Linux Audio

Check our new training course

Loading...
v6.13.7
   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
  29#include <linux/debugfs.h>
  30#include <linux/firmware.h>
  31#include <linux/module.h>
  32#include <linux/pci.h>
  33#include <linux/seq_file.h>
  34#include <linux/slab.h>
  35
 
  36#include <drm/drm_device.h>
  37#include <drm/drm_file.h>
  38#include <drm/drm_fourcc.h>
  39#include <drm/drm_framebuffer.h>
  40#include <drm/drm_vblank.h>
  41#include <drm/radeon_drm.h>
  42
  43#include "atom.h"
  44#include "r100_reg_safe.h"
  45#include "r100d.h"
  46#include "radeon.h"
  47#include "radeon_asic.h"
  48#include "radeon_reg.h"
  49#include "rn50_reg_safe.h"
  50#include "rs100d.h"
  51#include "rv200d.h"
  52#include "rv250d.h"
  53
  54/* Firmware Names */
  55#define FIRMWARE_R100		"radeon/R100_cp.bin"
  56#define FIRMWARE_R200		"radeon/R200_cp.bin"
  57#define FIRMWARE_R300		"radeon/R300_cp.bin"
  58#define FIRMWARE_R420		"radeon/R420_cp.bin"
  59#define FIRMWARE_RS690		"radeon/RS690_cp.bin"
  60#define FIRMWARE_RS600		"radeon/RS600_cp.bin"
  61#define FIRMWARE_R520		"radeon/R520_cp.bin"
  62
  63MODULE_FIRMWARE(FIRMWARE_R100);
  64MODULE_FIRMWARE(FIRMWARE_R200);
  65MODULE_FIRMWARE(FIRMWARE_R300);
  66MODULE_FIRMWARE(FIRMWARE_R420);
  67MODULE_FIRMWARE(FIRMWARE_RS690);
  68MODULE_FIRMWARE(FIRMWARE_RS600);
  69MODULE_FIRMWARE(FIRMWARE_R520);
  70
  71#include "r100_track.h"
  72
  73/* This files gather functions specifics to:
  74 * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
  75 * and others in some cases.
  76 */
  77
  78static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
  79{
  80	if (crtc == 0) {
  81		if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
  82			return true;
  83		else
  84			return false;
  85	} else {
  86		if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
  87			return true;
  88		else
  89			return false;
  90	}
  91}
  92
  93static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
  94{
  95	u32 vline1, vline2;
  96
  97	if (crtc == 0) {
  98		vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
  99		vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
 100	} else {
 101		vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
 102		vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
 103	}
 104	if (vline1 != vline2)
 105		return true;
 106	else
 107		return false;
 108}
 109
 110/**
 111 * r100_wait_for_vblank - vblank wait asic callback.
 112 *
 113 * @rdev: radeon_device pointer
 114 * @crtc: crtc to wait for vblank on
 115 *
 116 * Wait for vblank on the requested crtc (r1xx-r4xx).
 117 */
 118void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
 119{
 120	unsigned i = 0;
 121
 122	if (crtc >= rdev->num_crtc)
 123		return;
 124
 125	if (crtc == 0) {
 126		if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
 127			return;
 128	} else {
 129		if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
 130			return;
 131	}
 132
 133	/* depending on when we hit vblank, we may be close to active; if so,
 134	 * wait for another frame.
 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	while (!r100_is_in_vblank(rdev, crtc)) {
 144		if (i++ % 100 == 0) {
 145			if (!r100_is_counter_moving(rdev, crtc))
 146				break;
 147		}
 148	}
 149}
 150
 151/**
 152 * r100_page_flip - pageflip callback.
 153 *
 154 * @rdev: radeon_device pointer
 155 * @crtc_id: crtc to cleanup pageflip on
 156 * @crtc_base: new address of the crtc (GPU MC address)
 157 * @async: asynchronous flip
 158 *
 159 * Does the actual pageflip (r1xx-r4xx).
 160 * During vblank we take the crtc lock and wait for the update_pending
 161 * bit to go high, when it does, we release the lock, and allow the
 162 * double buffered update to take place.
 163 */
 164void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base, bool async)
 165{
 166	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
 167	uint32_t crtc_pitch, pitch_pixels;
 168	struct drm_framebuffer *fb = radeon_crtc->base.primary->fb;
 169	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
 170	int i;
 171
 172	/* Lock the graphics update lock */
 173	/* update the scanout addresses */
 174	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 175
 176	/* update pitch */
 177	pitch_pixels = fb->pitches[0] / fb->format->cpp[0];
 178	crtc_pitch = DIV_ROUND_UP(pitch_pixels * fb->format->cpp[0] * 8,
 179				  fb->format->cpp[0] * 8 * 8);
 180	crtc_pitch |= crtc_pitch << 16;
 181	WREG32(RADEON_CRTC_PITCH + radeon_crtc->crtc_offset, crtc_pitch);
 182
 183	/* Wait for update_pending to go high. */
 184	for (i = 0; i < rdev->usec_timeout; i++) {
 185		if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
 186			break;
 187		udelay(1);
 188	}
 189	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
 190
 191	/* Unlock the lock, so double-buffering can take place inside vblank */
 192	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
 193	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 194
 195}
 196
 197/**
 198 * r100_page_flip_pending - check if page flip is still pending
 199 *
 200 * @rdev: radeon_device pointer
 201 * @crtc_id: crtc to check
 202 *
 203 * Check if the last pagefilp is still pending (r1xx-r4xx).
 204 * Returns the current update pending status.
 205 */
 206bool r100_page_flip_pending(struct radeon_device *rdev, int crtc_id)
 207{
 208	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
 209
 210	/* Return current update_pending status: */
 211	return !!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) &
 212		RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET);
 213}
 214
 215/**
 216 * r100_pm_get_dynpm_state - look up dynpm power state callback.
 217 *
 218 * @rdev: radeon_device pointer
 219 *
 220 * Look up the optimal power state based on the
 221 * current state of the GPU (r1xx-r5xx).
 222 * Used for dynpm only.
 223 */
 224void r100_pm_get_dynpm_state(struct radeon_device *rdev)
 225{
 226	int i;
 227	rdev->pm.dynpm_can_upclock = true;
 228	rdev->pm.dynpm_can_downclock = true;
 229
 230	switch (rdev->pm.dynpm_planned_action) {
 231	case DYNPM_ACTION_MINIMUM:
 232		rdev->pm.requested_power_state_index = 0;
 233		rdev->pm.dynpm_can_downclock = false;
 234		break;
 235	case DYNPM_ACTION_DOWNCLOCK:
 236		if (rdev->pm.current_power_state_index == 0) {
 237			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 238			rdev->pm.dynpm_can_downclock = false;
 239		} else {
 240			if (rdev->pm.active_crtc_count > 1) {
 241				for (i = 0; i < rdev->pm.num_power_states; i++) {
 242					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 243						continue;
 244					else if (i >= rdev->pm.current_power_state_index) {
 245						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 246						break;
 247					} else {
 248						rdev->pm.requested_power_state_index = i;
 249						break;
 250					}
 251				}
 252			} else
 253				rdev->pm.requested_power_state_index =
 254					rdev->pm.current_power_state_index - 1;
 255		}
 256		/* don't use the power state if crtcs are active and no display flag is set */
 257		if ((rdev->pm.active_crtc_count > 0) &&
 258		    (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
 259		     RADEON_PM_MODE_NO_DISPLAY)) {
 260			rdev->pm.requested_power_state_index++;
 261		}
 262		break;
 263	case DYNPM_ACTION_UPCLOCK:
 264		if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
 265			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 266			rdev->pm.dynpm_can_upclock = false;
 267		} else {
 268			if (rdev->pm.active_crtc_count > 1) {
 269				for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
 270					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 271						continue;
 272					else if (i <= rdev->pm.current_power_state_index) {
 273						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 274						break;
 275					} else {
 276						rdev->pm.requested_power_state_index = i;
 277						break;
 278					}
 279				}
 280			} else
 281				rdev->pm.requested_power_state_index =
 282					rdev->pm.current_power_state_index + 1;
 283		}
 284		break;
 285	case DYNPM_ACTION_DEFAULT:
 286		rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
 287		rdev->pm.dynpm_can_upclock = false;
 288		break;
 289	case DYNPM_ACTION_NONE:
 290	default:
 291		DRM_ERROR("Requested mode for not defined action\n");
 292		return;
 293	}
 294	/* only one clock mode per power state */
 295	rdev->pm.requested_clock_mode_index = 0;
 296
 297	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
 298		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 299		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
 300		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 301		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
 302		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 303		  pcie_lanes);
 304}
 305
 306/**
 307 * r100_pm_init_profile - Initialize power profiles callback.
 308 *
 309 * @rdev: radeon_device pointer
 310 *
 311 * Initialize the power states used in profile mode
 312 * (r1xx-r3xx).
 313 * Used for profile mode only.
 314 */
 315void r100_pm_init_profile(struct radeon_device *rdev)
 316{
 317	/* default */
 318	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
 319	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 320	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
 321	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
 322	/* low sh */
 323	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
 324	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
 325	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
 326	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
 327	/* mid sh */
 328	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
 329	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
 330	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
 331	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
 332	/* high sh */
 333	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
 334	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 335	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
 336	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
 337	/* low mh */
 338	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
 339	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 340	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
 341	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
 342	/* mid mh */
 343	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
 344	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 345	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
 346	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
 347	/* high mh */
 348	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
 349	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 350	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
 351	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
 352}
 353
 354/**
 355 * r100_pm_misc - set additional pm hw parameters callback.
 356 *
 357 * @rdev: radeon_device pointer
 358 *
 359 * Set non-clock parameters associated with a power state
 360 * (voltage, pcie lanes, etc.) (r1xx-r4xx).
 361 */
 362void r100_pm_misc(struct radeon_device *rdev)
 363{
 364	int requested_index = rdev->pm.requested_power_state_index;
 365	struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
 366	struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
 367	u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
 368
 369	if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
 370		if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
 371			tmp = RREG32(voltage->gpio.reg);
 372			if (voltage->active_high)
 373				tmp |= voltage->gpio.mask;
 374			else
 375				tmp &= ~(voltage->gpio.mask);
 376			WREG32(voltage->gpio.reg, tmp);
 377			if (voltage->delay)
 378				udelay(voltage->delay);
 379		} else {
 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		}
 389	}
 390
 391	sclk_cntl = RREG32_PLL(SCLK_CNTL);
 392	sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
 393	sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
 394	sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
 395	sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
 396	if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
 397		sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
 398		if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
 399			sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
 400		else
 401			sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
 402		if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
 403			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
 404		else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
 405			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
 406	} else
 407		sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
 408
 409	if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
 410		sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
 411		if (voltage->delay) {
 412			sclk_more_cntl |= VOLTAGE_DROP_SYNC;
 413			switch (voltage->delay) {
 414			case 33:
 415				sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
 416				break;
 417			case 66:
 418				sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
 419				break;
 420			case 99:
 421				sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
 422				break;
 423			case 132:
 424				sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
 425				break;
 426			}
 427		} else
 428			sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
 429	} else
 430		sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
 431
 432	if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
 433		sclk_cntl &= ~FORCE_HDP;
 434	else
 435		sclk_cntl |= FORCE_HDP;
 436
 437	WREG32_PLL(SCLK_CNTL, sclk_cntl);
 438	WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
 439	WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
 440
 441	/* set pcie lanes */
 442	if ((rdev->flags & RADEON_IS_PCIE) &&
 443	    !(rdev->flags & RADEON_IS_IGP) &&
 444	    rdev->asic->pm.set_pcie_lanes &&
 445	    (ps->pcie_lanes !=
 446	     rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
 447		radeon_set_pcie_lanes(rdev,
 448				      ps->pcie_lanes);
 449		DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
 450	}
 451}
 452
 453/**
 454 * r100_pm_prepare - pre-power state change callback.
 455 *
 456 * @rdev: radeon_device pointer
 457 *
 458 * Prepare for a power state change (r1xx-r4xx).
 459 */
 460void r100_pm_prepare(struct radeon_device *rdev)
 461{
 462	struct drm_device *ddev = rdev_to_drm(rdev);
 463	struct drm_crtc *crtc;
 464	struct radeon_crtc *radeon_crtc;
 465	u32 tmp;
 466
 467	/* disable any active CRTCs */
 468	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 469		radeon_crtc = to_radeon_crtc(crtc);
 470		if (radeon_crtc->enabled) {
 471			if (radeon_crtc->crtc_id) {
 472				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 473				tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
 474				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 475			} else {
 476				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 477				tmp |= RADEON_CRTC_DISP_REQ_EN_B;
 478				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 479			}
 480		}
 481	}
 482}
 483
 484/**
 485 * r100_pm_finish - post-power state change callback.
 486 *
 487 * @rdev: radeon_device pointer
 488 *
 489 * Clean up after a power state change (r1xx-r4xx).
 490 */
 491void r100_pm_finish(struct radeon_device *rdev)
 492{
 493	struct drm_device *ddev = rdev_to_drm(rdev);
 494	struct drm_crtc *crtc;
 495	struct radeon_crtc *radeon_crtc;
 496	u32 tmp;
 497
 498	/* enable any active CRTCs */
 499	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 500		radeon_crtc = to_radeon_crtc(crtc);
 501		if (radeon_crtc->enabled) {
 502			if (radeon_crtc->crtc_id) {
 503				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 504				tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
 505				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 506			} else {
 507				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 508				tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
 509				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 510			}
 511		}
 512	}
 513}
 514
 515/**
 516 * r100_gui_idle - gui idle callback.
 517 *
 518 * @rdev: radeon_device pointer
 519 *
 520 * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
 521 * Returns true if idle, false if not.
 522 */
 523bool r100_gui_idle(struct radeon_device *rdev)
 524{
 525	if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
 526		return false;
 527	else
 528		return true;
 529}
 530
 531/* hpd for digital panel detect/disconnect */
 532/**
 533 * r100_hpd_sense - hpd sense callback.
 534 *
 535 * @rdev: radeon_device pointer
 536 * @hpd: hpd (hotplug detect) pin
 537 *
 538 * Checks if a digital monitor is connected (r1xx-r4xx).
 539 * Returns true if connected, false if not connected.
 540 */
 541bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
 542{
 543	bool connected = false;
 544
 545	switch (hpd) {
 546	case RADEON_HPD_1:
 547		if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
 548			connected = true;
 549		break;
 550	case RADEON_HPD_2:
 551		if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
 552			connected = true;
 553		break;
 554	default:
 555		break;
 556	}
 557	return connected;
 558}
 559
 560/**
 561 * r100_hpd_set_polarity - hpd set polarity callback.
 562 *
 563 * @rdev: radeon_device pointer
 564 * @hpd: hpd (hotplug detect) pin
 565 *
 566 * Set the polarity of the hpd pin (r1xx-r4xx).
 567 */
 568void r100_hpd_set_polarity(struct radeon_device *rdev,
 569			   enum radeon_hpd_id hpd)
 570{
 571	u32 tmp;
 572	bool connected = r100_hpd_sense(rdev, hpd);
 573
 574	switch (hpd) {
 575	case RADEON_HPD_1:
 576		tmp = RREG32(RADEON_FP_GEN_CNTL);
 577		if (connected)
 578			tmp &= ~RADEON_FP_DETECT_INT_POL;
 579		else
 580			tmp |= RADEON_FP_DETECT_INT_POL;
 581		WREG32(RADEON_FP_GEN_CNTL, tmp);
 582		break;
 583	case RADEON_HPD_2:
 584		tmp = RREG32(RADEON_FP2_GEN_CNTL);
 585		if (connected)
 586			tmp &= ~RADEON_FP2_DETECT_INT_POL;
 587		else
 588			tmp |= RADEON_FP2_DETECT_INT_POL;
 589		WREG32(RADEON_FP2_GEN_CNTL, tmp);
 590		break;
 591	default:
 592		break;
 593	}
 594}
 595
 596/**
 597 * r100_hpd_init - hpd setup callback.
 598 *
 599 * @rdev: radeon_device pointer
 600 *
 601 * Setup the hpd pins used by the card (r1xx-r4xx).
 602 * Set the polarity, and enable the hpd interrupts.
 603 */
 604void r100_hpd_init(struct radeon_device *rdev)
 605{
 606	struct drm_device *dev = rdev_to_drm(rdev);
 607	struct drm_connector *connector;
 608	unsigned enable = 0;
 609
 610	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 611		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 612		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
 613			enable |= 1 << radeon_connector->hpd.hpd;
 614		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
 615	}
 616	radeon_irq_kms_enable_hpd(rdev, enable);
 617}
 618
 619/**
 620 * r100_hpd_fini - hpd tear down callback.
 621 *
 622 * @rdev: radeon_device pointer
 623 *
 624 * Tear down the hpd pins used by the card (r1xx-r4xx).
 625 * Disable the hpd interrupts.
 626 */
 627void r100_hpd_fini(struct radeon_device *rdev)
 628{
 629	struct drm_device *dev = rdev_to_drm(rdev);
 630	struct drm_connector *connector;
 631	unsigned disable = 0;
 632
 633	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 634		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 635		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
 636			disable |= 1 << radeon_connector->hpd.hpd;
 637	}
 638	radeon_irq_kms_disable_hpd(rdev, disable);
 639}
 640
 641/*
 642 * PCI GART
 643 */
 644void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
 645{
 646	/* TODO: can we do somethings here ? */
 647	/* It seems hw only cache one entry so we should discard this
 648	 * entry otherwise if first GPU GART read hit this entry it
 649	 * could end up in wrong address. */
 650}
 651
 652int r100_pci_gart_init(struct radeon_device *rdev)
 653{
 654	int r;
 655
 656	if (rdev->gart.ptr) {
 657		WARN(1, "R100 PCI GART already initialized\n");
 658		return 0;
 659	}
 660	/* Initialize common gart structure */
 661	r = radeon_gart_init(rdev);
 662	if (r)
 663		return r;
 664	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 665	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
 666	rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
 667	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
 668	return radeon_gart_table_ram_alloc(rdev);
 669}
 670
 671int r100_pci_gart_enable(struct radeon_device *rdev)
 672{
 673	uint32_t tmp;
 674
 675	/* discard memory request outside of configured range */
 676	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 677	WREG32(RADEON_AIC_CNTL, tmp);
 678	/* set address range for PCI address translate */
 679	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
 680	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
 681	/* set PCI GART page-table base address */
 682	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
 683	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
 684	WREG32(RADEON_AIC_CNTL, tmp);
 685	r100_pci_gart_tlb_flush(rdev);
 686	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
 687		 (unsigned)(rdev->mc.gtt_size >> 20),
 688		 (unsigned long long)rdev->gart.table_addr);
 689	rdev->gart.ready = true;
 690	return 0;
 691}
 692
 693void r100_pci_gart_disable(struct radeon_device *rdev)
 694{
 695	uint32_t tmp;
 696
 697	/* discard memory request outside of configured range */
 698	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 699	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
 700	WREG32(RADEON_AIC_LO_ADDR, 0);
 701	WREG32(RADEON_AIC_HI_ADDR, 0);
 702}
 703
 704uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
 705{
 706	return addr;
 707}
 708
 709void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
 710			    uint64_t entry)
 711{
 712	u32 *gtt = rdev->gart.ptr;
 713	gtt[i] = cpu_to_le32(lower_32_bits(entry));
 714}
 715
 716void r100_pci_gart_fini(struct radeon_device *rdev)
 717{
 718	radeon_gart_fini(rdev);
 719	r100_pci_gart_disable(rdev);
 720	radeon_gart_table_ram_free(rdev);
 721}
 722
 723int r100_irq_set(struct radeon_device *rdev)
 724{
 725	uint32_t tmp = 0;
 726
 727	if (!rdev->irq.installed) {
 728		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
 729		WREG32(R_000040_GEN_INT_CNTL, 0);
 730		return -EINVAL;
 731	}
 732	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
 733		tmp |= RADEON_SW_INT_ENABLE;
 734	}
 735	if (rdev->irq.crtc_vblank_int[0] ||
 736	    atomic_read(&rdev->irq.pflip[0])) {
 737		tmp |= RADEON_CRTC_VBLANK_MASK;
 738	}
 739	if (rdev->irq.crtc_vblank_int[1] ||
 740	    atomic_read(&rdev->irq.pflip[1])) {
 741		tmp |= RADEON_CRTC2_VBLANK_MASK;
 742	}
 743	if (rdev->irq.hpd[0]) {
 744		tmp |= RADEON_FP_DETECT_MASK;
 745	}
 746	if (rdev->irq.hpd[1]) {
 747		tmp |= RADEON_FP2_DETECT_MASK;
 748	}
 749	WREG32(RADEON_GEN_INT_CNTL, tmp);
 750
 751	/* read back to post the write */
 752	RREG32(RADEON_GEN_INT_CNTL);
 753
 754	return 0;
 755}
 756
 757void r100_irq_disable(struct radeon_device *rdev)
 758{
 759	u32 tmp;
 760
 761	WREG32(R_000040_GEN_INT_CNTL, 0);
 762	/* Wait and acknowledge irq */
 763	mdelay(1);
 764	tmp = RREG32(R_000044_GEN_INT_STATUS);
 765	WREG32(R_000044_GEN_INT_STATUS, tmp);
 766}
 767
 768static uint32_t r100_irq_ack(struct radeon_device *rdev)
 769{
 770	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
 771	uint32_t irq_mask = RADEON_SW_INT_TEST |
 772		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
 773		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
 774
 775	if (irqs) {
 776		WREG32(RADEON_GEN_INT_STATUS, irqs);
 777	}
 778	return irqs & irq_mask;
 779}
 780
 781int r100_irq_process(struct radeon_device *rdev)
 782{
 783	uint32_t status, msi_rearm;
 784	bool queue_hotplug = false;
 785
 786	status = r100_irq_ack(rdev);
 787	if (!status) {
 788		return IRQ_NONE;
 789	}
 790	if (rdev->shutdown) {
 791		return IRQ_NONE;
 792	}
 793	while (status) {
 794		/* SW interrupt */
 795		if (status & RADEON_SW_INT_TEST) {
 796			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
 797		}
 798		/* Vertical blank interrupts */
 799		if (status & RADEON_CRTC_VBLANK_STAT) {
 800			if (rdev->irq.crtc_vblank_int[0]) {
 801				drm_handle_vblank(rdev_to_drm(rdev), 0);
 802				rdev->pm.vblank_sync = true;
 803				wake_up(&rdev->irq.vblank_queue);
 804			}
 805			if (atomic_read(&rdev->irq.pflip[0]))
 806				radeon_crtc_handle_vblank(rdev, 0);
 807		}
 808		if (status & RADEON_CRTC2_VBLANK_STAT) {
 809			if (rdev->irq.crtc_vblank_int[1]) {
 810				drm_handle_vblank(rdev_to_drm(rdev), 1);
 811				rdev->pm.vblank_sync = true;
 812				wake_up(&rdev->irq.vblank_queue);
 813			}
 814			if (atomic_read(&rdev->irq.pflip[1]))
 815				radeon_crtc_handle_vblank(rdev, 1);
 816		}
 817		if (status & RADEON_FP_DETECT_STAT) {
 818			queue_hotplug = true;
 819			DRM_DEBUG("HPD1\n");
 820		}
 821		if (status & RADEON_FP2_DETECT_STAT) {
 822			queue_hotplug = true;
 823			DRM_DEBUG("HPD2\n");
 824		}
 825		status = r100_irq_ack(rdev);
 826	}
 827	if (queue_hotplug)
 828		schedule_delayed_work(&rdev->hotplug_work, 0);
 829	if (rdev->msi_enabled) {
 830		switch (rdev->family) {
 831		case CHIP_RS400:
 832		case CHIP_RS480:
 833			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
 834			WREG32(RADEON_AIC_CNTL, msi_rearm);
 835			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
 836			break;
 837		default:
 838			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
 839			break;
 840		}
 841	}
 842	return IRQ_HANDLED;
 843}
 844
 845u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
 846{
 847	if (crtc == 0)
 848		return RREG32(RADEON_CRTC_CRNT_FRAME);
 849	else
 850		return RREG32(RADEON_CRTC2_CRNT_FRAME);
 851}
 852
 853/**
 854 * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
 855 * @rdev: radeon device structure
 856 * @ring: ring buffer struct for emitting packets
 857 */
 858static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
 859{
 860	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 861	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
 862				RADEON_HDP_READ_BUFFER_INVALIDATE);
 863	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 864	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
 865}
 866
 867/* Who ever call radeon_fence_emit should call ring_lock and ask
 868 * for enough space (today caller are ib schedule and buffer move) */
 869void r100_fence_ring_emit(struct radeon_device *rdev,
 870			  struct radeon_fence *fence)
 871{
 872	struct radeon_ring *ring = &rdev->ring[fence->ring];
 873
 874	/* We have to make sure that caches are flushed before
 875	 * CPU might read something from VRAM. */
 876	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
 877	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
 878	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
 879	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
 880	/* Wait until IDLE & CLEAN */
 881	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 882	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
 883	r100_ring_hdp_flush(rdev, ring);
 884	/* Emit fence sequence & fire IRQ */
 885	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
 886	radeon_ring_write(ring, fence->seq);
 887	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
 888	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
 889}
 890
 891bool r100_semaphore_ring_emit(struct radeon_device *rdev,
 892			      struct radeon_ring *ring,
 893			      struct radeon_semaphore *semaphore,
 894			      bool emit_wait)
 895{
 896	/* Unused on older asics, since we don't have semaphores or multiple rings */
 897	BUG();
 898	return false;
 899}
 900
 901struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
 902				    uint64_t src_offset,
 903				    uint64_t dst_offset,
 904				    unsigned num_gpu_pages,
 905				    struct dma_resv *resv)
 906{
 907	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 908	struct radeon_fence *fence;
 909	uint32_t cur_pages;
 910	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
 911	uint32_t pitch;
 912	uint32_t stride_pixels;
 913	unsigned ndw;
 914	int num_loops;
 915	int r = 0;
 916
 917	/* radeon limited to 16k stride */
 918	stride_bytes &= 0x3fff;
 919	/* radeon pitch is /64 */
 920	pitch = stride_bytes / 64;
 921	stride_pixels = stride_bytes / 4;
 922	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
 923
 924	/* Ask for enough room for blit + flush + fence */
 925	ndw = 64 + (10 * num_loops);
 926	r = radeon_ring_lock(rdev, ring, ndw);
 927	if (r) {
 928		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 929		return ERR_PTR(-EINVAL);
 930	}
 931	while (num_gpu_pages > 0) {
 932		cur_pages = num_gpu_pages;
 933		if (cur_pages > 8191) {
 934			cur_pages = 8191;
 935		}
 936		num_gpu_pages -= cur_pages;
 937
 938		/* pages are in Y direction - height
 939		   page width in X direction - width */
 940		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
 941		radeon_ring_write(ring,
 942				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 943				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 944				  RADEON_GMC_SRC_CLIPPING |
 945				  RADEON_GMC_DST_CLIPPING |
 946				  RADEON_GMC_BRUSH_NONE |
 947				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 948				  RADEON_GMC_SRC_DATATYPE_COLOR |
 949				  RADEON_ROP3_S |
 950				  RADEON_DP_SRC_SOURCE_MEMORY |
 951				  RADEON_GMC_CLR_CMP_CNTL_DIS |
 952				  RADEON_GMC_WR_MSK_DIS);
 953		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
 954		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
 955		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 956		radeon_ring_write(ring, 0);
 957		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 958		radeon_ring_write(ring, num_gpu_pages);
 959		radeon_ring_write(ring, num_gpu_pages);
 960		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
 961	}
 962	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 963	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
 964	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 965	radeon_ring_write(ring,
 966			  RADEON_WAIT_2D_IDLECLEAN |
 967			  RADEON_WAIT_HOST_IDLECLEAN |
 968			  RADEON_WAIT_DMA_GUI_IDLE);
 969	r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
 970	if (r) {
 971		radeon_ring_unlock_undo(rdev, ring);
 972		return ERR_PTR(r);
 973	}
 974	radeon_ring_unlock_commit(rdev, ring, false);
 975	return fence;
 976}
 977
 978static int r100_cp_wait_for_idle(struct radeon_device *rdev)
 979{
 980	unsigned i;
 981	u32 tmp;
 982
 983	for (i = 0; i < rdev->usec_timeout; i++) {
 984		tmp = RREG32(R_000E40_RBBM_STATUS);
 985		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
 986			return 0;
 987		}
 988		udelay(1);
 989	}
 990	return -1;
 991}
 992
 993void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
 994{
 995	int r;
 996
 997	r = radeon_ring_lock(rdev, ring, 2);
 998	if (r) {
 999		return;
1000	}
1001	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
1002	radeon_ring_write(ring,
1003			  RADEON_ISYNC_ANY2D_IDLE3D |
1004			  RADEON_ISYNC_ANY3D_IDLE2D |
1005			  RADEON_ISYNC_WAIT_IDLEGUI |
1006			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
1007	radeon_ring_unlock_commit(rdev, ring, false);
1008}
1009
1010
1011/* Load the microcode for the CP */
1012static int r100_cp_init_microcode(struct radeon_device *rdev)
1013{
1014	const char *fw_name = NULL;
1015	int err;
1016
1017	DRM_DEBUG_KMS("\n");
1018
1019	switch (rdev->family) {
1020	case CHIP_R100:
1021	case CHIP_RV100:
1022	case CHIP_RV200:
1023	case CHIP_RS100:
1024	case CHIP_RS200:
1025		DRM_INFO("Loading R100 Microcode\n");
1026		fw_name = FIRMWARE_R100;
1027		break;
1028
1029	case CHIP_R200:
1030	case CHIP_RV250:
1031	case CHIP_RV280:
1032	case CHIP_RS300:
1033		DRM_INFO("Loading R200 Microcode\n");
1034		fw_name = FIRMWARE_R200;
1035		break;
1036
1037	case CHIP_R300:
1038	case CHIP_R350:
1039	case CHIP_RV350:
1040	case CHIP_RV380:
1041	case CHIP_RS400:
1042	case CHIP_RS480:
1043		DRM_INFO("Loading R300 Microcode\n");
1044		fw_name = FIRMWARE_R300;
1045		break;
1046
1047	case CHIP_R420:
1048	case CHIP_R423:
1049	case CHIP_RV410:
1050		DRM_INFO("Loading R400 Microcode\n");
1051		fw_name = FIRMWARE_R420;
1052		break;
1053
1054	case CHIP_RS690:
1055	case CHIP_RS740:
1056		DRM_INFO("Loading RS690/RS740 Microcode\n");
1057		fw_name = FIRMWARE_RS690;
1058		break;
1059
1060	case CHIP_RS600:
1061		DRM_INFO("Loading RS600 Microcode\n");
1062		fw_name = FIRMWARE_RS600;
1063		break;
1064
1065	case CHIP_RV515:
1066	case CHIP_R520:
1067	case CHIP_RV530:
1068	case CHIP_R580:
1069	case CHIP_RV560:
1070	case CHIP_RV570:
1071		DRM_INFO("Loading R500 Microcode\n");
1072		fw_name = FIRMWARE_R520;
1073		break;
1074
1075	default:
1076		DRM_ERROR("Unsupported Radeon family %u\n", rdev->family);
1077		return -EINVAL;
1078	}
1079
1080	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1081	if (err) {
1082		pr_err("radeon_cp: Failed to load firmware \"%s\"\n", fw_name);
1083	} else if (rdev->me_fw->size % 8) {
1084		pr_err("radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1085		       rdev->me_fw->size, fw_name);
1086		err = -EINVAL;
1087		release_firmware(rdev->me_fw);
1088		rdev->me_fw = NULL;
1089	}
1090	return err;
1091}
1092
1093u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1094		      struct radeon_ring *ring)
1095{
1096	u32 rptr;
1097
1098	if (rdev->wb.enabled)
1099		rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1100	else
1101		rptr = RREG32(RADEON_CP_RB_RPTR);
1102
1103	return rptr;
1104}
1105
1106u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1107		      struct radeon_ring *ring)
1108{
1109	return RREG32(RADEON_CP_RB_WPTR);
1110}
1111
1112void r100_gfx_set_wptr(struct radeon_device *rdev,
1113		       struct radeon_ring *ring)
1114{
1115	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1116	(void)RREG32(RADEON_CP_RB_WPTR);
1117}
1118
1119static void r100_cp_load_microcode(struct radeon_device *rdev)
1120{
1121	const __be32 *fw_data;
1122	int i, size;
1123
1124	if (r100_gui_wait_for_idle(rdev)) {
1125		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1126	}
1127
1128	if (rdev->me_fw) {
1129		size = rdev->me_fw->size / 4;
1130		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1131		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1132		for (i = 0; i < size; i += 2) {
1133			WREG32(RADEON_CP_ME_RAM_DATAH,
1134			       be32_to_cpup(&fw_data[i]));
1135			WREG32(RADEON_CP_ME_RAM_DATAL,
1136			       be32_to_cpup(&fw_data[i + 1]));
1137		}
1138	}
1139}
1140
1141int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1142{
1143	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1144	unsigned rb_bufsz;
1145	unsigned rb_blksz;
1146	unsigned max_fetch;
1147	unsigned pre_write_timer;
1148	unsigned pre_write_limit;
1149	unsigned indirect2_start;
1150	unsigned indirect1_start;
1151	uint32_t tmp;
1152	int r;
1153
1154	r100_debugfs_cp_init(rdev);
 
