827	u32 tmp;
3828
3829	tmp = RREG8(R_0003C2_GENMO_WT);
3830	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3831}
3832
3833static void r100_debugfs(struct radeon_device *rdev)
3834{
3835	int r;
3836
3837	r = r100_debugfs_mc_info_init(rdev);
3838	if (r)
3839		dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3840}
3841
3842static void r100_mc_program(struct radeon_device *rdev)
3843{
3844	struct r100_mc_save save;
3845
3846	/* Stops all mc clients */
3847	r100_mc_stop(rdev, &save);
3848	if (rdev->flags & RADEON_IS_AGP) {
3849		WREG32(R_00014C_MC_AGP_LOCATION,
3850			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3851			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3852		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3853		if (rdev->family > CHIP_RV200)
3854			WREG32(R_00015C_AGP_BASE_2,
3855				upper_32_bits(rdev->mc.agp_base) & 0xff);
3856	} else {
3857		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3858		WREG32(R_000170_AGP_BASE, 0);
3859		if (rdev->family > CHIP_RV200)
3860			WREG32(R_00015C_AGP_BASE_2, 0);
3861	}
3862	/* Wait for mc idle */
3863	if (r100_mc_wait_for_idle(rdev))
3864		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3865	/* Program MC, should be a 32bits limited address space */
3866	WREG32(R_000148_MC_FB_LOCATION,
3867		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3868		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3869	r100_mc_resume(rdev, &save);
3870}
3871
3872static void r100_clock_startup(struct radeon_device *rdev)
3873{
3874	u32 tmp;
3875
3876	if (radeon_dynclks != -1 && radeon_dynclks)
3877		radeon_legacy_set_clock_gating(rdev, 1);
3878	/* We need to force on some of the block */
3879	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3880	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3881	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3882		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3883	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3884}
3885
3886static int r100_startup(struct radeon_device *rdev)
3887{
3888	int r;
3889
3890	/* set common regs */
3891	r100_set_common_regs(rdev);
3892	/* program mc */
3893	r100_mc_program(rdev);
3894	/* Resume clock */
3895	r100_clock_startup(rdev);
3896	/* Initialize GART (initialize after TTM so we can allocate
3897	 * memory through TTM but finalize after TTM) */
3898	r100_enable_bm(rdev);
3899	if (rdev->flags & RADEON_IS_PCI) {
3900		r = r100_pci_gart_enable(rdev);
3901		if (r)
3902			return r;
3903	}
3904
3905	/* allocate wb buffer */
3906	r = radeon_wb_init(rdev);
3907	if (r)
3908		return r;
3909
3910	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3911	if (r) {
3912		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3913		return r;
3914	}
3915
3916	/* Enable IRQ */
3917	if (!rdev->irq.installed) {
3918		r = radeon_irq_kms_init(rdev);
3919		if (r)
3920			return r;
3921	}
3922
3923	r100_irq_set(rdev);
3924	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3925	/* 1M ring buffer */
3926	r = r100_cp_init(rdev, 1024 * 1024);
3927	if (r) {
3928		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3929		return r;
3930	}
3931
3932	r = radeon_ib_pool_init(rdev);
3933	if (r) {
3934		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
 
 
 
3935		return r;
3936	}
3937
3938	return 0;
3939}
3940
3941int r100_resume(struct radeon_device *rdev)
3942{
3943	int r;
3944
3945	/* Make sur GART are not working */
3946	if (rdev->flags & RADEON_IS_PCI)
3947		r100_pci_gart_disable(rdev);
3948	/* Resume clock before doing reset */
3949	r100_clock_startup(rdev);
3950	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3951	if (radeon_asic_reset(rdev)) {
3952		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3953			RREG32(R_000E40_RBBM_STATUS),
3954			RREG32(R_0007C0_CP_STAT));
3955	}
3956	/* post */
3957	radeon_combios_asic_init(rdev->ddev);
3958	/* Resume clock after posting */
3959	r100_clock_startup(rdev);
3960	/* Initialize surface registers */
3961	radeon_surface_init(rdev);
3962
3963	rdev->accel_working = true;
3964	r = r100_startup(rdev);
3965	if (r) {
3966		rdev->accel_working = false;
3967	}
3968	return r;
3969}
3970
3971int r100_suspend(struct radeon_device *rdev)
3972{
3973	radeon_pm_suspend(rdev);
3974	r100_cp_disable(rdev);
3975	radeon_wb_disable(rdev);
3976	r100_irq_disable(rdev);
3977	if (rdev->flags & RADEON_IS_PCI)
3978		r100_pci_gart_disable(rdev);
3979	return 0;
3980}
3981
3982void r100_fini(struct radeon_device *rdev)
3983{
3984	radeon_pm_fini(rdev);
3985	r100_cp_fini(rdev);
3986	radeon_wb_fini(rdev);
3987	radeon_ib_pool_fini(rdev);
3988	radeon_gem_fini(rdev);
3989	if (rdev->flags & RADEON_IS_PCI)
3990		r100_pci_gart_fini(rdev);
3991	radeon_agp_fini(rdev);
3992	radeon_irq_kms_fini(rdev);
3993	radeon_fence_driver_fini(rdev);
3994	radeon_bo_fini(rdev);
3995	radeon_atombios_fini(rdev);
3996	kfree(rdev->bios);
3997	rdev->bios = NULL;
3998}
3999
4000/*
4001 * Due to how kexec works, it can leave the hw fully initialised when it
4002 * boots the new kernel. However doing our init sequence with the CP and
4003 * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
4004 * do some quick sanity checks and restore sane values to avoid this
4005 * problem.
4006 */
4007void r100_restore_sanity(struct radeon_device *rdev)
4008{
4009	u32 tmp;
4010
4011	tmp = RREG32(RADEON_CP_CSQ_CNTL);
4012	if (tmp) {
4013		WREG32(RADEON_CP_CSQ_CNTL, 0);
4014	}
4015	tmp = RREG32(RADEON_CP_RB_CNTL);
4016	if (tmp) {
4017		WREG32(RADEON_CP_RB_CNTL, 0);
4018	}
4019	tmp = RREG32(RADEON_SCRATCH_UMSK);
4020	if (tmp) {
4021		WREG32(RADEON_SCRATCH_UMSK, 0);
4022	}
4023}
4024
4025int r100_init(struct radeon_device *rdev)
4026{
4027	int r;
4028
4029	/* Register debugfs file specific to this group of asics */
4030	r100_debugfs(rdev);
4031	/* Disable VGA */
4032	r100_vga_render_disable(rdev);
4033	/* Initialize scratch registers */
4034	radeon_scratch_init(rdev);
4035	/* Initialize surface registers */
4036	radeon_surface_init(rdev);
4037	/* sanity check some register to avoid hangs like after kexec */
4038	r100_restore_sanity(rdev);
4039	/* TODO: disable VGA need to use VGA request */
4040	/* BIOS*/
4041	if (!radeon_get_bios(rdev)) {
4042		if (ASIC_IS_AVIVO(rdev))
4043			return -EINVAL;
4044	}
4045	if (rdev->is_atom_bios) {
4046		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4047		return -EINVAL;
4048	} else {
4049		r = radeon_combios_init(rdev);
4050		if (r)
4051			return r;
4052	}
4053	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4054	if (radeon_asic_reset(rdev)) {
4055		dev_warn(rdev->dev,
4056			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4057			RREG32(R_000E40_RBBM_STATUS),
4058			RREG32(R_0007C0_CP_STAT));
4059	}
4060	/* check if cards are posted or not */
4061	if (radeon_boot_test_post_card(rdev) == false)
4062		return -EINVAL;
4063	/* Set asic errata */
4064	r100_errata(rdev);
4065	/* Initialize clocks */
4066	radeon_get_clock_info(rdev->ddev);
4067	/* initialize AGP */
4068	if (rdev->flags & RADEON_IS_AGP) {
4069		r = radeon_agp_init(rdev);
4070		if (r) {
4071			radeon_agp_disable(rdev);
4072		}
4073	}
4074	/* initialize VRAM */
4075	r100_mc_init(rdev);
4076	/* Fence driver */
4077	r = radeon_fence_driver_init(rdev);
4078	if (r)
4079		return r;
 
 
 
4080	/* Memory manager */
4081	r = radeon_bo_init(rdev);
4082	if (r)
4083		return r;
4084	if (rdev->flags & RADEON_IS_PCI) {
4085		r = r100_pci_gart_init(rdev);
4086		if (r)
4087			return r;
4088	}
4089	r100_set_safe_registers(rdev);
4090
4091	/* Initialize power management */
4092	radeon_pm_init(rdev);
4093
4094	rdev->accel_working = true;
 
 
 
 
 
4095	r = r100_startup(rdev);
4096	if (r) {
4097		/* Somethings want wront with the accel init stop accel */
4098		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4099		r100_cp_fini(rdev);
4100		radeon_wb_fini(rdev);
4101		radeon_ib_pool_fini(rdev);
4102		radeon_irq_kms_fini(rdev);
4103		if (rdev->flags & RADEON_IS_PCI)
4104			r100_pci_gart_fini(rdev);
4105		rdev->accel_working = false;
4106	}
4107	return 0;
4108}
4109
4110uint32_t r100_mm_rreg_slow(struct radeon_device *rdev, uint32_t reg)
4111{
4112	unsigned long flags;
4113	uint32_t ret;
4114
4115	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4116	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4117	ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4118	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4119	return ret;
4120}
4121
4122void r100_mm_wreg_slow(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4123{
4124	unsigned long flags;
4125
4126	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4127	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4128	writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4129	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4130}
4131
4132u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4133{
4134	if (reg < rdev->rio_mem_size)
4135		return ioread32(rdev->rio_mem + reg);
4136	else {
4137		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4138		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4139	}
4140}
4141
4142void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4143{
4144	if (reg < rdev->rio_mem_size)
4145		iowrite32(v, rdev->rio_mem + reg);
4146	else {
4147		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4148		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4149	}
4150}
v3.5.6
   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 "drmP.h"
  31#include "drm.h"
  32#include "radeon_drm.h"
  33#include "radeon_reg.h"
  34#include "radeon.h"
  35#include "radeon_asic.h"
  36#include "r100d.h"
  37#include "rs100d.h"
  38#include "rv200d.h"
  39#include "rv250d.h"
  40#include "atom.h"
  41
  42#include <linux/firmware.h>
  43#include <linux/platform_device.h>
  44#include <linux/module.h>
  45
  46#include "r100_reg_safe.h"
  47#include "rn50_reg_safe.h"
  48
  49/* Firmware Names */
  50#define FIRMWARE_R100		"radeon/R100_cp.bin"
  51#define FIRMWARE_R200		"radeon/R200_cp.bin"
  52#define FIRMWARE_R300		"radeon/R300_cp.bin"
  53#define FIRMWARE_R420		"radeon/R420_cp.bin"
  54#define FIRMWARE_RS690		"radeon/RS690_cp.bin"
  55#define FIRMWARE_RS600		"radeon/RS600_cp.bin"
  56#define FIRMWARE_R520		"radeon/R520_cp.bin"
  57
  58MODULE_FIRMWARE(FIRMWARE_R100);
  59MODULE_FIRMWARE(FIRMWARE_R200);
  60MODULE_FIRMWARE(FIRMWARE_R300);
  61MODULE_FIRMWARE(FIRMWARE_R420);
  62MODULE_FIRMWARE(FIRMWARE_RS690);
  63MODULE_FIRMWARE(FIRMWARE_RS600);
  64MODULE_FIRMWARE(FIRMWARE_R520);
  65
  66#include "r100_track.h"
  67
  68void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
 
 
 
 
 
  69{
  70	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc];
  71	int i;
  72
  73	if (radeon_crtc->crtc_id == 0) {
  74		if (RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN) {
  75			for (i = 0; i < rdev->usec_timeout; i++) {
  76				if (!(RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR))
  77					break;
  78				udelay(1);
  79			}
  80			for (i = 0; i < rdev->usec_timeout; i++) {
  81				if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
  82					break;
  83				udelay(1);
  84			}
  85		}
  86	} else {
  87		if (RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN) {
  88			for (i = 0; i < rdev->usec_timeout; i++) {
  89				if (!(RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR))
  90					break;
  91				udelay(1);
  92			}
  93			for (i = 0; i < rdev->usec_timeout; i++) {
  94				if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
  95					break;
  96				udelay(1);
  97			}
  98		}
  99	}
 100}
 101
 102/* This files gather functions specifics to:
 103 * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
 104 */
 105
 106int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
 107			    struct radeon_cs_packet *pkt,
 108			    unsigned idx,
 109			    unsigned reg)
 110{
 111	int r;
 112	u32 tile_flags = 0;
 113	u32 tmp;
 114	struct radeon_cs_reloc *reloc;
 115	u32 value;
 116
 117	r = r100_cs_packet_next_reloc(p, &reloc);
 118	if (r) {
 119		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 120			  idx, reg);
 121		r100_cs_dump_packet(p, pkt);
 122		return r;
 123	}
 124
 125	value = radeon_get_ib_value(p, idx);
 126	tmp = value & 0x003fffff;
 127	tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
 128
 129	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
 130		if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 131			tile_flags |= RADEON_DST_TILE_MACRO;
 132		if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
 133			if (reg == RADEON_SRC_PITCH_OFFSET) {
 134				DRM_ERROR("Cannot src blit from microtiled surface\n");
 135				r100_cs_dump_packet(p, pkt);
 136				return -EINVAL;
 137			}
 138			tile_flags |= RADEON_DST_TILE_MICRO;
 139		}
 140
 141		tmp |= tile_flags;
 142		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
 143	} else
 144		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
 145	return 0;
 146}
 147
 148int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
 149			     struct radeon_cs_packet *pkt,
 150			     int idx)
 
 
 
 
 
 
 151{
 152	unsigned c, i;
 153	struct radeon_cs_reloc *reloc;
 154	struct r100_cs_track *track;
 155	int r = 0;
 156	volatile uint32_t *ib;
 157	u32 idx_value;
 158
 159	ib = p->ib.ptr;
 160	track = (struct r100_cs_track *)p->track;
 161	c = radeon_get_ib_value(p, idx++) & 0x1F;
 162	if (c > 16) {
 163	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
 164		      pkt->opcode);
 165	    r100_cs_dump_packet(p, pkt);
 166	    return -EINVAL;
 167	}
 168	track->num_arrays = c;
 169	for (i = 0; i < (c - 1); i+=2, idx+=3) {
 170		r = r100_cs_packet_next_reloc(p, &reloc);
 171		if (r) {
 172			DRM_ERROR("No reloc for packet3 %d\n",
 173				  pkt->opcode);
 174			r100_cs_dump_packet(p, pkt);
 175			return r;
 176		}
 177		idx_value = radeon_get_ib_value(p, idx);
 178		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
 179
 180		track->arrays[i + 0].esize = idx_value >> 8;
 181		track->arrays[i + 0].robj = reloc->robj;
 182		track->arrays[i + 0].esize &= 0x7F;
 183		r = r100_cs_packet_next_reloc(p, &reloc);
 184		if (r) {
 185			DRM_ERROR("No reloc for packet3 %d\n",
 186				  pkt->opcode);
 187			r100_cs_dump_packet(p, pkt);
 188			return r;
 189		}
 190		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->lobj.gpu_offset);
 191		track->arrays[i + 1].robj = reloc->robj;
 192		track->arrays[i + 1].esize = idx_value >> 24;
 193		track->arrays[i + 1].esize &= 0x7F;
 194	}
 195	if (c & 1) {
 196		r = r100_cs_packet_next_reloc(p, &reloc);
 197		if (r) {
 198			DRM_ERROR("No reloc for packet3 %d\n",
 199					  pkt->opcode);
 200			r100_cs_dump_packet(p, pkt);
 201			return r;
 202		}
 203		idx_value = radeon_get_ib_value(p, idx);
 204		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
 205		track->arrays[i + 0].robj = reloc->robj;
 206		track->arrays[i + 0].esize = idx_value >> 8;
 207		track->arrays[i + 0].esize &= 0x7F;
 208	}
 209	return r;
 210}
 211
 212void r100_pre_page_flip(struct radeon_device *rdev, int crtc)
 213{
 214	/* enable the pflip int */
 215	radeon_irq_kms_pflip_irq_get(rdev, crtc);
 216}
 217
 218void r100_post_page_flip(struct radeon_device *rdev, int crtc)
 219{
 220	/* disable the pflip int */
 221	radeon_irq_kms_pflip_irq_put(rdev, crtc);
 222}
 223
 224u32 r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
 225{
 226	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
 227	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
 228	int i;
 229
 230	/* Lock the graphics update lock */
 231	/* update the scanout addresses */
 232	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 233
 234	/* Wait for update_pending to go high. */
 235	for (i = 0; i < rdev->usec_timeout; i++) {
 236		if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
 237			break;
 238		udelay(1);
 239	}
 240	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
 241
 242	/* Unlock the lock, so double-buffering can take place inside vblank */
 243	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
 244	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 245
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 246	/* Return current update_pending status: */
 247	return RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET;
 
 248}
 249
 
 
 
 
 
 
 
 
 