 
1155	if (!rdev->me_fw) {
1156		r = r100_cp_init_microcode(rdev);
1157		if (r) {
1158			DRM_ERROR("Failed to load firmware!\n");
1159			return r;
1160		}
1161	}
1162
1163	/* Align ring size */
1164	rb_bufsz = order_base_2(ring_size / 8);
1165	ring_size = (1 << (rb_bufsz + 1)) * 4;
1166	r100_cp_load_microcode(rdev);
1167	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1168			     RADEON_CP_PACKET2);
1169	if (r) {
1170		return r;
1171	}
1172	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1173	 * the rptr copy in system ram */
1174	rb_blksz = 9;
1175	/* cp will read 128bytes at a time (4 dwords) */
1176	max_fetch = 1;
1177	ring->align_mask = 16 - 1;
1178	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1179	pre_write_timer = 64;
1180	/* Force CP_RB_WPTR write if written more than one time before the
1181	 * delay expire
1182	 */
1183	pre_write_limit = 0;
1184	/* Setup the cp cache like this (cache size is 96 dwords) :
1185	 *	RING		0  to 15
1186	 *	INDIRECT1	16 to 79
1187	 *	INDIRECT2	80 to 95
1188	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1189	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1190	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1191	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1192	 * so it gets the bigger cache.
1193	 */
1194	indirect2_start = 80;
1195	indirect1_start = 16;
1196	/* cp setup */
1197	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1198	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1199	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1200	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1201#ifdef __BIG_ENDIAN
1202	tmp |= RADEON_BUF_SWAP_32BIT;
1203#endif
1204	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1205
1206	/* Set ring address */
1207	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1208	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1209	/* Force read & write ptr to 0 */
1210	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1211	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1212	ring->wptr = 0;
1213	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1214
1215	/* set the wb address whether it's enabled or not */
1216	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1217		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1218	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1219
1220	if (rdev->wb.enabled)
1221		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1222	else {
1223		tmp |= RADEON_RB_NO_UPDATE;
1224		WREG32(R_000770_SCRATCH_UMSK, 0);
1225	}
1226
1227	WREG32(RADEON_CP_RB_CNTL, tmp);
1228	udelay(10);
1229	/* Set cp mode to bus mastering & enable cp*/
1230	WREG32(RADEON_CP_CSQ_MODE,
1231	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1232	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1233	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1234	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1235	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1236
1237	/* at this point everything should be setup correctly to enable master */
1238	pci_set_master(rdev->pdev);
1239
1240	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1241	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1242	if (r) {
1243		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1244		return r;
1245	}
1246	ring->ready = true;
1247	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1248
1249	if (!ring->rptr_save_reg /* not resuming from suspend */
1250	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1251		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1252		if (r) {
1253			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1254			ring->rptr_save_reg = 0;
1255		}
1256	}
1257	return 0;
1258}
1259
1260void r100_cp_fini(struct radeon_device *rdev)
1261{
1262	if (r100_cp_wait_for_idle(rdev)) {
1263		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1264	}
1265	/* Disable ring */
1266	r100_cp_disable(rdev);
1267	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1268	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1269	DRM_INFO("radeon: cp finalized\n");
1270}
1271
1272void r100_cp_disable(struct radeon_device *rdev)
1273{
1274	/* Disable ring */
1275	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1276	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1277	WREG32(RADEON_CP_CSQ_MODE, 0);
1278	WREG32(RADEON_CP_CSQ_CNTL, 0);
1279	WREG32(R_000770_SCRATCH_UMSK, 0);
1280	if (r100_gui_wait_for_idle(rdev)) {
1281		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1282	}
1283}
1284
1285/*
1286 * CS functions
1287 */
1288int r100_reloc_pitch_offset(struct radeon_cs_parser *p,
1289			    struct radeon_cs_packet *pkt,
1290			    unsigned idx,
1291			    unsigned reg)
1292{
1293	int r;
1294	u32 tile_flags = 0;
1295	u32 tmp;
1296	struct radeon_bo_list *reloc;
1297	u32 value;
1298
1299	r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1300	if (r) {
1301		DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1302			  idx, reg);
1303		radeon_cs_dump_packet(p, pkt);
1304		return r;
1305	}
1306
1307	value = radeon_get_ib_value(p, idx);
1308	tmp = value & 0x003fffff;
1309	tmp += (((u32)reloc->gpu_offset) >> 10);
1310
1311	if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1312		if (reloc->tiling_flags & RADEON_TILING_MACRO)
1313			tile_flags |= RADEON_DST_TILE_MACRO;
1314		if (reloc->tiling_flags & RADEON_TILING_MICRO) {
1315			if (reg == RADEON_SRC_PITCH_OFFSET) {
1316				DRM_ERROR("Cannot src blit from microtiled surface\n");
1317				radeon_cs_dump_packet(p, pkt);
1318				return -EINVAL;
1319			}
1320			tile_flags |= RADEON_DST_TILE_MICRO;
1321		}
1322
1323		tmp |= tile_flags;
1324		p->ib.ptr[idx] = (value & 0x3fc00000) | tmp;
1325	} else
1326		p->ib.ptr[idx] = (value & 0xffc00000) | tmp;
1327	return 0;
1328}
1329
1330int r100_packet3_load_vbpntr(struct radeon_cs_parser *p,
1331			     struct radeon_cs_packet *pkt,
1332			     int idx)
1333{
1334	unsigned c, i;
1335	struct radeon_bo_list *reloc;
1336	struct r100_cs_track *track;
1337	int r = 0;
1338	volatile uint32_t *ib;
1339	u32 idx_value;
1340
1341	ib = p->ib.ptr;
1342	track = (struct r100_cs_track *)p->track;
1343	c = radeon_get_ib_value(p, idx++) & 0x1F;
1344	if (c > 16) {
1345	    DRM_ERROR("Only 16 vertex buffers are allowed %d\n",
1346		      pkt->opcode);
1347	    radeon_cs_dump_packet(p, pkt);
1348	    return -EINVAL;
1349	}
1350	track->num_arrays = c;
1351	for (i = 0; i < (c - 1); i += 2, idx += 3) {
1352		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1353		if (r) {
1354			DRM_ERROR("No reloc for packet3 %d\n",
1355				  pkt->opcode);
1356			radeon_cs_dump_packet(p, pkt);
1357			return r;
1358		}
1359		idx_value = radeon_get_ib_value(p, idx);
1360		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1361
1362		track->arrays[i + 0].esize = idx_value >> 8;
1363		track->arrays[i + 0].robj = reloc->robj;
1364		track->arrays[i + 0].esize &= 0x7F;
1365		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1366		if (r) {
1367			DRM_ERROR("No reloc for packet3 %d\n",
1368				  pkt->opcode);
1369			radeon_cs_dump_packet(p, pkt);
1370			return r;
1371		}
1372		ib[idx+2] = radeon_get_ib_value(p, idx + 2) + ((u32)reloc->gpu_offset);
1373		track->arrays[i + 1].robj = reloc->robj;
1374		track->arrays[i + 1].esize = idx_value >> 24;
1375		track->arrays[i + 1].esize &= 0x7F;
1376	}
1377	if (c & 1) {
1378		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1379		if (r) {
1380			DRM_ERROR("No reloc for packet3 %d\n",
1381					  pkt->opcode);
1382			radeon_cs_dump_packet(p, pkt);
1383			return r;
1384		}
1385		idx_value = radeon_get_ib_value(p, idx);
1386		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1387		track->arrays[i + 0].robj = reloc->robj;
1388		track->arrays[i + 0].esize = idx_value >> 8;
1389		track->arrays[i + 0].esize &= 0x7F;
1390	}
1391	return r;
1392}
1393
1394int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1395			  struct radeon_cs_packet *pkt,
1396			  const unsigned *auth, unsigned n,
1397			  radeon_packet0_check_t check)
1398{
1399	unsigned reg;
1400	unsigned i, j, m;
1401	unsigned idx;
1402	int r;
1403
1404	idx = pkt->idx + 1;
1405	reg = pkt->reg;
1406	/* Check that register fall into register range
1407	 * determined by the number of entry (n) in the
1408	 * safe register bitmap.
1409	 */
1410	if (pkt->one_reg_wr) {
1411		if ((reg >> 7) > n) {
1412			return -EINVAL;
1413		}
1414	} else {
1415		if (((reg + (pkt->count << 2)) >> 7) > n) {
1416			return -EINVAL;
1417		}
1418	}
1419	for (i = 0; i <= pkt->count; i++, idx++) {
1420		j = (reg >> 7);
1421		m = 1 << ((reg >> 2) & 31);
1422		if (auth[j] & m) {
1423			r = check(p, pkt, idx, reg);
1424			if (r) {
1425				return r;
1426			}
1427		}
1428		if (pkt->one_reg_wr) {
1429			if (!(auth[j] & m)) {
1430				break;
1431			}
1432		} else {
1433			reg += 4;
1434		}
1435	}
1436	return 0;
1437}
1438
1439/**
1440 * r100_cs_packet_parse_vline() - parse userspace VLINE packet
1441 * @p:		parser structure holding parsing context.
1442 *
1443 * Userspace sends a special sequence for VLINE waits.
1444 * PACKET0 - VLINE_START_END + value
1445 * PACKET0 - WAIT_UNTIL +_value
1446 * RELOC (P3) - crtc_id in reloc.
1447 *
1448 * This function parses this and relocates the VLINE START END
1449 * and WAIT UNTIL packets to the correct crtc.
1450 * It also detects a switched off crtc and nulls out the
1451 * wait in that case.
1452 */
1453int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1454{
1455	struct drm_crtc *crtc;
1456	struct radeon_crtc *radeon_crtc;
1457	struct radeon_cs_packet p3reloc, waitreloc;
1458	int crtc_id;
1459	int r;
1460	uint32_t header, h_idx, reg;
1461	volatile uint32_t *ib;
1462
1463	ib = p->ib.ptr;
1464
1465	/* parse the wait until */
1466	r = radeon_cs_packet_parse(p, &waitreloc, p->idx);
1467	if (r)
1468		return r;
1469
1470	/* check its a wait until and only 1 count */
1471	if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1472	    waitreloc.count != 0) {
1473		DRM_ERROR("vline wait had illegal wait until segment\n");
1474		return -EINVAL;
1475	}
1476
1477	if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1478		DRM_ERROR("vline wait had illegal wait until\n");
1479		return -EINVAL;
1480	}
1481
1482	/* jump over the NOP */
1483	r = radeon_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1484	if (r)
1485		return r;
1486
1487	h_idx = p->idx - 2;
1488	p->idx += waitreloc.count + 2;
1489	p->idx += p3reloc.count + 2;
1490
1491	header = radeon_get_ib_value(p, h_idx);
1492	crtc_id = radeon_get_ib_value(p, h_idx + 5);
1493	reg = R100_CP_PACKET0_GET_REG(header);
1494	crtc = drm_crtc_find(rdev_to_drm(p->rdev), p->filp, crtc_id);
1495	if (!crtc) {
1496		DRM_ERROR("cannot find crtc %d\n", crtc_id);
1497		return -ENOENT;
1498	}
1499	radeon_crtc = to_radeon_crtc(crtc);
1500	crtc_id = radeon_crtc->crtc_id;
1501
1502	if (!crtc->enabled) {
1503		/* if the CRTC isn't enabled - we need to nop out the wait until */
1504		ib[h_idx + 2] = PACKET2(0);
1505		ib[h_idx + 3] = PACKET2(0);
1506	} else if (crtc_id == 1) {
1507		switch (reg) {
1508		case AVIVO_D1MODE_VLINE_START_END:
1509			header &= ~R300_CP_PACKET0_REG_MASK;
1510			header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1511			break;
1512		case RADEON_CRTC_GUI_TRIG_VLINE:
1513			header &= ~R300_CP_PACKET0_REG_MASK;
1514			header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1515			break;
1516		default:
1517			DRM_ERROR("unknown crtc reloc\n");
1518			return -EINVAL;
1519		}
1520		ib[h_idx] = header;
1521		ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1522	}
1523
1524	return 0;
1525}
1526
1527static int r100_get_vtx_size(uint32_t vtx_fmt)
1528{
1529	int vtx_size;
1530	vtx_size = 2;
1531	/* ordered according to bits in spec */
1532	if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1533		vtx_size++;
1534	if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1535		vtx_size += 3;
1536	if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1537		vtx_size++;
1538	if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1539		vtx_size++;
1540	if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1541		vtx_size += 3;
1542	if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1543		vtx_size++;
1544	if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1545		vtx_size++;
1546	if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1547		vtx_size += 2;
1548	if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1549		vtx_size += 2;
1550	if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1551		vtx_size++;
1552	if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1553		vtx_size += 2;
1554	if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1555		vtx_size++;
1556	if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1557		vtx_size += 2;
1558	if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1559		vtx_size++;
1560	if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1561		vtx_size++;
1562	/* blend weight */
1563	if (vtx_fmt & (0x7 << 15))
1564		vtx_size += (vtx_fmt >> 15) & 0x7;
1565	if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1566		vtx_size += 3;
1567	if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1568		vtx_size += 2;
1569	if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1570		vtx_size++;
1571	if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1572		vtx_size++;
1573	if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1574		vtx_size++;
1575	if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1576		vtx_size++;
1577	return vtx_size;
1578}
1579
1580static int r100_packet0_check(struct radeon_cs_parser *p,
1581			      struct radeon_cs_packet *pkt,
1582			      unsigned idx, unsigned reg)
1583{
1584	struct radeon_bo_list *reloc;
1585	struct r100_cs_track *track;
1586	volatile uint32_t *ib;
1587	uint32_t tmp;
1588	int r;
1589	int i, face;
1590	u32 tile_flags = 0;
1591	u32 idx_value;
1592
1593	ib = p->ib.ptr;
1594	track = (struct r100_cs_track *)p->track;
1595
1596	idx_value = radeon_get_ib_value(p, idx);
1597
1598	switch (reg) {
1599	case RADEON_CRTC_GUI_TRIG_VLINE:
1600		r = r100_cs_packet_parse_vline(p);
1601		if (r) {
1602			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1603				  idx, reg);
1604			radeon_cs_dump_packet(p, pkt);
1605			return r;
1606		}
1607		break;
1608		/* FIXME: only allow PACKET3 blit? easier to check for out of
1609		 * range access */
1610	case RADEON_DST_PITCH_OFFSET:
1611	case RADEON_SRC_PITCH_OFFSET:
1612		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1613		if (r)
1614			return r;
1615		break;
1616	case RADEON_RB3D_DEPTHOFFSET:
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		track->zb.robj = reloc->robj;
1625		track->zb.offset = idx_value;
1626		track->zb_dirty = true;
1627		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1628		break;
1629	case RADEON_RB3D_COLOROFFSET:
1630		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1631		if (r) {
1632			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1633				  idx, reg);
1634			radeon_cs_dump_packet(p, pkt);
1635			return r;
1636		}
1637		track->cb[0].robj = reloc->robj;
1638		track->cb[0].offset = idx_value;
1639		track->cb_dirty = true;
1640		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1641		break;
1642	case RADEON_PP_TXOFFSET_0:
1643	case RADEON_PP_TXOFFSET_1:
1644	case RADEON_PP_TXOFFSET_2:
1645		i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1646		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1647		if (r) {
1648			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1649				  idx, reg);
1650			radeon_cs_dump_packet(p, pkt);
1651			return r;
1652		}
1653		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1654			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1655				tile_flags |= RADEON_TXO_MACRO_TILE;
1656			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1657				tile_flags |= RADEON_TXO_MICRO_TILE_X2;
1658
1659			tmp = idx_value & ~(0x7 << 2);
1660			tmp |= tile_flags;
1661			ib[idx] = tmp + ((u32)reloc->gpu_offset);
1662		} else
1663			ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1664		track->textures[i].robj = reloc->robj;
1665		track->tex_dirty = true;
1666		break;
1667	case RADEON_PP_CUBIC_OFFSET_T0_0:
1668	case RADEON_PP_CUBIC_OFFSET_T0_1:
1669	case RADEON_PP_CUBIC_OFFSET_T0_2:
1670	case RADEON_PP_CUBIC_OFFSET_T0_3:
1671	case RADEON_PP_CUBIC_OFFSET_T0_4:
1672		i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1673		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1674		if (r) {
1675			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1676				  idx, reg);
1677			radeon_cs_dump_packet(p, pkt);
1678			return r;
1679		}
1680		track->textures[0].cube_info[i].offset = idx_value;
1681		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1682		track->textures[0].cube_info[i].robj = reloc->robj;
1683		track->tex_dirty = true;
1684		break;
1685	case RADEON_PP_CUBIC_OFFSET_T1_0:
1686	case RADEON_PP_CUBIC_OFFSET_T1_1:
1687	case RADEON_PP_CUBIC_OFFSET_T1_2:
1688	case RADEON_PP_CUBIC_OFFSET_T1_3:
1689	case RADEON_PP_CUBIC_OFFSET_T1_4:
1690		i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1691		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1692		if (r) {
1693			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1694				  idx, reg);
1695			radeon_cs_dump_packet(p, pkt);
1696			return r;
1697		}
1698		track->textures[1].cube_info[i].offset = idx_value;
1699		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1700		track->textures[1].cube_info[i].robj = reloc->robj;
1701		track->tex_dirty = true;
1702		break;
1703	case RADEON_PP_CUBIC_OFFSET_T2_0:
1704	case RADEON_PP_CUBIC_OFFSET_T2_1:
1705	case RADEON_PP_CUBIC_OFFSET_T2_2:
1706	case RADEON_PP_CUBIC_OFFSET_T2_3:
1707	case RADEON_PP_CUBIC_OFFSET_T2_4:
1708		i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1709		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1710		if (r) {
1711			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1712				  idx, reg);
1713			radeon_cs_dump_packet(p, pkt);
1714			return r;
1715		}
1716		track->textures[2].cube_info[i].offset = idx_value;
1717		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1718		track->textures[2].cube_info[i].robj = reloc->robj;
1719		track->tex_dirty = true;
1720		break;
1721	case RADEON_RE_WIDTH_HEIGHT:
1722		track->maxy = ((idx_value >> 16) & 0x7FF);
1723		track->cb_dirty = true;
1724		track->zb_dirty = true;
1725		break;
1726	case RADEON_RB3D_COLORPITCH:
1727		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1728		if (r) {
1729			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1730				  idx, reg);
1731			radeon_cs_dump_packet(p, pkt);
1732			return r;
1733		}
1734		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
1735			if (reloc->tiling_flags & RADEON_TILING_MACRO)
1736				tile_flags |= RADEON_COLOR_TILE_ENABLE;
1737			if (reloc->tiling_flags & RADEON_TILING_MICRO)
1738				tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1739
1740			tmp = idx_value & ~(0x7 << 16);
1741			tmp |= tile_flags;
1742			ib[idx] = tmp;
1743		} else
1744			ib[idx] = idx_value;
1745
1746		track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1747		track->cb_dirty = true;
1748		break;
1749	case RADEON_RB3D_DEPTHPITCH:
1750		track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1751		track->zb_dirty = true;
1752		break;
1753	case RADEON_RB3D_CNTL:
1754		switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1755		case 7:
1756		case 8:
1757		case 9:
1758		case 11:
1759		case 12:
1760			track->cb[0].cpp = 1;
1761			break;
1762		case 3:
1763		case 4:
1764		case 15:
1765			track->cb[0].cpp = 2;
1766			break;
1767		case 6:
1768			track->cb[0].cpp = 4;
1769			break;
1770		default:
1771			DRM_ERROR("Invalid color buffer format (%d) !\n",
1772				  ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1773			return -EINVAL;
1774		}
1775		track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1776		track->cb_dirty = true;
1777		track->zb_dirty = true;
1778		break;
1779	case RADEON_RB3D_ZSTENCILCNTL:
1780		switch (idx_value & 0xf) {
1781		case 0:
1782			track->zb.cpp = 2;
1783			break;
1784		case 2:
1785		case 3:
1786		case 4:
1787		case 5:
1788		case 9:
1789		case 11:
1790			track->zb.cpp = 4;
1791			break;
1792		default:
1793			break;
1794		}
1795		track->zb_dirty = true;
1796		break;
1797	case RADEON_RB3D_ZPASS_ADDR:
1798		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1799		if (r) {
1800			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1801				  idx, reg);
1802			radeon_cs_dump_packet(p, pkt);
1803			return r;
1804		}
1805		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1806		break;
1807	case RADEON_PP_CNTL:
1808		{
1809			uint32_t temp = idx_value >> 4;
1810			for (i = 0; i < track->num_texture; i++)
1811				track->textures[i].enabled = !!(temp & (1 << i));
1812			track->tex_dirty = true;
1813		}
1814		break;
1815	case RADEON_SE_VF_CNTL:
1816		track->vap_vf_cntl = idx_value;
1817		break;
1818	case RADEON_SE_VTX_FMT:
1819		track->vtx_size = r100_get_vtx_size(idx_value);
1820		break;
1821	case RADEON_PP_TEX_SIZE_0:
1822	case RADEON_PP_TEX_SIZE_1:
1823	case RADEON_PP_TEX_SIZE_2:
1824		i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1825		track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1826		track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1827		track->tex_dirty = true;
1828		break;
1829	case RADEON_PP_TEX_PITCH_0:
1830	case RADEON_PP_TEX_PITCH_1:
1831	case RADEON_PP_TEX_PITCH_2:
1832		i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1833		track->textures[i].pitch = idx_value + 32;
1834		track->tex_dirty = true;
1835		break;
1836	case RADEON_PP_TXFILTER_0:
1837	case RADEON_PP_TXFILTER_1:
1838	case RADEON_PP_TXFILTER_2:
1839		i = (reg - RADEON_PP_TXFILTER_0) / 24;
1840		track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1841						 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1842		tmp = (idx_value >> 23) & 0x7;
1843		if (tmp == 2 || tmp == 6)
1844			track->textures[i].roundup_w = false;
1845		tmp = (idx_value >> 27) & 0x7;
1846		if (tmp == 2 || tmp == 6)
1847			track->textures[i].roundup_h = false;
1848		track->tex_dirty = true;
1849		break;
1850	case RADEON_PP_TXFORMAT_0:
1851	case RADEON_PP_TXFORMAT_1:
1852	case RADEON_PP_TXFORMAT_2:
1853		i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1854		if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1855			track->textures[i].use_pitch = true;
1856		} else {
1857			track->textures[i].use_pitch = false;
1858			track->textures[i].width = 1 << ((idx_value & RADEON_TXFORMAT_WIDTH_MASK) >> RADEON_TXFORMAT_WIDTH_SHIFT);
1859			track->textures[i].height = 1 << ((idx_value & RADEON_TXFORMAT_HEIGHT_MASK) >> RADEON_TXFORMAT_HEIGHT_SHIFT);
1860		}
1861		if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1862			track->textures[i].tex_coord_type = 2;
1863		switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1864		case RADEON_TXFORMAT_I8:
1865		case RADEON_TXFORMAT_RGB332:
1866		case RADEON_TXFORMAT_Y8:
1867			track->textures[i].cpp = 1;
1868			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1869			break;
1870		case RADEON_TXFORMAT_AI88:
1871		case RADEON_TXFORMAT_ARGB1555:
1872		case RADEON_TXFORMAT_RGB565:
1873		case RADEON_TXFORMAT_ARGB4444:
1874		case RADEON_TXFORMAT_VYUY422:
1875		case RADEON_TXFORMAT_YVYU422:
1876		case RADEON_TXFORMAT_SHADOW16:
1877		case RADEON_TXFORMAT_LDUDV655:
1878		case RADEON_TXFORMAT_DUDV88:
1879			track->textures[i].cpp = 2;
1880			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1881			break;
1882		case RADEON_TXFORMAT_ARGB8888:
1883		case RADEON_TXFORMAT_RGBA8888:
1884		case RADEON_TXFORMAT_SHADOW32:
1885		case RADEON_TXFORMAT_LDUDUV8888:
1886			track->textures[i].cpp = 4;
1887			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1888			break;
1889		case RADEON_TXFORMAT_DXT1:
1890			track->textures[i].cpp = 1;
1891			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1892			break;
1893		case RADEON_TXFORMAT_DXT23:
1894		case RADEON_TXFORMAT_DXT45:
1895			track->textures[i].cpp = 1;
1896			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1897			break;
1898		}
1899		track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1900		track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1901		track->tex_dirty = true;
1902		break;
1903	case RADEON_PP_CUBIC_FACES_0:
1904	case RADEON_PP_CUBIC_FACES_1:
1905	case RADEON_PP_CUBIC_FACES_2:
1906		tmp = idx_value;
1907		i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1908		for (face = 0; face < 4; face++) {
1909			track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1910			track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1911		}
1912		track->tex_dirty = true;
1913		break;
1914	default:
1915		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
1916		return -EINVAL;
1917	}
1918	return 0;
1919}
1920
1921int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1922					 struct radeon_cs_packet *pkt,
1923					 struct radeon_bo *robj)
1924{
1925	unsigned idx;
1926	u32 value;
1927	idx = pkt->idx + 1;
1928	value = radeon_get_ib_value(p, idx + 2);
1929	if ((value + 1) > radeon_bo_size(robj)) {
1930		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1931			  "(need %u have %lu) !\n",
1932			  value + 1,
1933			  radeon_bo_size(robj));
1934		return -EINVAL;
1935	}
1936	return 0;
1937}
1938
1939static int r100_packet3_check(struct radeon_cs_parser *p,
1940			      struct radeon_cs_packet *pkt)
1941{
1942	struct radeon_bo_list *reloc;
1943	struct r100_cs_track *track;
1944	unsigned idx;
1945	volatile uint32_t *ib;
1946	int r;
1947
1948	ib = p->ib.ptr;
1949	idx = pkt->idx + 1;
1950	track = (struct r100_cs_track *)p->track;
1951	switch (pkt->opcode) {
1952	case PACKET3_3D_LOAD_VBPNTR:
1953		r = r100_packet3_load_vbpntr(p, pkt, idx);
1954		if (r)
1955			return r;
1956		break;
1957	case PACKET3_INDX_BUFFER:
1958		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1959		if (r) {
1960			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1961			radeon_cs_dump_packet(p, pkt);
1962			return r;
1963		}
1964		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1965		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1966		if (r) {
1967			return r;
1968		}
1969		break;
1970	case 0x23:
1971		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1972		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1973		if (r) {
1974			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1975			radeon_cs_dump_packet(p, pkt);
1976			return r;
1977		}
1978		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1979		track->num_arrays = 1;
1980		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1981
1982		track->arrays[0].robj = reloc->robj;
1983		track->arrays[0].esize = track->vtx_size;
1984
1985		track->max_indx = radeon_get_ib_value(p, idx+1);
1986
1987		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1988		track->immd_dwords = pkt->count - 1;
1989		r = r100_cs_track_check(p->rdev, track);
1990		if (r)
1991			return r;
1992		break;
1993	case PACKET3_3D_DRAW_IMMD:
1994		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1995			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1996			return -EINVAL;
1997		}
1998		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1999		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2000		track->immd_dwords = pkt->count - 1;
2001		r = r100_cs_track_check(p->rdev, track);
2002		if (r)
2003			return r;
2004		break;
2005		/* triggers drawing using in-packet vertex data */
2006	case PACKET3_3D_DRAW_IMMD_2:
2007		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
2008			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
2009			return -EINVAL;
2010		}
2011		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
2012		track->immd_dwords = pkt->count;
2013		r = r100_cs_track_check(p->rdev, track);
2014		if (r)
2015			return r;
2016		break;
2017		/* triggers drawing using in-packet vertex data */
2018	case PACKET3_3D_DRAW_VBUF_2:
2019		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
2020		r = r100_cs_track_check(p->rdev, track);
2021		if (r)
2022			return r;
2023		break;
2024		/* triggers drawing of vertex buffers setup elsewhere */
2025	case PACKET3_3D_DRAW_INDX_2:
2026		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
2027		r = r100_cs_track_check(p->rdev, track);
2028		if (r)
2029			return r;
2030		break;
2031		/* triggers drawing using indices to vertex buffer */
2032	case PACKET3_3D_DRAW_VBUF:
2033		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2034		r = r100_cs_track_check(p->rdev, track);
2035		if (r)
2036			return r;
2037		break;
2038		/* triggers drawing of vertex buffers setup elsewhere */
2039	case PACKET3_3D_DRAW_INDX:
2040		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2041		r = r100_cs_track_check(p->rdev, track);
2042		if (r)
2043			return r;
2044		break;
2045		/* triggers drawing using indices to vertex buffer */
2046	case PACKET3_3D_CLEAR_HIZ:
2047	case PACKET3_3D_CLEAR_ZMASK:
2048		if (p->rdev->hyperz_filp != p->filp)
2049			return -EINVAL;
2050		break;
2051	case PACKET3_NOP:
2052		break;
2053	default:
2054		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2055		return -EINVAL;
2056	}
2057	return 0;
2058}
2059
2060int r100_cs_parse(struct radeon_cs_parser *p)
2061{
2062	struct radeon_cs_packet pkt;
2063	struct r100_cs_track *track;
2064	int r;
2065
2066	track = kzalloc(sizeof(*track), GFP_KERNEL);
2067	if (!track)
2068		return -ENOMEM;
2069	r100_cs_track_clear(p->rdev, track);
2070	p->track = track;
2071	do {
2072		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2073		if (r) {
2074			return r;
2075		}
2076		p->idx += pkt.count + 2;
2077		switch (pkt.type) {
2078		case RADEON_PACKET_TYPE0:
2079			if (p->rdev->family >= CHIP_R200)
2080				r = r100_cs_parse_packet0(p, &pkt,
2081					p->rdev->config.r100.reg_safe_bm,
2082					p->rdev->config.r100.reg_safe_bm_size,
2083					&r200_packet0_check);
2084			else
2085				r = r100_cs_parse_packet0(p, &pkt,
2086					p->rdev->config.r100.reg_safe_bm,
2087					p->rdev->config.r100.reg_safe_bm_size,
2088					&r100_packet0_check);
2089			break;
2090		case RADEON_PACKET_TYPE2:
2091			break;
2092		case RADEON_PACKET_TYPE3:
2093			r = r100_packet3_check(p, &pkt);
2094			break;
2095		default:
2096			DRM_ERROR("Unknown packet type %d !\n",
2097				  pkt.type);
2098			return -EINVAL;
2099		}
2100		if (r)
2101			return r;
2102	} while (p->idx < p->chunk_ib->length_dw);
2103	return 0;
2104}
2105
2106static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2107{
2108	DRM_ERROR("pitch                      %d\n", t->pitch);
2109	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2110	DRM_ERROR("width                      %d\n", t->width);
2111	DRM_ERROR("width_11                   %d\n", t->width_11);
2112	DRM_ERROR("height                     %d\n", t->height);
2113	DRM_ERROR("height_11                  %d\n", t->height_11);
2114	DRM_ERROR("num levels                 %d\n", t->num_levels);
2115	DRM_ERROR("depth                      %d\n", t->txdepth);
2116	DRM_ERROR("bpp                        %d\n", t->cpp);
2117	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2118	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2119	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2120	DRM_ERROR("compress format            %d\n", t->compress_format);
2121}
2122
2123static int r100_track_compress_size(int compress_format, int w, int h)
2124{
2125	int block_width, block_height, block_bytes;
2126	int wblocks, hblocks;
2127	int min_wblocks;
2128	int sz;
2129
2130	block_width = 4;
2131	block_height = 4;
2132
2133	switch (compress_format) {
2134	case R100_TRACK_COMP_DXT1:
2135		block_bytes = 8;
2136		min_wblocks = 4;
2137		break;
2138	default:
2139	case R100_TRACK_COMP_DXT35:
2140		block_bytes = 16;
2141		min_wblocks = 2;
2142		break;
2143	}
2144
2145	hblocks = (h + block_height - 1) / block_height;
2146	wblocks = (w + block_width - 1) / block_width;
2147	if (wblocks < min_wblocks)
2148		wblocks = min_wblocks;
2149	sz = wblocks * hblocks * block_bytes;
2150	return sz;
2151}
2152
2153static int r100_cs_track_cube(struct radeon_device *rdev,
2154			      struct r100_cs_track *track, unsigned idx)
2155{
2156	unsigned face, w, h;
2157	struct radeon_bo *cube_robj;
2158	unsigned long size;
2159	unsigned compress_format = track->textures[idx].compress_format;
2160
2161	for (face = 0; face < 5; face++) {
2162		cube_robj = track->textures[idx].cube_info[face].robj;
2163		w = track->textures[idx].cube_info[face].width;
2164		h = track->textures[idx].cube_info[face].height;
2165
2166		if (compress_format) {
2167			size = r100_track_compress_size(compress_format, w, h);
2168		} else
2169			size = w * h;
2170		size *= track->textures[idx].cpp;
2171
2172		size += track->textures[idx].cube_info[face].offset;
2173
2174		if (size > radeon_bo_size(cube_robj)) {
2175			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2176				  size, radeon_bo_size(cube_robj));
2177			r100_cs_track_texture_print(&track->textures[idx]);
2178			return -1;
2179		}
2180	}
2181	return 0;
2182}
2183
2184static int r100_cs_track_texture_check(struct radeon_device *rdev,
2185				       struct r100_cs_track *track)
2186{
2187	struct radeon_bo *robj;
2188	unsigned long size;
2189	unsigned u, i, w, h, d;
2190	int ret;
2191
2192	for (u = 0; u < track->num_texture; u++) {
2193		if (!track->textures[u].enabled)
2194			continue;
2195		if (track->textures[u].lookup_disable)
2196			continue;
2197		robj = track->textures[u].robj;
2198		if (robj == NULL) {
2199			DRM_ERROR("No texture bound to unit %u\n", u);
2200			return -EINVAL;
2201		}
2202		size = 0;
2203		for (i = 0; i <= track->textures[u].num_levels; i++) {
2204			if (track->textures[u].use_pitch) {
2205				if (rdev->family < CHIP_R300)
2206					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2207				else
2208					w = track->textures[u].pitch / (1 << i);
2209			} else {
2210				w = track->textures[u].width;
2211				if (rdev->family >= CHIP_RV515)
2212					w |= track->textures[u].width_11;
2213				w = w / (1 << i);
2214				if (track->textures[u].roundup_w)
2215					w = roundup_pow_of_two(w);
2216			}
2217			h = track->textures[u].height;
2218			if (rdev->family >= CHIP_RV515)
2219				h |= track->textures[u].height_11;
2220			h = h / (1 << i);
2221			if (track->textures[u].roundup_h)
2222				h = roundup_pow_of_two(h);
2223			if (track->textures[u].tex_coord_type == 1) {
2224				d = (1 << track->textures[u].txdepth) / (1 << i);
2225				if (!d)
2226					d = 1;
2227			} else {
2228				d = 1;
2229			}
2230			if (track->textures[u].compress_format) {
2231
2232				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2233				/* compressed textures are block based */
2234			} else
2235				size += w * h * d;
2236		}
2237		size *= track->textures[u].cpp;
2238
2239		switch (track->textures[u].tex_coord_type) {
2240		case 0:
2241		case 1:
2242			break;
2243		case 2:
2244			if (track->separate_cube) {
2245				ret = r100_cs_track_cube(rdev, track, u);
2246				if (ret)
2247					return ret;
2248			} else
2249				size *= 6;
2250			break;
2251		default:
2252			DRM_ERROR("Invalid texture coordinate type %u for unit "
2253				  "%u\n", track->textures[u].tex_coord_type, u);
2254			return -EINVAL;
2255		}
2256		if (size > radeon_bo_size(robj)) {
2257			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2258				  "%lu\n", u, size, radeon_bo_size(robj));
2259			r100_cs_track_texture_print(&track->textures[u]);
2260			return -EINVAL;
2261		}
2262	}
2263	return 0;
2264}
2265
2266int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2267{
2268	unsigned i;
2269	unsigned long size;
2270	unsigned prim_walk;
2271	unsigned nverts;
2272	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2273
2274	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2275	    !track->blend_read_enable)
2276		num_cb = 0;
2277
2278	for (i = 0; i < num_cb; i++) {
2279		if (track->cb[i].robj == NULL) {
2280			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2281			return -EINVAL;
2282		}
2283		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2284		size += track->cb[i].offset;
2285		if (size > radeon_bo_size(track->cb[i].robj)) {
2286			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2287				  "(need %lu have %lu) !\n", i, size,
2288				  radeon_bo_size(track->cb[i].robj));
2289			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2290				  i, track->cb[i].pitch, track->cb[i].cpp,
2291				  track->cb[i].offset, track->maxy);
2292			return -EINVAL;
2293		}
2294	}
2295	track->cb_dirty = false;
2296
2297	if (track->zb_dirty && track->z_enabled) {
2298		if (track->zb.robj == NULL) {
2299			DRM_ERROR("[drm] No buffer for z buffer !\n");
2300			return -EINVAL;
2301		}
2302		size = track->zb.pitch * track->zb.cpp * track->maxy;
2303		size += track->zb.offset;
2304		if (size > radeon_bo_size(track->zb.robj)) {
2305			DRM_ERROR("[drm] Buffer too small for z buffer "
2306				  "(need %lu have %lu) !\n", size,
2307				  radeon_bo_size(track->zb.robj));
2308			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2309				  track->zb.pitch, track->zb.cpp,
2310				  track->zb.offset, track->maxy);
2311			return -EINVAL;
2312		}
2313	}
2314	track->zb_dirty = false;
2315
2316	if (track->aa_dirty && track->aaresolve) {
2317		if (track->aa.robj == NULL) {
2318			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2319			return -EINVAL;
2320		}
2321		/* I believe the format comes from colorbuffer0. */
2322		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2323		size += track->aa.offset;
2324		if (size > radeon_bo_size(track->aa.robj)) {
2325			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2326				  "(need %lu have %lu) !\n", i, size,
2327				  radeon_bo_size(track->aa.robj));
2328			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2329				  i, track->aa.pitch, track->cb[0].cpp,
2330				  track->aa.offset, track->maxy);
2331			return -EINVAL;
2332		}
2333	}
2334	track->aa_dirty = false;
2335
2336	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2337	if (track->vap_vf_cntl & (1 << 14)) {
2338		nverts = track->vap_alt_nverts;
2339	} else {
2340		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2341	}
2342	switch (prim_walk) {
2343	case 1:
2344		for (i = 0; i < track->num_arrays; i++) {
2345			size = track->arrays[i].esize * track->max_indx * 4UL;
2346			if (track->arrays[i].robj == NULL) {
2347				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2348					  "bound\n", prim_walk, i);
2349				return -EINVAL;
2350			}
2351			if (size > radeon_bo_size(track->arrays[i].robj)) {
2352				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2353					"need %lu dwords have %lu dwords\n",
2354					prim_walk, i, size >> 2,
2355					radeon_bo_size(track->arrays[i].robj)
2356					>> 2);
2357				DRM_ERROR("Max indices %u\n", track->max_indx);
2358				return -EINVAL;
2359			}
2360		}
2361		break;
2362	case 2:
2363		for (i = 0; i < track->num_arrays; i++) {
2364			size = track->arrays[i].esize * (nverts - 1) * 4UL;
2365			if (track->arrays[i].robj == NULL) {
2366				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2367					  "bound\n", prim_walk, i);
2368				return -EINVAL;
2369			}
2370			if (size > radeon_bo_size(track->arrays[i].robj)) {
2371				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2372					"need %lu dwords have %lu dwords\n",
2373					prim_walk, i, size >> 2,
2374					radeon_bo_size(track->arrays[i].robj)
2375					>> 2);
2376				return -EINVAL;
2377			}
2378		}
2379		break;
2380	case 3:
2381		size = track->vtx_size * nverts;
2382		if (size != track->immd_dwords) {
2383			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2384				  track->immd_dwords, size);
2385			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2386				  nverts, track->vtx_size);
2387			return -EINVAL;
2388		}
2389		break;
2390	default:
2391		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2392			  prim_walk);
2393		return -EINVAL;
2394	}
2395
2396	if (track->tex_dirty) {
2397		track->tex_dirty = false;
2398		return r100_cs_track_texture_check(rdev, track);
2399	}
2400	return 0;
2401}
2402
2403void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2404{
2405	unsigned i, face;
2406
2407	track->cb_dirty = true;
2408	track->zb_dirty = true;
2409	track->tex_dirty = true;
2410	track->aa_dirty = true;
2411
2412	if (rdev->family < CHIP_R300) {
2413		track->num_cb = 1;
2414		if (rdev->family <= CHIP_RS200)
2415			track->num_texture = 3;
2416		else
2417			track->num_texture = 6;
2418		track->maxy = 2048;
2419		track->separate_cube = true;
2420	} else {
2421		track->num_cb = 4;
2422		track->num_texture = 16;
2423		track->maxy = 4096;
2424		track->separate_cube = false;
2425		track->aaresolve = false;
2426		track->aa.robj = NULL;
2427	}
2428
2429	for (i = 0; i < track->num_cb; i++) {
2430		track->cb[i].robj = NULL;
2431		track->cb[i].pitch = 8192;
2432		track->cb[i].cpp = 16;
2433		track->cb[i].offset = 0;
2434	}
2435	track->z_enabled = true;
2436	track->zb.robj = NULL;
2437	track->zb.pitch = 8192;
2438	track->zb.cpp = 4;
2439	track->zb.offset = 0;
2440	track->vtx_size = 0x7F;
2441	track->immd_dwords = 0xFFFFFFFFUL;
2442	track->num_arrays = 11;
2443	track->max_indx = 0x00FFFFFFUL;
2444	for (i = 0; i < track->num_arrays; i++) {
2445		track->arrays[i].robj = NULL;
2446		track->arrays[i].esize = 0x7F;
2447	}
2448	for (i = 0; i < track->num_texture; i++) {
2449		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2450		track->textures[i].pitch = 16536;
2451		track->textures[i].width = 16536;
2452		track->textures[i].height = 16536;
2453		track->textures[i].width_11 = 1 << 11;
2454		track->textures[i].height_11 = 1 << 11;
2455		track->textures[i].num_levels = 12;
2456		if (rdev->family <= CHIP_RS200) {
2457			track->textures[i].tex_coord_type = 0;
2458			track->textures[i].txdepth = 0;
2459		} else {
2460			track->textures[i].txdepth = 16;
2461			track->textures[i].tex_coord_type = 1;
2462		}
2463		track->textures[i].cpp = 64;
2464		track->textures[i].robj = NULL;
2465		/* CS IB emission code makes sure texture unit are disabled */
2466		track->textures[i].enabled = false;
2467		track->textures[i].lookup_disable = false;
2468		track->textures[i].roundup_w = true;
2469		track->textures[i].roundup_h = true;
2470		if (track->separate_cube)
2471			for (face = 0; face < 5; face++) {
2472				track->textures[i].cube_info[face].robj = NULL;
2473				track->textures[i].cube_info[face].width = 16536;
2474				track->textures[i].cube_info[face].height = 16536;
2475				track->textures[i].cube_info[face].offset = 0;
2476			}
2477	}
2478}
2479
2480/*
2481 * Global GPU functions
2482 */
2483static void r100_errata(struct radeon_device *rdev)
2484{
2485	rdev->pll_errata = 0;
2486
2487	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2488		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2489	}
2490
2491	if (rdev->family == CHIP_RV100 ||
2492	    rdev->family == CHIP_RS100 ||
2493	    rdev->family == CHIP_RS200) {
2494		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2495	}
2496}
2497
2498static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2499{
2500	unsigned i;
2501	uint32_t tmp;
2502
2503	for (i = 0; i < rdev->usec_timeout; i++) {
2504		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2505		if (tmp >= n) {
2506			return 0;
2507		}
2508		udelay(1);
2509	}
2510	return -1;
2511}
2512
2513int r100_gui_wait_for_idle(struct radeon_device *rdev)
2514{
2515	unsigned i;
2516	uint32_t tmp;
2517
2518	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2519		pr_warn("radeon: wait for empty RBBM fifo failed! Bad things might happen.\n");
2520	}
2521	for (i = 0; i < rdev->usec_timeout; i++) {
2522		tmp = RREG32(RADEON_RBBM_STATUS);
2523		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2524			return 0;
2525		}
2526		udelay(1);
2527	}
2528	return -1;
2529}
2530
2531int r100_mc_wait_for_idle(struct radeon_device *rdev)
2532{
2533	unsigned i;
2534	uint32_t tmp;
2535
2536	for (i = 0; i < rdev->usec_timeout; i++) {
2537		/* read MC_STATUS */
2538		tmp = RREG32(RADEON_MC_STATUS);
2539		if (tmp & RADEON_MC_IDLE) {
2540			return 0;
2541		}
2542		udelay(1);
2543	}
2544	return -1;
2545}
2546
2547bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2548{
2549	u32 rbbm_status;
2550
2551	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2552	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2553		radeon_ring_lockup_update(rdev, ring);
2554		return false;
2555	}
2556	return radeon_ring_test_lockup(rdev, ring);
2557}
2558
2559/* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2560void r100_enable_bm(struct radeon_device *rdev)
2561{
2562	uint32_t tmp;
2563	/* Enable bus mastering */
2564	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2565	WREG32(RADEON_BUS_CNTL, tmp);
2566}
2567
2568void r100_bm_disable(struct radeon_device *rdev)
2569{
2570	u32 tmp;
2571
2572	/* disable bus mastering */
2573	tmp = RREG32(R_000030_BUS_CNTL);
2574	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2575	mdelay(1);
2576	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2577	mdelay(1);
2578	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2579	tmp = RREG32(RADEON_BUS_CNTL);
2580	mdelay(1);
2581	pci_clear_master(rdev->pdev);
2582	mdelay(1);
2583}
2584
2585int r100_asic_reset(struct radeon_device *rdev, bool hard)
2586{
2587	struct r100_mc_save save;
2588	u32 status, tmp;
2589	int ret = 0;
2590
2591	status = RREG32(R_000E40_RBBM_STATUS);
2592	if (!G_000E40_GUI_ACTIVE(status)) {
2593		return 0;
2594	}
2595	r100_mc_stop(rdev, &save);
2596	status = RREG32(R_000E40_RBBM_STATUS);
2597	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2598	/* stop CP */
2599	WREG32(RADEON_CP_CSQ_CNTL, 0);
2600	tmp = RREG32(RADEON_CP_RB_CNTL);
2601	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2602	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2603	WREG32(RADEON_CP_RB_WPTR, 0);
2604	WREG32(RADEON_CP_RB_CNTL, tmp);
2605	/* save PCI state */
2606	pci_save_state(rdev->pdev);
2607	/* disable bus mastering */
2608	r100_bm_disable(rdev);
2609	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2610					S_0000F0_SOFT_RESET_RE(1) |
2611					S_0000F0_SOFT_RESET_PP(1) |
2612					S_0000F0_SOFT_RESET_RB(1));
2613	RREG32(R_0000F0_RBBM_SOFT_RESET);
2614	mdelay(500);
2615	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2616	mdelay(1);
2617	status = RREG32(R_000E40_RBBM_STATUS);
2618	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2619	/* reset CP */
2620	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2621	RREG32(R_0000F0_RBBM_SOFT_RESET);
2622	mdelay(500);
2623	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2624	mdelay(1);
2625	status = RREG32(R_000E40_RBBM_STATUS);
2626	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2627	/* restore PCI & busmastering */
2628	pci_restore_state(rdev->pdev);
2629	r100_enable_bm(rdev);
2630	/* Check if GPU is idle */
2631	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2632		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2633		dev_err(rdev->dev, "failed to reset GPU\n");
2634		ret = -1;
2635	} else
2636		dev_info(rdev->dev, "GPU reset succeed\n");
2637	r100_mc_resume(rdev, &save);
2638	return ret;
2639}
2640
2641void r100_set_common_regs(struct radeon_device *rdev)
2642{
 