 250void r100_pm_get_dynpm_state(struct radeon_device *rdev)
 251{
 252	int i;
 253	rdev->pm.dynpm_can_upclock = true;
 254	rdev->pm.dynpm_can_downclock = true;
 255
 256	switch (rdev->pm.dynpm_planned_action) {
 257	case DYNPM_ACTION_MINIMUM:
 258		rdev->pm.requested_power_state_index = 0;
 259		rdev->pm.dynpm_can_downclock = false;
 260		break;
 261	case DYNPM_ACTION_DOWNCLOCK:
 262		if (rdev->pm.current_power_state_index == 0) {
 263			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 264			rdev->pm.dynpm_can_downclock = false;
 265		} else {
 266			if (rdev->pm.active_crtc_count > 1) {
 267				for (i = 0; i < rdev->pm.num_power_states; i++) {
 268					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 269						continue;
 270					else if (i >= rdev->pm.current_power_state_index) {
 271						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 272						break;
 273					} else {
 274						rdev->pm.requested_power_state_index = i;
 275						break;
 276					}
 277				}
 278			} else
 279				rdev->pm.requested_power_state_index =
 280					rdev->pm.current_power_state_index - 1;
 281		}
 282		/* don't use the power state if crtcs are active and no display flag is set */
 283		if ((rdev->pm.active_crtc_count > 0) &&
 284		    (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
 285		     RADEON_PM_MODE_NO_DISPLAY)) {
 286			rdev->pm.requested_power_state_index++;
 287		}
 288		break;
 289	case DYNPM_ACTION_UPCLOCK:
 290		if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
 291			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 292			rdev->pm.dynpm_can_upclock = false;
 293		} else {
 294			if (rdev->pm.active_crtc_count > 1) {
 295				for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
 296					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 297						continue;
 298					else if (i <= rdev->pm.current_power_state_index) {
 299						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 300						break;
 301					} else {
 302						rdev->pm.requested_power_state_index = i;
 303						break;
 304					}
 305				}
 306			} else
 307				rdev->pm.requested_power_state_index =
 308					rdev->pm.current_power_state_index + 1;
 309		}
 310		break;
 311	case DYNPM_ACTION_DEFAULT:
 312		rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
 313		rdev->pm.dynpm_can_upclock = false;
 314		break;
 315	case DYNPM_ACTION_NONE:
 316	default:
 317		DRM_ERROR("Requested mode for not defined action\n");
 318		return;
 319	}
 320	/* only one clock mode per power state */
 321	rdev->pm.requested_clock_mode_index = 0;
 322
 323	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
 324		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 325		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
 326		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 327		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
 328		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 329		  pcie_lanes);
 330}
 331
 
 
 
 
 
 
 
 
 
 332void r100_pm_init_profile(struct radeon_device *rdev)
 333{
 334	/* default */
 335	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
 336	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 337	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
 338	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
 339	/* low sh */
 340	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
 341	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
 342	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
 343	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
 344	/* mid sh */
 345	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
 346	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
 347	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
 348	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
 349	/* high sh */
 350	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
 351	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 352	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
 353	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
 354	/* low mh */
 355	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
 356	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 357	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
 358	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
 359	/* mid mh */
 360	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
 361	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 362	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
 363	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
 364	/* high mh */
 365	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
 366	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 367	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
 368	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
 369}
 370
 
 
 
 
 
 
 
 
 371void r100_pm_misc(struct radeon_device *rdev)
 372{
 373	int requested_index = rdev->pm.requested_power_state_index;
 374	struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
 375	struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
 376	u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
 377
 378	if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
 379		if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
 380			tmp = RREG32(voltage->gpio.reg);
 381			if (voltage->active_high)
 382				tmp |= voltage->gpio.mask;
 383			else
 384				tmp &= ~(voltage->gpio.mask);
 385			WREG32(voltage->gpio.reg, tmp);
 386			if (voltage->delay)
 387				udelay(voltage->delay);
 388		} else {
 389			tmp = RREG32(voltage->gpio.reg);
 390			if (voltage->active_high)
 391				tmp &= ~voltage->gpio.mask;
 392			else
 393				tmp |= voltage->gpio.mask;
 394			WREG32(voltage->gpio.reg, tmp);
 395			if (voltage->delay)
 396				udelay(voltage->delay);
 397		}
 398	}
 399
 400	sclk_cntl = RREG32_PLL(SCLK_CNTL);
 401	sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
 402	sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
 403	sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
 404	sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
 405	if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
 406		sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
 407		if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
 408			sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
 409		else
 410			sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
 411		if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
 412			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
 413		else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
 414			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
 415	} else
 416		sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
 417
 418	if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
 419		sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
 420		if (voltage->delay) {
 421			sclk_more_cntl |= VOLTAGE_DROP_SYNC;
 422			switch (voltage->delay) {
 423			case 33:
 424				sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
 425				break;
 426			case 66:
 427				sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
 428				break;
 429			case 99:
 430				sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
 431				break;
 432			case 132:
 433				sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
 434				break;
 435			}
 436		} else
 437			sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
 438	} else
 439		sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
 440
 441	if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
 442		sclk_cntl &= ~FORCE_HDP;
 443	else
 444		sclk_cntl |= FORCE_HDP;
 445
 446	WREG32_PLL(SCLK_CNTL, sclk_cntl);
 447	WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
 448	WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
 449
 450	/* set pcie lanes */
 451	if ((rdev->flags & RADEON_IS_PCIE) &&
 452	    !(rdev->flags & RADEON_IS_IGP) &&
 453	    rdev->asic->pm.set_pcie_lanes &&
 454	    (ps->pcie_lanes !=
 455	     rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
 456		radeon_set_pcie_lanes(rdev,
 457				      ps->pcie_lanes);
 458		DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
 459	}
 460}
 461
 
 
 
 
 
 
 
 462void r100_pm_prepare(struct radeon_device *rdev)
 463{
 464	struct drm_device *ddev = rdev->ddev;
 465	struct drm_crtc *crtc;
 466	struct radeon_crtc *radeon_crtc;
 467	u32 tmp;
 468
 469	/* disable any active CRTCs */
 470	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 471		radeon_crtc = to_radeon_crtc(crtc);
 472		if (radeon_crtc->enabled) {
 473			if (radeon_crtc->crtc_id) {
 474				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 475				tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
 476				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 477			} else {
 478				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 479				tmp |= RADEON_CRTC_DISP_REQ_EN_B;
 480				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 481			}
 482		}
 483	}
 484}
 485
 
 
 
 
 
 
 
 486void r100_pm_finish(struct radeon_device *rdev)
 487{
 488	struct drm_device *ddev = rdev->ddev;
 489	struct drm_crtc *crtc;
 490	struct radeon_crtc *radeon_crtc;
 491	u32 tmp;
 492
 493	/* enable any active CRTCs */
 494	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 495		radeon_crtc = to_radeon_crtc(crtc);
 496		if (radeon_crtc->enabled) {
 497			if (radeon_crtc->crtc_id) {
 498				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 499				tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
 500				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 501			} else {
 502				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 503				tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
 504				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 505			}
 506		}
 507	}
 508}
 509
 
 
 
 
 
 
 
 
 510bool r100_gui_idle(struct radeon_device *rdev)
 511{
 512	if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
 513		return false;
 514	else
 515		return true;
 516}
 517
 518/* hpd for digital panel detect/disconnect */
 
 
 
 
 
 
 
 
 
 519bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
 520{
 521	bool connected = false;
 522
 523	switch (hpd) {
 524	case RADEON_HPD_1:
 525		if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
 526			connected = true;
 527		break;
 528	case RADEON_HPD_2:
 529		if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
 530			connected = true;
 531		break;
 532	default:
 533		break;
 534	}
 535	return connected;
 536}
 537
 
 
 
 
 
 
 
 
 538void r100_hpd_set_polarity(struct radeon_device *rdev,
 539			   enum radeon_hpd_id hpd)
 540{
 541	u32 tmp;
 542	bool connected = r100_hpd_sense(rdev, hpd);
 543
 544	switch (hpd) {
 545	case RADEON_HPD_1:
 546		tmp = RREG32(RADEON_FP_GEN_CNTL);
 547		if (connected)
 548			tmp &= ~RADEON_FP_DETECT_INT_POL;
 549		else
 550			tmp |= RADEON_FP_DETECT_INT_POL;
 551		WREG32(RADEON_FP_GEN_CNTL, tmp);
 552		break;
 553	case RADEON_HPD_2:
 554		tmp = RREG32(RADEON_FP2_GEN_CNTL);
 555		if (connected)
 556			tmp &= ~RADEON_FP2_DETECT_INT_POL;
 557		else
 558			tmp |= RADEON_FP2_DETECT_INT_POL;
 559		WREG32(RADEON_FP2_GEN_CNTL, tmp);
 560		break;
 561	default:
 562		break;
 563	}
 564}
 565
 
 
 
 
 
 
 
 
 566void r100_hpd_init(struct radeon_device *rdev)
 567{
 568	struct drm_device *dev = rdev->ddev;
 569	struct drm_connector *connector;
 
 570
 571	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 572		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 573		switch (radeon_connector->hpd.hpd) {
 574		case RADEON_HPD_1:
 575			rdev->irq.hpd[0] = true;
 576			break;
 577		case RADEON_HPD_2:
 578			rdev->irq.hpd[1] = true;
 579			break;
 580		default:
 581			break;
 582		}
 583		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
 584	}
 585	if (rdev->irq.installed)
 586		r100_irq_set(rdev);
 587}
 588
 
 
 
 
 
 
 
 
 589void r100_hpd_fini(struct radeon_device *rdev)
 590{
 591	struct drm_device *dev = rdev->ddev;
 592	struct drm_connector *connector;
 
 593
 594	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 595		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 596		switch (radeon_connector->hpd.hpd) {
 597		case RADEON_HPD_1:
 598			rdev->irq.hpd[0] = false;
 599			break;
 600		case RADEON_HPD_2:
 601			rdev->irq.hpd[1] = false;
 602			break;
 603		default:
 604			break;
 605		}
 606	}
 
 607}
 608
 609/*
 610 * PCI GART
 611 */
 612void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
 613{
 614	/* TODO: can we do somethings here ? */
 615	/* It seems hw only cache one entry so we should discard this
 616	 * entry otherwise if first GPU GART read hit this entry it
 617	 * could end up in wrong address. */
 618}
 619
 620int r100_pci_gart_init(struct radeon_device *rdev)
 621{
 622	int r;
 623
 624	if (rdev->gart.ptr) {
 625		WARN(1, "R100 PCI GART already initialized\n");
 626		return 0;
 627	}
 628	/* Initialize common gart structure */
 629	r = radeon_gart_init(rdev);
 630	if (r)
 631		return r;
 632	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 633	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
 
 634	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
 635	return radeon_gart_table_ram_alloc(rdev);
 636}
 637
 638/* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
 639void r100_enable_bm(struct radeon_device *rdev)
 640{
 641	uint32_t tmp;
 642	/* Enable bus mastering */
 643	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
 644	WREG32(RADEON_BUS_CNTL, tmp);
 645}
 646
 647int r100_pci_gart_enable(struct radeon_device *rdev)
 648{
 649	uint32_t tmp;
 650
 651	radeon_gart_restore(rdev);
 652	/* discard memory request outside of configured range */
 653	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 654	WREG32(RADEON_AIC_CNTL, tmp);
 655	/* set address range for PCI address translate */
 656	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
 657	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
 658	/* set PCI GART page-table base address */
 659	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
 660	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
 661	WREG32(RADEON_AIC_CNTL, tmp);
 662	r100_pci_gart_tlb_flush(rdev);
 663	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
 664		 (unsigned)(rdev->mc.gtt_size >> 20),
 665		 (unsigned long long)rdev->gart.table_addr);
 666	rdev->gart.ready = true;
 667	return 0;
 668}
 669
 670void r100_pci_gart_disable(struct radeon_device *rdev)
 671{
 672	uint32_t tmp;
 673
 674	/* discard memory request outside of configured range */
 675	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 676	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
 677	WREG32(RADEON_AIC_LO_ADDR, 0);
 678	WREG32(RADEON_AIC_HI_ADDR, 0);
 679}
 680
 681int r100_pci_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
 
 
 
 
 
 
 682{
 683	u32 *gtt = rdev->gart.ptr;
 684
 685	if (i < 0 || i > rdev->gart.num_gpu_pages) {
 686		return -EINVAL;
 687	}
 688	gtt[i] = cpu_to_le32(lower_32_bits(addr));
 689	return 0;
 690}
 691
 692void r100_pci_gart_fini(struct radeon_device *rdev)
 693{
 694	radeon_gart_fini(rdev);
 695	r100_pci_gart_disable(rdev);
 696	radeon_gart_table_ram_free(rdev);
 697}
 698
 699int r100_irq_set(struct radeon_device *rdev)
 700{
 701	uint32_t tmp = 0;
 702
 703	if (!rdev->irq.installed) {
 704		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
 705		WREG32(R_000040_GEN_INT_CNTL, 0);
 706		return -EINVAL;
 707	}
 708	if (rdev->irq.sw_int[RADEON_RING_TYPE_GFX_INDEX]) {
 709		tmp |= RADEON_SW_INT_ENABLE;
 710	}
 711	if (rdev->irq.gui_idle) {
 712		tmp |= RADEON_GUI_IDLE_MASK;
 713	}
 714	if (rdev->irq.crtc_vblank_int[0] ||
 715	    rdev->irq.pflip[0]) {
 716		tmp |= RADEON_CRTC_VBLANK_MASK;
 717	}
 718	if (rdev->irq.crtc_vblank_int[1] ||
 719	    rdev->irq.pflip[1]) {
 720		tmp |= RADEON_CRTC2_VBLANK_MASK;
 721	}
 722	if (rdev->irq.hpd[0]) {
 723		tmp |= RADEON_FP_DETECT_MASK;
 724	}
 725	if (rdev->irq.hpd[1]) {
 726		tmp |= RADEON_FP2_DETECT_MASK;
 727	}
 728	WREG32(RADEON_GEN_INT_CNTL, tmp);
 
 
 
 
 729	return 0;
 730}
 731
 732void r100_irq_disable(struct radeon_device *rdev)
 733{
 734	u32 tmp;
 735
 736	WREG32(R_000040_GEN_INT_CNTL, 0);
 737	/* Wait and acknowledge irq */
 738	mdelay(1);
 739	tmp = RREG32(R_000044_GEN_INT_STATUS);
 740	WREG32(R_000044_GEN_INT_STATUS, tmp);
 741}
 742
 743static uint32_t r100_irq_ack(struct radeon_device *rdev)
 744{
 745	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
 746	uint32_t irq_mask = RADEON_SW_INT_TEST |
 747		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
 748		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
 749
 750	/* the interrupt works, but the status bit is permanently asserted */
 751	if (rdev->irq.gui_idle && radeon_gui_idle(rdev)) {
 752		if (!rdev->irq.gui_idle_acked)
 753			irq_mask |= RADEON_GUI_IDLE_STAT;
 754	}
 755
 756	if (irqs) {
 757		WREG32(RADEON_GEN_INT_STATUS, irqs);
 758	}
 759	return irqs & irq_mask;
 760}
 761
 762int r100_irq_process(struct radeon_device *rdev)
 763{
 764	uint32_t status, msi_rearm;
 765	bool queue_hotplug = false;
 766
 767	/* reset gui idle ack.  the status bit is broken */
 768	rdev->irq.gui_idle_acked = false;
 769
 770	status = r100_irq_ack(rdev);
 771	if (!status) {
 772		return IRQ_NONE;
 773	}
 774	if (rdev->shutdown) {
 775		return IRQ_NONE;
 776	}
 777	while (status) {
 778		/* SW interrupt */
 779		if (status & RADEON_SW_INT_TEST) {
 780			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
 781		}
 782		/* gui idle interrupt */
 783		if (status & RADEON_GUI_IDLE_STAT) {
 784			rdev->irq.gui_idle_acked = true;
 785			rdev->pm.gui_idle = true;
 786			wake_up(&rdev->irq.idle_queue);
 787		}
 788		/* Vertical blank interrupts */
 789		if (status & RADEON_CRTC_VBLANK_STAT) {
 790			if (rdev->irq.crtc_vblank_int[0]) {
 791				drm_handle_vblank(rdev->ddev, 0);
 792				rdev->pm.vblank_sync = true;
 793				wake_up(&rdev->irq.vblank_queue);
 794			}
 795			if (rdev->irq.pflip[0])
 796				radeon_crtc_handle_flip(rdev, 0);
 797		}
 798		if (status & RADEON_CRTC2_VBLANK_STAT) {
 799			if (rdev->irq.crtc_vblank_int[1]) {
 800				drm_handle_vblank(rdev->ddev, 1);
 801				rdev->pm.vblank_sync = true;
 802				wake_up(&rdev->irq.vblank_queue);
 803			}
 804			if (rdev->irq.pflip[1])
 805				radeon_crtc_handle_flip(rdev, 1);
 806		}
 807		if (status & RADEON_FP_DETECT_STAT) {
 808			queue_hotplug = true;
 809			DRM_DEBUG("HPD1\n");
 810		}
 811		if (status & RADEON_FP2_DETECT_STAT) {
 812			queue_hotplug = true;
 813			DRM_DEBUG("HPD2\n");
 814		}
 815		status = r100_irq_ack(rdev);
 816	}
 817	/* reset gui idle ack.  the status bit is broken */
 818	rdev->irq.gui_idle_acked = false;
 819	if (queue_hotplug)
 820		schedule_work(&rdev->hotplug_work);
 821	if (rdev->msi_enabled) {
 822		switch (rdev->family) {
 823		case CHIP_RS400:
 824		case CHIP_RS480:
 825			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
 826			WREG32(RADEON_AIC_CNTL, msi_rearm);
 827			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
 828			break;
 829		default:
 830			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
 831			break;
 832		}
 833	}
 834	return IRQ_HANDLED;
 835}
 836
 837u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
 838{
 839	if (crtc == 0)
 840		return RREG32(RADEON_CRTC_CRNT_FRAME);
 841	else
 842		return RREG32(RADEON_CRTC2_CRNT_FRAME);
 843}
 844
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 845/* Who ever call radeon_fence_emit should call ring_lock and ask
 846 * for enough space (today caller are ib schedule and buffer move) */
 847void r100_fence_ring_emit(struct radeon_device *rdev,
 848			  struct radeon_fence *fence)
 849{
 850	struct radeon_ring *ring = &rdev->ring[fence->ring];
 851
 852	/* We have to make sure that caches are flushed before
 853	 * CPU might read something from VRAM. */
 854	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
 855	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
 856	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
 857	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
 858	/* Wait until IDLE & CLEAN */
 859	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 860	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
 861	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 862	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
 863				RADEON_HDP_READ_BUFFER_INVALIDATE);
 864	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 865	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
 866	/* Emit fence sequence & fire IRQ */
 867	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
 868	radeon_ring_write(ring, fence->seq);
 869	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
 870	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
 871}
 872
 873void r100_semaphore_ring_emit(struct radeon_device *rdev,
 874			      struct radeon_ring *ring,
 875			      struct radeon_semaphore *semaphore,
 876			      bool emit_wait)
 877{
 878	/* Unused on older asics, since we don't have semaphores or multiple rings */
 879	BUG();
 