2643	bool force_dac2 = false;
2644	u32 tmp;
2645
2646	/* set these so they don't interfere with anything */
2647	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2648	WREG32(RADEON_SUBPIC_CNTL, 0);
2649	WREG32(RADEON_VIPH_CONTROL, 0);
2650	WREG32(RADEON_I2C_CNTL_1, 0);
2651	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2652	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2653	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2654
2655	/* always set up dac2 on rn50 and some rv100 as lots
2656	 * of servers seem to wire it up to a VGA port but
2657	 * don't report it in the bios connector
2658	 * table.
2659	 */
2660	switch (rdev->pdev->device) {
2661		/* RN50 */
2662	case 0x515e:
2663	case 0x5969:
2664		force_dac2 = true;
2665		break;
2666		/* RV100*/
2667	case 0x5159:
2668	case 0x515a:
2669		/* DELL triple head servers */
2670		if ((rdev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2671		    ((rdev->pdev->subsystem_device == 0x016c) ||
2672		     (rdev->pdev->subsystem_device == 0x016d) ||
2673		     (rdev->pdev->subsystem_device == 0x016e) ||
2674		     (rdev->pdev->subsystem_device == 0x016f) ||
2675		     (rdev->pdev->subsystem_device == 0x0170) ||
2676		     (rdev->pdev->subsystem_device == 0x017d) ||
2677		     (rdev->pdev->subsystem_device == 0x017e) ||
2678		     (rdev->pdev->subsystem_device == 0x0183) ||
2679		     (rdev->pdev->subsystem_device == 0x018a) ||
2680		     (rdev->pdev->subsystem_device == 0x019a)))
2681			force_dac2 = true;
2682		break;
2683	}
2684
2685	if (force_dac2) {
2686		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2687		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2688		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2689
2690		/* For CRT on DAC2, don't turn it on if BIOS didn't
2691		   enable it, even it's detected.
2692		*/
2693
2694		/* force it to crtc0 */
2695		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2696		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2697		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2698
2699		/* set up the TV DAC */
2700		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2701				 RADEON_TV_DAC_STD_MASK |
2702				 RADEON_TV_DAC_RDACPD |
2703				 RADEON_TV_DAC_GDACPD |
2704				 RADEON_TV_DAC_BDACPD |
2705				 RADEON_TV_DAC_BGADJ_MASK |
2706				 RADEON_TV_DAC_DACADJ_MASK);
2707		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2708				RADEON_TV_DAC_NHOLD |
2709				RADEON_TV_DAC_STD_PS2 |
2710				(0x58 << 16));
2711
2712		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2713		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2714		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2715	}
2716
2717	/* switch PM block to ACPI mode */
2718	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2719	tmp &= ~RADEON_PM_MODE_SEL;
2720	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2721
2722}
2723
2724/*
2725 * VRAM info
2726 */
2727static void r100_vram_get_type(struct radeon_device *rdev)
2728{
2729	uint32_t tmp;
2730
2731	rdev->mc.vram_is_ddr = false;
2732	if (rdev->flags & RADEON_IS_IGP)
2733		rdev->mc.vram_is_ddr = true;
2734	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2735		rdev->mc.vram_is_ddr = true;
2736	if ((rdev->family == CHIP_RV100) ||
2737	    (rdev->family == CHIP_RS100) ||
2738	    (rdev->family == CHIP_RS200)) {
2739		tmp = RREG32(RADEON_MEM_CNTL);
2740		if (tmp & RV100_HALF_MODE) {
2741			rdev->mc.vram_width = 32;
2742		} else {
2743			rdev->mc.vram_width = 64;
2744		}
2745		if (rdev->flags & RADEON_SINGLE_CRTC) {
2746			rdev->mc.vram_width /= 4;
2747			rdev->mc.vram_is_ddr = true;
2748		}
2749	} else if (rdev->family <= CHIP_RV280) {
2750		tmp = RREG32(RADEON_MEM_CNTL);
2751		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2752			rdev->mc.vram_width = 128;
2753		} else {
2754			rdev->mc.vram_width = 64;
2755		}
2756	} else {
2757		/* newer IGPs */
2758		rdev->mc.vram_width = 128;
2759	}
2760}
2761
2762static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2763{
2764	u32 aper_size;
2765	u8 byte;
2766
2767	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2768
2769	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2770	 * that is has the 2nd generation multifunction PCI interface
2771	 */
2772	if (rdev->family == CHIP_RV280 ||
2773	    rdev->family >= CHIP_RV350) {
2774		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2775		       ~RADEON_HDP_APER_CNTL);
2776		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2777		return aper_size * 2;
2778	}
2779
2780	/* Older cards have all sorts of funny issues to deal with. First
2781	 * check if it's a multifunction card by reading the PCI config
2782	 * header type... Limit those to one aperture size
2783	 */
2784	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2785	if (byte & 0x80) {
2786		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2787		DRM_INFO("Limiting VRAM to one aperture\n");
2788		return aper_size;
2789	}
2790
2791	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2792	 * have set it up. We don't write this as it's broken on some ASICs but
2793	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2794	 */
2795	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2796		return aper_size * 2;
2797	return aper_size;
2798}
2799
2800void r100_vram_init_sizes(struct radeon_device *rdev)
2801{
2802	u64 config_aper_size;
2803
2804	/* work out accessible VRAM */
2805	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2806	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2807	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2808	/* FIXME we don't use the second aperture yet when we could use it */
2809	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2810		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2811	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2812	if (rdev->flags & RADEON_IS_IGP) {
2813		uint32_t tom;
2814		/* read NB_TOM to get the amount of ram stolen for the GPU */
2815		tom = RREG32(RADEON_NB_TOM);
2816		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2817		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2818		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2819	} else {
2820		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2821		/* Some production boards of m6 will report 0
2822		 * if it's 8 MB
2823		 */
2824		if (rdev->mc.real_vram_size == 0) {
2825			rdev->mc.real_vram_size = 8192 * 1024;
2826			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2827		}
2828		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2829		 * Novell bug 204882 + along with lots of ubuntu ones
2830		 */
2831		if (rdev->mc.aper_size > config_aper_size)
2832			config_aper_size = rdev->mc.aper_size;
2833
2834		if (config_aper_size > rdev->mc.real_vram_size)
2835			rdev->mc.mc_vram_size = config_aper_size;
2836		else
2837			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2838	}
2839}
2840
2841void r100_vga_set_state(struct radeon_device *rdev, bool state)
2842{
2843	uint32_t temp;
2844
2845	temp = RREG32(RADEON_CONFIG_CNTL);
2846	if (!state) {
2847		temp &= ~RADEON_CFG_VGA_RAM_EN;
2848		temp |= RADEON_CFG_VGA_IO_DIS;
2849	} else {
2850		temp &= ~RADEON_CFG_VGA_IO_DIS;
2851	}
2852	WREG32(RADEON_CONFIG_CNTL, temp);
2853}
2854
2855static void r100_mc_init(struct radeon_device *rdev)
2856{
2857	u64 base;
2858
2859	r100_vram_get_type(rdev);
2860	r100_vram_init_sizes(rdev);
2861	base = rdev->mc.aper_base;
2862	if (rdev->flags & RADEON_IS_IGP)
2863		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2864	radeon_vram_location(rdev, &rdev->mc, base);
2865	rdev->mc.gtt_base_align = 0;
2866	if (!(rdev->flags & RADEON_IS_AGP))
2867		radeon_gtt_location(rdev, &rdev->mc);
2868	radeon_update_bandwidth_info(rdev);
2869}
2870
2871
2872/*
2873 * Indirect registers accessor
2874 */
2875void r100_pll_errata_after_index(struct radeon_device *rdev)
2876{
2877	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2878		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2879		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2880	}
2881}
2882
2883static void r100_pll_errata_after_data(struct radeon_device *rdev)
2884{
2885	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2886	 * or the chip could hang on a subsequent access
2887	 */
2888	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2889		mdelay(5);
2890	}
2891
2892	/* This function is required to workaround a hardware bug in some (all?)
2893	 * revisions of the R300.  This workaround should be called after every
2894	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2895	 * may not be correct.
2896	 */
2897	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2898		uint32_t save, tmp;
2899
2900		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2901		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2902		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2903		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2904		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2905	}
2906}
2907
2908uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2909{
2910	unsigned long flags;
2911	uint32_t data;
2912
2913	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2914	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2915	r100_pll_errata_after_index(rdev);
2916	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2917	r100_pll_errata_after_data(rdev);
2918	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2919	return data;
2920}
2921
2922void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2923{
2924	unsigned long flags;
2925
2926	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2927	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2928	r100_pll_errata_after_index(rdev);
2929	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2930	r100_pll_errata_after_data(rdev);
2931	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2932}
2933
2934static void r100_set_safe_registers(struct radeon_device *rdev)
2935{
2936	if (ASIC_IS_RN50(rdev)) {
2937		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2938		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2939	} else if (rdev->family < CHIP_R200) {
2940		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2941		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2942	} else {
2943		r200_set_safe_registers(rdev);
2944	}
2945}
2946
2947/*
2948 * Debugfs info
2949 */
2950#if defined(CONFIG_DEBUG_FS)
2951static int r100_debugfs_rbbm_info_show(struct seq_file *m, void *unused)
2952{
2953	struct radeon_device *rdev = m->private;
 