 880}
 881
 882int r100_copy_blit(struct radeon_device *rdev,
 883		   uint64_t src_offset,
 884		   uint64_t dst_offset,
 885		   unsigned num_gpu_pages,
 886		   struct radeon_fence *fence)
 887{
 888	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 
 889	uint32_t cur_pages;
 890	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
 891	uint32_t pitch;
 892	uint32_t stride_pixels;
 893	unsigned ndw;
 894	int num_loops;
 895	int r = 0;
 896
 897	/* radeon limited to 16k stride */
 898	stride_bytes &= 0x3fff;
 899	/* radeon pitch is /64 */
 900	pitch = stride_bytes / 64;
 901	stride_pixels = stride_bytes / 4;
 902	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
 903
 904	/* Ask for enough room for blit + flush + fence */
 905	ndw = 64 + (10 * num_loops);
 906	r = radeon_ring_lock(rdev, ring, ndw);
 907	if (r) {
 908		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 909		return -EINVAL;
 910	}
 911	while (num_gpu_pages > 0) {
 912		cur_pages = num_gpu_pages;
 913		if (cur_pages > 8191) {
 914			cur_pages = 8191;
 915		}
 916		num_gpu_pages -= cur_pages;
 917
 918		/* pages are in Y direction - height
 919		   page width in X direction - width */
 920		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
 921		radeon_ring_write(ring,
 922				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 923				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 924				  RADEON_GMC_SRC_CLIPPING |
 925				  RADEON_GMC_DST_CLIPPING |
 926				  RADEON_GMC_BRUSH_NONE |
 927				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 928				  RADEON_GMC_SRC_DATATYPE_COLOR |
 929				  RADEON_ROP3_S |
 930				  RADEON_DP_SRC_SOURCE_MEMORY |
 931				  RADEON_GMC_CLR_CMP_CNTL_DIS |
 932				  RADEON_GMC_WR_MSK_DIS);
 933		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
 934		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
 935		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 936		radeon_ring_write(ring, 0);
 937		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 938		radeon_ring_write(ring, num_gpu_pages);
 939		radeon_ring_write(ring, num_gpu_pages);
 940		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
 941	}
 942	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 943	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
 944	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 945	radeon_ring_write(ring,
 946			  RADEON_WAIT_2D_IDLECLEAN |
 947			  RADEON_WAIT_HOST_IDLECLEAN |
 948			  RADEON_WAIT_DMA_GUI_IDLE);
 949	if (fence) {
 950		r = radeon_fence_emit(rdev, fence);
 
 
 951	}
 952	radeon_ring_unlock_commit(rdev, ring);
 953	return r;
 954}
 955
 956static int r100_cp_wait_for_idle(struct radeon_device *rdev)
 957{
 958	unsigned i;
 959	u32 tmp;
 960
 961	for (i = 0; i < rdev->usec_timeout; i++) {
 962		tmp = RREG32(R_000E40_RBBM_STATUS);
 963		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
 964			return 0;
 965		}
 966		udelay(1);
 967	}
 968	return -1;
 969}
 970
 971void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
 972{
 973	int r;
 974
 975	r = radeon_ring_lock(rdev, ring, 2);
 976	if (r) {
 977		return;
 978	}
 979	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
 980	radeon_ring_write(ring,
 981			  RADEON_ISYNC_ANY2D_IDLE3D |
 982			  RADEON_ISYNC_ANY3D_IDLE2D |
 983			  RADEON_ISYNC_WAIT_IDLEGUI |
 984			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 985	radeon_ring_unlock_commit(rdev, ring);
 986}
 987
 988
 989/* Load the microcode for the CP */
 990static int r100_cp_init_microcode(struct radeon_device *rdev)
 991{
 992	struct platform_device *pdev;
 993	const char *fw_name = NULL;
 994	int err;
 995
 996	DRM_DEBUG_KMS("\n");
 997
 998	pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
 999	err = IS_ERR(pdev);
1000	if (err) {
1001		printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
1002		return -EINVAL;
1003	}
1004	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
1005	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
1006	    (rdev->family == CHIP_RS200)) {
1007		DRM_INFO("Loading R100 Microcode\n");
1008		fw_name = FIRMWARE_R100;
1009	} else if ((rdev->family == CHIP_R200) ||
1010		   (rdev->family == CHIP_RV250) ||
1011		   (rdev->family == CHIP_RV280) ||
1012		   (rdev->family == CHIP_RS300)) {
1013		DRM_INFO("Loading R200 Microcode\n");
1014		fw_name = FIRMWARE_R200;
1015	} else if ((rdev->family == CHIP_R300) ||
1016		   (rdev->family == CHIP_R350) ||
1017		   (rdev->family == CHIP_RV350) ||
1018		   (rdev->family == CHIP_RV380) ||
1019		   (rdev->family == CHIP_RS400) ||
1020		   (rdev->family == CHIP_RS480)) {
1021		DRM_INFO("Loading R300 Microcode\n");
1022		fw_name = FIRMWARE_R300;
1023	} else if ((rdev->family == CHIP_R420) ||
1024		   (rdev->family == CHIP_R423) ||
1025		   (rdev->family == CHIP_RV410)) {
1026		DRM_INFO("Loading R400 Microcode\n");
1027		fw_name = FIRMWARE_R420;
1028	} else if ((rdev->family == CHIP_RS690) ||
1029		   (rdev->family == CHIP_RS740)) {
1030		DRM_INFO("Loading RS690/RS740 Microcode\n");
1031		fw_name = FIRMWARE_RS690;
1032	} else if (rdev->family == CHIP_RS600) {
1033		DRM_INFO("Loading RS600 Microcode\n");
1034		fw_name = FIRMWARE_RS600;
1035	} else if ((rdev->family == CHIP_RV515) ||
1036		   (rdev->family == CHIP_R520) ||
1037		   (rdev->family == CHIP_RV530) ||
1038		   (rdev->family == CHIP_R580) ||
1039		   (rdev->family == CHIP_RV560) ||
1040		   (rdev->family == CHIP_RV570)) {
1041		DRM_INFO("Loading R500 Microcode\n");
1042		fw_name = FIRMWARE_R520;
1043	}
1044
1045	err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
1046	platform_device_unregister(pdev);
1047	if (err) {
1048		printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
1049		       fw_name);
1050	} else if (rdev->me_fw->size % 8) {
1051		printk(KERN_ERR
1052		       "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1053		       rdev->me_fw->size, fw_name);
1054		err = -EINVAL;
1055		release_firmware(rdev->me_fw);
1056		rdev->me_fw = NULL;
1057	}
1058	return err;
1059}
1060
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1061static void r100_cp_load_microcode(struct radeon_device *rdev)
1062{
1063	const __be32 *fw_data;
1064	int i, size;
1065
1066	if (r100_gui_wait_for_idle(rdev)) {
1067		printk(KERN_WARNING "Failed to wait GUI idle while "
1068		       "programming pipes. Bad things might happen.\n");
1069	}
1070
1071	if (rdev->me_fw) {
1072		size = rdev->me_fw->size / 4;
1073		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1074		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1075		for (i = 0; i < size; i += 2) {
1076			WREG32(RADEON_CP_ME_RAM_DATAH,
1077			       be32_to_cpup(&fw_data[i]));
1078			WREG32(RADEON_CP_ME_RAM_DATAL,
1079			       be32_to_cpup(&fw_data[i + 1]));
1080		}
1081	}
1082}
1083
1084int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1085{
1086	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1087	unsigned rb_bufsz;
1088	unsigned rb_blksz;
1089	unsigned max_fetch;
1090	unsigned pre_write_timer;
1091	unsigned pre_write_limit;
1092	unsigned indirect2_start;
1093	unsigned indirect1_start;
1094	uint32_t tmp;
1095	int r;
1096
1097	if (r100_debugfs_cp_init(rdev)) {
1098		DRM_ERROR("Failed to register debugfs file for CP !\n");
1099	}
1100	if (!rdev->me_fw) {
1101		r = r100_cp_init_microcode(rdev);
1102		if (r) {
1103			DRM_ERROR("Failed to load firmware!\n");
1104			return r;
1105		}
1106	}
1107
1108	/* Align ring size */
1109	rb_bufsz = drm_order(ring_size / 8);
1110	ring_size = (1 << (rb_bufsz + 1)) * 4;
1111	r100_cp_load_microcode(rdev);
1112	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1113			     RADEON_CP_RB_RPTR, RADEON_CP_RB_WPTR,
1114			     0, 0x7fffff, RADEON_CP_PACKET2);
1115	if (r) {
1116		return r;
1117	}
1118	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1119	 * the rptr copy in system ram */
1120	rb_blksz = 9;
1121	/* cp will read 128bytes at a time (4 dwords) */
1122	max_fetch = 1;
1123	ring->align_mask = 16 - 1;
1124	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1125	pre_write_timer = 64;
1126	/* Force CP_RB_WPTR write if written more than one time before the
1127	 * delay expire
1128	 */
1129	pre_write_limit = 0;
1130	/* Setup the cp cache like this (cache size is 96 dwords) :
1131	 *	RING		0  to 15
1132	 *	INDIRECT1	16 to 79
1133	 *	INDIRECT2	80 to 95
1134	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1135	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1136	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1137	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1138	 * so it gets the bigger cache.
1139	 */
1140	indirect2_start = 80;
1141	indirect1_start = 16;
1142	/* cp setup */
1143	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1144	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1145	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1146	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1147#ifdef __BIG_ENDIAN
1148	tmp |= RADEON_BUF_SWAP_32BIT;
1149#endif
1150	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1151
1152	/* Set ring address */
1153	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1154	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1155	/* Force read & write ptr to 0 */
1156	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1157	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1158	ring->wptr = 0;
1159	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1160
1161	/* set the wb address whether it's enabled or not */
1162	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1163		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1164	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1165
1166	if (rdev->wb.enabled)
1167		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1168	else {
1169		tmp |= RADEON_RB_NO_UPDATE;
1170		WREG32(R_000770_SCRATCH_UMSK, 0);
1171	}
1172
1173	WREG32(RADEON_CP_RB_CNTL, tmp);
1174	udelay(10);
1175	ring->rptr = RREG32(RADEON_CP_RB_RPTR);
1176	/* Set cp mode to bus mastering & enable cp*/
1177	WREG32(RADEON_CP_CSQ_MODE,
1178	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1179	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1180	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1181	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1182	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1183
1184	/* at this point everything should be setup correctly to enable master */
1185	pci_set_master(rdev->pdev);
1186
1187	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1188	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1189	if (r) {
1190		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1191		return r;
1192	}
1193	ring->ready = true;
1194	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
 
 
 
 
 
 
 
 
 
1195	return 0;
1196}
1197
1198void r100_cp_fini(struct radeon_device *rdev)
1199{
1200	if (r100_cp_wait_for_idle(rdev)) {
1201		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1202	}
1203	/* Disable ring */
1204	r100_cp_disable(rdev);
 