 
2954	uint32_t reg, value;
2955	unsigned i;
2956
2957	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2958	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2959	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2960	for (i = 0; i < 64; i++) {
2961		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2962		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2963		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2964		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2965		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2966	}
2967	return 0;
2968}
2969
2970static int r100_debugfs_cp_ring_info_show(struct seq_file *m, void *unused)
2971{
2972	struct radeon_device *rdev = m->private;
 
 
2973	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2974	uint32_t rdp, wdp;
2975	unsigned count, i, j;
2976
2977	radeon_ring_free_size(rdev, ring);
2978	rdp = RREG32(RADEON_CP_RB_RPTR);
2979	wdp = RREG32(RADEON_CP_RB_WPTR);
2980	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2981	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2982	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2983	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2984	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2985	seq_printf(m, "%u dwords in ring\n", count);
2986	if (ring->ready) {
2987		for (j = 0; j <= count; j++) {
2988			i = (rdp + j) & ring->ptr_mask;
2989			seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2990		}
2991	}
2992	return 0;
2993}
2994
2995
2996static int r100_debugfs_cp_csq_fifo_show(struct seq_file *m, void *unused)
2997{
2998	struct radeon_device *rdev = m->private;
 
 
2999	uint32_t csq_stat, csq2_stat, tmp;
3000	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
3001	unsigned i;
3002
3003	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
3004	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
3005	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
3006	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
3007	r_rptr = (csq_stat >> 0) & 0x3ff;
3008	r_wptr = (csq_stat >> 10) & 0x3ff;
3009	ib1_rptr = (csq_stat >> 20) & 0x3ff;
3010	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
3011	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
3012	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
3013	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
3014	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
3015	seq_printf(m, "Ring rptr %u\n", r_rptr);
3016	seq_printf(m, "Ring wptr %u\n", r_wptr);
3017	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
3018	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
3019	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
3020	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
3021	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
3022	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
3023	seq_printf(m, "Ring fifo:\n");
3024	for (i = 0; i < 256; i++) {
3025		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3026		tmp = RREG32(RADEON_CP_CSQ_DATA);
3027		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3028	}
3029	seq_printf(m, "Indirect1 fifo:\n");
3030	for (i = 256; i <= 512; i++) {
3031		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3032		tmp = RREG32(RADEON_CP_CSQ_DATA);
3033		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3034	}
3035	seq_printf(m, "Indirect2 fifo:\n");
3036	for (i = 640; i < ib1_wptr; i++) {
3037		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3038		tmp = RREG32(RADEON_CP_CSQ_DATA);
3039		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3040	}
3041	return 0;
3042}
3043
3044static int r100_debugfs_mc_info_show(struct seq_file *m, void *unused)
3045{
3046	struct radeon_device *rdev = m->private;
 
 
3047	uint32_t tmp;
3048
3049	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3050	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3051	tmp = RREG32(RADEON_MC_FB_LOCATION);
3052	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3053	tmp = RREG32(RADEON_BUS_CNTL);
3054	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3055	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3056	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3057	tmp = RREG32(RADEON_AGP_BASE);
3058	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3059	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3060	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3061	tmp = RREG32(0x01D0);
3062	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3063	tmp = RREG32(RADEON_AIC_LO_ADDR);
3064	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3065	tmp = RREG32(RADEON_AIC_HI_ADDR);
3066	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3067	tmp = RREG32(0x01E4);
3068	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3069	return 0;
3070}
3071
3072DEFINE_SHOW_ATTRIBUTE(r100_debugfs_rbbm_info);
3073DEFINE_SHOW_ATTRIBUTE(r100_debugfs_cp_ring_info);
3074DEFINE_SHOW_ATTRIBUTE(r100_debugfs_cp_csq_fifo);
3075DEFINE_SHOW_ATTRIBUTE(r100_debugfs_mc_info);
3076
 
 
 
 
 
 
 
3077#endif
3078
3079void  r100_debugfs_rbbm_init(struct radeon_device *rdev)
3080{
3081#if defined(CONFIG_DEBUG_FS)
3082	struct dentry *root = rdev_to_drm(rdev)->primary->debugfs_root;
3083
3084	debugfs_create_file("r100_rbbm_info", 0444, root, rdev,
3085			    &r100_debugfs_rbbm_info_fops);
3086#endif
3087}
3088
3089void r100_debugfs_cp_init(struct radeon_device *rdev)
3090{
3091#if defined(CONFIG_DEBUG_FS)
3092	struct dentry *root = rdev_to_drm(rdev)->primary->debugfs_root;
3093
3094	debugfs_create_file("r100_cp_ring_info", 0444, root, rdev,
3095			    &r100_debugfs_cp_ring_info_fops);
3096	debugfs_create_file("r100_cp_csq_fifo", 0444, root, rdev,
3097			    &r100_debugfs_cp_csq_fifo_fops);
3098#endif
3099}
3100
3101void  r100_debugfs_mc_info_init(struct radeon_device *rdev)
3102{
3103#if defined(CONFIG_DEBUG_FS)
3104	struct dentry *root = rdev_to_drm(rdev)->primary->debugfs_root;
3105
3106	debugfs_create_file("r100_mc_info", 0444, root, rdev,
3107			    &r100_debugfs_mc_info_fops);
3108#endif
3109}
3110
3111int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3112			 uint32_t tiling_flags, uint32_t pitch,
3113			 uint32_t offset, uint32_t obj_size)
3114{
3115	int surf_index = reg * 16;
3116	int flags = 0;
3117
3118	if (rdev->family <= CHIP_RS200) {
3119		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3120				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3121			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3122		if (tiling_flags & RADEON_TILING_MACRO)
3123			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3124		/* setting pitch to 0 disables tiling */
3125		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3126				== 0)
3127			pitch = 0;
3128	} else if (rdev->family <= CHIP_RV280) {
3129		if (tiling_flags & (RADEON_TILING_MACRO))
3130			flags |= R200_SURF_TILE_COLOR_MACRO;
3131		if (tiling_flags & RADEON_TILING_MICRO)
3132			flags |= R200_SURF_TILE_COLOR_MICRO;
3133	} else {
3134		if (tiling_flags & RADEON_TILING_MACRO)
3135			flags |= R300_SURF_TILE_MACRO;
3136		if (tiling_flags & RADEON_TILING_MICRO)
3137			flags |= R300_SURF_TILE_MICRO;
3138	}
3139
3140	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3141		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3142	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3143		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3144
3145	/* r100/r200 divide by 16 */
3146	if (rdev->family < CHIP_R300)
3147		flags |= pitch / 16;
3148	else
3149		flags |= pitch / 8;
3150
3151
3152	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3153	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3154	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3155	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3156	return 0;
3157}
3158
3159void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3160{
3161	int surf_index = reg * 16;
3162	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3163}
3164
3165void r100_bandwidth_update(struct radeon_device *rdev)
3166{
3167	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3168	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3169	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
3170	fixed20_12 crit_point_ff = {0};
3171	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3172	fixed20_12 memtcas_ff[8] = {
3173		dfixed_init(1),
3174		dfixed_init(2),
3175		dfixed_init(3),
3176		dfixed_init(0),
3177		dfixed_init_half(1),
3178		dfixed_init_half(2),
3179		dfixed_init(0),
3180	};
3181	fixed20_12 memtcas_rs480_ff[8] = {
3182		dfixed_init(0),
3183		dfixed_init(1),
3184		dfixed_init(2),
3185		dfixed_init(3),
3186		dfixed_init(0),
3187		dfixed_init_half(1),
3188		dfixed_init_half(2),
3189		dfixed_init_half(3),
3190	};
3191	fixed20_12 memtcas2_ff[8] = {
3192		dfixed_init(0),
3193		dfixed_init(1),
3194		dfixed_init(2),
3195		dfixed_init(3),
3196		dfixed_init(4),
3197		dfixed_init(5),
3198		dfixed_init(6),
3199		dfixed_init(7),
3200	};
3201	fixed20_12 memtrbs[8] = {
3202		dfixed_init(1),
3203		dfixed_init_half(1),
3204		dfixed_init(2),
3205		dfixed_init_half(2),
3206		dfixed_init(3),
3207		dfixed_init_half(3),
3208		dfixed_init(4),
3209		dfixed_init_half(4)
3210	};
3211	fixed20_12 memtrbs_r4xx[8] = {
3212		dfixed_init(4),
3213		dfixed_init(5),
3214		dfixed_init(6),
3215		dfixed_init(7),
3216		dfixed_init(8),
3217		dfixed_init(9),
3218		dfixed_init(10),
3219		dfixed_init(11)
3220	};
3221	fixed20_12 min_mem_eff;
3222	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3223	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3224	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate = {0},
3225		disp_drain_rate2, read_return_rate;
3226	fixed20_12 time_disp1_drop_priority;
3227	int c;
3228	int cur_size = 16;       /* in octawords */
3229	int critical_point = 0, critical_point2;
3230/* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3231	int stop_req, max_stop_req;
3232	struct drm_display_mode *mode1 = NULL;
3233	struct drm_display_mode *mode2 = NULL;
3234	uint32_t pixel_bytes1 = 0;
3235	uint32_t pixel_bytes2 = 0;
3236
3237	/* Guess line buffer size to be 8192 pixels */
3238	u32 lb_size = 8192;
3239
3240	if (!rdev->mode_info.mode_config_initialized)
3241		return;
3242
3243	radeon_update_display_priority(rdev);
3244
3245	if (rdev->mode_info.crtcs[0]->base.enabled) {
3246		const struct drm_framebuffer *fb =
3247			rdev->mode_info.crtcs[0]->base.primary->fb;
3248
3249		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3250		pixel_bytes1 = fb->format->cpp[0];
3251	}
3252	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3253		if (rdev->mode_info.crtcs[1]->base.enabled) {
3254			const struct drm_framebuffer *fb =
3255				rdev->mode_info.crtcs[1]->base.primary->fb;
3256
3257			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3258			pixel_bytes2 = fb->format->cpp[0];
3259		}
3260	}
3261
3262	min_mem_eff.full = dfixed_const_8(0);
3263	/* get modes */
3264	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3265		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3266		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3267		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3268		/* check crtc enables */
3269		if (mode2)
3270			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3271		if (mode1)
3272			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3273		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3274	}
3275
3276	/*
3277	 * determine is there is enough bw for current mode
3278	 */
3279	sclk_ff = rdev->pm.sclk;
3280	mclk_ff = rdev->pm.mclk;
3281
3282	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3283	temp_ff.full = dfixed_const(temp);
3284	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3285
3286	pix_clk.full = 0;
3287	pix_clk2.full = 0;
3288	peak_disp_bw.full = 0;
3289	if (mode1) {
3290		temp_ff.full = dfixed_const(1000);
3291		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3292		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3293		temp_ff.full = dfixed_const(pixel_bytes1);
3294		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3295	}
3296	if (mode2) {
3297		temp_ff.full = dfixed_const(1000);
3298		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3299		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3300		temp_ff.full = dfixed_const(pixel_bytes2);
3301		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3302	}
3303
3304	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3305	if (peak_disp_bw.full >= mem_bw.full) {
3306		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3307			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3308	}
3309
3310	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3311	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3312	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3313		mem_trcd = ((temp >> 2) & 0x3) + 1;
3314		mem_trp  = ((temp & 0x3)) + 1;
3315		mem_tras = ((temp & 0x70) >> 4) + 1;
3316	} else if (rdev->family == CHIP_R300 ||
3317		   rdev->family == CHIP_R350) { /* r300, r350 */
3318		mem_trcd = (temp & 0x7) + 1;
3319		mem_trp = ((temp >> 8) & 0x7) + 1;
3320		mem_tras = ((temp >> 11) & 0xf) + 4;
3321	} else if (rdev->family == CHIP_RV350 ||
3322		   rdev->family == CHIP_RV380) {
3323		/* rv3x0 */
3324		mem_trcd = (temp & 0x7) + 3;
3325		mem_trp = ((temp >> 8) & 0x7) + 3;
3326		mem_tras = ((temp >> 11) & 0xf) + 6;
3327	} else if (rdev->family == CHIP_R420 ||
3328		   rdev->family == CHIP_R423 ||
3329		   rdev->family == CHIP_RV410) {
3330		/* r4xx */
3331		mem_trcd = (temp & 0xf) + 3;
3332		if (mem_trcd > 15)
3333			mem_trcd = 15;
3334		mem_trp = ((temp >> 8) & 0xf) + 3;
3335		if (mem_trp > 15)
3336			mem_trp = 15;
3337		mem_tras = ((temp >> 12) & 0x1f) + 6;
3338		if (mem_tras > 31)
3339			mem_tras = 31;
3340	} else { /* RV200, R200 */
3341		mem_trcd = (temp & 0x7) + 1;
3342		mem_trp = ((temp >> 8) & 0x7) + 1;
3343		mem_tras = ((temp >> 12) & 0xf) + 4;
3344	}
3345	/* convert to FF */
3346	trcd_ff.full = dfixed_const(mem_trcd);
3347	trp_ff.full = dfixed_const(mem_trp);
3348	tras_ff.full = dfixed_const(mem_tras);
3349
3350	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3351	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3352	data = (temp & (7 << 20)) >> 20;
3353	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3354		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3355			tcas_ff = memtcas_rs480_ff[data];
3356		else
3357			tcas_ff = memtcas_ff[data];
3358	} else
3359		tcas_ff = memtcas2_ff[data];
3360
3361	if (rdev->family == CHIP_RS400 ||
3362	    rdev->family == CHIP_RS480) {
3363		/* extra cas latency stored in bits 23-25 0-4 clocks */
3364		data = (temp >> 23) & 0x7;
3365		if (data < 5)
3366			tcas_ff.full += dfixed_const(data);
3367	}
3368
3369	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3370		/* on the R300, Tcas is included in Trbs.
3371		 */
3372		temp = RREG32(RADEON_MEM_CNTL);
3373		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3374		if (data == 1) {
3375			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3376				temp = RREG32(R300_MC_IND_INDEX);
3377				temp &= ~R300_MC_IND_ADDR_MASK;
3378				temp |= R300_MC_READ_CNTL_CD_mcind;
3379				WREG32(R300_MC_IND_INDEX, temp);
3380				temp = RREG32(R300_MC_IND_DATA);
3381				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3382			} else {
3383				temp = RREG32(R300_MC_READ_CNTL_AB);
3384				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3385			}
3386		} else {
3387			temp = RREG32(R300_MC_READ_CNTL_AB);
3388			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3389		}
3390		if (rdev->family == CHIP_RV410 ||
3391		    rdev->family == CHIP_R420 ||
3392		    rdev->family == CHIP_R423)
3393			trbs_ff = memtrbs_r4xx[data];
3394		else
3395			trbs_ff = memtrbs[data];
3396		tcas_ff.full += trbs_ff.full;
3397	}
3398
3399	sclk_eff_ff.full = sclk_ff.full;
3400
3401	if (rdev->flags & RADEON_IS_AGP) {
3402		fixed20_12 agpmode_ff;
3403		agpmode_ff.full = dfixed_const(radeon_agpmode);
3404		temp_ff.full = dfixed_const_666(16);
3405		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3406	}
3407	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3408
3409	if (ASIC_IS_R300(rdev)) {
3410		sclk_delay_ff.full = dfixed_const(250);
3411	} else {
3412		if ((rdev->family == CHIP_RV100) ||
3413		    rdev->flags & RADEON_IS_IGP) {
3414			if (rdev->mc.vram_is_ddr)
3415				sclk_delay_ff.full = dfixed_const(41);
3416			else
3417				sclk_delay_ff.full = dfixed_const(33);
3418		} else {
3419			if (rdev->mc.vram_width == 128)
3420				sclk_delay_ff.full = dfixed_const(57);
3421			else
3422				sclk_delay_ff.full = dfixed_const(41);
3423		}
3424	}
3425
3426	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3427
3428	if (rdev->mc.vram_is_ddr) {
3429		if (rdev->mc.vram_width == 32) {
3430			k1.full = dfixed_const(40);
3431			c  = 3;
3432		} else {
3433			k1.full = dfixed_const(20);
3434			c  = 1;
3435		}
3436	} else {
3437		k1.full = dfixed_const(40);
3438		c  = 3;
3439	}
3440
3441	temp_ff.full = dfixed_const(2);
3442	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3443	temp_ff.full = dfixed_const(c);
3444	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3445	temp_ff.full = dfixed_const(4);
3446	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3447	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3448	mc_latency_mclk.full += k1.full;
3449
3450	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3451	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3452
3453	/*
3454	  HW cursor time assuming worst case of full size colour cursor.
3455	*/
3456	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3457	temp_ff.full += trcd_ff.full;
3458	if (temp_ff.full < tras_ff.full)
3459		temp_ff.full = tras_ff.full;
3460	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3461
3462	temp_ff.full = dfixed_const(cur_size);
3463	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3464	/*
3465	  Find the total latency for the display data.
3466	*/
3467	disp_latency_overhead.full = dfixed_const(8);
3468	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3469	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3470	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3471
3472	if (mc_latency_mclk.full > mc_latency_sclk.full)
3473		disp_latency.full = mc_latency_mclk.full;
3474	else
3475		disp_latency.full = mc_latency_sclk.full;
3476
3477	/* setup Max GRPH_STOP_REQ default value */
3478	if (ASIC_IS_RV100(rdev))
3479		max_stop_req = 0x5c;
3480	else
3481		max_stop_req = 0x7c;
3482
3483	if (mode1) {
3484		/*  CRTC1
3485		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3486		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3487		*/
3488		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3489
3490		if (stop_req > max_stop_req)
3491			stop_req = max_stop_req;
3492
3493		/*
3494		  Find the drain rate of the display buffer.
3495		*/
3496		temp_ff.full = dfixed_const((16/pixel_bytes1));
3497		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3498
3499		/*
3500		  Find the critical point of the display buffer.
3501		*/
3502		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3503		crit_point_ff.full += dfixed_const_half(0);
3504
3505		critical_point = dfixed_trunc(crit_point_ff);
3506
3507		if (rdev->disp_priority == 2) {
3508			critical_point = 0;
3509		}
3510
3511		/*
3512		  The critical point should never be above max_stop_req-4.  Setting
3513		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3514		*/
3515		if (max_stop_req - critical_point < 4)
3516			critical_point = 0;
3517
3518		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3519			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3520			critical_point = 0x10;
3521		}
3522
3523		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3524		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3525		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3526		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3527		if ((rdev->family == CHIP_R350) &&
3528		    (stop_req > 0x15)) {
3529			stop_req -= 0x10;
3530		}
3531		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3532		temp |= RADEON_GRPH_BUFFER_SIZE;
3533		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3534			  RADEON_GRPH_CRITICAL_AT_SOF |
3535			  RADEON_GRPH_STOP_CNTL);
3536		/*
3537		  Write the result into the register.
3538		*/
3539		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3540						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3541
3542#if 0
3543		if ((rdev->family == CHIP_RS400) ||
3544		    (rdev->family == CHIP_RS480)) {
3545			/* attempt to program RS400 disp regs correctly ??? */
3546			temp = RREG32(RS400_DISP1_REG_CNTL);
3547			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3548				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3549			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3550						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3551						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3552			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3553			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3554				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3555			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3556						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3557						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3558		}
3559#endif
3560
3561		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3562			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3563			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3564	}
3565
3566	if (mode2) {
3567		u32 grph2_cntl;
3568		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3569
3570		if (stop_req > max_stop_req)
3571			stop_req = max_stop_req;
3572
3573		/*
3574		  Find the drain rate of the display buffer.
3575		*/
3576		temp_ff.full = dfixed_const((16/pixel_bytes2));
3577		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3578
3579		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3580		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3581		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3582		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3583		if ((rdev->family == CHIP_R350) &&
3584		    (stop_req > 0x15)) {
3585			stop_req -= 0x10;
3586		}
3587		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3588		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3589		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3590			  RADEON_GRPH_CRITICAL_AT_SOF |
3591			  RADEON_GRPH_STOP_CNTL);
3592
3593		if ((rdev->family == CHIP_RS100) ||
3594		    (rdev->family == CHIP_RS200))
3595			critical_point2 = 0;
3596		else {
3597			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3598			temp_ff.full = dfixed_const(temp);
3599			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3600			if (sclk_ff.full < temp_ff.full)
3601				temp_ff.full = sclk_ff.full;
3602
3603			read_return_rate.full = temp_ff.full;
3604
3605			if (mode1) {
3606				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3607				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3608			} else {
3609				time_disp1_drop_priority.full = 0;
3610			}
3611			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3612			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3613			crit_point_ff.full += dfixed_const_half(0);
3614
3615			critical_point2 = dfixed_trunc(crit_point_ff);
3616
3617			if (rdev->disp_priority == 2) {
3618				critical_point2 = 0;
3619			}
3620
3621			if (max_stop_req - critical_point2 < 4)
3622				critical_point2 = 0;
3623
3624		}
3625
3626		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3627			/* some R300 cards have problem with this set to 0 */
3628			critical_point2 = 0x10;
3629		}
3630
3631		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3632						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3633
3634		if ((rdev->family == CHIP_RS400) ||
3635		    (rdev->family == CHIP_RS480)) {
3636#if 0
3637			/* attempt to program RS400 disp2 regs correctly ??? */
3638			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3639			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3640				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3641			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3642						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3643						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3644			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3645			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3646				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3647			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3648						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3649						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3650#endif
3651			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3652			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3653			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3654			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3655		}
3656
3657		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3658			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3659	}
3660
3661	/* Save number of lines the linebuffer leads before the scanout */
3662	if (mode1)
3663	    rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
3664
3665	if (mode2)
3666	    rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
3667}
3668
3669int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3670{
3671	uint32_t scratch;
3672	uint32_t tmp = 0;
3673	unsigned i;
3674	int r;
3675
3676	r = radeon_scratch_get(rdev, &scratch);
3677	if (r) {
3678		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3679		return r;
3680	}
3681	WREG32(scratch, 0xCAFEDEAD);
3682	r = radeon_ring_lock(rdev, ring, 2);
3683	if (r) {
3684		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3685		radeon_scratch_free(rdev, scratch);
3686		return r;
3687	}
3688	radeon_ring_write(ring, PACKET0(scratch, 0));
3689	radeon_ring_write(ring, 0xDEADBEEF);
3690	radeon_ring_unlock_commit(rdev, ring, false);
3691	for (i = 0; i < rdev->usec_timeout; i++) {
3692		tmp = RREG32(scratch);
3693		if (tmp == 0xDEADBEEF) {
3694			break;
3695		}
3696		udelay(1);
3697	}
3698	if (i < rdev->usec_timeout) {
3699		DRM_INFO("ring test succeeded in %d usecs\n", i);
3700	} else {
3701		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3702			  scratch, tmp);
3703		r = -EINVAL;
3704	}
3705	radeon_scratch_free(rdev, scratch);
3706	return r;
3707}
3708
3709void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3710{
3711	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3712
3713	if (ring->rptr_save_reg) {
3714		u32 next_rptr = ring->wptr + 2 + 3;
3715		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3716		radeon_ring_write(ring, next_rptr);
3717	}
3718
3719	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3720	radeon_ring_write(ring, ib->gpu_addr);
3721	radeon_ring_write(ring, ib->length_dw);
3722}
3723
3724int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3725{
3726	struct radeon_ib ib;
3727	uint32_t scratch;
3728	uint32_t tmp = 0;
3729	unsigned i;
3730	int r;
3731
3732	r = radeon_scratch_get(rdev, &scratch);
3733	if (r) {
3734		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3735		return r;
3736	}
3737	WREG32(scratch, 0xCAFEDEAD);
3738	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3739	if (r) {
3740		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3741		goto free_scratch;
3742	}
3743	ib.ptr[0] = PACKET0(scratch, 0);
3744	ib.ptr[1] = 0xDEADBEEF;
3745	ib.ptr[2] = PACKET2(0);
3746	ib.ptr[3] = PACKET2(0);
3747	ib.ptr[4] = PACKET2(0);
3748	ib.ptr[5] = PACKET2(0);
3749	ib.ptr[6] = PACKET2(0);
3750	ib.ptr[7] = PACKET2(0);
3751	ib.length_dw = 8;
3752	r = radeon_ib_schedule(rdev, &ib, NULL, false);
3753	if (r) {
3754		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3755		goto free_ib;
3756	}
3757	r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
3758		RADEON_USEC_IB_TEST_TIMEOUT));
3759	if (r < 0) {
3760		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3761		goto free_ib;
3762	} else if (r == 0) {
3763		DRM_ERROR("radeon: fence wait timed out.\n");
3764		r = -ETIMEDOUT;
3765		goto free_ib;
3766	}
3767	r = 0;
3768	for (i = 0; i < rdev->usec_timeout; i++) {
3769		tmp = RREG32(scratch);
3770		if (tmp == 0xDEADBEEF) {
3771			break;
3772		}
3773		udelay(1);
3774	}
3775	if (i < rdev->usec_timeout) {
3776		DRM_INFO("ib test succeeded in %u usecs\n", i);
3777	} else {
3778		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3779			  scratch, tmp);
3780		r = -EINVAL;
3781	}
3782free_ib:
3783	radeon_ib_free(rdev, &ib);
3784free_scratch:
3785	radeon_scratch_free(rdev, scratch);
3786	return r;
3787}
3788
3789void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3790{
3791	/* Shutdown CP we shouldn't need to do that but better be safe than
3792	 * sorry
3793	 */
3794	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3795	WREG32(R_000740_CP_CSQ_CNTL, 0);
3796
3797	/* Save few CRTC registers */
3798	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3799	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3800	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3801	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3802	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3803		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3804		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3805	}
3806
3807	/* Disable VGA aperture access */
3808	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3809	/* Disable cursor, overlay, crtc */
3810	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3811	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3812					S_000054_CRTC_DISPLAY_DIS(1));
3813	WREG32(R_000050_CRTC_GEN_CNTL,
3814			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3815			S_000050_CRTC_DISP_REQ_EN_B(1));
3816	WREG32(R_000420_OV0_SCALE_CNTL,
3817		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3818	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3819	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3820		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3821						S_000360_CUR2_LOCK(1));
3822		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3823			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3824			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3825			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3826		WREG32(R_000360_CUR2_OFFSET,
3827			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3828	}
3829}
3830
3831void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3832{
3833	/* Update base address for crtc */
3834	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3835	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3836		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3837	}
3838	/* Restore CRTC registers */
3839	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3840	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3841	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3842	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3843		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3844	}
3845}
3846
3847void r100_vga_render_disable(struct radeon_device *rdev)
3848{
3849	u32 tmp;
3850
3851	tmp = RREG8(R_0003C2_GENMO_WT);
3852	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3853}
3854
 