1205	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1206	DRM_INFO("radeon: cp finalized\n");
1207}
1208
1209void r100_cp_disable(struct radeon_device *rdev)
1210{
1211	/* Disable ring */
1212	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1213	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1214	WREG32(RADEON_CP_CSQ_MODE, 0);
1215	WREG32(RADEON_CP_CSQ_CNTL, 0);
1216	WREG32(R_000770_SCRATCH_UMSK, 0);
1217	if (r100_gui_wait_for_idle(rdev)) {
1218		printk(KERN_WARNING "Failed to wait GUI idle while "
1219		       "programming pipes. Bad things might happen.\n");
1220	}
1221}
1222
1223/*
1224 * CS functions
1225 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1226int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1227			  struct radeon_cs_packet *pkt,
1228			  const unsigned *auth, unsigned n,
1229			  radeon_packet0_check_t check)
1230{
1231	unsigned reg;
1232	unsigned i, j, m;
1233	unsigned idx;
1234	int r;
1235
1236	idx = pkt->idx + 1;
1237	reg = pkt->reg;
1238	/* Check that register fall into register range
1239	 * determined by the number of entry (n) in the
1240	 * safe register bitmap.
1241	 */
1242	if (pkt->one_reg_wr) {
1243		if ((reg >> 7) > n) {
1244			return -EINVAL;
1245		}
1246	} else {
1247		if (((reg + (pkt->count << 2)) >> 7) > n) {
1248			return -EINVAL;
1249		}
1250	}
1251	for (i = 0; i <= pkt->count; i++, idx++) {
1252		j = (reg >> 7);
1253		m = 1 << ((reg >> 2) & 31);
1254		if (auth[j] & m) {
1255			r = check(p, pkt, idx, reg);
1256			if (r) {
1257				return r;
1258			}
1259		}
1260		if (pkt->one_reg_wr) {
1261			if (!(auth[j] & m)) {
1262				break;
1263			}
1264		} else {
1265			reg += 4;
1266		}
1267	}
1268	return 0;
1269}
1270
1271void r100_cs_dump_packet(struct radeon_cs_parser *p,
1272			 struct radeon_cs_packet *pkt)
1273{
1274	volatile uint32_t *ib;
1275	unsigned i;
1276	unsigned idx;
1277
1278	ib = p->ib.ptr;
1279	idx = pkt->idx;
1280	for (i = 0; i <= (pkt->count + 1); i++, idx++) {
1281		DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
1282	}
1283}
1284
1285/**
1286 * r100_cs_packet_parse() - parse cp packet and point ib index to next packet
1287 * @parser:	parser structure holding parsing context.
1288 * @pkt:	where to store packet informations
1289 *
1290 * Assume that chunk_ib_index is properly set. Will return -EINVAL
1291 * if packet is bigger than remaining ib size. or if packets is unknown.
1292 **/
1293int r100_cs_packet_parse(struct radeon_cs_parser *p,
1294			 struct radeon_cs_packet *pkt,
1295			 unsigned idx)
1296{
1297	struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
1298	uint32_t header;
1299
1300	if (idx >= ib_chunk->length_dw) {
1301		DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
1302			  idx, ib_chunk->length_dw);
1303		return -EINVAL;
1304	}
1305	header = radeon_get_ib_value(p, idx);
1306	pkt->idx = idx;
1307	pkt->type = CP_PACKET_GET_TYPE(header);
1308	pkt->count = CP_PACKET_GET_COUNT(header);
1309	switch (pkt->type) {
1310	case PACKET_TYPE0:
1311		pkt->reg = CP_PACKET0_GET_REG(header);
1312		pkt->one_reg_wr = CP_PACKET0_GET_ONE_REG_WR(header);
1313		break;
1314	case PACKET_TYPE3:
1315		pkt->opcode = CP_PACKET3_GET_OPCODE(header);
1316		break;
1317	case PACKET_TYPE2:
1318		pkt->count = -1;
1319		break;
1320	default:
1321		DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
1322		return -EINVAL;
1323	}
1324	if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
1325		DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
1326			  pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
1327		return -EINVAL;
1328	}
1329	return 0;
1330}
1331
1332/**
1333 * r100_cs_packet_next_vline() - parse userspace VLINE packet
1334 * @parser:		parser structure holding parsing context.
1335 *
1336 * Userspace sends a special sequence for VLINE waits.
1337 * PACKET0 - VLINE_START_END + value
1338 * PACKET0 - WAIT_UNTIL +_value
1339 * RELOC (P3) - crtc_id in reloc.
1340 *
1341 * This function parses this and relocates the VLINE START END
1342 * and WAIT UNTIL packets to the correct crtc.
1343 * It also detects a switched off crtc and nulls out the
1344 * wait in that case.
1345 */
1346int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1347{
1348	struct drm_mode_object *obj;
1349	struct drm_crtc *crtc;
1350	struct radeon_crtc *radeon_crtc;
1351	struct radeon_cs_packet p3reloc, waitreloc;
1352	int crtc_id;
1353	int r;
1354	uint32_t header, h_idx, reg;
1355	volatile uint32_t *ib;
1356
1357	ib = p->ib.ptr;
1358
1359	/* parse the wait until */
1360	r = r100_cs_packet_parse(p, &waitreloc, p->idx);
1361	if (r)
1362		return r;
1363
1364	/* check its a wait until and only 1 count */
1365	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1366	    waitreloc.count != 0) {
1367		DRM_ERROR("vline wait had illegal wait until segment\n");
1368		return -EINVAL;
1369	}
1370
1371	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1372		DRM_ERROR("vline wait had illegal wait until\n");
1373		return -EINVAL;
1374	}
1375
1376	/* jump over the NOP */
1377	r = r100_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1378	if (r)
1379		return r;
1380
1381	h_idx = p->idx - 2;
1382	p->idx += waitreloc.count + 2;
1383	p->idx += p3reloc.count + 2;
1384
1385	header = radeon_get_ib_value(p, h_idx);
1386	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1387	reg = CP_PACKET0_GET_REG(header);
1388	obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
1389	if (!obj) {
1390		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1391		return -EINVAL;
1392	}
1393	crtc = obj_to_crtc(obj);
1394	radeon_crtc = to_radeon_crtc(crtc);
1395	crtc_id = radeon_crtc->crtc_id;
1396
1397	if (!crtc->enabled) {
1398		/* if the CRTC isn't enabled - we need to nop out the wait until */
1399		ib[h_idx + 2] = PACKET2(0);
1400		ib[h_idx + 3] = PACKET2(0);
1401	} else if (crtc_id == 1) {
1402		switch (reg) {
1403		case AVIVO_D1MODE_VLINE_START_END:
1404			header &= ~R300_CP_PACKET0_REG_MASK;
1405			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1406			break;
1407		case RADEON_CRTC_GUI_TRIG_VLINE:
1408			header &= ~R300_CP_PACKET0_REG_MASK;
1409			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1410			break;
1411		default:
1412			DRM_ERROR("unknown crtc reloc\n");
1413			return -EINVAL;
1414		}
1415		ib[h_idx] = header;
1416		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1417	}
1418
1419	return 0;
1420}
1421
1422/**
1423 * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3
1424 * @parser:		parser structure holding parsing context.
1425 * @data:		pointer to relocation data
1426 * @offset_start:	starting offset
1427 * @offset_mask:	offset mask (to align start offset on)
1428 * @reloc:		reloc informations
1429 *
1430 * Check next packet is relocation packet3, do bo validation and compute
1431 * GPU offset using the provided start.
1432 **/
1433int r100_cs_packet_next_reloc(struct radeon_cs_parser *p,
1434			      struct radeon_cs_reloc **cs_reloc)
1435{
1436	struct radeon_cs_chunk *relocs_chunk;
1437	struct radeon_cs_packet p3reloc;
1438	unsigned idx;
1439	int r;
1440
1441	if (p->chunk_relocs_idx == -1) {
1442		DRM_ERROR("No relocation chunk !\n");
1443		return -EINVAL;
1444	}
1445	*cs_reloc = NULL;
1446	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
1447	r = r100_cs_packet_parse(p, &p3reloc, p->idx);
1448	if (r) {
1449		return r;
1450	}
1451	p->idx += p3reloc.count + 2;
1452	if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
1453		DRM_ERROR("No packet3 for relocation for packet at %d.\n",
1454			  p3reloc.idx);
1455		r100_cs_dump_packet(p, &p3reloc);
1456		return -EINVAL;
1457	}
1458	idx = radeon_get_ib_value(p, p3reloc.idx + 1);
1459	if (idx >= relocs_chunk->length_dw) {
1460		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
1461			  idx, relocs_chunk->length_dw);
1462		r100_cs_dump_packet(p, &p3reloc);
1463		return -EINVAL;
1464	}
1465	/* FIXME: we assume reloc size is 4 dwords */
1466	*cs_reloc = p->relocs_ptr[(idx / 4)];
1467	return 0;
1468}
1469
1470static int r100_get_vtx_size(uint32_t vtx_fmt)
1471{
1472	int vtx_size;
1473	vtx_size = 2;
1474	/* ordered according to bits in spec */
1475	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1476		vtx_size++;
1477	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1478		vtx_size += 3;
1479	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1480		vtx_size++;
1481	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1482		vtx_size++;
1483	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1484		vtx_size += 3;
1485	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1486		vtx_size++;
1487	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1488		vtx_size++;
1489	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1490		vtx_size += 2;
1491	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1492		vtx_size += 2;
1493	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1494		vtx_size++;
1495	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1496		vtx_size += 2;
1497	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1498		vtx_size++;
1499	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1500		vtx_size += 2;
1501	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1502		vtx_size++;
1503	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1504		vtx_size++;
1505	/* blend weight */
1506	if (vtx_fmt & (0x7 << 15))
1507		vtx_size += (vtx_fmt >> 15) & 0x7;
1508	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1509		vtx_size += 3;
1510	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1511		vtx_size += 2;
1512	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1513		vtx_size++;
1514	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1515		vtx_size++;
1516	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1517		vtx_size++;
1518	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1519		vtx_size++;
1520	return vtx_size;
1521}
1522
1523static int r100_packet0_check(struct radeon_cs_parser *p,
1524			      struct radeon_cs_packet *pkt,
1525			      unsigned idx, unsigned reg)
1526{
1527	struct radeon_cs_reloc *reloc;
1528	struct r100_cs_track *track;
1529	volatile uint32_t *ib;
1530	uint32_t tmp;
1531	int r;
1532	int i, face;
1533	u32 tile_flags = 0;
1534	u32 idx_value;
1535
1536	ib = p->ib.ptr;
1537	track = (struct r100_cs_track *)p->track;
1538
1539	idx_value = radeon_get_ib_value(p, idx);
1540
1541	switch (reg) {
1542	case RADEON_CRTC_GUI_TRIG_VLINE:
1543		r = r100_cs_packet_parse_vline(p);
1544		if (r) {
1545			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1546				  idx, reg);
1547			r100_cs_dump_packet(p, pkt);
1548			return r;
1549		}
1550		break;
1551		/* FIXME: only allow PACKET3 blit? easier to check for out of
1552		 * range access */
1553	case RADEON_DST_PITCH_OFFSET:
1554	case RADEON_SRC_PITCH_OFFSET:
1555		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1556		if (r)
1557			return r;
1558		break;
1559	case RADEON_RB3D_DEPTHOFFSET:
1560		r = r100_cs_packet_next_reloc(p, &reloc);
1561		if (r) {
1562			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1563				  idx, reg);
1564			r100_cs_dump_packet(p, pkt);
1565			return r;
1566		}
1567		track->zb.robj = reloc->robj;
1568		track->zb.offset = idx_value;
1569		track->zb_dirty = true;
1570		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1571		break;
1572	case RADEON_RB3D_COLOROFFSET:
1573		r = r100_cs_packet_next_reloc(p, &reloc);
1574		if (r) {
1575			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1576				  idx, reg);
1577			r100_cs_dump_packet(p, pkt);
1578			return r;
1579		}
1580		track->cb[0].robj = reloc->robj;
1581		track->cb[0].offset = idx_value;
1582		track->cb_dirty = true;
1583		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1584		break;
1585	case RADEON_PP_TXOFFSET_0:
1586	case RADEON_PP_TXOFFSET_1:
1587	case RADEON_PP_TXOFFSET_2:
1588		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1589		r = r100_cs_packet_next_reloc(p, &reloc);
1590		if (r) {
1591			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1592				  idx, reg);
1593			r100_cs_dump_packet(p, pkt);
1594			return r;
1595		}
1596		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1597			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1598				tile_flags |= RADEON_TXO_MACRO_TILE;
1599			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1600				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1601
1602			tmp = idx_value & ~(0x7 << 2);
1603			tmp |= tile_flags;
1604			ib[idx] = tmp + ((u32)reloc->lobj.gpu_offset);
1605		} else
1606			ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1607		track->textures[i].robj = reloc->robj;
1608		track->tex_dirty = true;
1609		break;
1610	case RADEON_PP_CUBIC_OFFSET_T0_0:
1611	case RADEON_PP_CUBIC_OFFSET_T0_1:
1612	case RADEON_PP_CUBIC_OFFSET_T0_2:
1613	case RADEON_PP_CUBIC_OFFSET_T0_3:
1614	case RADEON_PP_CUBIC_OFFSET_T0_4:
1615		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1616		r = r100_cs_packet_next_reloc(p, &reloc);
1617		if (r) {
1618			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1619				  idx, reg);
1620			r100_cs_dump_packet(p, pkt);
1621			return r;
1622		}
1623		track->textures[0].cube_info[i].offset = idx_value;
1624		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1625		track->textures[0].cube_info[i].robj = reloc->robj;
1626		track->tex_dirty = true;
1627		break;
1628	case RADEON_PP_CUBIC_OFFSET_T1_0:
1629	case RADEON_PP_CUBIC_OFFSET_T1_1:
1630	case RADEON_PP_CUBIC_OFFSET_T1_2:
1631	case RADEON_PP_CUBIC_OFFSET_T1_3:
1632	case RADEON_PP_CUBIC_OFFSET_T1_4:
1633		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1634		r = r100_cs_packet_next_reloc(p, &reloc);
1635		if (r) {
1636			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1637				  idx, reg);
1638			r100_cs_dump_packet(p, pkt);
1639			return r;
1640		}
1641		track->textures[1].cube_info[i].offset = idx_value;
1642		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1643		track->textures[1].cube_info[i].robj = reloc->robj;
1644		track->tex_dirty = true;
1645		break;
1646	case RADEON_PP_CUBIC_OFFSET_T2_0:
1647	case RADEON_PP_CUBIC_OFFSET_T2_1:
1648	case RADEON_PP_CUBIC_OFFSET_T2_2:
1649	case RADEON_PP_CUBIC_OFFSET_T2_3:
1650	case RADEON_PP_CUBIC_OFFSET_T2_4:
1651		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1652		r = r100_cs_packet_next_reloc(p, &reloc);
1653		if (r) {
1654			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1655				  idx, reg);
1656			r100_cs_dump_packet(p, pkt);
1657			return r;
1658		}
1659		track->textures[2].cube_info[i].offset = idx_value;
1660		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1661		track->textures[2].cube_info[i].robj = reloc->robj;
1662		track->tex_dirty = true;
1663		break;
1664	case RADEON_RE_WIDTH_HEIGHT:
1665		track->maxy = ((idx_value >> 16) & 0x7FF);
1666		track->cb_dirty = true;
1667		track->zb_dirty = true;
1668		break;
1669	case RADEON_RB3D_COLORPITCH:
1670		r = r100_cs_packet_next_reloc(p, &reloc);
1671		if (r) {
1672			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1673				  idx, reg);
1674			r100_cs_dump_packet(p, pkt);
1675			return r;
1676		}
1677		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1678			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1679				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1680			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1681				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1682
1683			tmp = idx_value & ~(0x7 << 16);
1684			tmp |= tile_flags;
1685			ib[idx] = tmp;
1686		} else
1687			ib[idx] = idx_value;
1688
1689		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1690		track->cb_dirty = true;
1691		break;
1692	case RADEON_RB3D_DEPTHPITCH:
1693		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1694		track->zb_dirty = true;
1695		break;
1696	case RADEON_RB3D_CNTL:
1697		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1698		case 7:
1699		case 8:
1700		case 9:
1701		case 11:
1702		case 12:
1703			track->cb[0].cpp = 1;
1704			break;
1705		case 3:
1706		case 4:
1707		case 15:
1708			track->cb[0].cpp = 2;
1709			break;
1710		case 6:
1711			track->cb[0].cpp = 4;
1712			break;
1713		default:
1714			DRM_ERROR("Invalid color buffer format (%d) !\n",
1715				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1716			return -EINVAL;
1717		}
1718		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1719		track->cb_dirty = true;
1720		track->zb_dirty = true;
1721		break;
1722	case RADEON_RB3D_ZSTENCILCNTL:
1723		switch (idx_value & 0xf) {
1724		case 0:
1725			track->zb.cpp = 2;
1726			break;
1727		case 2:
1728		case 3:
1729		case 4:
1730		case 5:
1731		case 9:
1732		case 11:
1733			track->zb.cpp = 4;
1734			break;
1735		default:
1736			break;
1737		}
1738		track->zb_dirty = true;
1739		break;
1740	case RADEON_RB3D_ZPASS_ADDR:
1741		r = r100_cs_packet_next_reloc(p, &reloc);
1742		if (r) {
1743			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1744				  idx, reg);
1745			r100_cs_dump_packet(p, pkt);
1746			return r;
1747		}
1748		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1749		break;
1750	case RADEON_PP_CNTL:
1751		{
1752			uint32_t temp = idx_value >> 4;
1753			for (i = 0; i < track->num_texture; i++)
1754				track->textures[i].enabled = !!(temp & (1 << i));
1755			track->tex_dirty = true;
1756		}
1757		break;
1758	case RADEON_SE_VF_CNTL:
1759		track->vap_vf_cntl = idx_value;
1760		break;
1761	case RADEON_SE_VTX_FMT:
1762		track->vtx_size = r100_get_vtx_size(idx_value);
1763		break;
1764	case RADEON_PP_TEX_SIZE_0:
1765	case RADEON_PP_TEX_SIZE_1:
1766	case RADEON_PP_TEX_SIZE_2:
1767		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1768		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1769		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1770		track->tex_dirty = true;
1771		break;
1772	case RADEON_PP_TEX_PITCH_0:
1773	case RADEON_PP_TEX_PITCH_1:
1774	case RADEON_PP_TEX_PITCH_2:
1775		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1776		track->textures[i].pitch = idx_value + 32;
1777		track->tex_dirty = true;
1778		break;
1779	case RADEON_PP_TXFILTER_0:
1780	case RADEON_PP_TXFILTER_1:
1781	case RADEON_PP_TXFILTER_2:
1782		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1783		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1784						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1785		tmp = (idx_value >> 23) & 0x7;
1786		if (tmp == 2 || tmp == 6)
1787			track->textures[i].roundup_w = false;
1788		tmp = (idx_value >> 27) & 0x7;
1789		if (tmp == 2 || tmp == 6)
1790			track->textures[i].roundup_h = false;
1791		track->tex_dirty = true;
1792		break;
1793	case RADEON_PP_TXFORMAT_0:
1794	case RADEON_PP_TXFORMAT_1:
1795	case RADEON_PP_TXFORMAT_2:
1796		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1797		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1798			track->textures[i].use_pitch = 1;
1799		} else {
1800			track->textures[i].use_pitch = 0;
1801			track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1802			track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1803		}
1804		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1805			track->textures[i].tex_coord_type = 2;
1806		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1807		case RADEON_TXFORMAT_I8:
1808		case RADEON_TXFORMAT_RGB332:
1809		case RADEON_TXFORMAT_Y8:
1810			track->textures[i].cpp = 1;
1811			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1812			break;
1813		case RADEON_TXFORMAT_AI88:
1814		case RADEON_TXFORMAT_ARGB1555:
1815		case RADEON_TXFORMAT_RGB565:
1816		case RADEON_TXFORMAT_ARGB4444:
1817		case RADEON_TXFORMAT_VYUY422:
1818		case RADEON_TXFORMAT_YVYU422:
1819		case RADEON_TXFORMAT_SHADOW16:
1820		case RADEON_TXFORMAT_LDUDV655:
1821		case RADEON_TXFORMAT_DUDV88:
1822			track->textures[i].cpp = 2;
1823			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1824			break;
1825		case RADEON_TXFORMAT_ARGB8888:
1826		case RADEON_TXFORMAT_RGBA8888:
1827		case RADEON_TXFORMAT_SHADOW32:
1828		case RADEON_TXFORMAT_LDUDUV8888:
1829			track->textures[i].cpp = 4;
1830			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1831			break;
1832		case RADEON_TXFORMAT_DXT1:
1833			track->textures[i].cpp = 1;
1834			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1835			break;
1836		case RADEON_TXFORMAT_DXT23:
1837		case RADEON_TXFORMAT_DXT45:
1838			track->textures[i].cpp = 1;
1839			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1840			break;
1841		}
1842		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1843		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1844		track->tex_dirty = true;
1845		break;
1846	case RADEON_PP_CUBIC_FACES_0:
1847	case RADEON_PP_CUBIC_FACES_1:
1848	case RADEON_PP_CUBIC_FACES_2:
1849		tmp = idx_value;
1850		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1851		for (face = 0; face < 4; face++) {
1852			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1853			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1854		}
1855		track->tex_dirty = true;
1856		break;
1857	default:
1858		printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1859		       reg, idx);
1860		return -EINVAL;
1861	}
1862	return 0;
1863}
1864
1865int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1866					 struct radeon_cs_packet *pkt,
1867					 struct radeon_bo *robj)
1868{
1869	unsigned idx;
1870	u32 value;
1871	idx = pkt->idx + 1;
1872	value = radeon_get_ib_value(p, idx + 2);
1873	if ((value + 1) > radeon_bo_size(robj)) {
1874		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1875			  "(need %u have %lu) !\n",
1876			  value + 1,
1877			  radeon_bo_size(robj));
1878		return -EINVAL;
1879	}
1880	return 0;
1881}
1882
1883static int r100_packet3_check(struct radeon_cs_parser *p,
1884			      struct radeon_cs_packet *pkt)
1885{
1886	struct radeon_cs_reloc *reloc;
1887	struct r100_cs_track *track;
1888	unsigned idx;
1889	volatile uint32_t *ib;
1890	int r;
1891
1892	ib = p->ib.ptr;
1893	idx = pkt->idx + 1;
1894	track = (struct r100_cs_track *)p->track;
1895	switch (pkt->opcode) {
1896	case PACKET3_3D_LOAD_VBPNTR:
1897		r = r100_packet3_load_vbpntr(p, pkt, idx);
1898		if (r)
1899			return r;
1900		break;
1901	case PACKET3_INDX_BUFFER:
1902		r = r100_cs_packet_next_reloc(p, &reloc);
1903		if (r) {
1904			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1905			r100_cs_dump_packet(p, pkt);
1906			return r;
1907		}
1908		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->lobj.gpu_offset);
1909		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1910		if (r) {
1911			return r;
1912		}
1913		break;
1914	case 0x23:
1915		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1916		r = r100_cs_packet_next_reloc(p, &reloc);
1917		if (r) {
1918			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1919			r100_cs_dump_packet(p, pkt);
1920			return r;
1921		}
1922		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->lobj.gpu_offset);
1923		track->num_arrays = 1;
1924		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1925
1926		track->arrays[0].robj = reloc->robj;
1927		track->arrays[0].esize = track->vtx_size;
1928
1929		track->max_indx = radeon_get_ib_value(p, idx+1);
1930
1931		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1932		track->immd_dwords = pkt->count - 1;
1933		r = r100_cs_track_check(p->rdev, track);
1934		if (r)
1935			return r;
1936		break;
1937	case PACKET3_3D_DRAW_IMMD:
1938		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1939			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1940			return -EINVAL;
1941		}
1942		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1943		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1944		track->immd_dwords = pkt->count - 1;
1945		r = r100_cs_track_check(p->rdev, track);
1946		if (r)
1947			return r;
1948		break;
1949		/* triggers drawing using in-packet vertex data */
1950	case PACKET3_3D_DRAW_IMMD_2:
1951		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1952			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1953			return -EINVAL;
1954		}
1955		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1956		track->immd_dwords = pkt->count;
1957		r = r100_cs_track_check(p->rdev, track);
1958		if (r)
1959			return r;
1960		break;
1961		/* triggers drawing using in-packet vertex data */
1962	case PACKET3_3D_DRAW_VBUF_2:
1963		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1964		r = r100_cs_track_check(p->rdev, track);
1965		if (r)
1966			return r;
1967		break;
1968		/* triggers drawing of vertex buffers setup elsewhere */
1969	case PACKET3_3D_DRAW_INDX_2:
1970		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1971		r = r100_cs_track_check(p->rdev, track);
1972		if (r)
1973			return r;
1974		break;
1975		/* triggers drawing using indices to vertex buffer */
1976	case PACKET3_3D_DRAW_VBUF:
1977		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1978		r = r100_cs_track_check(p->rdev, track);
1979		if (r)
1980			return r;
1981		break;
1982		/* triggers drawing of vertex buffers setup elsewhere */
1983	case PACKET3_3D_DRAW_INDX:
1984		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1985		r = r100_cs_track_check(p->rdev, track);
1986		if (r)
1987			return r;
1988		break;
1989		/* triggers drawing using indices to vertex buffer */
1990	case PACKET3_3D_CLEAR_HIZ:
1991	case PACKET3_3D_CLEAR_ZMASK:
1992		if (p->rdev->hyperz_filp != p->filp)
1993			return -EINVAL;
1994		break;
1995	case PACKET3_NOP:
1996		break;
1997	default:
1998		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1999		return -EINVAL;
2000	}
2001	return 0;
2002}
2003
2004int r100_cs_parse(struct radeon_cs_parser *p)
2005{
2006	struct radeon_cs_packet pkt;
2007	struct r100_cs_track *track;
2008	int r;
2009
2010	track = kzalloc(sizeof(*track), GFP_KERNEL);
2011	if (!track)
2012		return -ENOMEM;
2013	r100_cs_track_clear(p->rdev, track);
2014	p->track = track;
2015	do {
2016		r = r100_cs_packet_parse(p, &pkt, p->idx);
2017		if (r) {
2018			return r;
2019		}
2020		p->idx += pkt.count + 2;
2021		switch (pkt.type) {
2022			case PACKET_TYPE0:
2023				if (p->rdev->family >= CHIP_R200)
2024					r = r100_cs_parse_packet0(p, &pkt,
2025								  p->rdev->config.r100.reg_safe_bm,
2026								  p->rdev->config.r100.reg_safe_bm_size,
2027								  &r200_packet0_check);
2028				else
2029					r = r100_cs_parse_packet0(p, &pkt,
2030								  p->rdev->config.r100.reg_safe_bm,
2031								  p->rdev->config.r100.reg_safe_bm_size,
2032								  &r100_packet0_check);
2033				break;
2034			case PACKET_TYPE2:
2035				break;
2036			case PACKET_TYPE3:
2037				r = r100_packet3_check(p, &pkt);
2038				break;
2039			default:
2040				DRM_ERROR("Unknown packet type %d !\n",
2041					  pkt.type);
2042				return -EINVAL;
2043		}
2044		if (r) {
2045			return r;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2046		}
2047	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
2048	return 0;
2049}
2050
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2051
2052/*
2053 * Global GPU functions
2054 */
2055void r100_errata(struct radeon_device *rdev)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2056{
2057	rdev->pll_errata = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2058
2059	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2060		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2061	}
2062
2063	if (rdev->family == CHIP_RV100 ||
2064	    rdev->family == CHIP_RS100 ||
2065	    rdev->family == CHIP_RS200) {
2066		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2067	}
 
2068}
2069
2070/* Wait for vertical sync on primary CRTC */
2071void r100_gpu_wait_for_vsync(struct radeon_device *rdev)
2072{
2073	uint32_t crtc_gen_cntl, tmp;
2074	int i;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2075
2076	crtc_gen_cntl = RREG32(RADEON_CRTC_GEN_CNTL);
2077	if ((crtc_gen_cntl & RADEON_CRTC_DISP_REQ_EN_B) ||
2078	    !(crtc_gen_cntl & RADEON_CRTC_EN)) {
2079		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2080	}
2081	/* Clear the CRTC_VBLANK_SAVE bit */
2082	WREG32(RADEON_CRTC_STATUS, RADEON_CRTC_VBLANK_SAVE_CLEAR);
2083	for (i = 0; i < rdev->usec_timeout; i++) {
2084		tmp = RREG32(RADEON_CRTC_STATUS);
2085		if (tmp & RADEON_CRTC_VBLANK_SAVE) {
2086			return;
 
 
 
 
 
 
 
 
2087		}
2088		DRM_UDELAY(1);
 
 
 
 
 
 
 
 
 
 
 
 
 
2089	}
2090}
2091
2092/* Wait for vertical sync on secondary CRTC */
2093void r100_gpu_wait_for_vsync2(struct radeon_device *rdev)
 
 
2094{
2095	uint32_t crtc2_gen_cntl, tmp;
2096	int i;
2097
2098	crtc2_gen_cntl = RREG32(RADEON_CRTC2_GEN_CNTL);
2099	if ((crtc2_gen_cntl & RADEON_CRTC2_DISP_REQ_EN_B) ||
2100	    !(crtc2_gen_cntl & RADEON_CRTC2_EN))
2101		return;
2102
2103	/* Clear the CRTC_VBLANK_SAVE bit */
2104	WREG32(RADEON_CRTC2_STATUS, RADEON_CRTC2_VBLANK_SAVE_CLEAR);
2105	for (i = 0; i < rdev->usec_timeout; i++) {
2106		tmp = RREG32(RADEON_CRTC2_STATUS);
2107		if (tmp & RADEON_CRTC2_VBLANK_SAVE) {
2108			return;
2109		}
2110		DRM_UDELAY(1);
2111	}
2112}
2113
2114int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2115{
2116	unsigned i;
2117	uint32_t tmp;
2118
2119	for (i = 0; i < rdev->usec_timeout; i++) {
2120		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2121		if (tmp >= n) {
2122			return 0;
2123		}
2124		DRM_UDELAY(1);
2125	}
2126	return -1;
2127}
2128
2129int r100_gui_wait_for_idle(struct radeon_device *rdev)
2130{
2131	unsigned i;
2132	uint32_t tmp;
2133
2134	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2135		printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
2136		       " Bad things might happen.\n");
2137	}
2138	for (i = 0; i < rdev->usec_timeout; i++) {
2139		tmp = RREG32(RADEON_RBBM_STATUS);
2140		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2141			return 0;
2142		}
2143		DRM_UDELAY(1);
2144	}
2145	return -1;
2146}
2147
2148int r100_mc_wait_for_idle(struct radeon_device *rdev)
2149{
2150	unsigned i;
2151	uint32_t tmp;
2152
2153	for (i = 0; i < rdev->usec_timeout; i++) {
2154		/* read MC_STATUS */
2155		tmp = RREG32(RADEON_MC_STATUS);
2156		if (tmp & RADEON_MC_IDLE) {
2157			return 0;
2158		}
2159		DRM_UDELAY(1);
2160	}
2161	return -1;
2162}
2163
2164bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2165{
2166	u32 rbbm_status;
2167
2168	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2169	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2170		radeon_ring_lockup_update(ring);
2171		return false;
2172	}
2173	/* force CP activities */
2174	radeon_ring_force_activity(rdev, ring);
2175	return radeon_ring_test_lockup(rdev, ring);
2176}
2177
 
 
 
 
 
 
 
 
 
2178void r100_bm_disable(struct radeon_device *rdev)
2179{
2180	u32 tmp;
2181
2182	/* disable bus mastering */
2183	tmp = RREG32(R_000030_BUS_CNTL);
2184	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2185	mdelay(1);
2186	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2187	mdelay(1);
2188	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2189	tmp = RREG32(RADEON_BUS_CNTL);
2190	mdelay(1);
2191	pci_clear_master(rdev->pdev);
2192	mdelay(1);
2193}
2194
2195int r100_asic_reset(struct radeon_device *rdev)
2196{
2197	struct r100_mc_save save;
2198	u32 status, tmp;
2199	int ret = 0;
2200
2201	status = RREG32(R_000E40_RBBM_STATUS);
2202	if (!G_000E40_GUI_ACTIVE(status)) {
2203		return 0;
2204	}
2205	r100_mc_stop(rdev, &save);
2206	status = RREG32(R_000E40_RBBM_STATUS);
2207	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2208	/* stop CP */
2209	WREG32(RADEON_CP_CSQ_CNTL, 0);
2210	tmp = RREG32(RADEON_CP_RB_CNTL);
2211	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2212	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2213	WREG32(RADEON_CP_RB_WPTR, 0);
2214	WREG32(RADEON_CP_RB_CNTL, tmp);
2215	/* save PCI state */
2216	pci_save_state(rdev->pdev);
2217	/* disable bus mastering */
2218	r100_bm_disable(rdev);
2219	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2220					S_0000F0_SOFT_RESET_RE(1) |
2221					S_0000F0_SOFT_RESET_PP(1) |
2222					S_0000F0_SOFT_RESET_RB(1));
2223	RREG32(R_0000F0_RBBM_SOFT_RESET);
2224	mdelay(500);
2225	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2226	mdelay(1);
2227	status = RREG32(R_000E40_RBBM_STATUS);
2228	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2229	/* reset CP */
2230	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2231	RREG32(R_0000F0_RBBM_SOFT_RESET);
2232	mdelay(500);
2233	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2234	mdelay(1);
2235	status = RREG32(R_000E40_RBBM_STATUS);
2236	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2237	/* restore PCI & busmastering */
2238	pci_restore_state(rdev->pdev);
2239	r100_enable_bm(rdev);
2240	/* Check if GPU is idle */
2241	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2242		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2243		dev_err(rdev->dev, "failed to reset GPU\n");
2244		ret = -1;
2245	} else
2246		dev_info(rdev->dev, "GPU reset succeed\n");
2247	r100_mc_resume(rdev, &save);
2248	return ret;
2249}
2250
2251void r100_set_common_regs(struct radeon_device *rdev)
2252{
2253	struct drm_device *dev = rdev->ddev;
2254	bool force_dac2 = false;
2255	u32 tmp;
2256
2257	/* set these so they don't interfere with anything */
2258	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2259	WREG32(RADEON_SUBPIC_CNTL, 0);
2260	WREG32(RADEON_VIPH_CONTROL, 0);
2261	WREG32(RADEON_I2C_CNTL_1, 0);
2262	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2263	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2264	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2265
2266	/* always set up dac2 on rn50 and some rv100 as lots
2267	 * of servers seem to wire it up to a VGA port but
2268	 * don't report it in the bios connector
2269	 * table.
2270	 */
2271	switch (dev->pdev->device) {
2272		/* RN50 */
2273	case 0x515e:
2274	case 0x5969:
2275		force_dac2 = true;
2276		break;
2277		/* RV100*/
2278	case 0x5159:
2279	case 0x515a:
2280		/* DELL triple head servers */
2281		if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2282		    ((dev->pdev->subsystem_device == 0x016c) ||
2283		     (dev->pdev->subsystem_device == 0x016d) ||
2284		     (dev->pdev->subsystem_device == 0x016e) ||
2285		     (dev->pdev->subsystem_device == 0x016f) ||
2286		     (dev->pdev->subsystem_device == 0x0170) ||
2287		     (dev->pdev->subsystem_device == 0x017d) ||
2288		     (dev->pdev->subsystem_device == 0x017e) ||
2289		     (dev->pdev->subsystem_device == 0x0183) ||
2290		     (dev->pdev->subsystem_device == 0x018a) ||
2291		     (dev->pdev->subsystem_device == 0x019a)))
2292			force_dac2 = true;
2293		break;
2294	}
2295
2296	if (force_dac2) {
2297		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2298		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2299		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2300
2301		/* For CRT on DAC2, don't turn it on if BIOS didn't
2302		   enable it, even it's detected.
2303		*/
2304
2305		/* force it to crtc0 */
2306		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2307		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2308		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2309
2310		/* set up the TV DAC */
2311		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2312				 RADEON_TV_DAC_STD_MASK |
2313				 RADEON_TV_DAC_RDACPD |
2314				 RADEON_TV_DAC_GDACPD |
2315				 RADEON_TV_DAC_BDACPD |
2316				 RADEON_TV_DAC_BGADJ_MASK |
2317				 RADEON_TV_DAC_DACADJ_MASK);
2318		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2319				RADEON_TV_DAC_NHOLD |
2320				RADEON_TV_DAC_STD_PS2 |
2321				(0x58 << 16));
2322
2323		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2324		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2325		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2326	}
2327
2328	/* switch PM block to ACPI mode */
2329	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2330	tmp &= ~RADEON_PM_MODE_SEL;
2331	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2332
2333}
2334
2335/*
2336 * VRAM info
2337 */
2338static void r100_vram_get_type(struct radeon_device *rdev)
2339{
2340	uint32_t tmp;
2341
2342	rdev->mc.vram_is_ddr = false;
2343	if (rdev->flags & RADEON_IS_IGP)
2344		rdev->mc.vram_is_ddr = true;
2345	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2346		rdev->mc.vram_is_ddr = true;
2347	if ((rdev->family == CHIP_RV100) ||
2348	    (rdev->family == CHIP_RS100) ||
2349	    (rdev->family == CHIP_RS200)) {
2350		tmp = RREG32(RADEON_MEM_CNTL);
2351		if (tmp & RV100_HALF_MODE) {
2352			rdev->mc.vram_width = 32;
2353		} else {
2354			rdev->mc.vram_width = 64;
2355		}
2356		if (rdev->flags & RADEON_SINGLE_CRTC) {
2357			rdev->mc.vram_width /= 4;
2358			rdev->mc.vram_is_ddr = true;
2359		}
2360	} else if (rdev->family <= CHIP_RV280) {
2361		tmp = RREG32(RADEON_MEM_CNTL);
2362		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2363			rdev->mc.vram_width = 128;
2364		} else {
2365			rdev->mc.vram_width = 64;
2366		}
2367	} else {
2368		/* newer IGPs */
2369		rdev->mc.vram_width = 128;
2370	}
2371}
2372
2373static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2374{
2375	u32 aper_size;
2376	u8 byte;
2377
2378	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2379
2380	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2381	 * that is has the 2nd generation multifunction PCI interface
2382	 */
2383	if (rdev->family == CHIP_RV280 ||
2384	    rdev->family >= CHIP_RV350) {
2385		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2386		       ~RADEON_HDP_APER_CNTL);
2387		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2388		return aper_size * 2;
2389	}
2390
2391	/* Older cards have all sorts of funny issues to deal with. First
2392	 * check if it's a multifunction card by reading the PCI config
2393	 * header type... Limit those to one aperture size
2394	 */
2395	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2396	if (byte & 0x80) {
2397		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2398		DRM_INFO("Limiting VRAM to one aperture\n");
2399		return aper_size;
2400	}
2401
2402	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2403	 * have set it up. We don't write this as it's broken on some ASICs but
2404	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2405	 */
2406	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2407		return aper_size * 2;
2408	return aper_size;
2409}
2410
2411void r100_vram_init_sizes(struct radeon_device *rdev)
2412{
2413	u64 config_aper_size;
2414
2415	/* work out accessible VRAM */
2416	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2417	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2418	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2419	/* FIXME we don't use the second aperture yet when we could use it */
2420	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2421		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2422	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2423	if (rdev->flags & RADEON_IS_IGP) {
2424		uint32_t tom;
2425		/* read NB_TOM to get the amount of ram stolen for the GPU */
2426		tom = RREG32(RADEON_NB_TOM);
2427		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2428		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2429		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2430	} else {
2431		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2432		/* Some production boards of m6 will report 0
2433		 * if it's 8 MB
2434		 */
2435		if (rdev->mc.real_vram_size == 0) {
2436			rdev->mc.real_vram_size = 8192 * 1024;
2437			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2438		}
2439		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM - 
2440		 * Novell bug 204882 + along with lots of ubuntu ones
2441		 */
2442		if (rdev->mc.aper_size > config_aper_size)
2443			config_aper_size = rdev->mc.aper_size;
2444
2445		if (config_aper_size > rdev->mc.real_vram_size)
2446			rdev->mc.mc_vram_size = config_aper_size;
2447		else
2448			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2449	}
2450}
2451
2452void r100_vga_set_state(struct radeon_device *rdev, bool state)
2453{
2454	uint32_t temp;
2455
2456	temp = RREG32(RADEON_CONFIG_CNTL);
2457	if (state == false) {
2458		temp &= ~RADEON_CFG_VGA_RAM_EN;
2459		temp |= RADEON_CFG_VGA_IO_DIS;
2460	} else {
2461		temp &= ~RADEON_CFG_VGA_IO_DIS;
2462	}
2463	WREG32(RADEON_CONFIG_CNTL, temp);
2464}
2465
2466void r100_mc_init(struct radeon_device *rdev)
2467{
2468	u64 base;
2469
2470	r100_vram_get_type(rdev);
2471	r100_vram_init_sizes(rdev);
2472	base = rdev->mc.aper_base;
2473	if (rdev->flags & RADEON_IS_IGP)
2474		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2475	radeon_vram_location(rdev, &rdev->mc, base);
2476	rdev->mc.gtt_base_align = 0;
2477	if (!(rdev->flags & RADEON_IS_AGP))
2478		radeon_gtt_location(rdev, &rdev->mc);
2479	radeon_update_bandwidth_info(rdev);
2480}
2481
2482
2483/*
2484 * Indirect registers accessor
2485 */
2486void r100_pll_errata_after_index(struct radeon_device *rdev)
2487{
2488	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2489		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2490		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2491	}
2492}
2493
2494static void r100_pll_errata_after_data(struct radeon_device *rdev)
2495{
2496	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2497	 * or the chip could hang on a subsequent access
2498	 */
2499	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2500		mdelay(5);
2501	}
2502
2503	/* This function is required to workaround a hardware bug in some (all?)
2504	 * revisions of the R300.  This workaround should be called after every
2505	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2506	 * may not be correct.
2507	 */
2508	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2509		uint32_t save, tmp;
2510
2511		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2512		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2513		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2514		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2515		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2516	}
2517}
2518
2519uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2520{
 