 
 
 
 
 
 
 
 
3855static void r100_mc_program(struct radeon_device *rdev)
3856{
3857	struct r100_mc_save save;
3858
3859	/* Stops all mc clients */
3860	r100_mc_stop(rdev, &save);
3861	if (rdev->flags & RADEON_IS_AGP) {
3862		WREG32(R_00014C_MC_AGP_LOCATION,
3863			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3864			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3865		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3866		if (rdev->family > CHIP_RV200)
3867			WREG32(R_00015C_AGP_BASE_2,
3868				upper_32_bits(rdev->mc.agp_base) & 0xff);
3869	} else {
3870		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3871		WREG32(R_000170_AGP_BASE, 0);
3872		if (rdev->family > CHIP_RV200)
3873			WREG32(R_00015C_AGP_BASE_2, 0);
3874	}
3875	/* Wait for mc idle */
3876	if (r100_mc_wait_for_idle(rdev))
3877		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3878	/* Program MC, should be a 32bits limited address space */
3879	WREG32(R_000148_MC_FB_LOCATION,
3880		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3881		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3882	r100_mc_resume(rdev, &save);
3883}
3884
3885static void r100_clock_startup(struct radeon_device *rdev)
3886{
3887	u32 tmp;
3888
3889	if (radeon_dynclks != -1 && radeon_dynclks)
3890		radeon_legacy_set_clock_gating(rdev, 1);
3891	/* We need to force on some of the block */
3892	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3893	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3894	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3895		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3896	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3897}
3898
3899static int r100_startup(struct radeon_device *rdev)
3900{
3901	int r;
3902
3903	/* set common regs */
3904	r100_set_common_regs(rdev);
3905	/* program mc */
3906	r100_mc_program(rdev);
3907	/* Resume clock */
3908	r100_clock_startup(rdev);
3909	/* Initialize GART (initialize after TTM so we can allocate
3910	 * memory through TTM but finalize after TTM) */
3911	r100_enable_bm(rdev);
3912	if (rdev->flags & RADEON_IS_PCI) {
3913		r = r100_pci_gart_enable(rdev);
3914		if (r)
3915			return r;
3916	}
3917
3918	/* allocate wb buffer */
3919	r = radeon_wb_init(rdev);
3920	if (r)
3921		return r;
3922
3923	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3924	if (r) {
3925		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3926		return r;
3927	}
3928
3929	/* Enable IRQ */
3930	if (!rdev->irq.installed) {
3931		r = radeon_irq_kms_init(rdev);
3932		if (r)
3933			return r;
3934	}
3935
3936	r100_irq_set(rdev);
3937	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3938	/* 1M ring buffer */
3939	r = r100_cp_init(rdev, 1024 * 1024);
3940	if (r) {
3941		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3942		return r;
3943	}
3944
3945	r = radeon_ib_pool_init(rdev);
3946	if (r) {
3947		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3948		return r;
3949	}
3950
3951	return 0;
3952}
3953
3954int r100_resume(struct radeon_device *rdev)
3955{
3956	int r;
3957
3958	/* Make sur GART are not working */
3959	if (rdev->flags & RADEON_IS_PCI)
3960		r100_pci_gart_disable(rdev);
3961	/* Resume clock before doing reset */
3962	r100_clock_startup(rdev);
3963	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3964	if (radeon_asic_reset(rdev)) {
3965		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3966			RREG32(R_000E40_RBBM_STATUS),
3967			RREG32(R_0007C0_CP_STAT));
3968	}
3969	/* post */
3970	radeon_combios_asic_init(rdev_to_drm(rdev));
3971	/* Resume clock after posting */
3972	r100_clock_startup(rdev);
3973	/* Initialize surface registers */
3974	radeon_surface_init(rdev);
3975
3976	rdev->accel_working = true;
3977	r = r100_startup(rdev);
3978	if (r) {
3979		rdev->accel_working = false;
3980	}
3981	return r;
3982}
3983
3984int r100_suspend(struct radeon_device *rdev)
3985{
3986	radeon_pm_suspend(rdev);
3987	r100_cp_disable(rdev);
3988	radeon_wb_disable(rdev);
3989	r100_irq_disable(rdev);
3990	if (rdev->flags & RADEON_IS_PCI)
3991		r100_pci_gart_disable(rdev);
3992	return 0;
3993}
3994
3995void r100_fini(struct radeon_device *rdev)
3996{
3997	radeon_pm_fini(rdev);
3998	r100_cp_fini(rdev);
3999	radeon_wb_fini(rdev);
4000	radeon_ib_pool_fini(rdev);
4001	radeon_gem_fini(rdev);
4002	if (rdev->flags & RADEON_IS_PCI)
4003		r100_pci_gart_fini(rdev);
4004	radeon_agp_fini(rdev);
4005	radeon_irq_kms_fini(rdev);
4006	radeon_fence_driver_fini(rdev);
4007	radeon_bo_fini(rdev);
4008	radeon_atombios_fini(rdev);
4009	kfree(rdev->bios);
4010	rdev->bios = NULL;
4011}
4012
4013/*
4014 * Due to how kexec works, it can leave the hw fully initialised when it
4015 * boots the new kernel. However doing our init sequence with the CP and
4016 * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
4017 * do some quick sanity checks and restore sane values to avoid this
4018 * problem.
4019 */
4020void r100_restore_sanity(struct radeon_device *rdev)
4021{
4022	u32 tmp;
4023
4024	tmp = RREG32(RADEON_CP_CSQ_CNTL);
4025	if (tmp) {
4026		WREG32(RADEON_CP_CSQ_CNTL, 0);
4027	}
4028	tmp = RREG32(RADEON_CP_RB_CNTL);
4029	if (tmp) {
4030		WREG32(RADEON_CP_RB_CNTL, 0);
4031	}
4032	tmp = RREG32(RADEON_SCRATCH_UMSK);
4033	if (tmp) {
4034		WREG32(RADEON_SCRATCH_UMSK, 0);
4035	}
4036}
4037
4038int r100_init(struct radeon_device *rdev)
4039{
4040	int r;
4041
4042	/* Register debugfs file specific to this group of asics */
4043	r100_debugfs_mc_info_init(rdev);
4044	/* Disable VGA */
4045	r100_vga_render_disable(rdev);
4046	/* Initialize scratch registers */
4047	radeon_scratch_init(rdev);
4048	/* Initialize surface registers */
4049	radeon_surface_init(rdev);
4050	/* sanity check some register to avoid hangs like after kexec */
4051	r100_restore_sanity(rdev);
4052	/* TODO: disable VGA need to use VGA request */
4053	/* BIOS*/
4054	if (!radeon_get_bios(rdev)) {
4055		if (ASIC_IS_AVIVO(rdev))
4056			return -EINVAL;
4057	}
4058	if (rdev->is_atom_bios) {
4059		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4060		return -EINVAL;
4061	} else {
4062		r = radeon_combios_init(rdev);
4063		if (r)
4064			return r;
4065	}
4066	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4067	if (radeon_asic_reset(rdev)) {
4068		dev_warn(rdev->dev,
4069			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4070			RREG32(R_000E40_RBBM_STATUS),
4071			RREG32(R_0007C0_CP_STAT));
4072	}
4073	/* check if cards are posted or not */
4074	if (radeon_boot_test_post_card(rdev) == false)
4075		return -EINVAL;
4076	/* Set asic errata */
4077	r100_errata(rdev);
4078	/* Initialize clocks */
4079	radeon_get_clock_info(rdev_to_drm(rdev));
4080	/* initialize AGP */
4081	if (rdev->flags & RADEON_IS_AGP) {
4082		r = radeon_agp_init(rdev);
4083		if (r) {
4084			radeon_agp_disable(rdev);
4085		}
4086	}
4087	/* initialize VRAM */
4088	r100_mc_init(rdev);
4089	/* Fence driver */
4090	radeon_fence_driver_init(rdev);
 
 
4091	/* Memory manager */
4092	r = radeon_bo_init(rdev);
4093	if (r)
4094		return r;
4095	if (rdev->flags & RADEON_IS_PCI) {
4096		r = r100_pci_gart_init(rdev);
4097		if (r)
4098			return r;
4099	}
4100	r100_set_safe_registers(rdev);
4101
4102	/* Initialize power management */
4103	radeon_pm_init(rdev);
4104
4105	rdev->accel_working = true;
4106	r = r100_startup(rdev);
4107	if (r) {
4108		/* Somethings want wront with the accel init stop accel */
4109		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4110		r100_cp_fini(rdev);
4111		radeon_wb_fini(rdev);
4112		radeon_ib_pool_fini(rdev);
4113		radeon_irq_kms_fini(rdev);
4114		if (rdev->flags & RADEON_IS_PCI)
4115			r100_pci_gart_fini(rdev);
4116		rdev->accel_working = false;
4117	}
4118	return 0;
4119}
4120
4121uint32_t r100_mm_rreg_slow(struct radeon_device *rdev, uint32_t reg)
4122{
4123	unsigned long flags;
4124	uint32_t ret;
4125
4126	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4127	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4128	ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4129	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4130	return ret;
4131}
4132
4133void r100_mm_wreg_slow(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4134{
4135	unsigned long flags;
4136
4137	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4138	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4139	writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4140	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4141}
4142
4143u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4144{
4145	if (reg < rdev->rio_mem_size)
4146		return ioread32(rdev->rio_mem + reg);
4147	else {
4148		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4149		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4150	}
4151}
4152
4153void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4154{
4155	if (reg < rdev->rio_mem_size)
4156		iowrite32(v, rdev->rio_mem + reg);
4157	else {
4158		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4159		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4160	}
4161}
v5.9
   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
 
  29#include <linux/firmware.h>
  30#include <linux/module.h>
  31#include <linux/pci.h>
  32#include <linux/seq_file.h>
  33#include <linux/slab.h>
  34
  35#include <drm/drm_debugfs.h>
  36#include <drm/drm_device.h>
  37#include <drm/drm_file.h>
  38#include <drm/drm_fourcc.h>
 
  39#include <drm/drm_vblank.h>
  40#include <drm/radeon_drm.h>
  41
  42#include "atom.h"
  43#include "r100_reg_safe.h"
  44#include "r100d.h"
  45#include "radeon.h"
  46#include "radeon_asic.h"
  47#include "radeon_reg.h"
  48#include "rn50_reg_safe.h"
  49#include "rs100d.h"
  50#include "rv200d.h"
  51#include "rv250d.h"
  52
  53/* Firmware Names */
  54#define FIRMWARE_R100		"radeon/R100_cp.bin"
  55#define FIRMWARE_R200		"radeon/R200_cp.bin"
  56#define FIRMWARE_R300		"radeon/R300_cp.bin"
  57#define FIRMWARE_R420		"radeon/R420_cp.bin"
  58#define FIRMWARE_RS690		"radeon/RS690_cp.bin"
  59#define FIRMWARE_RS600		"radeon/RS600_cp.bin"
  60#define FIRMWARE_R520		"radeon/R520_cp.bin"
  61
  62MODULE_FIRMWARE(FIRMWARE_R100);
  63MODULE_FIRMWARE(FIRMWARE_R200);
  64MODULE_FIRMWARE(FIRMWARE_R300);
  65MODULE_FIRMWARE(FIRMWARE_R420);
  66MODULE_FIRMWARE(FIRMWARE_RS690);
  67MODULE_FIRMWARE(FIRMWARE_RS600);
  68MODULE_FIRMWARE(FIRMWARE_R520);
  69
  70#include "r100_track.h"
  71
  72/* This files gather functions specifics to:
  73 * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
  74 * and others in some cases.
  75 */
  76
  77static bool r100_is_in_vblank(struct radeon_device *rdev, int crtc)
  78{
  79	if (crtc == 0) {
  80		if (RREG32(RADEON_CRTC_STATUS) & RADEON_CRTC_VBLANK_CUR)
  81			return true;
  82		else
  83			return false;
  84	} else {
  85		if (RREG32(RADEON_CRTC2_STATUS) & RADEON_CRTC2_VBLANK_CUR)
  86			return true;
  87		else
  88			return false;
  89	}
  90}
  91
  92static bool r100_is_counter_moving(struct radeon_device *rdev, int crtc)
  93{
  94	u32 vline1, vline2;
  95
  96	if (crtc == 0) {
  97		vline1 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
  98		vline2 = (RREG32(RADEON_CRTC_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
  99	} else {
 100		vline1 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
 101		vline2 = (RREG32(RADEON_CRTC2_VLINE_CRNT_VLINE) >> 16) & RADEON_CRTC_V_TOTAL;
 102	}
 103	if (vline1 != vline2)
 104		return true;
 105	else
 106		return false;
 107}
 108
 109/**
 110 * r100_wait_for_vblank - vblank wait asic callback.
 111 *
 112 * @rdev: radeon_device pointer
 113 * @crtc: crtc to wait for vblank on
 114 *
 115 * Wait for vblank on the requested crtc (r1xx-r4xx).
 116 */
 117void r100_wait_for_vblank(struct radeon_device *rdev, int crtc)
 118{
 119	unsigned i = 0;
 120
 121	if (crtc >= rdev->num_crtc)
 122		return;
 123
 124	if (crtc == 0) {
 125		if (!(RREG32(RADEON_CRTC_GEN_CNTL) & RADEON_CRTC_EN))
 126			return;
 127	} else {
 128		if (!(RREG32(RADEON_CRTC2_GEN_CNTL) & RADEON_CRTC2_EN))
 129			return;
 130	}
 131
 132	/* depending on when we hit vblank, we may be close to active; if so,
 133	 * wait for another frame.
 134	 */
 135	while (r100_is_in_vblank(rdev, crtc)) {
 136		if (i++ % 100 == 0) {
 137			if (!r100_is_counter_moving(rdev, crtc))
 138				break;
 139		}
 140	}
 141
 142	while (!r100_is_in_vblank(rdev, crtc)) {
 143		if (i++ % 100 == 0) {
 144			if (!r100_is_counter_moving(rdev, crtc))
 145				break;
 146		}
 147	}
 148}
 149
 150/**
 151 * r100_page_flip - pageflip callback.
 152 *
 153 * @rdev: radeon_device pointer
 154 * @crtc_id: crtc to cleanup pageflip on
 155 * @crtc_base: new address of the crtc (GPU MC address)
 
 156 *
 157 * Does the actual pageflip (r1xx-r4xx).
 158 * During vblank we take the crtc lock and wait for the update_pending
 159 * bit to go high, when it does, we release the lock, and allow the
 160 * double buffered update to take place.
 161 */
 162void r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base, bool async)
 163{
 164	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
 
 
 165	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
 166	int i;
 167
 168	/* Lock the graphics update lock */
 169	/* update the scanout addresses */
 170	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 171
 
 
 
 
 
 
 