2521	uint32_t data;
2522
 
2523	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2524	r100_pll_errata_after_index(rdev);
2525	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2526	r100_pll_errata_after_data(rdev);
 
2527	return data;
2528}
2529
2530void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2531{
 
 
 
2532	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2533	r100_pll_errata_after_index(rdev);
2534	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2535	r100_pll_errata_after_data(rdev);
 
2536}
2537
2538void r100_set_safe_registers(struct radeon_device *rdev)
2539{
2540	if (ASIC_IS_RN50(rdev)) {
2541		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2542		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2543	} else if (rdev->family < CHIP_R200) {
2544		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2545		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2546	} else {
2547		r200_set_safe_registers(rdev);
2548	}
2549}
2550
2551/*
2552 * Debugfs info
2553 */
2554#if defined(CONFIG_DEBUG_FS)
2555static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2556{
2557	struct drm_info_node *node = (struct drm_info_node *) m->private;
2558	struct drm_device *dev = node->minor->dev;
2559	struct radeon_device *rdev = dev->dev_private;
2560	uint32_t reg, value;
2561	unsigned i;
2562
2563	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2564	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2565	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2566	for (i = 0; i < 64; i++) {
2567		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2568		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2569		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2570		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2571		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2572	}
2573	return 0;
2574}
2575
2576static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2577{
2578	struct drm_info_node *node = (struct drm_info_node *) m->private;
2579	struct drm_device *dev = node->minor->dev;
2580	struct radeon_device *rdev = dev->dev_private;
2581	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2582	uint32_t rdp, wdp;
2583	unsigned count, i, j;
2584
2585	radeon_ring_free_size(rdev, ring);
2586	rdp = RREG32(RADEON_CP_RB_RPTR);
2587	wdp = RREG32(RADEON_CP_RB_WPTR);
2588	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2589	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2590	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2591	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2592	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2593	seq_printf(m, "%u dwords in ring\n", count);
2594	for (j = 0; j <= count; j++) {
2595		i = (rdp + j) & ring->ptr_mask;
2596		seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
 
 
2597	}
2598	return 0;
2599}
2600
2601
2602static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2603{
2604	struct drm_info_node *node = (struct drm_info_node *) m->private;
2605	struct drm_device *dev = node->minor->dev;
2606	struct radeon_device *rdev = dev->dev_private;
2607	uint32_t csq_stat, csq2_stat, tmp;
2608	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2609	unsigned i;
2610
2611	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2612	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2613	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2614	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2615	r_rptr = (csq_stat >> 0) & 0x3ff;
2616	r_wptr = (csq_stat >> 10) & 0x3ff;
2617	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2618	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2619	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2620	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2621	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2622	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2623	seq_printf(m, "Ring rptr %u\n", r_rptr);
2624	seq_printf(m, "Ring wptr %u\n", r_wptr);
2625	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2626	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2627	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2628	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2629	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
2630	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
2631	seq_printf(m, "Ring fifo:\n");
2632	for (i = 0; i < 256; i++) {
2633		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2634		tmp = RREG32(RADEON_CP_CSQ_DATA);
2635		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
2636	}
2637	seq_printf(m, "Indirect1 fifo:\n");
2638	for (i = 256; i <= 512; i++) {
2639		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2640		tmp = RREG32(RADEON_CP_CSQ_DATA);
2641		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
2642	}
2643	seq_printf(m, "Indirect2 fifo:\n");
2644	for (i = 640; i < ib1_wptr; i++) {
2645		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2646		tmp = RREG32(RADEON_CP_CSQ_DATA);
2647		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
2648	}
2649	return 0;
2650}
2651
2652static int r100_debugfs_mc_info(struct seq_file *m, void *data)
2653{
2654	struct drm_info_node *node = (struct drm_info_node *) m->private;
2655	struct drm_device *dev = node->minor->dev;
2656	struct radeon_device *rdev = dev->dev_private;
2657	uint32_t tmp;
2658
2659	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
2660	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
2661	tmp = RREG32(RADEON_MC_FB_LOCATION);
2662	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
2663	tmp = RREG32(RADEON_BUS_CNTL);
2664	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
2665	tmp = RREG32(RADEON_MC_AGP_LOCATION);
2666	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
2667	tmp = RREG32(RADEON_AGP_BASE);
2668	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
2669	tmp = RREG32(RADEON_HOST_PATH_CNTL);
2670	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
2671	tmp = RREG32(0x01D0);
2672	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
2673	tmp = RREG32(RADEON_AIC_LO_ADDR);
2674	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
2675	tmp = RREG32(RADEON_AIC_HI_ADDR);
2676	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
2677	tmp = RREG32(0x01E4);
2678	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
2679	return 0;
2680}
2681
2682static struct drm_info_list r100_debugfs_rbbm_list[] = {
2683	{"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
2684};
2685
2686static struct drm_info_list r100_debugfs_cp_list[] = {
2687	{"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
2688	{"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
2689};
2690
2691static struct drm_info_list r100_debugfs_mc_info_list[] = {
2692	{"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
2693};
2694#endif
2695
2696int r100_debugfs_rbbm_init(struct radeon_device *rdev)
2697{
2698#if defined(CONFIG_DEBUG_FS)
2699	return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
2700#else
2701	return 0;
2702#endif
2703}
2704
2705int r100_debugfs_cp_init(struct radeon_device *rdev)
2706{
2707#if defined(CONFIG_DEBUG_FS)
2708	return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
2709#else
2710	return 0;
2711#endif
2712}
2713
2714int r100_debugfs_mc_info_init(struct radeon_device *rdev)
2715{
2716#if defined(CONFIG_DEBUG_FS)
2717	return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
2718#else
2719	return 0;
2720#endif
2721}
2722
2723int r100_set_surface_reg(struct radeon_device *rdev, int reg,
2724			 uint32_t tiling_flags, uint32_t pitch,
2725			 uint32_t offset, uint32_t obj_size)
2726{
2727	int surf_index = reg * 16;
2728	int flags = 0;
2729
2730	if (rdev->family <= CHIP_RS200) {
2731		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
2732				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
2733			flags |= RADEON_SURF_TILE_COLOR_BOTH;
2734		if (tiling_flags & RADEON_TILING_MACRO)
2735			flags |= RADEON_SURF_TILE_COLOR_MACRO;
 
 
 
 
2736	} else if (rdev->family <= CHIP_RV280) {
2737		if (tiling_flags & (RADEON_TILING_MACRO))
2738			flags |= R200_SURF_TILE_COLOR_MACRO;
2739		if (tiling_flags & RADEON_TILING_MICRO)
2740			flags |= R200_SURF_TILE_COLOR_MICRO;
2741	} else {
2742		if (tiling_flags & RADEON_TILING_MACRO)
2743			flags |= R300_SURF_TILE_MACRO;
2744		if (tiling_flags & RADEON_TILING_MICRO)
2745			flags |= R300_SURF_TILE_MICRO;
2746	}
2747
2748	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
2749		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
2750	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
2751		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
2752
2753	/* when we aren't tiling the pitch seems to needs to be furtherdivided down. - tested on power5 + rn50 server */
2754	if (tiling_flags & (RADEON_TILING_SWAP_16BIT | RADEON_TILING_SWAP_32BIT)) {
2755		if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO)))
2756			if (ASIC_IS_RN50(rdev))
2757				pitch /= 16;
2758	}
2759
2760	/* r100/r200 divide by 16 */
2761	if (rdev->family < CHIP_R300)
2762		flags |= pitch / 16;
2763	else
2764		flags |= pitch / 8;
2765
2766
2767	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
2768	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
2769	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
2770	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
2771	return 0;
2772}
2773
2774void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
2775{
2776	int surf_index = reg * 16;
2777	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
2778}
2779
2780void r100_bandwidth_update(struct radeon_device *rdev)
2781{
2782	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
2783	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
2784	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
 
2785	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
2786	fixed20_12 memtcas_ff[8] = {
2787		dfixed_init(1),
2788		dfixed_init(2),
2789		dfixed_init(3),
2790		dfixed_init(0),
2791		dfixed_init_half(1),
2792		dfixed_init_half(2),
2793		dfixed_init(0),
2794	};
2795	fixed20_12 memtcas_rs480_ff[8] = {
2796		dfixed_init(0),
2797		dfixed_init(1),
2798		dfixed_init(2),
2799		dfixed_init(3),
2800		dfixed_init(0),
2801		dfixed_init_half(1),
2802		dfixed_init_half(2),
2803		dfixed_init_half(3),
2804	};
2805	fixed20_12 memtcas2_ff[8] = {
2806		dfixed_init(0),
2807		dfixed_init(1),
2808		dfixed_init(2),
2809		dfixed_init(3),
2810		dfixed_init(4),
2811		dfixed_init(5),
2812		dfixed_init(6),
2813		dfixed_init(7),
2814	};
2815	fixed20_12 memtrbs[8] = {
2816		dfixed_init(1),
2817		dfixed_init_half(1),
2818		dfixed_init(2),
2819		dfixed_init_half(2),
2820		dfixed_init(3),
2821		dfixed_init_half(3),
2822		dfixed_init(4),
2823		dfixed_init_half(4)
2824	};
2825	fixed20_12 memtrbs_r4xx[8] = {
2826		dfixed_init(4),
2827		dfixed_init(5),
2828		dfixed_init(6),
2829		dfixed_init(7),
2830		dfixed_init(8),
2831		dfixed_init(9),
2832		dfixed_init(10),
2833		dfixed_init(11)
2834	};
2835	fixed20_12 min_mem_eff;
2836	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
2837	fixed20_12 cur_latency_mclk, cur_latency_sclk;
2838	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
2839		disp_drain_rate2, read_return_rate;
2840	fixed20_12 time_disp1_drop_priority;
2841	int c;
2842	int cur_size = 16;       /* in octawords */
2843	int critical_point = 0, critical_point2;
2844/* 	uint32_t read_return_rate, time_disp1_drop_priority; */
2845	int stop_req, max_stop_req;
2846	struct drm_display_mode *mode1 = NULL;
2847	struct drm_display_mode *mode2 = NULL;
2848	uint32_t pixel_bytes1 = 0;
2849	uint32_t pixel_bytes2 = 0;
2850
 
 
 
 
 
 
2851	radeon_update_display_priority(rdev);
2852
2853	if (rdev->mode_info.crtcs[0]->base.enabled) {
2854		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
2855		pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
2856	}
2857	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
2858		if (rdev->mode_info.crtcs[1]->base.enabled) {
2859			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
2860			pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
2861		}
2862	}
2863
2864	min_mem_eff.full = dfixed_const_8(0);
2865	/* get modes */
2866	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
2867		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
2868		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
2869		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
2870		/* check crtc enables */
2871		if (mode2)
2872			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
2873		if (mode1)
2874			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
2875		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
2876	}
2877
2878	/*
2879	 * determine is there is enough bw for current mode
2880	 */
2881	sclk_ff = rdev->pm.sclk;
2882	mclk_ff = rdev->pm.mclk;
2883
2884	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
2885	temp_ff.full = dfixed_const(temp);
2886	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
2887
2888	pix_clk.full = 0;
2889	pix_clk2.full = 0;
2890	peak_disp_bw.full = 0;
2891	if (mode1) {
2892		temp_ff.full = dfixed_const(1000);
2893		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
2894		pix_clk.full = dfixed_div(pix_clk, temp_ff);
2895		temp_ff.full = dfixed_const(pixel_bytes1);
2896		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
2897	}
2898	if (mode2) {
2899		temp_ff.full = dfixed_const(1000);
2900		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
2901		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
2902		temp_ff.full = dfixed_const(pixel_bytes2);
2903		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
2904	}
2905
2906	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
2907	if (peak_disp_bw.full >= mem_bw.full) {
2908		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
2909			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
2910	}
2911
2912	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
2913	temp = RREG32(RADEON_MEM_TIMING_CNTL);
2914	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
2915		mem_trcd = ((temp >> 2) & 0x3) + 1;
2916		mem_trp  = ((temp & 0x3)) + 1;
2917		mem_tras = ((temp & 0x70) >> 4) + 1;
2918	} else if (rdev->family == CHIP_R300 ||
2919		   rdev->family == CHIP_R350) { /* r300, r350 */
2920		mem_trcd = (temp & 0x7) + 1;
2921		mem_trp = ((temp >> 8) & 0x7) + 1;
2922		mem_tras = ((temp >> 11) & 0xf) + 4;
2923	} else if (rdev->family == CHIP_RV350 ||
2924		   rdev->family <= CHIP_RV380) {
2925		/* rv3x0 */
2926		mem_trcd = (temp & 0x7) + 3;
2927		mem_trp = ((temp >> 8) & 0x7) + 3;
2928		mem_tras = ((temp >> 11) & 0xf) + 6;
2929	} else if (rdev->family == CHIP_R420 ||
2930		   rdev->family == CHIP_R423 ||
2931		   rdev->family == CHIP_RV410) {
2932		/* r4xx */
2933		mem_trcd = (temp & 0xf) + 3;
2934		if (mem_trcd > 15)
2935			mem_trcd = 15;
2936		mem_trp = ((temp >> 8) & 0xf) + 3;
2937		if (mem_trp > 15)
2938			mem_trp = 15;
2939		mem_tras = ((temp >> 12) & 0x1f) + 6;
2940		if (mem_tras > 31)
2941			mem_tras = 31;
2942	} else { /* RV200, R200 */
2943		mem_trcd = (temp & 0x7) + 1;
2944		mem_trp = ((temp >> 8) & 0x7) + 1;
2945		mem_tras = ((temp >> 12) & 0xf) + 4;
2946	}
2947	/* convert to FF */
2948	trcd_ff.full = dfixed_const(mem_trcd);
2949	trp_ff.full = dfixed_const(mem_trp);
2950	tras_ff.full = dfixed_const(mem_tras);
2951
2952	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
2953	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
2954	data = (temp & (7 << 20)) >> 20;
2955	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
2956		if (rdev->family == CHIP_RS480) /* don't think rs400 */
2957			tcas_ff = memtcas_rs480_ff[data];
2958		else
2959			tcas_ff = memtcas_ff[data];
2960	} else
2961		tcas_ff = memtcas2_ff[data];
2962
2963	if (rdev->family == CHIP_RS400 ||
2964	    rdev->family == CHIP_RS480) {
2965		/* extra cas latency stored in bits 23-25 0-4 clocks */
2966		data = (temp >> 23) & 0x7;
2967		if (data < 5)
2968			tcas_ff.full += dfixed_const(data);
2969	}
2970
2971	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
2972		/* on the R300, Tcas is included in Trbs.
2973		 */
2974		temp = RREG32(RADEON_MEM_CNTL);
2975		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
2976		if (data == 1) {
2977			if (R300_MEM_USE_CD_CH_ONLY & temp) {
2978				temp = RREG32(R300_MC_IND_INDEX);
2979				temp &= ~R300_MC_IND_ADDR_MASK;
2980				temp |= R300_MC_READ_CNTL_CD_mcind;
2981				WREG32(R300_MC_IND_INDEX, temp);
2982				temp = RREG32(R300_MC_IND_DATA);
2983				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
2984			} else {
2985				temp = RREG32(R300_MC_READ_CNTL_AB);
2986				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2987			}
2988		} else {
2989			temp = RREG32(R300_MC_READ_CNTL_AB);
2990			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2991		}
2992		if (rdev->family == CHIP_RV410 ||
2993		    rdev->family == CHIP_R420 ||
2994		    rdev->family == CHIP_R423)
2995			trbs_ff = memtrbs_r4xx[data];
2996		else
2997			trbs_ff = memtrbs[data];
2998		tcas_ff.full += trbs_ff.full;
2999	}
3000
3001	sclk_eff_ff.full = sclk_ff.full;
3002
3003	if (rdev->flags & RADEON_IS_AGP) {
3004		fixed20_12 agpmode_ff;
3005		agpmode_ff.full = dfixed_const(radeon_agpmode);
3006		temp_ff.full = dfixed_const_666(16);
3007		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3008	}
3009	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3010
3011	if (ASIC_IS_R300(rdev)) {
3012		sclk_delay_ff.full = dfixed_const(250);
3013	} else {
3014		if ((rdev->family == CHIP_RV100) ||
3015		    rdev->flags & RADEON_IS_IGP) {
3016			if (rdev->mc.vram_is_ddr)
3017				sclk_delay_ff.full = dfixed_const(41);
3018			else
3019				sclk_delay_ff.full = dfixed_const(33);
3020		} else {
3021			if (rdev->mc.vram_width == 128)
3022				sclk_delay_ff.full = dfixed_const(57);
3023			else
3024				sclk_delay_ff.full = dfixed_const(41);
3025		}
3026	}
3027
3028	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3029
3030	if (rdev->mc.vram_is_ddr) {
3031		if (rdev->mc.vram_width == 32) {
3032			k1.full = dfixed_const(40);
3033			c  = 3;
3034		} else {
3035			k1.full = dfixed_const(20);
3036			c  = 1;
3037		}
3038	} else {
3039		k1.full = dfixed_const(40);
3040		c  = 3;
3041	}
3042
3043	temp_ff.full = dfixed_const(2);
3044	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3045	temp_ff.full = dfixed_const(c);
3046	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3047	temp_ff.full = dfixed_const(4);
3048	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3049	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3050	mc_latency_mclk.full += k1.full;
3051
3052	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3053	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3054
3055	/*
3056	  HW cursor time assuming worst case of full size colour cursor.
3057	*/
3058	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3059	temp_ff.full += trcd_ff.full;
3060	if (temp_ff.full < tras_ff.full)
3061		temp_ff.full = tras_ff.full;
3062	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3063
3064	temp_ff.full = dfixed_const(cur_size);
3065	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3066	/*
3067	  Find the total latency for the display data.
3068	*/
3069	disp_latency_overhead.full = dfixed_const(8);
3070	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3071	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3072	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3073
3074	if (mc_latency_mclk.full > mc_latency_sclk.full)
3075		disp_latency.full = mc_latency_mclk.full;
3076	else
3077		disp_latency.full = mc_latency_sclk.full;
3078
3079	/* setup Max GRPH_STOP_REQ default value */
3080	if (ASIC_IS_RV100(rdev))
3081		max_stop_req = 0x5c;
3082	else
3083		max_stop_req = 0x7c;
3084
3085	if (mode1) {
3086		/*  CRTC1
3087		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3088		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3089		*/
3090		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3091
3092		if (stop_req > max_stop_req)
3093			stop_req = max_stop_req;
3094
3095		/*
3096		  Find the drain rate of the display buffer.
3097		*/
3098		temp_ff.full = dfixed_const((16/pixel_bytes1));
3099		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3100
3101		/*
3102		  Find the critical point of the display buffer.
3103		*/
3104		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3105		crit_point_ff.full += dfixed_const_half(0);
3106
3107		critical_point = dfixed_trunc(crit_point_ff);
3108
3109		if (rdev->disp_priority == 2) {
3110			critical_point = 0;
3111		}
3112
3113		/*
3114		  The critical point should never be above max_stop_req-4.  Setting
3115		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3116		*/
3117		if (max_stop_req - critical_point < 4)
3118			critical_point = 0;
3119
3120		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3121			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3122			critical_point = 0x10;
3123		}
3124
3125		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3126		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3127		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3128		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3129		if ((rdev->family == CHIP_R350) &&
3130		    (stop_req > 0x15)) {
3131			stop_req -= 0x10;
3132		}
3133		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3134		temp |= RADEON_GRPH_BUFFER_SIZE;
3135		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3136			  RADEON_GRPH_CRITICAL_AT_SOF |
3137			  RADEON_GRPH_STOP_CNTL);
3138		/*
3139		  Write the result into the register.
3140		*/
3141		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3142						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3143
3144#if 0
3145		if ((rdev->family == CHIP_RS400) ||
3146		    (rdev->family == CHIP_RS480)) {
3147			/* attempt to program RS400 disp regs correctly ??? */
3148			temp = RREG32(RS400_DISP1_REG_CNTL);
3149			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3150				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3151			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3152						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3153						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3154			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3155			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3156				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3157			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3158						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3159						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3160		}
3161#endif
3162
3163		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3164			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3165			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3166	}
3167
3168	if (mode2) {
3169		u32 grph2_cntl;
3170		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3171
3172		if (stop_req > max_stop_req)
3173			stop_req = max_stop_req;
3174
3175		/*
3176		  Find the drain rate of the display buffer.
3177		*/
3178		temp_ff.full = dfixed_const((16/pixel_bytes2));
3179		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3180
3181		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3182		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3183		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3184		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3185		if ((rdev->family == CHIP_R350) &&
3186		    (stop_req > 0x15)) {
3187			stop_req -= 0x10;
3188		}
3189		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3190		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3191		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3192			  RADEON_GRPH_CRITICAL_AT_SOF |
3193			  RADEON_GRPH_STOP_CNTL);
3194
3195		if ((rdev->family == CHIP_RS100) ||
3196		    (rdev->family == CHIP_RS200))
3197			critical_point2 = 0;
3198		else {
3199			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3200			temp_ff.full = dfixed_const(temp);
3201			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3202			if (sclk_ff.full < temp_ff.full)
3203				temp_ff.full = sclk_ff.full;
3204
3205			read_return_rate.full = temp_ff.full;
3206
3207			if (mode1) {
3208				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3209				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3210			} else {
3211				time_disp1_drop_priority.full = 0;
3212			}
3213			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3214			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3215			crit_point_ff.full += dfixed_const_half(0);
3216
3217			critical_point2 = dfixed_trunc(crit_point_ff);
3218
3219			if (rdev->disp_priority == 2) {
3220				critical_point2 = 0;
3221			}
3222
3223			if (max_stop_req - critical_point2 < 4)
3224				critical_point2 = 0;
3225
3226		}
3227
3228		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3229			/* some R300 cards have problem with this set to 0 */
3230			critical_point2 = 0x10;
3231		}
3232
3233		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3234						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3235
3236		if ((rdev->family == CHIP_RS400) ||
3237		    (rdev->family == CHIP_RS480)) {
3238#if 0
3239			/* attempt to program RS400 disp2 regs correctly ??? */
3240			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3241			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3242				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3243			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3244						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3245						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3246			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3247			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3248				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3249			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3250						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3251						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3252#endif
3253			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3254			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3255			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3256			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3257		}
3258
3259		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3260			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3261	}
3262}
3263
3264static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
3265{
3266	DRM_ERROR("pitch                      %d\n", t->pitch);
3267	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
3268	DRM_ERROR("width                      %d\n", t->width);
3269	DRM_ERROR("width_11                   %d\n", t->width_11);
3270	DRM_ERROR("height                     %d\n", t->height);
3271	DRM_ERROR("height_11                  %d\n", t->height_11);
3272	DRM_ERROR("num levels                 %d\n", t->num_levels);
3273	DRM_ERROR("depth                      %d\n", t->txdepth);
3274	DRM_ERROR("bpp                        %d\n", t->cpp);
3275	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
3276	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
3277	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
3278	DRM_ERROR("compress format            %d\n", t->compress_format);
3279}
3280
3281static int r100_track_compress_size(int compress_format, int w, int h)
3282{
3283	int block_width, block_height, block_bytes;
3284	int wblocks, hblocks;
3285	int min_wblocks;
3286	int sz;
3287
3288	block_width = 4;
3289	block_height = 4;
3290
3291	switch (compress_format) {
3292	case R100_TRACK_COMP_DXT1:
3293		block_bytes = 8;
3294		min_wblocks = 4;
3295		break;
3296	default:
3297	case R100_TRACK_COMP_DXT35:
3298		block_bytes = 16;
3299		min_wblocks = 2;
3300		break;
3301	}
3302
3303	hblocks = (h + block_height - 1) / block_height;
3304	wblocks = (w + block_width - 1) / block_width;
3305	if (wblocks < min_wblocks)
3306		wblocks = min_wblocks;
3307	sz = wblocks * hblocks * block_bytes;
3308	return sz;
3309}
3310
3311static int r100_cs_track_cube(struct radeon_device *rdev,
3312			      struct r100_cs_track *track, unsigned idx)
3313{
3314	unsigned face, w, h;
3315	struct radeon_bo *cube_robj;
3316	unsigned long size;
3317	unsigned compress_format = track->textures[idx].compress_format;
3318
3319	for (face = 0; face < 5; face++) {
3320		cube_robj = track->textures[idx].cube_info[face].robj;
3321		w = track->textures[idx].cube_info[face].width;
3322		h = track->textures[idx].cube_info[face].height;
3323
3324		if (compress_format) {
3325			size = r100_track_compress_size(compress_format, w, h);
3326		} else
3327			size = w * h;
3328		size *= track->textures[idx].cpp;
3329
3330		size += track->textures[idx].cube_info[face].offset;
3331
3332		if (size > radeon_bo_size(cube_robj)) {
3333			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
3334				  size, radeon_bo_size(cube_robj));
3335			r100_cs_track_texture_print(&track->textures[idx]);
3336			return -1;
3337		}
3338	}
3339	return 0;
3340}
3341
3342static int r100_cs_track_texture_check(struct radeon_device *rdev,
3343				       struct r100_cs_track *track)
3344{
3345	struct radeon_bo *robj;
3346	unsigned long size;
3347	unsigned u, i, w, h, d;
3348	int ret;
3349
3350	for (u = 0; u < track->num_texture; u++) {
3351		if (!track->textures[u].enabled)
3352			continue;
3353		if (track->textures[u].lookup_disable)
3354			continue;
3355		robj = track->textures[u].robj;
3356		if (robj == NULL) {
3357			DRM_ERROR("No texture bound to unit %u\n", u);
3358			return -EINVAL;
3359		}
3360		size = 0;
3361		for (i = 0; i <= track->textures[u].num_levels; i++) {
3362			if (track->textures[u].use_pitch) {
3363				if (rdev->family < CHIP_R300)
3364					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
3365				else
3366					w = track->textures[u].pitch / (1 << i);
3367			} else {
3368				w = track->textures[u].width;
3369				if (rdev->family >= CHIP_RV515)
3370					w |= track->textures[u].width_11;
3371				w = w / (1 << i);
3372				if (track->textures[u].roundup_w)
3373					w = roundup_pow_of_two(w);
3374			}
3375			h = track->textures[u].height;
3376			if (rdev->family >= CHIP_RV515)
3377				h |= track->textures[u].height_11;
3378			h = h / (1 << i);
3379			if (track->textures[u].roundup_h)
3380				h = roundup_pow_of_two(h);
3381			if (track->textures[u].tex_coord_type == 1) {
3382				d = (1 << track->textures[u].txdepth) / (1 << i);
3383				if (!d)
3384					d = 1;
3385			} else {
3386				d = 1;
3387			}
3388			if (track->textures[u].compress_format) {
3389
3390				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
3391				/* compressed textures are block based */
3392			} else
3393				size += w * h * d;
3394		}
3395		size *= track->textures[u].cpp;
3396
3397		switch (track->textures[u].tex_coord_type) {
3398		case 0:
3399		case 1:
3400			break;
3401		case 2:
3402			if (track->separate_cube) {
3403				ret = r100_cs_track_cube(rdev, track, u);
3404				if (ret)
3405					return ret;
3406			} else
3407				size *= 6;
3408			break;
3409		default:
3410			DRM_ERROR("Invalid texture coordinate type %u for unit "
3411				  "%u\n", track->textures[u].tex_coord_type, u);
3412			return -EINVAL;
3413		}
3414		if (size > radeon_bo_size(robj)) {
3415			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
3416				  "%lu\n", u, size, radeon_bo_size(robj));
3417			r100_cs_track_texture_print(&track->textures[u]);
3418			return -EINVAL;
3419		}
3420	}
3421	return 0;
3422}
3423
3424int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
3425{
3426	unsigned i;
3427	unsigned long size;
3428	unsigned prim_walk;
3429	unsigned nverts;
3430	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
3431
3432	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
3433	    !track->blend_read_enable)
3434		num_cb = 0;
3435
3436	for (i = 0; i < num_cb; i++) {
3437		if (track->cb[i].robj == NULL) {
3438			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
3439			return -EINVAL;
3440		}
3441		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
3442		size += track->cb[i].offset;
3443		if (size > radeon_bo_size(track->cb[i].robj)) {
3444			DRM_ERROR("[drm] Buffer too small for color buffer %d "
3445				  "(need %lu have %lu) !\n", i, size,
3446				  radeon_bo_size(track->cb[i].robj));
3447			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
3448				  i, track->cb[i].pitch, track->cb[i].cpp,
3449				  track->cb[i].offset, track->maxy);
3450			return -EINVAL;
3451		}
3452	}
3453	track->cb_dirty = false;
3454
3455	if (track->zb_dirty && track->z_enabled) {
3456		if (track->zb.robj == NULL) {
3457			DRM_ERROR("[drm] No buffer for z buffer !\n");
3458			return -EINVAL;
3459		}
3460		size = track->zb.pitch * track->zb.cpp * track->maxy;
3461		size += track->zb.offset;
3462		if (size > radeon_bo_size(track->zb.robj)) {
3463			DRM_ERROR("[drm] Buffer too small for z buffer "
3464				  "(need %lu have %lu) !\n", size,
3465				  radeon_bo_size(track->zb.robj));
3466			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
3467				  track->zb.pitch, track->zb.cpp,
3468				  track->zb.offset, track->maxy);
3469			return -EINVAL;
3470		}
3471	}
3472	track->zb_dirty = false;
3473
3474	if (track->aa_dirty && track->aaresolve) {
3475		if (track->aa.robj == NULL) {
3476			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
3477			return -EINVAL;
3478		}
3479		/* I believe the format comes from colorbuffer0. */
3480		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
3481		size += track->aa.offset;
3482		if (size > radeon_bo_size(track->aa.robj)) {
3483			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
3484				  "(need %lu have %lu) !\n", i, size,
3485				  radeon_bo_size(track->aa.robj));
3486			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
3487				  i, track->aa.pitch, track->cb[0].cpp,
3488				  track->aa.offset, track->maxy);
3489			return -EINVAL;
3490		}
3491	}
3492	track->aa_dirty = false;
3493
3494	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
3495	if (track->vap_vf_cntl & (1 << 14)) {
3496		nverts = track->vap_alt_nverts;
3497	} else {
3498		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
3499	}
3500	switch (prim_walk) {
3501	case 1:
3502		for (i = 0; i < track->num_arrays; i++) {
3503			size = track->arrays[i].esize * track->max_indx * 4;
3504			if (track->arrays[i].robj == NULL) {
3505				DRM_ERROR("(PW %u) Vertex array %u no buffer "
3506					  "bound\n", prim_walk, i);
3507				return -EINVAL;
3508			}
3509			if (size > radeon_bo_size(track->arrays[i].robj)) {
3510				dev_err(rdev->dev, "(PW %u) Vertex array %u "
3511					"need %lu dwords have %lu dwords\n",
3512					prim_walk, i, size >> 2,
3513					radeon_bo_size(track->arrays[i].robj)
3514					>> 2);
3515				DRM_ERROR("Max indices %u\n", track->max_indx);
3516				return -EINVAL;
3517			}
3518		}
3519		break;
3520	case 2:
3521		for (i = 0; i < track->num_arrays; i++) {
3522			size = track->arrays[i].esize * (nverts - 1) * 4;
3523			if (track->arrays[i].robj == NULL) {
3524				DRM_ERROR("(PW %u) Vertex array %u no buffer "
3525					  "bound\n", prim_walk, i);
3526				return -EINVAL;
3527			}
3528			if (size > radeon_bo_size(track->arrays[i].robj)) {
3529				dev_err(rdev->dev, "(PW %u) Vertex array %u "
3530					"need %lu dwords have %lu dwords\n",
3531					prim_walk, i, size >> 2,
3532					radeon_bo_size(track->arrays[i].robj)
3533					>> 2);
3534				return -EINVAL;
3535			}
3536		}
3537		break;
3538	case 3:
3539		size = track->vtx_size * nverts;
3540		if (size != track->immd_dwords) {
3541			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
3542				  track->immd_dwords, size);
3543			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
3544				  nverts, track->vtx_size);
3545			return -EINVAL;
3546		}
3547		break;
3548	default:
3549		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
3550			  prim_walk);
3551		return -EINVAL;
3552	}
3553
3554	if (track->tex_dirty) {
3555		track->tex_dirty = false;
3556		return r100_cs_track_texture_check(rdev, track);
3557	}
3558	return 0;
3559}
3560
3561void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
3562{
3563	unsigned i, face;
3564
3565	track->cb_dirty = true;
3566	track->zb_dirty = true;
3567	track->tex_dirty = true;
3568	track->aa_dirty = true;
3569
3570	if (rdev->family < CHIP_R300) {
3571		track->num_cb = 1;
3572		if (rdev->family <= CHIP_RS200)
3573			track->num_texture = 3;
3574		else
3575			track->num_texture = 6;
3576		track->maxy = 2048;
3577		track->separate_cube = 1;
3578	} else {
3579		track->num_cb = 4;
3580		track->num_texture = 16;
3581		track->maxy = 4096;
3582		track->separate_cube = 0;
3583		track->aaresolve = false;
3584		track->aa.robj = NULL;
3585	}
3586
3587	for (i = 0; i < track->num_cb; i++) {
3588		track->cb[i].robj = NULL;
3589		track->cb[i].pitch = 8192;
3590		track->cb[i].cpp = 16;
3591		track->cb[i].offset = 0;
3592	}
3593	track->z_enabled = true;
3594	track->zb.robj = NULL;
3595	track->zb.pitch = 8192;
3596	track->zb.cpp = 4;
3597	track->zb.offset = 0;
3598	track->vtx_size = 0x7F;
3599	track->immd_dwords = 0xFFFFFFFFUL;
3600	track->num_arrays = 11;
3601	track->max_indx = 0x00FFFFFFUL;
3602	for (i = 0; i < track->num_arrays; i++) {
3603		track->arrays[i].robj = NULL;
3604		track->arrays[i].esize = 0x7F;
3605	}
3606	for (i = 0; i < track->num_texture; i++) {
3607		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
3608		track->textures[i].pitch = 16536;
3609		track->textures[i].width = 16536;
3610		track->textures[i].height = 16536;
3611		track->textures[i].width_11 = 1 << 11;
3612		track->textures[i].height_11 = 1 << 11;
3613		track->textures[i].num_levels = 12;
3614		if (rdev->family <= CHIP_RS200) {
3615			track->textures[i].tex_coord_type = 0;
3616			track->textures[i].txdepth = 0;
3617		} else {
3618			track->textures[i].txdepth = 16;
3619			track->textures[i].tex_coord_type = 1;
3620		}
3621		track->textures[i].cpp = 64;
3622		track->textures[i].robj = NULL;
3623		/* CS IB emission code makes sure texture unit are disabled */
3624		track->textures[i].enabled = false;
3625		track->textures[i].lookup_disable = false;
3626		track->textures[i].roundup_w = true;
3627		track->textures[i].roundup_h = true;
3628		if (track->separate_cube)
3629			for (face = 0; face < 5; face++) {
3630				track->textures[i].cube_info[face].robj = NULL;
3631				track->textures[i].cube_info[face].width = 16536;
3632				track->textures[i].cube_info[face].height = 16536;
3633				track->textures[i].cube_info[face].offset = 0;
3634			}
3635	}
3636}
3637
3638int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3639{
3640	uint32_t scratch;
3641	uint32_t tmp = 0;
3642	unsigned i;
3643	int r;
3644
3645	r = radeon_scratch_get(rdev, &scratch);
3646	if (r) {
3647		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3648		return r;
3649	}
3650	WREG32(scratch, 0xCAFEDEAD);
3651	r = radeon_ring_lock(rdev, ring, 2);
3652	if (r) {
3653		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3654		radeon_scratch_free(rdev, scratch);
3655		return r;
3656	}
3657	radeon_ring_write(ring, PACKET0(scratch, 0));
3658	radeon_ring_write(ring, 0xDEADBEEF);
3659	radeon_ring_unlock_commit(rdev, ring);
3660	for (i = 0; i < rdev->usec_timeout; i++) {
3661		tmp = RREG32(scratch);
3662		if (tmp == 0xDEADBEEF) {
3663			break;
3664		}
3665		DRM_UDELAY(1);
3666	}
3667	if (i < rdev->usec_timeout) {
3668		DRM_INFO("ring test succeeded in %d usecs\n", i);
3669	} else {
3670		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3671			  scratch, tmp);
3672		r = -EINVAL;
3673	}
3674	radeon_scratch_free(rdev, scratch);
3675	return r;
3676}
3677
3678void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3679{
3680	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3681
 