 172	/* Wait for update_pending to go high. */
 173	for (i = 0; i < rdev->usec_timeout; i++) {
 174		if (RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET)
 175			break;
 176		udelay(1);
 177	}
 178	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
 179
 180	/* Unlock the lock, so double-buffering can take place inside vblank */
 181	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
 182	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
 183
 184}
 185
 186/**
 187 * r100_page_flip_pending - check if page flip is still pending
 188 *
 189 * @rdev: radeon_device pointer
 190 * @crtc_id: crtc to check
 191 *
 192 * Check if the last pagefilp is still pending (r1xx-r4xx).
 193 * Returns the current update pending status.
 194 */
 195bool r100_page_flip_pending(struct radeon_device *rdev, int crtc_id)
 196{
 197	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
 198
 199	/* Return current update_pending status: */
 200	return !!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) &
 201		RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET);
 202}
 203
 204/**
 205 * r100_pm_get_dynpm_state - look up dynpm power state callback.
 206 *
 207 * @rdev: radeon_device pointer
 208 *
 209 * Look up the optimal power state based on the
 210 * current state of the GPU (r1xx-r5xx).
 211 * Used for dynpm only.
 212 */
 213void r100_pm_get_dynpm_state(struct radeon_device *rdev)
 214{
 215	int i;
 216	rdev->pm.dynpm_can_upclock = true;
 217	rdev->pm.dynpm_can_downclock = true;
 218
 219	switch (rdev->pm.dynpm_planned_action) {
 220	case DYNPM_ACTION_MINIMUM:
 221		rdev->pm.requested_power_state_index = 0;
 222		rdev->pm.dynpm_can_downclock = false;
 223		break;
 224	case DYNPM_ACTION_DOWNCLOCK:
 225		if (rdev->pm.current_power_state_index == 0) {
 226			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 227			rdev->pm.dynpm_can_downclock = false;
 228		} else {
 229			if (rdev->pm.active_crtc_count > 1) {
 230				for (i = 0; i < rdev->pm.num_power_states; i++) {
 231					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 232						continue;
 233					else if (i >= rdev->pm.current_power_state_index) {
 234						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 235						break;
 236					} else {
 237						rdev->pm.requested_power_state_index = i;
 238						break;
 239					}
 240				}
 241			} else
 242				rdev->pm.requested_power_state_index =
 243					rdev->pm.current_power_state_index - 1;
 244		}
 245		/* don't use the power state if crtcs are active and no display flag is set */
 246		if ((rdev->pm.active_crtc_count > 0) &&
 247		    (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
 248		     RADEON_PM_MODE_NO_DISPLAY)) {
 249			rdev->pm.requested_power_state_index++;
 250		}
 251		break;
 252	case DYNPM_ACTION_UPCLOCK:
 253		if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
 254			rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 255			rdev->pm.dynpm_can_upclock = false;
 256		} else {
 257			if (rdev->pm.active_crtc_count > 1) {
 258				for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
 259					if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
 260						continue;
 261					else if (i <= rdev->pm.current_power_state_index) {
 262						rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
 263						break;
 264					} else {
 265						rdev->pm.requested_power_state_index = i;
 266						break;
 267					}
 268				}
 269			} else
 270				rdev->pm.requested_power_state_index =
 271					rdev->pm.current_power_state_index + 1;
 272		}
 273		break;
 274	case DYNPM_ACTION_DEFAULT:
 275		rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
 276		rdev->pm.dynpm_can_upclock = false;
 277		break;
 278	case DYNPM_ACTION_NONE:
 279	default:
 280		DRM_ERROR("Requested mode for not defined action\n");
 281		return;
 282	}
 283	/* only one clock mode per power state */
 284	rdev->pm.requested_clock_mode_index = 0;
 285
 286	DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
 287		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 288		  clock_info[rdev->pm.requested_clock_mode_index].sclk,
 289		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 290		  clock_info[rdev->pm.requested_clock_mode_index].mclk,
 291		  rdev->pm.power_state[rdev->pm.requested_power_state_index].
 292		  pcie_lanes);
 293}
 294
 295/**
 296 * r100_pm_init_profile - Initialize power profiles callback.
 297 *
 298 * @rdev: radeon_device pointer
 299 *
 300 * Initialize the power states used in profile mode
 301 * (r1xx-r3xx).
 302 * Used for profile mode only.
 303 */
 304void r100_pm_init_profile(struct radeon_device *rdev)
 305{
 306	/* default */
 307	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
 308	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 309	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
 310	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
 311	/* low sh */
 312	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
 313	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
 314	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
 315	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
 316	/* mid sh */
 317	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
 318	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
 319	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
 320	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
 321	/* high sh */
 322	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
 323	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 324	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
 325	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
 326	/* low mh */
 327	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
 328	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 329	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
 330	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
 331	/* mid mh */
 332	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
 333	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 334	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
 335	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
 336	/* high mh */
 337	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
 338	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
 339	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
 340	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
 341}
 342
 343/**
 344 * r100_pm_misc - set additional pm hw parameters callback.
 345 *
 346 * @rdev: radeon_device pointer
 347 *
 348 * Set non-clock parameters associated with a power state
 349 * (voltage, pcie lanes, etc.) (r1xx-r4xx).
 350 */
 351void r100_pm_misc(struct radeon_device *rdev)
 352{
 353	int requested_index = rdev->pm.requested_power_state_index;
 354	struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
 355	struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
 356	u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
 357
 358	if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
 359		if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
 360			tmp = RREG32(voltage->gpio.reg);
 361			if (voltage->active_high)
 362				tmp |= voltage->gpio.mask;
 363			else
 364				tmp &= ~(voltage->gpio.mask);
 365			WREG32(voltage->gpio.reg, tmp);
 366			if (voltage->delay)
 367				udelay(voltage->delay);
 368		} else {
 369			tmp = RREG32(voltage->gpio.reg);
 370			if (voltage->active_high)
 371				tmp &= ~voltage->gpio.mask;
 372			else
 373				tmp |= voltage->gpio.mask;
 374			WREG32(voltage->gpio.reg, tmp);
 375			if (voltage->delay)
 376				udelay(voltage->delay);
 377		}
 378	}
 379
 380	sclk_cntl = RREG32_PLL(SCLK_CNTL);
 381	sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
 382	sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
 383	sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
 384	sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
 385	if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
 386		sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
 387		if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
 388			sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
 389		else
 390			sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
 391		if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
 392			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
 393		else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
 394			sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
 395	} else
 396		sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
 397
 398	if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
 399		sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
 400		if (voltage->delay) {
 401			sclk_more_cntl |= VOLTAGE_DROP_SYNC;
 402			switch (voltage->delay) {
 403			case 33:
 404				sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
 405				break;
 406			case 66:
 407				sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
 408				break;
 409			case 99:
 410				sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
 411				break;
 412			case 132:
 413				sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
 414				break;
 415			}
 416		} else
 417			sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
 418	} else
 419		sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
 420
 421	if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
 422		sclk_cntl &= ~FORCE_HDP;
 423	else
 424		sclk_cntl |= FORCE_HDP;
 425
 426	WREG32_PLL(SCLK_CNTL, sclk_cntl);
 427	WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
 428	WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
 429
 430	/* set pcie lanes */
 431	if ((rdev->flags & RADEON_IS_PCIE) &&
 432	    !(rdev->flags & RADEON_IS_IGP) &&
 433	    rdev->asic->pm.set_pcie_lanes &&
 434	    (ps->pcie_lanes !=
 435	     rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
 436		radeon_set_pcie_lanes(rdev,
 437				      ps->pcie_lanes);
 438		DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
 439	}
 440}
 441
 442/**
 443 * r100_pm_prepare - pre-power state change callback.
 444 *
 445 * @rdev: radeon_device pointer
 446 *
 447 * Prepare for a power state change (r1xx-r4xx).
 448 */
 449void r100_pm_prepare(struct radeon_device *rdev)
 450{
 451	struct drm_device *ddev = rdev->ddev;
 452	struct drm_crtc *crtc;
 453	struct radeon_crtc *radeon_crtc;
 454	u32 tmp;
 455
 456	/* disable any active CRTCs */
 457	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 458		radeon_crtc = to_radeon_crtc(crtc);
 459		if (radeon_crtc->enabled) {
 460			if (radeon_crtc->crtc_id) {
 461				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 462				tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
 463				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 464			} else {
 465				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 466				tmp |= RADEON_CRTC_DISP_REQ_EN_B;
 467				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 468			}
 469		}
 470	}
 471}
 472
 473/**
 474 * r100_pm_finish - post-power state change callback.
 475 *
 476 * @rdev: radeon_device pointer
 477 *
 478 * Clean up after a power state change (r1xx-r4xx).
 479 */
 480void r100_pm_finish(struct radeon_device *rdev)
 481{
 482	struct drm_device *ddev = rdev->ddev;
 483	struct drm_crtc *crtc;
 484	struct radeon_crtc *radeon_crtc;
 485	u32 tmp;
 486
 487	/* enable any active CRTCs */
 488	list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
 489		radeon_crtc = to_radeon_crtc(crtc);
 490		if (radeon_crtc->enabled) {
 491			if (radeon_crtc->crtc_id) {
 492				tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
 493				tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
 494				WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
 495			} else {
 496				tmp = RREG32(RADEON_CRTC_GEN_CNTL);
 497				tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
 498				WREG32(RADEON_CRTC_GEN_CNTL, tmp);
 499			}
 500		}
 501	}
 502}
 503
 504/**
 505 * r100_gui_idle - gui idle callback.
 506 *
 507 * @rdev: radeon_device pointer
 508 *
 509 * Check of the GUI (2D/3D engines) are idle (r1xx-r5xx).
 510 * Returns true if idle, false if not.
 511 */
 512bool r100_gui_idle(struct radeon_device *rdev)
 513{
 514	if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
 515		return false;
 516	else
 517		return true;
 518}
 519
 520/* hpd for digital panel detect/disconnect */
 521/**
 522 * r100_hpd_sense - hpd sense callback.
 523 *
 524 * @rdev: radeon_device pointer
 525 * @hpd: hpd (hotplug detect) pin
 526 *
 527 * Checks if a digital monitor is connected (r1xx-r4xx).
 528 * Returns true if connected, false if not connected.
 529 */
 530bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
 531{
 532	bool connected = false;
 533
 534	switch (hpd) {
 535	case RADEON_HPD_1:
 536		if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
 537			connected = true;
 538		break;
 539	case RADEON_HPD_2:
 540		if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
 541			connected = true;
 542		break;
 543	default:
 544		break;
 545	}
 546	return connected;
 547}
 548
 549/**
 550 * r100_hpd_set_polarity - hpd set polarity callback.
 551 *
 552 * @rdev: radeon_device pointer
 553 * @hpd: hpd (hotplug detect) pin
 554 *
 555 * Set the polarity of the hpd pin (r1xx-r4xx).
 556 */
 557void r100_hpd_set_polarity(struct radeon_device *rdev,
 558			   enum radeon_hpd_id hpd)
 559{
 560	u32 tmp;
 561	bool connected = r100_hpd_sense(rdev, hpd);
 562
 563	switch (hpd) {
 564	case RADEON_HPD_1:
 565		tmp = RREG32(RADEON_FP_GEN_CNTL);
 566		if (connected)
 567			tmp &= ~RADEON_FP_DETECT_INT_POL;
 568		else
 569			tmp |= RADEON_FP_DETECT_INT_POL;
 570		WREG32(RADEON_FP_GEN_CNTL, tmp);
 571		break;
 572	case RADEON_HPD_2:
 573		tmp = RREG32(RADEON_FP2_GEN_CNTL);
 574		if (connected)
 575			tmp &= ~RADEON_FP2_DETECT_INT_POL;
 576		else
 577			tmp |= RADEON_FP2_DETECT_INT_POL;
 578		WREG32(RADEON_FP2_GEN_CNTL, tmp);
 579		break;
 580	default:
 581		break;
 582	}
 583}
 584
 585/**
 586 * r100_hpd_init - hpd setup callback.
 587 *
 588 * @rdev: radeon_device pointer
 589 *
 590 * Setup the hpd pins used by the card (r1xx-r4xx).
 591 * Set the polarity, and enable the hpd interrupts.
 592 */
 593void r100_hpd_init(struct radeon_device *rdev)
 594{
 595	struct drm_device *dev = rdev->ddev;
 596	struct drm_connector *connector;
 597	unsigned enable = 0;
 598
 599	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 600		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 601		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
 602			enable |= 1 << radeon_connector->hpd.hpd;
 603		radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
 604	}
 605	radeon_irq_kms_enable_hpd(rdev, enable);
 606}
 607
 608/**
 609 * r100_hpd_fini - hpd tear down callback.
 610 *
 611 * @rdev: radeon_device pointer
 612 *
 613 * Tear down the hpd pins used by the card (r1xx-r4xx).
 614 * Disable the hpd interrupts.
 615 */
 616void r100_hpd_fini(struct radeon_device *rdev)
 617{
 618	struct drm_device *dev = rdev->ddev;
 619	struct drm_connector *connector;
 620	unsigned disable = 0;
 621
 622	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 623		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 624		if (radeon_connector->hpd.hpd != RADEON_HPD_NONE)
 625			disable |= 1 << radeon_connector->hpd.hpd;
 626	}
 627	radeon_irq_kms_disable_hpd(rdev, disable);
 628}
 629
 630/*
 631 * PCI GART
 632 */
 633void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
 634{
 635	/* TODO: can we do somethings here ? */
 636	/* It seems hw only cache one entry so we should discard this
 637	 * entry otherwise if first GPU GART read hit this entry it
 638	 * could end up in wrong address. */
 639}
 640
 641int r100_pci_gart_init(struct radeon_device *rdev)
 642{
 643	int r;
 644
 645	if (rdev->gart.ptr) {
 646		WARN(1, "R100 PCI GART already initialized\n");
 647		return 0;
 648	}
 649	/* Initialize common gart structure */
 650	r = radeon_gart_init(rdev);
 651	if (r)
 652		return r;
 653	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 654	rdev->asic->gart.tlb_flush = &r100_pci_gart_tlb_flush;
 655	rdev->asic->gart.get_page_entry = &r100_pci_gart_get_page_entry;
 656	rdev->asic->gart.set_page = &r100_pci_gart_set_page;
 657	return radeon_gart_table_ram_alloc(rdev);
 658}
 659
 660int r100_pci_gart_enable(struct radeon_device *rdev)
 661{
 662	uint32_t tmp;
 663
 664	/* discard memory request outside of configured range */
 665	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 666	WREG32(RADEON_AIC_CNTL, tmp);
 667	/* set address range for PCI address translate */
 668	WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
 669	WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
 670	/* set PCI GART page-table base address */
 671	WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
 672	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
 673	WREG32(RADEON_AIC_CNTL, tmp);
 674	r100_pci_gart_tlb_flush(rdev);
 675	DRM_INFO("PCI GART of %uM enabled (table at 0x%016llX).\n",
 676		 (unsigned)(rdev->mc.gtt_size >> 20),
 677		 (unsigned long long)rdev->gart.table_addr);
 678	rdev->gart.ready = true;
 679	return 0;
 680}
 681
 682void r100_pci_gart_disable(struct radeon_device *rdev)
 683{
 684	uint32_t tmp;
 685
 686	/* discard memory request outside of configured range */
 687	tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 688	WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
 689	WREG32(RADEON_AIC_LO_ADDR, 0);
 690	WREG32(RADEON_AIC_HI_ADDR, 0);
 691}
 692
 693uint64_t r100_pci_gart_get_page_entry(uint64_t addr, uint32_t flags)
 694{
 695	return addr;
 696}
 697
 698void r100_pci_gart_set_page(struct radeon_device *rdev, unsigned i,
 699			    uint64_t entry)
 700{
 701	u32 *gtt = rdev->gart.ptr;
 702	gtt[i] = cpu_to_le32(lower_32_bits(entry));
 703}
 704
 705void r100_pci_gart_fini(struct radeon_device *rdev)
 706{
 707	radeon_gart_fini(rdev);
 708	r100_pci_gart_disable(rdev);
 709	radeon_gart_table_ram_free(rdev);
 710}
 711
 712int r100_irq_set(struct radeon_device *rdev)
 713{
 714	uint32_t tmp = 0;
 715
 716	if (!rdev->irq.installed) {
 717		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
 718		WREG32(R_000040_GEN_INT_CNTL, 0);
 719		return -EINVAL;
 720	}
 721	if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
 722		tmp |= RADEON_SW_INT_ENABLE;
 723	}
 724	if (rdev->irq.crtc_vblank_int[0] ||
 725	    atomic_read(&rdev->irq.pflip[0])) {
 726		tmp |= RADEON_CRTC_VBLANK_MASK;
 727	}
 728	if (rdev->irq.crtc_vblank_int[1] ||
 729	    atomic_read(&rdev->irq.pflip[1])) {
 730		tmp |= RADEON_CRTC2_VBLANK_MASK;
 731	}
 732	if (rdev->irq.hpd[0]) {
 733		tmp |= RADEON_FP_DETECT_MASK;
 734	}
 735	if (rdev->irq.hpd[1]) {
 736		tmp |= RADEON_FP2_DETECT_MASK;
 737	}
 738	WREG32(RADEON_GEN_INT_CNTL, tmp);
 739
 740	/* read back to post the write */
 741	RREG32(RADEON_GEN_INT_CNTL);
 742
 743	return 0;
 744}
 745
 746void r100_irq_disable(struct radeon_device *rdev)
 747{
 748	u32 tmp;
 749
 750	WREG32(R_000040_GEN_INT_CNTL, 0);
 751	/* Wait and acknowledge irq */
 752	mdelay(1);
 753	tmp = RREG32(R_000044_GEN_INT_STATUS);
 754	WREG32(R_000044_GEN_INT_STATUS, tmp);
 755}
 756
 757static uint32_t r100_irq_ack(struct radeon_device *rdev)
 758{
 759	uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
 760	uint32_t irq_mask = RADEON_SW_INT_TEST |
 761		RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
 762		RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
 763
 764	if (irqs) {
 765		WREG32(RADEON_GEN_INT_STATUS, irqs);
 766	}
 767	return irqs & irq_mask;
 768}
 769
 770int r100_irq_process(struct radeon_device *rdev)
 771{
 772	uint32_t status, msi_rearm;
 773	bool queue_hotplug = false;
 774
 775	status = r100_irq_ack(rdev);
 776	if (!status) {
 777		return IRQ_NONE;
 778	}
 779	if (rdev->shutdown) {
 780		return IRQ_NONE;
 781	}
 782	while (status) {
 783		/* SW interrupt */
 784		if (status & RADEON_SW_INT_TEST) {
 785			radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
 786		}
 787		/* Vertical blank interrupts */
 788		if (status & RADEON_CRTC_VBLANK_STAT) {
 789			if (rdev->irq.crtc_vblank_int[0]) {
 790				drm_handle_vblank(rdev->ddev, 0);
 791				rdev->pm.vblank_sync = true;
 792				wake_up(&rdev->irq.vblank_queue);
 793			}
 794			if (atomic_read(&rdev->irq.pflip[0]))
 795				radeon_crtc_handle_vblank(rdev, 0);
 796		}
 797		if (status & RADEON_CRTC2_VBLANK_STAT) {
 798			if (rdev->irq.crtc_vblank_int[1]) {
 799				drm_handle_vblank(rdev->ddev, 1);
 800				rdev->pm.vblank_sync = true;
 801				wake_up(&rdev->irq.vblank_queue);
 802			}
 803			if (atomic_read(&rdev->irq.pflip[1]))
 804				radeon_crtc_handle_vblank(rdev, 1);
 805		}
 806		if (status & RADEON_FP_DETECT_STAT) {
 807			queue_hotplug = true;
 808			DRM_DEBUG("HPD1\n");
 809		}
 810		if (status & RADEON_FP2_DETECT_STAT) {
 811			queue_hotplug = true;
 812			DRM_DEBUG("HPD2\n");
 813		}
 814		status = r100_irq_ack(rdev);
 815	}
 816	if (queue_hotplug)
 817		schedule_delayed_work(&rdev->hotplug_work, 0);
 818	if (rdev->msi_enabled) {
 819		switch (rdev->family) {
 820		case CHIP_RS400:
 821		case CHIP_RS480:
 822			msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
 823			WREG32(RADEON_AIC_CNTL, msi_rearm);
 824			WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
 825			break;
 826		default:
 827			WREG32(RADEON_MSI_REARM_EN, RV370_MSI_REARM_EN);
 828			break;
 829		}
 830	}
 831	return IRQ_HANDLED;
 832}
 833
 834u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
 835{
 836	if (crtc == 0)
 837		return RREG32(RADEON_CRTC_CRNT_FRAME);
 838	else
 839		return RREG32(RADEON_CRTC2_CRNT_FRAME);
 840}
 841
 842/**
 843 * r100_ring_hdp_flush - flush Host Data Path via the ring buffer
 844 * rdev: radeon device structure
 845 * ring: ring buffer struct for emitting packets
 846 */
 847static void r100_ring_hdp_flush(struct radeon_device *rdev, struct radeon_ring *ring)
 848{
 849	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 850	radeon_ring_write(ring, rdev->config.r100.hdp_cntl |
 851				RADEON_HDP_READ_BUFFER_INVALIDATE);
 852	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 853	radeon_ring_write(ring, rdev->config.r100.hdp_cntl);
 854}
 855
 856/* Who ever call radeon_fence_emit should call ring_lock and ask
 857 * for enough space (today caller are ib schedule and buffer move) */
 858void r100_fence_ring_emit(struct radeon_device *rdev,
 859			  struct radeon_fence *fence)
 860{
 861	struct radeon_ring *ring = &rdev->ring[fence->ring];
 862
 863	/* We have to make sure that caches are flushed before
 864	 * CPU might read something from VRAM. */
 865	radeon_ring_write(ring, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
 866	radeon_ring_write(ring, RADEON_RB3D_DC_FLUSH_ALL);
 867	radeon_ring_write(ring, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
 868	radeon_ring_write(ring, RADEON_RB3D_ZC_FLUSH_ALL);
 869	/* Wait until IDLE & CLEAN */
 870	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 871	radeon_ring_write(ring, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
 872	r100_ring_hdp_flush(rdev, ring);
 873	/* Emit fence sequence & fire IRQ */
 874	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
 875	radeon_ring_write(ring, fence->seq);
 876	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
 877	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
 878}
 879
 880bool r100_semaphore_ring_emit(struct radeon_device *rdev,
 881			      struct radeon_ring *ring,
 882			      struct radeon_semaphore *semaphore,
 883			      bool emit_wait)
 884{
 885	/* Unused on older asics, since we don't have semaphores or multiple rings */
 886	BUG();
 887	return false;
 888}
 889
 890struct radeon_fence *r100_copy_blit(struct radeon_device *rdev,
 891				    uint64_t src_offset,
 892				    uint64_t dst_offset,
 893				    unsigned num_gpu_pages,
 894				    struct dma_resv *resv)
 895{
 896	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 897	struct radeon_fence *fence;
 898	uint32_t cur_pages;
 899	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
 900	uint32_t pitch;
 901	uint32_t stride_pixels;
 902	unsigned ndw;
 903	int num_loops;
 904	int r = 0;
 905
 906	/* radeon limited to 16k stride */
 907	stride_bytes &= 0x3fff;
 908	/* radeon pitch is /64 */
 909	pitch = stride_bytes / 64;
 910	stride_pixels = stride_bytes / 4;
 911	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
 912
 913	/* Ask for enough room for blit + flush + fence */
 914	ndw = 64 + (10 * num_loops);
 915	r = radeon_ring_lock(rdev, ring, ndw);
 916	if (r) {
 917		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 918		return ERR_PTR(-EINVAL);
 919	}
 920	while (num_gpu_pages > 0) {
 921		cur_pages = num_gpu_pages;
 922		if (cur_pages > 8191) {
 923			cur_pages = 8191;
 924		}
 925		num_gpu_pages -= cur_pages;
 926
 927		/* pages are in Y direction - height
 928		   page width in X direction - width */
 929		radeon_ring_write(ring, PACKET3(PACKET3_BITBLT_MULTI, 8));
 930		radeon_ring_write(ring,
 931				  RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 932				  RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 933				  RADEON_GMC_SRC_CLIPPING |
 934				  RADEON_GMC_DST_CLIPPING |
 935				  RADEON_GMC_BRUSH_NONE |
 936				  (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 937				  RADEON_GMC_SRC_DATATYPE_COLOR |
 938				  RADEON_ROP3_S |
 939				  RADEON_DP_SRC_SOURCE_MEMORY |
 940				  RADEON_GMC_CLR_CMP_CNTL_DIS |
 941				  RADEON_GMC_WR_MSK_DIS);
 942		radeon_ring_write(ring, (pitch << 22) | (src_offset >> 10));
 943		radeon_ring_write(ring, (pitch << 22) | (dst_offset >> 10));
 944		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 945		radeon_ring_write(ring, 0);
 946		radeon_ring_write(ring, (0x1fff) | (0x1fff << 16));
 947		radeon_ring_write(ring, num_gpu_pages);
 948		radeon_ring_write(ring, num_gpu_pages);
 949		radeon_ring_write(ring, cur_pages | (stride_pixels << 16));
 950	}
 951	radeon_ring_write(ring, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 952	radeon_ring_write(ring, RADEON_RB2D_DC_FLUSH_ALL);
 953	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 954	radeon_ring_write(ring,
 955			  RADEON_WAIT_2D_IDLECLEAN |
 956			  RADEON_WAIT_HOST_IDLECLEAN |
 957			  RADEON_WAIT_DMA_GUI_IDLE);
 958	r = radeon_fence_emit(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
 959	if (r) {
 960		radeon_ring_unlock_undo(rdev, ring);
 961		return ERR_PTR(r);
 962	}
 963	radeon_ring_unlock_commit(rdev, ring, false);
 964	return fence;
 965}
 966
 967static int r100_cp_wait_for_idle(struct radeon_device *rdev)
 968{
 969	unsigned i;
 970	u32 tmp;
 971
 972	for (i = 0; i < rdev->usec_timeout; i++) {
 973		tmp = RREG32(R_000E40_RBBM_STATUS);
 974		if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
 975			return 0;
 976		}
 977		udelay(1);
 978	}
 979	return -1;
 980}
 981
 982void r100_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
 983{
 984	int r;
 985
 986	r = radeon_ring_lock(rdev, ring, 2);
 987	if (r) {
 988		return;
 989	}
 990	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
 991	radeon_ring_write(ring,
 992			  RADEON_ISYNC_ANY2D_IDLE3D |
 993			  RADEON_ISYNC_ANY3D_IDLE2D |
 994			  RADEON_ISYNC_WAIT_IDLEGUI |
 995			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 996	radeon_ring_unlock_commit(rdev, ring, false);
 997}
 998
 999
1000/* Load the microcode for the CP */
1001static int r100_cp_init_microcode(struct radeon_device *rdev)
1002{
1003	const char *fw_name = NULL;
1004	int err;
1005
1006	DRM_DEBUG_KMS("\n");
1007
1008	if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
1009	    (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
1010	    (rdev->family == CHIP_RS200)) {
 
 
 
1011		DRM_INFO("Loading R100 Microcode\n");
1012		fw_name = FIRMWARE_R100;
1013	} else if ((rdev->family == CHIP_R200) ||
1014		   (rdev->family == CHIP_RV250) ||
1015		   (rdev->family == CHIP_RV280) ||
1016		   (rdev->family == CHIP_RS300)) {
 
 
1017		DRM_INFO("Loading R200 Microcode\n");
1018		fw_name = FIRMWARE_R200;
1019	} else if ((rdev->family == CHIP_R300) ||
1020		   (rdev->family == CHIP_R350) ||
1021		   (rdev->family == CHIP_RV350) ||
1022		   (rdev->family == CHIP_RV380) ||
1023		   (rdev->family == CHIP_RS400) ||
1024		   (rdev->family == CHIP_RS480)) {
 
 
1025		DRM_INFO("Loading R300 Microcode\n");
1026		fw_name = FIRMWARE_R300;
1027	} else if ((rdev->family == CHIP_R420) ||
1028		   (rdev->family == CHIP_R423) ||
1029		   (rdev->family == CHIP_RV410)) {
 
 
1030		DRM_INFO("Loading R400 Microcode\n");
1031		fw_name = FIRMWARE_R420;
1032	} else if ((rdev->family == CHIP_RS690) ||
1033		   (rdev->family == CHIP_RS740)) {
 
 
1034		DRM_INFO("Loading RS690/RS740 Microcode\n");
1035		fw_name = FIRMWARE_RS690;
1036	} else if (rdev->family == CHIP_RS600) {
 
 
1037		DRM_INFO("Loading RS600 Microcode\n");
1038		fw_name = FIRMWARE_RS600;
1039	} else if ((rdev->family == CHIP_RV515) ||
1040		   (rdev->family == CHIP_R520) ||
1041		   (rdev->family == CHIP_RV530) ||
1042		   (rdev->family == CHIP_R580) ||
1043		   (rdev->family == CHIP_RV560) ||
1044		   (rdev->family == CHIP_RV570)) {
 
 
1045		DRM_INFO("Loading R500 Microcode\n");
1046		fw_name = FIRMWARE_R520;
 
 
 
 
 