 
 
 
 
 
3682	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3683	radeon_ring_write(ring, ib->gpu_addr);
3684	radeon_ring_write(ring, ib->length_dw);
3685}
3686
3687int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3688{
3689	struct radeon_ib ib;
3690	uint32_t scratch;
3691	uint32_t tmp = 0;
3692	unsigned i;
3693	int r;
3694
3695	r = radeon_scratch_get(rdev, &scratch);
3696	if (r) {
3697		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3698		return r;
3699	}
3700	WREG32(scratch, 0xCAFEDEAD);
3701	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, 256);
3702	if (r) {
3703		return r;
 
3704	}
3705	ib.ptr[0] = PACKET0(scratch, 0);
3706	ib.ptr[1] = 0xDEADBEEF;
3707	ib.ptr[2] = PACKET2(0);
3708	ib.ptr[3] = PACKET2(0);
3709	ib.ptr[4] = PACKET2(0);
3710	ib.ptr[5] = PACKET2(0);
3711	ib.ptr[6] = PACKET2(0);
3712	ib.ptr[7] = PACKET2(0);
3713	ib.length_dw = 8;
3714	r = radeon_ib_schedule(rdev, &ib);
3715	if (r) {
3716		radeon_scratch_free(rdev, scratch);
3717		radeon_ib_free(rdev, &ib);
3718		return r;
3719	}
3720	r = radeon_fence_wait(ib.fence, false);
3721	if (r) {
3722		return r;
 
 
 
 
 
 
3723	}
 
3724	for (i = 0; i < rdev->usec_timeout; i++) {
3725		tmp = RREG32(scratch);
3726		if (tmp == 0xDEADBEEF) {
3727			break;
3728		}
3729		DRM_UDELAY(1);
3730	}
3731	if (i < rdev->usec_timeout) {
3732		DRM_INFO("ib test succeeded in %u usecs\n", i);
3733	} else {
3734		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3735			  scratch, tmp);
3736		r = -EINVAL;
3737	}
 
 
 
3738	radeon_scratch_free(rdev, scratch);
3739	radeon_ib_free(rdev, &ib);
3740	return r;
3741}
3742
3743void r100_ib_fini(struct radeon_device *rdev)
3744{
3745	radeon_ib_pool_suspend(rdev);
3746	radeon_ib_pool_fini(rdev);
3747}
3748
3749void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3750{
3751	/* Shutdown CP we shouldn't need to do that but better be safe than
3752	 * sorry
3753	 */
3754	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3755	WREG32(R_000740_CP_CSQ_CNTL, 0);
3756
3757	/* Save few CRTC registers */
3758	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3759	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3760	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3761	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3762	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3763		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3764		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3765	}
3766
3767	/* Disable VGA aperture access */
3768	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3769	/* Disable cursor, overlay, crtc */
3770	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3771	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3772					S_000054_CRTC_DISPLAY_DIS(1));
3773	WREG32(R_000050_CRTC_GEN_CNTL,
3774			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3775			S_000050_CRTC_DISP_REQ_EN_B(1));
3776	WREG32(R_000420_OV0_SCALE_CNTL,
3777		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3778	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3779	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3780		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3781						S_000360_CUR2_LOCK(1));
3782		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3783			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3784			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3785			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3786		WREG32(R_000360_CUR2_OFFSET,
3787			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3788	}
3789}
3790
3791void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3792{
3793	/* Update base address for crtc */
3794	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3795	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3796		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3797	}
3798	/* Restore CRTC registers */
3799	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3800	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3801	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3802	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3803		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3804	}
3805}
3806
3807void r100_vga_render_disable(struct radeon_device *rdev)
3808{
3809	u32 tmp;
3810
3811	tmp = RREG8(R_0003C2_GENMO_WT);
3812	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3813}
3814
3815static void r100_debugfs(struct radeon_device *rdev)
3816{
3817	int r;
3818
3819	r = r100_debugfs_mc_info_init(rdev);
3820	if (r)
3821		dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3822}
3823
3824static void r100_mc_program(struct radeon_device *rdev)
3825{
3826	struct r100_mc_save save;
3827
3828	/* Stops all mc clients */
3829	r100_mc_stop(rdev, &save);
3830	if (rdev->flags & RADEON_IS_AGP) {
3831		WREG32(R_00014C_MC_AGP_LOCATION,
3832			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3833			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3834		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3835		if (rdev->family > CHIP_RV200)
3836			WREG32(R_00015C_AGP_BASE_2,
3837				upper_32_bits(rdev->mc.agp_base) & 0xff);
3838	} else {
3839		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3840		WREG32(R_000170_AGP_BASE, 0);
3841		if (rdev->family > CHIP_RV200)
3842			WREG32(R_00015C_AGP_BASE_2, 0);
3843	}
3844	/* Wait for mc idle */
3845	if (r100_mc_wait_for_idle(rdev))
3846		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3847	/* Program MC, should be a 32bits limited address space */
3848	WREG32(R_000148_MC_FB_LOCATION,
3849		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3850		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3851	r100_mc_resume(rdev, &save);
3852}
3853
3854void r100_clock_startup(struct radeon_device *rdev)
3855{
3856	u32 tmp;
3857
3858	if (radeon_dynclks != -1 && radeon_dynclks)
3859		radeon_legacy_set_clock_gating(rdev, 1);
3860	/* We need to force on some of the block */
3861	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3862	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3863	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3864		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3865	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3866}
3867
3868static int r100_startup(struct radeon_device *rdev)
3869{
3870	int r;
3871
3872	/* set common regs */
3873	r100_set_common_regs(rdev);
3874	/* program mc */
3875	r100_mc_program(rdev);
3876	/* Resume clock */
3877	r100_clock_startup(rdev);
3878	/* Initialize GART (initialize after TTM so we can allocate
3879	 * memory through TTM but finalize after TTM) */
3880	r100_enable_bm(rdev);
3881	if (rdev->flags & RADEON_IS_PCI) {
3882		r = r100_pci_gart_enable(rdev);
3883		if (r)
3884			return r;
3885	}
3886
3887	/* allocate wb buffer */
3888	r = radeon_wb_init(rdev);
3889	if (r)
3890		return r;
3891
3892	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3893	if (r) {
3894		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3895		return r;
3896	}
3897
3898	/* Enable IRQ */
 
 
 
 
 
 
3899	r100_irq_set(rdev);
3900	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3901	/* 1M ring buffer */
3902	r = r100_cp_init(rdev, 1024 * 1024);
3903	if (r) {
3904		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3905		return r;
3906	}
3907
3908	r = radeon_ib_pool_start(rdev);
3909	if (r)
3910		return r;
3911
3912	r = radeon_ib_ring_tests(rdev);
3913	if (r)
3914		return r;
 
3915
3916	return 0;
3917}
3918
3919int r100_resume(struct radeon_device *rdev)
3920{
3921	int r;
3922
3923	/* Make sur GART are not working */
3924	if (rdev->flags & RADEON_IS_PCI)
3925		r100_pci_gart_disable(rdev);
3926	/* Resume clock before doing reset */
3927	r100_clock_startup(rdev);
3928	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3929	if (radeon_asic_reset(rdev)) {
3930		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3931			RREG32(R_000E40_RBBM_STATUS),
3932			RREG32(R_0007C0_CP_STAT));
3933	}
3934	/* post */
3935	radeon_combios_asic_init(rdev->ddev);
3936	/* Resume clock after posting */
3937	r100_clock_startup(rdev);
3938	/* Initialize surface registers */
3939	radeon_surface_init(rdev);
3940
3941	rdev->accel_working = true;
3942	r = r100_startup(rdev);
3943	if (r) {
3944		rdev->accel_working = false;
3945	}
3946	return r;
3947}
3948
3949int r100_suspend(struct radeon_device *rdev)
3950{
3951	radeon_ib_pool_suspend(rdev);
3952	r100_cp_disable(rdev);
3953	radeon_wb_disable(rdev);
3954	r100_irq_disable(rdev);
3955	if (rdev->flags & RADEON_IS_PCI)
3956		r100_pci_gart_disable(rdev);
3957	return 0;
3958}
3959
3960void r100_fini(struct radeon_device *rdev)
3961{
 
3962	r100_cp_fini(rdev);
3963	radeon_wb_fini(rdev);
3964	r100_ib_fini(rdev);
3965	radeon_gem_fini(rdev);
3966	if (rdev->flags & RADEON_IS_PCI)
3967		r100_pci_gart_fini(rdev);
3968	radeon_agp_fini(rdev);
3969	radeon_irq_kms_fini(rdev);
3970	radeon_fence_driver_fini(rdev);
3971	radeon_bo_fini(rdev);
3972	radeon_atombios_fini(rdev);
3973	kfree(rdev->bios);
3974	rdev->bios = NULL;
3975}
3976
3977/*
3978 * Due to how kexec works, it can leave the hw fully initialised when it
3979 * boots the new kernel. However doing our init sequence with the CP and
3980 * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3981 * do some quick sanity checks and restore sane values to avoid this
3982 * problem.
3983 */
3984void r100_restore_sanity(struct radeon_device *rdev)
3985{
3986	u32 tmp;
3987
3988	tmp = RREG32(RADEON_CP_CSQ_CNTL);
3989	if (tmp) {
3990		WREG32(RADEON_CP_CSQ_CNTL, 0);
3991	}
3992	tmp = RREG32(RADEON_CP_RB_CNTL);
3993	if (tmp) {
3994		WREG32(RADEON_CP_RB_CNTL, 0);
3995	}
3996	tmp = RREG32(RADEON_SCRATCH_UMSK);
3997	if (tmp) {
3998		WREG32(RADEON_SCRATCH_UMSK, 0);
3999	}
4000}
4001
4002int r100_init(struct radeon_device *rdev)
4003{
4004	int r;
4005
4006	/* Register debugfs file specific to this group of asics */
4007	r100_debugfs(rdev);
4008	/* Disable VGA */
4009	r100_vga_render_disable(rdev);
4010	/* Initialize scratch registers */
4011	radeon_scratch_init(rdev);
4012	/* Initialize surface registers */
4013	radeon_surface_init(rdev);
4014	/* sanity check some register to avoid hangs like after kexec */
4015	r100_restore_sanity(rdev);
4016	/* TODO: disable VGA need to use VGA request */
4017	/* BIOS*/
4018	if (!radeon_get_bios(rdev)) {
4019		if (ASIC_IS_AVIVO(rdev))
4020			return -EINVAL;
4021	}
4022	if (rdev->is_atom_bios) {
4023		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4024		return -EINVAL;
4025	} else {
4026		r = radeon_combios_init(rdev);
4027		if (r)
4028			return r;
4029	}
4030	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4031	if (radeon_asic_reset(rdev)) {
4032		dev_warn(rdev->dev,
4033			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4034			RREG32(R_000E40_RBBM_STATUS),
4035			RREG32(R_0007C0_CP_STAT));
4036	}
4037	/* check if cards are posted or not */
4038	if (radeon_boot_test_post_card(rdev) == false)
4039		return -EINVAL;
4040	/* Set asic errata */
4041	r100_errata(rdev);
4042	/* Initialize clocks */
4043	radeon_get_clock_info(rdev->ddev);
4044	/* initialize AGP */
4045	if (rdev->flags & RADEON_IS_AGP) {
4046		r = radeon_agp_init(rdev);
4047		if (r) {
4048			radeon_agp_disable(rdev);
4049		}
4050	}
4051	/* initialize VRAM */
4052	r100_mc_init(rdev);
4053	/* Fence driver */
4054	r = radeon_fence_driver_init(rdev);
4055	if (r)
4056		return r;
4057	r = radeon_irq_kms_init(rdev);
4058	if (r)
4059		return r;
4060	/* Memory manager */
4061	r = radeon_bo_init(rdev);
4062	if (r)
4063		return r;
4064	if (rdev->flags & RADEON_IS_PCI) {
4065		r = r100_pci_gart_init(rdev);
4066		if (r)
4067			return r;
4068	}
4069	r100_set_safe_registers(rdev);
4070
4071	r = radeon_ib_pool_init(rdev);
 
 
4072	rdev->accel_working = true;
4073	if (r) {
4074		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
4075		rdev->accel_working = false;
4076	}
4077
4078	r = r100_startup(rdev);
4079	if (r) {
4080		/* Somethings want wront with the accel init stop accel */
4081		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4082		r100_cp_fini(rdev);
4083		radeon_wb_fini(rdev);
4084		r100_ib_fini(rdev);
4085		radeon_irq_kms_fini(rdev);
4086		if (rdev->flags & RADEON_IS_PCI)
4087			r100_pci_gart_fini(rdev);
4088		rdev->accel_working = false;
4089	}
4090	return 0;
4091}
4092
4093uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg)
4094{
4095	if (reg < rdev->rmmio_size)
4096		return readl(((void __iomem *)rdev->rmmio) + reg);
4097	else {
4098		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4099		return readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4100	}
 
 
4101}
4102
4103void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4104{
4105	if (reg < rdev->rmmio_size)
4106		writel(v, ((void __iomem *)rdev->rmmio) + reg);
4107	else {
4108		writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4109		writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4110	}
4111}
4112
4113u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4114{
4115	if (reg < rdev->rio_mem_size)
4116		return ioread32(rdev->rio_mem + reg);
4117	else {
4118		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4119		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4120	}
4121}
4122
4123void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4124{
4125	if (reg < rdev->rio_mem_size)
4126		iowrite32(v, rdev->rio_mem + reg);
4127	else {
4128		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4129		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4130	}
4131}