1047	}
1048
1049	err = request_firmware(&rdev->me_fw, fw_name, rdev->dev);
1050	if (err) {
1051		pr_err("radeon_cp: Failed to load firmware \"%s\"\n", fw_name);
1052	} else if (rdev->me_fw->size % 8) {
1053		pr_err("radeon_cp: Bogus length %zu in firmware \"%s\"\n",
1054		       rdev->me_fw->size, fw_name);
1055		err = -EINVAL;
1056		release_firmware(rdev->me_fw);
1057		rdev->me_fw = NULL;
1058	}
1059	return err;
1060}
1061
1062u32 r100_gfx_get_rptr(struct radeon_device *rdev,
1063		      struct radeon_ring *ring)
1064{
1065	u32 rptr;
1066
1067	if (rdev->wb.enabled)
1068		rptr = le32_to_cpu(rdev->wb.wb[ring->rptr_offs/4]);
1069	else
1070		rptr = RREG32(RADEON_CP_RB_RPTR);
1071
1072	return rptr;
1073}
1074
1075u32 r100_gfx_get_wptr(struct radeon_device *rdev,
1076		      struct radeon_ring *ring)
1077{
1078	return RREG32(RADEON_CP_RB_WPTR);
1079}
1080
1081void r100_gfx_set_wptr(struct radeon_device *rdev,
1082		       struct radeon_ring *ring)
1083{
1084	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1085	(void)RREG32(RADEON_CP_RB_WPTR);
1086}
1087
1088static void r100_cp_load_microcode(struct radeon_device *rdev)
1089{
1090	const __be32 *fw_data;
1091	int i, size;
1092
1093	if (r100_gui_wait_for_idle(rdev)) {
1094		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
1095	}
1096
1097	if (rdev->me_fw) {
1098		size = rdev->me_fw->size / 4;
1099		fw_data = (const __be32 *)&rdev->me_fw->data[0];
1100		WREG32(RADEON_CP_ME_RAM_ADDR, 0);
1101		for (i = 0; i < size; i += 2) {
1102			WREG32(RADEON_CP_ME_RAM_DATAH,
1103			       be32_to_cpup(&fw_data[i]));
1104			WREG32(RADEON_CP_ME_RAM_DATAL,
1105			       be32_to_cpup(&fw_data[i + 1]));
1106		}
1107	}
1108}
1109
1110int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
1111{
1112	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1113	unsigned rb_bufsz;
1114	unsigned rb_blksz;
1115	unsigned max_fetch;
1116	unsigned pre_write_timer;
1117	unsigned pre_write_limit;
1118	unsigned indirect2_start;
1119	unsigned indirect1_start;
1120	uint32_t tmp;
1121	int r;
1122
1123	if (r100_debugfs_cp_init(rdev)) {
1124		DRM_ERROR("Failed to register debugfs file for CP !\n");
1125	}
1126	if (!rdev->me_fw) {
1127		r = r100_cp_init_microcode(rdev);
1128		if (r) {
1129			DRM_ERROR("Failed to load firmware!\n");
1130			return r;
1131		}
1132	}
1133
1134	/* Align ring size */
1135	rb_bufsz = order_base_2(ring_size / 8);
1136	ring_size = (1 << (rb_bufsz + 1)) * 4;
1137	r100_cp_load_microcode(rdev);
1138	r = radeon_ring_init(rdev, ring, ring_size, RADEON_WB_CP_RPTR_OFFSET,
1139			     RADEON_CP_PACKET2);
1140	if (r) {
1141		return r;
1142	}
1143	/* Each time the cp read 1024 bytes (16 dword/quadword) update
1144	 * the rptr copy in system ram */
1145	rb_blksz = 9;
1146	/* cp will read 128bytes at a time (4 dwords) */
1147	max_fetch = 1;
1148	ring->align_mask = 16 - 1;
1149	/* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
1150	pre_write_timer = 64;
1151	/* Force CP_RB_WPTR write if written more than one time before the
1152	 * delay expire
1153	 */
1154	pre_write_limit = 0;
1155	/* Setup the cp cache like this (cache size is 96 dwords) :
1156	 *	RING		0  to 15
1157	 *	INDIRECT1	16 to 79
1158	 *	INDIRECT2	80 to 95
1159	 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1160	 *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
1161	 *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
1162	 * Idea being that most of the gpu cmd will be through indirect1 buffer
1163	 * so it gets the bigger cache.
1164	 */
1165	indirect2_start = 80;
1166	indirect1_start = 16;
1167	/* cp setup */
1168	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
1169	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
1170	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
1171	       REG_SET(RADEON_MAX_FETCH, max_fetch));
1172#ifdef __BIG_ENDIAN
1173	tmp |= RADEON_BUF_SWAP_32BIT;
1174#endif
1175	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1176
1177	/* Set ring address */
1178	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)ring->gpu_addr);
1179	WREG32(RADEON_CP_RB_BASE, ring->gpu_addr);
1180	/* Force read & write ptr to 0 */
1181	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1182	WREG32(RADEON_CP_RB_RPTR_WR, 0);
1183	ring->wptr = 0;
1184	WREG32(RADEON_CP_RB_WPTR, ring->wptr);
1185
1186	/* set the wb address whether it's enabled or not */
1187	WREG32(R_00070C_CP_RB_RPTR_ADDR,
1188		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1189	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1190
1191	if (rdev->wb.enabled)
1192		WREG32(R_000770_SCRATCH_UMSK, 0xff);
1193	else {
1194		tmp |= RADEON_RB_NO_UPDATE;
1195		WREG32(R_000770_SCRATCH_UMSK, 0);
1196	}
1197
1198	WREG32(RADEON_CP_RB_CNTL, tmp);
1199	udelay(10);
1200	/* Set cp mode to bus mastering & enable cp*/
1201	WREG32(RADEON_CP_CSQ_MODE,
1202	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1203	       REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1204	WREG32(RADEON_CP_RB_WPTR_DELAY, 0);
1205	WREG32(RADEON_CP_CSQ_MODE, 0x00004D4D);
1206	WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1207
1208	/* at this point everything should be setup correctly to enable master */
1209	pci_set_master(rdev->pdev);
1210
1211	radeon_ring_start(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1212	r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1213	if (r) {
1214		DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1215		return r;
1216	}
1217	ring->ready = true;
1218	radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size);
1219
1220	if (!ring->rptr_save_reg /* not resuming from suspend */
1221	    && radeon_ring_supports_scratch_reg(rdev, ring)) {
1222		r = radeon_scratch_get(rdev, &ring->rptr_save_reg);
1223		if (r) {
1224			DRM_ERROR("failed to get scratch reg for rptr save (%d).\n", r);
1225			ring->rptr_save_reg = 0;
1226		}
1227	}
1228	return 0;
1229}
1230
1231void r100_cp_fini(struct radeon_device *rdev)
1232{
1233	if (r100_cp_wait_for_idle(rdev)) {
1234		DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1235	}
1236	/* Disable ring */
1237	r100_cp_disable(rdev);
1238	radeon_scratch_free(rdev, rdev->ring[RADEON_RING_TYPE_GFX_INDEX].rptr_save_reg);
1239	radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1240	DRM_INFO("radeon: cp finalized\n");
1241}
1242
1243void r100_cp_disable(struct radeon_device *rdev)
1244{
1245	/* Disable ring */
1246	radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1247	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1248	WREG32(RADEON_CP_CSQ_MODE, 0);
1249	WREG32(RADEON_CP_CSQ_CNTL, 0);
1250	WREG32(R_000770_SCRATCH_UMSK, 0);
1251	if (r100_gui_wait_for_idle(rdev)) {
1252		pr_warn("Failed to wait GUI idle while 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, p->filp, 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 = true;
1827		} else {
1828			track->textures[i].use_pitch = false;
1829			track->textures[i].width = 1 << ((idx_value & RADEON_TXFORMAT_WIDTH_MASK) >> RADEON_TXFORMAT_WIDTH_SHIFT);
1830			track->textures[i].height = 1 << ((idx_value & RADEON_TXFORMAT_HEIGHT_MASK) >> RADEON_TXFORMAT_HEIGHT_SHIFT);
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		pr_err("Forbidden register 0x%04X in cs at %d\n", reg, idx);
1887		return -EINVAL;
1888	}
1889	return 0;
1890}
1891
1892int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1893					 struct radeon_cs_packet *pkt,
1894					 struct radeon_bo *robj)
1895{
1896	unsigned idx;
1897	u32 value;
1898	idx = pkt->idx + 1;
1899	value = radeon_get_ib_value(p, idx + 2);
1900	if ((value + 1) > radeon_bo_size(robj)) {
1901		DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1902			  "(need %u have %lu) !\n",
1903			  value + 1,
1904			  radeon_bo_size(robj));
1905		return -EINVAL;
1906	}
1907	return 0;
1908}
1909
1910static int r100_packet3_check(struct radeon_cs_parser *p,
1911			      struct radeon_cs_packet *pkt)
1912{
1913	struct radeon_bo_list *reloc;
1914	struct r100_cs_track *track;
1915	unsigned idx;
1916	volatile uint32_t *ib;
1917	int r;
1918
1919	ib = p->ib.ptr;
1920	idx = pkt->idx + 1;
1921	track = (struct r100_cs_track *)p->track;
1922	switch (pkt->opcode) {
1923	case PACKET3_3D_LOAD_VBPNTR:
1924		r = r100_packet3_load_vbpntr(p, pkt, idx);
1925		if (r)
1926			return r;
1927		break;
1928	case PACKET3_INDX_BUFFER:
1929		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1930		if (r) {
1931			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1932			radeon_cs_dump_packet(p, pkt);
1933			return r;
1934		}
1935		ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->gpu_offset);
1936		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1937		if (r) {
1938			return r;
1939		}
1940		break;
1941	case 0x23:
1942		/* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1943		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1944		if (r) {
1945			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1946			radeon_cs_dump_packet(p, pkt);
1947			return r;
1948		}
1949		ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->gpu_offset);
1950		track->num_arrays = 1;
1951		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1952
1953		track->arrays[0].robj = reloc->robj;
1954		track->arrays[0].esize = track->vtx_size;
1955
1956		track->max_indx = radeon_get_ib_value(p, idx+1);
1957
1958		track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1959		track->immd_dwords = pkt->count - 1;
1960		r = r100_cs_track_check(p->rdev, track);
1961		if (r)
1962			return r;
1963		break;
1964	case PACKET3_3D_DRAW_IMMD:
1965		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1966			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1967			return -EINVAL;
1968		}
1969		track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1970		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1971		track->immd_dwords = pkt->count - 1;
1972		r = r100_cs_track_check(p->rdev, track);
1973		if (r)
1974			return r;
1975		break;
1976		/* triggers drawing using in-packet vertex data */
1977	case PACKET3_3D_DRAW_IMMD_2:
1978		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1979			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1980			return -EINVAL;
1981		}
1982		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1983		track->immd_dwords = pkt->count;
1984		r = r100_cs_track_check(p->rdev, track);
1985		if (r)
1986			return r;
1987		break;
1988		/* triggers drawing using in-packet vertex data */
1989	case PACKET3_3D_DRAW_VBUF_2:
1990		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1991		r = r100_cs_track_check(p->rdev, track);
1992		if (r)
1993			return r;
1994		break;
1995		/* triggers drawing of vertex buffers setup elsewhere */
1996	case PACKET3_3D_DRAW_INDX_2:
1997		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1998		r = r100_cs_track_check(p->rdev, track);
1999		if (r)
2000			return r;
2001		break;
2002		/* triggers drawing using indices to vertex buffer */
2003	case PACKET3_3D_DRAW_VBUF:
2004		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2005		r = r100_cs_track_check(p->rdev, track);
2006		if (r)
2007			return r;
2008		break;
2009		/* triggers drawing of vertex buffers setup elsewhere */
2010	case PACKET3_3D_DRAW_INDX:
2011		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
2012		r = r100_cs_track_check(p->rdev, track);
2013		if (r)
2014			return r;
2015		break;
2016		/* triggers drawing using indices to vertex buffer */
2017	case PACKET3_3D_CLEAR_HIZ:
2018	case PACKET3_3D_CLEAR_ZMASK:
2019		if (p->rdev->hyperz_filp != p->filp)
2020			return -EINVAL;
2021		break;
2022	case PACKET3_NOP:
2023		break;
2024	default:
2025		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
2026		return -EINVAL;
2027	}
2028	return 0;
2029}
2030
2031int r100_cs_parse(struct radeon_cs_parser *p)
2032{
2033	struct radeon_cs_packet pkt;
2034	struct r100_cs_track *track;
2035	int r;
2036
2037	track = kzalloc(sizeof(*track), GFP_KERNEL);
2038	if (!track)
2039		return -ENOMEM;
2040	r100_cs_track_clear(p->rdev, track);
2041	p->track = track;
2042	do {
2043		r = radeon_cs_packet_parse(p, &pkt, p->idx);
2044		if (r) {
2045			return r;
2046		}
2047		p->idx += pkt.count + 2;
2048		switch (pkt.type) {
2049		case RADEON_PACKET_TYPE0:
2050			if (p->rdev->family >= CHIP_R200)
2051				r = r100_cs_parse_packet0(p, &pkt,
2052					p->rdev->config.r100.reg_safe_bm,
2053					p->rdev->config.r100.reg_safe_bm_size,
2054					&r200_packet0_check);
2055			else
2056				r = r100_cs_parse_packet0(p, &pkt,
2057					p->rdev->config.r100.reg_safe_bm,
2058					p->rdev->config.r100.reg_safe_bm_size,
2059					&r100_packet0_check);
2060			break;
2061		case RADEON_PACKET_TYPE2:
2062			break;
2063		case RADEON_PACKET_TYPE3:
2064			r = r100_packet3_check(p, &pkt);
2065			break;
2066		default:
2067			DRM_ERROR("Unknown packet type %d !\n",
2068				  pkt.type);
2069			return -EINVAL;
2070		}
2071		if (r)
2072			return r;
2073	} while (p->idx < p->chunk_ib->length_dw);
2074	return 0;
2075}
2076
2077static void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2078{
2079	DRM_ERROR("pitch                      %d\n", t->pitch);
2080	DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2081	DRM_ERROR("width                      %d\n", t->width);
2082	DRM_ERROR("width_11                   %d\n", t->width_11);
2083	DRM_ERROR("height                     %d\n", t->height);
2084	DRM_ERROR("height_11                  %d\n", t->height_11);
2085	DRM_ERROR("num levels                 %d\n", t->num_levels);
2086	DRM_ERROR("depth                      %d\n", t->txdepth);
2087	DRM_ERROR("bpp                        %d\n", t->cpp);
2088	DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2089	DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2090	DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2091	DRM_ERROR("compress format            %d\n", t->compress_format);
2092}
2093
2094static int r100_track_compress_size(int compress_format, int w, int h)
2095{
2096	int block_width, block_height, block_bytes;
2097	int wblocks, hblocks;
2098	int min_wblocks;
2099	int sz;
2100
2101	block_width = 4;
2102	block_height = 4;
2103
2104	switch (compress_format) {
2105	case R100_TRACK_COMP_DXT1:
2106		block_bytes = 8;
2107		min_wblocks = 4;
2108		break;
2109	default:
2110	case R100_TRACK_COMP_DXT35:
2111		block_bytes = 16;
2112		min_wblocks = 2;
2113		break;
2114	}
2115
2116	hblocks = (h + block_height - 1) / block_height;
2117	wblocks = (w + block_width - 1) / block_width;
2118	if (wblocks < min_wblocks)
2119		wblocks = min_wblocks;
2120	sz = wblocks * hblocks * block_bytes;
2121	return sz;
2122}
2123
2124static int r100_cs_track_cube(struct radeon_device *rdev,
2125			      struct r100_cs_track *track, unsigned idx)
2126{
2127	unsigned face, w, h;
2128	struct radeon_bo *cube_robj;
2129	unsigned long size;
2130	unsigned compress_format = track->textures[idx].compress_format;
2131
2132	for (face = 0; face < 5; face++) {
2133		cube_robj = track->textures[idx].cube_info[face].robj;
2134		w = track->textures[idx].cube_info[face].width;
2135		h = track->textures[idx].cube_info[face].height;
2136
2137		if (compress_format) {
2138			size = r100_track_compress_size(compress_format, w, h);
2139		} else
2140			size = w * h;
2141		size *= track->textures[idx].cpp;
2142
2143		size += track->textures[idx].cube_info[face].offset;
2144
2145		if (size > radeon_bo_size(cube_robj)) {
2146			DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2147				  size, radeon_bo_size(cube_robj));
2148			r100_cs_track_texture_print(&track->textures[idx]);
2149			return -1;
2150		}
2151	}
2152	return 0;
2153}
2154
2155static int r100_cs_track_texture_check(struct radeon_device *rdev,
2156				       struct r100_cs_track *track)
2157{
2158	struct radeon_bo *robj;
2159	unsigned long size;
2160	unsigned u, i, w, h, d;
2161	int ret;
2162
2163	for (u = 0; u < track->num_texture; u++) {
2164		if (!track->textures[u].enabled)
2165			continue;
2166		if (track->textures[u].lookup_disable)
2167			continue;
2168		robj = track->textures[u].robj;
2169		if (robj == NULL) {
2170			DRM_ERROR("No texture bound to unit %u\n", u);
2171			return -EINVAL;
2172		}
2173		size = 0;
2174		for (i = 0; i <= track->textures[u].num_levels; i++) {
2175			if (track->textures[u].use_pitch) {
2176				if (rdev->family < CHIP_R300)
2177					w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2178				else
2179					w = track->textures[u].pitch / (1 << i);
2180			} else {
2181				w = track->textures[u].width;
2182				if (rdev->family >= CHIP_RV515)
2183					w |= track->textures[u].width_11;
2184				w = w / (1 << i);
2185				if (track->textures[u].roundup_w)
2186					w = roundup_pow_of_two(w);
2187			}
2188			h = track->textures[u].height;
2189			if (rdev->family >= CHIP_RV515)
2190				h |= track->textures[u].height_11;
2191			h = h / (1 << i);
2192			if (track->textures[u].roundup_h)
2193				h = roundup_pow_of_two(h);
2194			if (track->textures[u].tex_coord_type == 1) {
2195				d = (1 << track->textures[u].txdepth) / (1 << i);
2196				if (!d)
2197					d = 1;
2198			} else {
2199				d = 1;
2200			}
2201			if (track->textures[u].compress_format) {
2202
2203				size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
2204				/* compressed textures are block based */
2205			} else
2206				size += w * h * d;
2207		}
2208		size *= track->textures[u].cpp;
2209
2210		switch (track->textures[u].tex_coord_type) {
2211		case 0:
2212		case 1:
2213			break;
2214		case 2:
2215			if (track->separate_cube) {
2216				ret = r100_cs_track_cube(rdev, track, u);
2217				if (ret)
2218					return ret;
2219			} else
2220				size *= 6;
2221			break;
2222		default:
2223			DRM_ERROR("Invalid texture coordinate type %u for unit "
2224				  "%u\n", track->textures[u].tex_coord_type, u);
2225			return -EINVAL;
2226		}
2227		if (size > radeon_bo_size(robj)) {
2228			DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2229				  "%lu\n", u, size, radeon_bo_size(robj));
2230			r100_cs_track_texture_print(&track->textures[u]);
2231			return -EINVAL;
2232		}
2233	}
2234	return 0;
2235}
2236
2237int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2238{
2239	unsigned i;
2240	unsigned long size;
2241	unsigned prim_walk;
2242	unsigned nverts;
2243	unsigned num_cb = track->cb_dirty ? track->num_cb : 0;
2244
2245	if (num_cb && !track->zb_cb_clear && !track->color_channel_mask &&
2246	    !track->blend_read_enable)
2247		num_cb = 0;
2248
2249	for (i = 0; i < num_cb; i++) {
2250		if (track->cb[i].robj == NULL) {
2251			DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2252			return -EINVAL;
2253		}
2254		size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2255		size += track->cb[i].offset;
2256		if (size > radeon_bo_size(track->cb[i].robj)) {
2257			DRM_ERROR("[drm] Buffer too small for color buffer %d "
2258				  "(need %lu have %lu) !\n", i, size,
2259				  radeon_bo_size(track->cb[i].robj));
2260			DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2261				  i, track->cb[i].pitch, track->cb[i].cpp,
2262				  track->cb[i].offset, track->maxy);
2263			return -EINVAL;
2264		}
2265	}
2266	track->cb_dirty = false;
2267
2268	if (track->zb_dirty && track->z_enabled) {
2269		if (track->zb.robj == NULL) {
2270			DRM_ERROR("[drm] No buffer for z buffer !\n");
2271			return -EINVAL;
2272		}
2273		size = track->zb.pitch * track->zb.cpp * track->maxy;
2274		size += track->zb.offset;
2275		if (size > radeon_bo_size(track->zb.robj)) {
2276			DRM_ERROR("[drm] Buffer too small for z buffer "
2277				  "(need %lu have %lu) !\n", size,
2278				  radeon_bo_size(track->zb.robj));
2279			DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
2280				  track->zb.pitch, track->zb.cpp,
2281				  track->zb.offset, track->maxy);
2282			return -EINVAL;
2283		}
2284	}
2285	track->zb_dirty = false;
2286
2287	if (track->aa_dirty && track->aaresolve) {
2288		if (track->aa.robj == NULL) {
2289			DRM_ERROR("[drm] No buffer for AA resolve buffer %d !\n", i);
2290			return -EINVAL;
2291		}
2292		/* I believe the format comes from colorbuffer0. */
2293		size = track->aa.pitch * track->cb[0].cpp * track->maxy;
2294		size += track->aa.offset;
2295		if (size > radeon_bo_size(track->aa.robj)) {
2296			DRM_ERROR("[drm] Buffer too small for AA resolve buffer %d "
2297				  "(need %lu have %lu) !\n", i, size,
2298				  radeon_bo_size(track->aa.robj));
2299			DRM_ERROR("[drm] AA resolve buffer %d (%u %u %u %u)\n",
2300				  i, track->aa.pitch, track->cb[0].cpp,
2301				  track->aa.offset, track->maxy);
2302			return -EINVAL;
2303		}
2304	}
2305	track->aa_dirty = false;
2306
2307	prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
2308	if (track->vap_vf_cntl & (1 << 14)) {
2309		nverts = track->vap_alt_nverts;
2310	} else {
2311		nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
2312	}
2313	switch (prim_walk) {
2314	case 1:
2315		for (i = 0; i < track->num_arrays; i++) {
2316			size = track->arrays[i].esize * track->max_indx * 4;
2317			if (track->arrays[i].robj == NULL) {
2318				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2319					  "bound\n", prim_walk, i);
2320				return -EINVAL;
2321			}
2322			if (size > radeon_bo_size(track->arrays[i].robj)) {
2323				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2324					"need %lu dwords have %lu dwords\n",
2325					prim_walk, i, size >> 2,
2326					radeon_bo_size(track->arrays[i].robj)
2327					>> 2);
2328				DRM_ERROR("Max indices %u\n", track->max_indx);
2329				return -EINVAL;
2330			}
2331		}
2332		break;
2333	case 2:
2334		for (i = 0; i < track->num_arrays; i++) {
2335			size = track->arrays[i].esize * (nverts - 1) * 4;
2336			if (track->arrays[i].robj == NULL) {
2337				DRM_ERROR("(PW %u) Vertex array %u no buffer "
2338					  "bound\n", prim_walk, i);
2339				return -EINVAL;
2340			}
2341			if (size > radeon_bo_size(track->arrays[i].robj)) {
2342				dev_err(rdev->dev, "(PW %u) Vertex array %u "
2343					"need %lu dwords have %lu dwords\n",
2344					prim_walk, i, size >> 2,
2345					radeon_bo_size(track->arrays[i].robj)
2346					>> 2);
2347				return -EINVAL;
2348			}
2349		}
2350		break;
2351	case 3:
2352		size = track->vtx_size * nverts;
2353		if (size != track->immd_dwords) {
2354			DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
2355				  track->immd_dwords, size);
2356			DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
2357				  nverts, track->vtx_size);
2358			return -EINVAL;
2359		}
2360		break;
2361	default:
2362		DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
2363			  prim_walk);
2364		return -EINVAL;
2365	}
2366
2367	if (track->tex_dirty) {
2368		track->tex_dirty = false;
2369		return r100_cs_track_texture_check(rdev, track);
2370	}
2371	return 0;
2372}
2373
2374void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
2375{
2376	unsigned i, face;
2377
2378	track->cb_dirty = true;
2379	track->zb_dirty = true;
2380	track->tex_dirty = true;
2381	track->aa_dirty = true;
2382
2383	if (rdev->family < CHIP_R300) {
2384		track->num_cb = 1;
2385		if (rdev->family <= CHIP_RS200)
2386			track->num_texture = 3;
2387		else
2388			track->num_texture = 6;
2389		track->maxy = 2048;
2390		track->separate_cube = true;
2391	} else {
2392		track->num_cb = 4;
2393		track->num_texture = 16;
2394		track->maxy = 4096;
2395		track->separate_cube = false;
2396		track->aaresolve = false;
2397		track->aa.robj = NULL;
2398	}
2399
2400	for (i = 0; i < track->num_cb; i++) {
2401		track->cb[i].robj = NULL;
2402		track->cb[i].pitch = 8192;
2403		track->cb[i].cpp = 16;
2404		track->cb[i].offset = 0;
2405	}
2406	track->z_enabled = true;
2407	track->zb.robj = NULL;
2408	track->zb.pitch = 8192;
2409	track->zb.cpp = 4;
2410	track->zb.offset = 0;
2411	track->vtx_size = 0x7F;
2412	track->immd_dwords = 0xFFFFFFFFUL;
2413	track->num_arrays = 11;
2414	track->max_indx = 0x00FFFFFFUL;
2415	for (i = 0; i < track->num_arrays; i++) {
2416		track->arrays[i].robj = NULL;
2417		track->arrays[i].esize = 0x7F;
2418	}
2419	for (i = 0; i < track->num_texture; i++) {
2420		track->textures[i].compress_format = R100_TRACK_COMP_NONE;
2421		track->textures[i].pitch = 16536;
2422		track->textures[i].width = 16536;
2423		track->textures[i].height = 16536;
2424		track->textures[i].width_11 = 1 << 11;
2425		track->textures[i].height_11 = 1 << 11;
2426		track->textures[i].num_levels = 12;
2427		if (rdev->family <= CHIP_RS200) {
2428			track->textures[i].tex_coord_type = 0;
2429			track->textures[i].txdepth = 0;
2430		} else {
2431			track->textures[i].txdepth = 16;
2432			track->textures[i].tex_coord_type = 1;
2433		}
2434		track->textures[i].cpp = 64;
2435		track->textures[i].robj = NULL;
2436		/* CS IB emission code makes sure texture unit are disabled */
2437		track->textures[i].enabled = false;
2438		track->textures[i].lookup_disable = false;
2439		track->textures[i].roundup_w = true;
2440		track->textures[i].roundup_h = true;
2441		if (track->separate_cube)
2442			for (face = 0; face < 5; face++) {
2443				track->textures[i].cube_info[face].robj = NULL;
2444				track->textures[i].cube_info[face].width = 16536;
2445				track->textures[i].cube_info[face].height = 16536;
2446				track->textures[i].cube_info[face].offset = 0;
2447			}
2448	}
2449}
2450
2451/*
2452 * Global GPU functions
2453 */
2454static void r100_errata(struct radeon_device *rdev)
2455{
2456	rdev->pll_errata = 0;
2457
2458	if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
2459		rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
2460	}
2461
2462	if (rdev->family == CHIP_RV100 ||
2463	    rdev->family == CHIP_RS100 ||
2464	    rdev->family == CHIP_RS200) {
2465		rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
2466	}
2467}
2468
2469static int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
2470{
2471	unsigned i;
2472	uint32_t tmp;
2473
2474	for (i = 0; i < rdev->usec_timeout; i++) {
2475		tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
2476		if (tmp >= n) {
2477			return 0;
2478		}
2479		udelay(1);
2480	}
2481	return -1;
2482}
2483
2484int r100_gui_wait_for_idle(struct radeon_device *rdev)
2485{
2486	unsigned i;
2487	uint32_t tmp;
2488
2489	if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
2490		pr_warn("radeon: wait for empty RBBM fifo failed! Bad things might happen.\n");
2491	}
2492	for (i = 0; i < rdev->usec_timeout; i++) {
2493		tmp = RREG32(RADEON_RBBM_STATUS);
2494		if (!(tmp & RADEON_RBBM_ACTIVE)) {
2495			return 0;
2496		}
2497		udelay(1);
2498	}
2499	return -1;
2500}
2501
2502int r100_mc_wait_for_idle(struct radeon_device *rdev)
2503{
2504	unsigned i;
2505	uint32_t tmp;
2506
2507	for (i = 0; i < rdev->usec_timeout; i++) {
2508		/* read MC_STATUS */
2509		tmp = RREG32(RADEON_MC_STATUS);
2510		if (tmp & RADEON_MC_IDLE) {
2511			return 0;
2512		}
2513		udelay(1);
2514	}
2515	return -1;
2516}
2517
2518bool r100_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2519{
2520	u32 rbbm_status;
2521
2522	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2523	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2524		radeon_ring_lockup_update(rdev, ring);
2525		return false;
2526	}
2527	return radeon_ring_test_lockup(rdev, ring);
2528}
2529
2530/* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
2531void r100_enable_bm(struct radeon_device *rdev)
2532{
2533	uint32_t tmp;
2534	/* Enable bus mastering */
2535	tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
2536	WREG32(RADEON_BUS_CNTL, tmp);
2537}
2538
2539void r100_bm_disable(struct radeon_device *rdev)
2540{
2541	u32 tmp;
2542
2543	/* disable bus mastering */
2544	tmp = RREG32(R_000030_BUS_CNTL);
2545	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2546	mdelay(1);
2547	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2548	mdelay(1);
2549	WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2550	tmp = RREG32(RADEON_BUS_CNTL);
2551	mdelay(1);
2552	pci_clear_master(rdev->pdev);
2553	mdelay(1);
2554}
2555
2556int r100_asic_reset(struct radeon_device *rdev, bool hard)
2557{
2558	struct r100_mc_save save;
2559	u32 status, tmp;
2560	int ret = 0;
2561
2562	status = RREG32(R_000E40_RBBM_STATUS);
2563	if (!G_000E40_GUI_ACTIVE(status)) {
2564		return 0;
2565	}
2566	r100_mc_stop(rdev, &save);
2567	status = RREG32(R_000E40_RBBM_STATUS);
2568	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2569	/* stop CP */
2570	WREG32(RADEON_CP_CSQ_CNTL, 0);
2571	tmp = RREG32(RADEON_CP_RB_CNTL);
2572	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2573	WREG32(RADEON_CP_RB_RPTR_WR, 0);
2574	WREG32(RADEON_CP_RB_WPTR, 0);
2575	WREG32(RADEON_CP_RB_CNTL, tmp);
2576	/* save PCI state */
2577	pci_save_state(rdev->pdev);
2578	/* disable bus mastering */
2579	r100_bm_disable(rdev);
2580	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2581					S_0000F0_SOFT_RESET_RE(1) |
2582					S_0000F0_SOFT_RESET_PP(1) |
2583					S_0000F0_SOFT_RESET_RB(1));
2584	RREG32(R_0000F0_RBBM_SOFT_RESET);
2585	mdelay(500);
2586	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2587	mdelay(1);
2588	status = RREG32(R_000E40_RBBM_STATUS);
2589	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2590	/* reset CP */
2591	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2592	RREG32(R_0000F0_RBBM_SOFT_RESET);
2593	mdelay(500);
2594	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2595	mdelay(1);
2596	status = RREG32(R_000E40_RBBM_STATUS);
2597	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2598	/* restore PCI & busmastering */
2599	pci_restore_state(rdev->pdev);
2600	r100_enable_bm(rdev);
2601	/* Check if GPU is idle */
2602	if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2603		G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2604		dev_err(rdev->dev, "failed to reset GPU\n");
2605		ret = -1;
2606	} else
2607		dev_info(rdev->dev, "GPU reset succeed\n");
2608	r100_mc_resume(rdev, &save);
2609	return ret;
2610}
2611
2612void r100_set_common_regs(struct radeon_device *rdev)
2613{
2614	struct drm_device *dev = rdev->ddev;
2615	bool force_dac2 = false;
2616	u32 tmp;
2617
2618	/* set these so they don't interfere with anything */
2619	WREG32(RADEON_OV0_SCALE_CNTL, 0);
2620	WREG32(RADEON_SUBPIC_CNTL, 0);
2621	WREG32(RADEON_VIPH_CONTROL, 0);
2622	WREG32(RADEON_I2C_CNTL_1, 0);
2623	WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2624	WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2625	WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2626
2627	/* always set up dac2 on rn50 and some rv100 as lots
2628	 * of servers seem to wire it up to a VGA port but
2629	 * don't report it in the bios connector
2630	 * table.
2631	 */
2632	switch (dev->pdev->device) {
2633		/* RN50 */
2634	case 0x515e:
2635	case 0x5969:
2636		force_dac2 = true;
2637		break;
2638		/* RV100*/
2639	case 0x5159:
2640	case 0x515a:
2641		/* DELL triple head servers */
2642		if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2643		    ((dev->pdev->subsystem_device == 0x016c) ||
2644		     (dev->pdev->subsystem_device == 0x016d) ||
2645		     (dev->pdev->subsystem_device == 0x016e) ||
2646		     (dev->pdev->subsystem_device == 0x016f) ||
2647		     (dev->pdev->subsystem_device == 0x0170) ||
2648		     (dev->pdev->subsystem_device == 0x017d) ||
2649		     (dev->pdev->subsystem_device == 0x017e) ||
2650		     (dev->pdev->subsystem_device == 0x0183) ||
2651		     (dev->pdev->subsystem_device == 0x018a) ||
2652		     (dev->pdev->subsystem_device == 0x019a)))
2653			force_dac2 = true;
2654		break;
2655	}
2656
2657	if (force_dac2) {
2658		u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2659		u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2660		u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2661
2662		/* For CRT on DAC2, don't turn it on if BIOS didn't
2663		   enable it, even it's detected.
2664		*/
2665
2666		/* force it to crtc0 */
2667		dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2668		dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2669		disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2670
2671		/* set up the TV DAC */
2672		tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2673				 RADEON_TV_DAC_STD_MASK |
2674				 RADEON_TV_DAC_RDACPD |
2675				 RADEON_TV_DAC_GDACPD |
2676				 RADEON_TV_DAC_BDACPD |
2677				 RADEON_TV_DAC_BGADJ_MASK |
2678				 RADEON_TV_DAC_DACADJ_MASK);
2679		tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2680				RADEON_TV_DAC_NHOLD |
2681				RADEON_TV_DAC_STD_PS2 |
2682				(0x58 << 16));
2683
2684		WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2685		WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2686		WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2687	}
2688
2689	/* switch PM block to ACPI mode */
2690	tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2691	tmp &= ~RADEON_PM_MODE_SEL;
2692	WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2693
2694}
2695
2696/*
2697 * VRAM info
2698 */
2699static void r100_vram_get_type(struct radeon_device *rdev)
2700{
2701	uint32_t tmp;
2702
2703	rdev->mc.vram_is_ddr = false;
2704	if (rdev->flags & RADEON_IS_IGP)
2705		rdev->mc.vram_is_ddr = true;
2706	else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2707		rdev->mc.vram_is_ddr = true;
2708	if ((rdev->family == CHIP_RV100) ||
2709	    (rdev->family == CHIP_RS100) ||
2710	    (rdev->family == CHIP_RS200)) {
2711		tmp = RREG32(RADEON_MEM_CNTL);
2712		if (tmp & RV100_HALF_MODE) {
2713			rdev->mc.vram_width = 32;
2714		} else {
2715			rdev->mc.vram_width = 64;
2716		}
2717		if (rdev->flags & RADEON_SINGLE_CRTC) {
2718			rdev->mc.vram_width /= 4;
2719			rdev->mc.vram_is_ddr = true;
2720		}
2721	} else if (rdev->family <= CHIP_RV280) {
2722		tmp = RREG32(RADEON_MEM_CNTL);
2723		if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2724			rdev->mc.vram_width = 128;
2725		} else {
2726			rdev->mc.vram_width = 64;
2727		}
2728	} else {
2729		/* newer IGPs */
2730		rdev->mc.vram_width = 128;
2731	}
2732}
2733
2734static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2735{
2736	u32 aper_size;
2737	u8 byte;
2738
2739	aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2740
2741	/* Set HDP_APER_CNTL only on cards that are known not to be broken,
2742	 * that is has the 2nd generation multifunction PCI interface
2743	 */
2744	if (rdev->family == CHIP_RV280 ||
2745	    rdev->family >= CHIP_RV350) {
2746		WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2747		       ~RADEON_HDP_APER_CNTL);
2748		DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2749		return aper_size * 2;
2750	}
2751
2752	/* Older cards have all sorts of funny issues to deal with. First
2753	 * check if it's a multifunction card by reading the PCI config
2754	 * header type... Limit those to one aperture size
2755	 */
2756	pci_read_config_byte(rdev->pdev, 0xe, &byte);
2757	if (byte & 0x80) {
2758		DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2759		DRM_INFO("Limiting VRAM to one aperture\n");
2760		return aper_size;
2761	}
2762
2763	/* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2764	 * have set it up. We don't write this as it's broken on some ASICs but
2765	 * we expect the BIOS to have done the right thing (might be too optimistic...)
2766	 */
2767	if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2768		return aper_size * 2;
2769	return aper_size;
2770}
2771
2772void r100_vram_init_sizes(struct radeon_device *rdev)
2773{
2774	u64 config_aper_size;
2775
2776	/* work out accessible VRAM */
2777	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2778	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2779	rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2780	/* FIXME we don't use the second aperture yet when we could use it */
2781	if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2782		rdev->mc.visible_vram_size = rdev->mc.aper_size;
2783	config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2784	if (rdev->flags & RADEON_IS_IGP) {
2785		uint32_t tom;
2786		/* read NB_TOM to get the amount of ram stolen for the GPU */
2787		tom = RREG32(RADEON_NB_TOM);
2788		rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2789		WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2790		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2791	} else {
2792		rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2793		/* Some production boards of m6 will report 0
2794		 * if it's 8 MB
2795		 */
2796		if (rdev->mc.real_vram_size == 0) {
2797			rdev->mc.real_vram_size = 8192 * 1024;
2798			WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2799		}
2800		/* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM - 
2801		 * Novell bug 204882 + along with lots of ubuntu ones
2802		 */
2803		if (rdev->mc.aper_size > config_aper_size)
2804			config_aper_size = rdev->mc.aper_size;
2805
2806		if (config_aper_size > rdev->mc.real_vram_size)
2807			rdev->mc.mc_vram_size = config_aper_size;
2808		else
2809			rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2810	}
2811}
2812
2813void r100_vga_set_state(struct radeon_device *rdev, bool state)
2814{
2815	uint32_t temp;
2816
2817	temp = RREG32(RADEON_CONFIG_CNTL);
2818	if (!state) {
2819		temp &= ~RADEON_CFG_VGA_RAM_EN;
2820		temp |= RADEON_CFG_VGA_IO_DIS;
2821	} else {
2822		temp &= ~RADEON_CFG_VGA_IO_DIS;
2823	}
2824	WREG32(RADEON_CONFIG_CNTL, temp);
2825}
2826
2827static void r100_mc_init(struct radeon_device *rdev)
2828{
2829	u64 base;
2830
2831	r100_vram_get_type(rdev);
2832	r100_vram_init_sizes(rdev);
2833	base = rdev->mc.aper_base;
2834	if (rdev->flags & RADEON_IS_IGP)
2835		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2836	radeon_vram_location(rdev, &rdev->mc, base);
2837	rdev->mc.gtt_base_align = 0;
2838	if (!(rdev->flags & RADEON_IS_AGP))
2839		radeon_gtt_location(rdev, &rdev->mc);
2840	radeon_update_bandwidth_info(rdev);
2841}
2842
2843
2844/*
2845 * Indirect registers accessor
2846 */
2847void r100_pll_errata_after_index(struct radeon_device *rdev)
2848{
2849	if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2850		(void)RREG32(RADEON_CLOCK_CNTL_DATA);
2851		(void)RREG32(RADEON_CRTC_GEN_CNTL);
2852	}
2853}
2854
2855static void r100_pll_errata_after_data(struct radeon_device *rdev)
2856{
2857	/* This workarounds is necessary on RV100, RS100 and RS200 chips
2858	 * or the chip could hang on a subsequent access
2859	 */
2860	if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2861		mdelay(5);
2862	}
2863
2864	/* This function is required to workaround a hardware bug in some (all?)
2865	 * revisions of the R300.  This workaround should be called after every
2866	 * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2867	 * may not be correct.
2868	 */
2869	if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2870		uint32_t save, tmp;
2871
2872		save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2873		tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2874		WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2875		tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2876		WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2877	}
2878}
2879
2880uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2881{
2882	unsigned long flags;
2883	uint32_t data;
2884
2885	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2886	WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2887	r100_pll_errata_after_index(rdev);
2888	data = RREG32(RADEON_CLOCK_CNTL_DATA);
2889	r100_pll_errata_after_data(rdev);
2890	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2891	return data;
2892}
2893
2894void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2895{
2896	unsigned long flags;
2897
2898	spin_lock_irqsave(&rdev->pll_idx_lock, flags);
2899	WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2900	r100_pll_errata_after_index(rdev);
2901	WREG32(RADEON_CLOCK_CNTL_DATA, v);
2902	r100_pll_errata_after_data(rdev);
2903	spin_unlock_irqrestore(&rdev->pll_idx_lock, flags);
2904}
2905
2906static void r100_set_safe_registers(struct radeon_device *rdev)
2907{
2908	if (ASIC_IS_RN50(rdev)) {
2909		rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2910		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2911	} else if (rdev->family < CHIP_R200) {
2912		rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2913		rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2914	} else {
2915		r200_set_safe_registers(rdev);
2916	}
2917}
2918
2919/*
2920 * Debugfs info
2921 */
2922#if defined(CONFIG_DEBUG_FS)
2923static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2924{
2925	struct drm_info_node *node = (struct drm_info_node *) m->private;
2926	struct drm_device *dev = node->minor->dev;
2927	struct radeon_device *rdev = dev->dev_private;
2928	uint32_t reg, value;
2929	unsigned i;
2930
2931	seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2932	seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2933	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2934	for (i = 0; i < 64; i++) {
2935		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2936		reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2937		WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2938		value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2939		seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2940	}
2941	return 0;
2942}
2943
2944static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2945{
2946	struct drm_info_node *node = (struct drm_info_node *) m->private;
2947	struct drm_device *dev = node->minor->dev;
2948	struct radeon_device *rdev = dev->dev_private;
2949	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2950	uint32_t rdp, wdp;
2951	unsigned count, i, j;
2952
2953	radeon_ring_free_size(rdev, ring);
2954	rdp = RREG32(RADEON_CP_RB_RPTR);
2955	wdp = RREG32(RADEON_CP_RB_WPTR);
2956	count = (rdp + ring->ring_size - wdp) & ring->ptr_mask;
2957	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2958	seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2959	seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2960	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
2961	seq_printf(m, "%u dwords in ring\n", count);
2962	if (ring->ready) {
2963		for (j = 0; j <= count; j++) {
2964			i = (rdp + j) & ring->ptr_mask;
2965			seq_printf(m, "r[%04d]=0x%08x\n", i, ring->ring[i]);
2966		}
2967	}
2968	return 0;
2969}
2970
2971
2972static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2973{
2974	struct drm_info_node *node = (struct drm_info_node *) m->private;
2975	struct drm_device *dev = node->minor->dev;
2976	struct radeon_device *rdev = dev->dev_private;
2977	uint32_t csq_stat, csq2_stat, tmp;
2978	unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2979	unsigned i;
2980
2981	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2982	seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2983	csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2984	csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2985	r_rptr = (csq_stat >> 0) & 0x3ff;
2986	r_wptr = (csq_stat >> 10) & 0x3ff;
2987	ib1_rptr = (csq_stat >> 20) & 0x3ff;
2988	ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2989	ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2990	ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2991	seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2992	seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2993	seq_printf(m, "Ring rptr %u\n", r_rptr);
2994	seq_printf(m, "Ring wptr %u\n", r_wptr);
2995	seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2996	seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2997	seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2998	seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2999	/* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
3000	 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
3001	seq_printf(m, "Ring fifo:\n");
3002	for (i = 0; i < 256; i++) {
3003		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3004		tmp = RREG32(RADEON_CP_CSQ_DATA);
3005		seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
3006	}
3007	seq_printf(m, "Indirect1 fifo:\n");
3008	for (i = 256; i <= 512; i++) {
3009		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3010		tmp = RREG32(RADEON_CP_CSQ_DATA);
3011		seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
3012	}
3013	seq_printf(m, "Indirect2 fifo:\n");
3014	for (i = 640; i < ib1_wptr; i++) {
3015		WREG32(RADEON_CP_CSQ_ADDR, i << 2);
3016		tmp = RREG32(RADEON_CP_CSQ_DATA);
3017		seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
3018	}
3019	return 0;
3020}
3021
3022static int r100_debugfs_mc_info(struct seq_file *m, void *data)
3023{
3024	struct drm_info_node *node = (struct drm_info_node *) m->private;
3025	struct drm_device *dev = node->minor->dev;
3026	struct radeon_device *rdev = dev->dev_private;
3027	uint32_t tmp;
3028
3029	tmp = RREG32(RADEON_CONFIG_MEMSIZE);
3030	seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
3031	tmp = RREG32(RADEON_MC_FB_LOCATION);
3032	seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
3033	tmp = RREG32(RADEON_BUS_CNTL);
3034	seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
3035	tmp = RREG32(RADEON_MC_AGP_LOCATION);
3036	seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
3037	tmp = RREG32(RADEON_AGP_BASE);
3038	seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
3039	tmp = RREG32(RADEON_HOST_PATH_CNTL);
3040	seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
3041	tmp = RREG32(0x01D0);
3042	seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
3043	tmp = RREG32(RADEON_AIC_LO_ADDR);
3044	seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
3045	tmp = RREG32(RADEON_AIC_HI_ADDR);
3046	seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
3047	tmp = RREG32(0x01E4);
3048	seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
3049	return 0;
3050}
3051
3052static struct drm_info_list r100_debugfs_rbbm_list[] = {
3053	{"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
3054};
3055
3056static struct drm_info_list r100_debugfs_cp_list[] = {
3057	{"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
3058	{"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
3059};
3060
3061static struct drm_info_list r100_debugfs_mc_info_list[] = {
3062	{"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
3063};
3064#endif
3065
3066int r100_debugfs_rbbm_init(struct radeon_device *rdev)
3067{
3068#if defined(CONFIG_DEBUG_FS)
3069	return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
3070#else
3071	return 0;
 
3072#endif
3073}
3074
3075int r100_debugfs_cp_init(struct radeon_device *rdev)
3076{
3077#if defined(CONFIG_DEBUG_FS)
3078	return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
3079#else
3080	return 0;
 
 
 
3081#endif
3082}
3083
3084int r100_debugfs_mc_info_init(struct radeon_device *rdev)
3085{
3086#if defined(CONFIG_DEBUG_FS)
3087	return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
3088#else
3089	return 0;
 
3090#endif
3091}
3092
3093int r100_set_surface_reg(struct radeon_device *rdev, int reg,
3094			 uint32_t tiling_flags, uint32_t pitch,
3095			 uint32_t offset, uint32_t obj_size)
3096{
3097	int surf_index = reg * 16;
3098	int flags = 0;
3099
3100	if (rdev->family <= CHIP_RS200) {
3101		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3102				 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3103			flags |= RADEON_SURF_TILE_COLOR_BOTH;
3104		if (tiling_flags & RADEON_TILING_MACRO)
3105			flags |= RADEON_SURF_TILE_COLOR_MACRO;
3106		/* setting pitch to 0 disables tiling */
3107		if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
3108				== 0)
3109			pitch = 0;
3110	} else if (rdev->family <= CHIP_RV280) {
3111		if (tiling_flags & (RADEON_TILING_MACRO))
3112			flags |= R200_SURF_TILE_COLOR_MACRO;
3113		if (tiling_flags & RADEON_TILING_MICRO)
3114			flags |= R200_SURF_TILE_COLOR_MICRO;
3115	} else {
3116		if (tiling_flags & RADEON_TILING_MACRO)
3117			flags |= R300_SURF_TILE_MACRO;
3118		if (tiling_flags & RADEON_TILING_MICRO)
3119			flags |= R300_SURF_TILE_MICRO;
3120	}
3121
3122	if (tiling_flags & RADEON_TILING_SWAP_16BIT)
3123		flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
3124	if (tiling_flags & RADEON_TILING_SWAP_32BIT)
3125		flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
3126
3127	/* r100/r200 divide by 16 */
3128	if (rdev->family < CHIP_R300)
3129		flags |= pitch / 16;
3130	else
3131		flags |= pitch / 8;
3132
3133
3134	DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
3135	WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
3136	WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
3137	WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
3138	return 0;
3139}
3140
3141void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
3142{
3143	int surf_index = reg * 16;
3144	WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
3145}
3146
3147void r100_bandwidth_update(struct radeon_device *rdev)
3148{
3149	fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
3150	fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
3151	fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff;
3152	fixed20_12 crit_point_ff = {0};
3153	uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
3154	fixed20_12 memtcas_ff[8] = {
3155		dfixed_init(1),
3156		dfixed_init(2),
3157		dfixed_init(3),
3158		dfixed_init(0),
3159		dfixed_init_half(1),
3160		dfixed_init_half(2),
3161		dfixed_init(0),
3162	};
3163	fixed20_12 memtcas_rs480_ff[8] = {
3164		dfixed_init(0),
3165		dfixed_init(1),
3166		dfixed_init(2),
3167		dfixed_init(3),
3168		dfixed_init(0),
3169		dfixed_init_half(1),
3170		dfixed_init_half(2),
3171		dfixed_init_half(3),
3172	};
3173	fixed20_12 memtcas2_ff[8] = {
3174		dfixed_init(0),
3175		dfixed_init(1),
3176		dfixed_init(2),
3177		dfixed_init(3),
3178		dfixed_init(4),
3179		dfixed_init(5),
3180		dfixed_init(6),
3181		dfixed_init(7),
3182	};
3183	fixed20_12 memtrbs[8] = {
3184		dfixed_init(1),
3185		dfixed_init_half(1),
3186		dfixed_init(2),
3187		dfixed_init_half(2),
3188		dfixed_init(3),
3189		dfixed_init_half(3),
3190		dfixed_init(4),
3191		dfixed_init_half(4)
3192	};
3193	fixed20_12 memtrbs_r4xx[8] = {
3194		dfixed_init(4),
3195		dfixed_init(5),
3196		dfixed_init(6),
3197		dfixed_init(7),
3198		dfixed_init(8),
3199		dfixed_init(9),
3200		dfixed_init(10),
3201		dfixed_init(11)
3202	};
3203	fixed20_12 min_mem_eff;
3204	fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
3205	fixed20_12 cur_latency_mclk, cur_latency_sclk;
3206	fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate = {0},
3207		disp_drain_rate2, read_return_rate;
3208	fixed20_12 time_disp1_drop_priority;
3209	int c;
3210	int cur_size = 16;       /* in octawords */
3211	int critical_point = 0, critical_point2;
3212/* 	uint32_t read_return_rate, time_disp1_drop_priority; */
3213	int stop_req, max_stop_req;
3214	struct drm_display_mode *mode1 = NULL;
3215	struct drm_display_mode *mode2 = NULL;
3216	uint32_t pixel_bytes1 = 0;
3217	uint32_t pixel_bytes2 = 0;
3218
3219	/* Guess line buffer size to be 8192 pixels */
3220	u32 lb_size = 8192;
3221
3222	if (!rdev->mode_info.mode_config_initialized)
3223		return;
3224
3225	radeon_update_display_priority(rdev);
3226
3227	if (rdev->mode_info.crtcs[0]->base.enabled) {
3228		const struct drm_framebuffer *fb =
3229			rdev->mode_info.crtcs[0]->base.primary->fb;
3230
3231		mode1 = &rdev->mode_info.crtcs[0]->base.mode;
3232		pixel_bytes1 = fb->format->cpp[0];
3233	}
3234	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3235		if (rdev->mode_info.crtcs[1]->base.enabled) {
3236			const struct drm_framebuffer *fb =
3237				rdev->mode_info.crtcs[1]->base.primary->fb;
3238
3239			mode2 = &rdev->mode_info.crtcs[1]->base.mode;
3240			pixel_bytes2 = fb->format->cpp[0];
3241		}
3242	}
3243
3244	min_mem_eff.full = dfixed_const_8(0);
3245	/* get modes */
3246	if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
3247		uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
3248		mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
3249		mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
3250		/* check crtc enables */
3251		if (mode2)
3252			mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
3253		if (mode1)
3254			mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
3255		WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
3256	}
3257
3258	/*
3259	 * determine is there is enough bw for current mode
3260	 */
3261	sclk_ff = rdev->pm.sclk;
3262	mclk_ff = rdev->pm.mclk;
3263
3264	temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
3265	temp_ff.full = dfixed_const(temp);
3266	mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
3267
3268	pix_clk.full = 0;
3269	pix_clk2.full = 0;
3270	peak_disp_bw.full = 0;
3271	if (mode1) {
3272		temp_ff.full = dfixed_const(1000);
3273		pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
3274		pix_clk.full = dfixed_div(pix_clk, temp_ff);
3275		temp_ff.full = dfixed_const(pixel_bytes1);
3276		peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
3277	}
3278	if (mode2) {
3279		temp_ff.full = dfixed_const(1000);
3280		pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
3281		pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
3282		temp_ff.full = dfixed_const(pixel_bytes2);
3283		peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
3284	}
3285
3286	mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
3287	if (peak_disp_bw.full >= mem_bw.full) {
3288		DRM_ERROR("You may not have enough display bandwidth for current mode\n"
3289			  "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
3290	}
3291
3292	/*  Get values from the EXT_MEM_CNTL register...converting its contents. */
3293	temp = RREG32(RADEON_MEM_TIMING_CNTL);
3294	if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
3295		mem_trcd = ((temp >> 2) & 0x3) + 1;
3296		mem_trp  = ((temp & 0x3)) + 1;
3297		mem_tras = ((temp & 0x70) >> 4) + 1;
3298	} else if (rdev->family == CHIP_R300 ||
3299		   rdev->family == CHIP_R350) { /* r300, r350 */
3300		mem_trcd = (temp & 0x7) + 1;
3301		mem_trp = ((temp >> 8) & 0x7) + 1;
3302		mem_tras = ((temp >> 11) & 0xf) + 4;
3303	} else if (rdev->family == CHIP_RV350 ||
3304		   rdev->family == CHIP_RV380) {
3305		/* rv3x0 */
3306		mem_trcd = (temp & 0x7) + 3;
3307		mem_trp = ((temp >> 8) & 0x7) + 3;
3308		mem_tras = ((temp >> 11) & 0xf) + 6;
3309	} else if (rdev->family == CHIP_R420 ||
3310		   rdev->family == CHIP_R423 ||
3311		   rdev->family == CHIP_RV410) {
3312		/* r4xx */
3313		mem_trcd = (temp & 0xf) + 3;
3314		if (mem_trcd > 15)
3315			mem_trcd = 15;
3316		mem_trp = ((temp >> 8) & 0xf) + 3;
3317		if (mem_trp > 15)
3318			mem_trp = 15;
3319		mem_tras = ((temp >> 12) & 0x1f) + 6;
3320		if (mem_tras > 31)
3321			mem_tras = 31;
3322	} else { /* RV200, R200 */
3323		mem_trcd = (temp & 0x7) + 1;
3324		mem_trp = ((temp >> 8) & 0x7) + 1;
3325		mem_tras = ((temp >> 12) & 0xf) + 4;
3326	}
3327	/* convert to FF */
3328	trcd_ff.full = dfixed_const(mem_trcd);
3329	trp_ff.full = dfixed_const(mem_trp);
3330	tras_ff.full = dfixed_const(mem_tras);
3331
3332	/* Get values from the MEM_SDRAM_MODE_REG register...converting its */
3333	temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
3334	data = (temp & (7 << 20)) >> 20;
3335	if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
3336		if (rdev->family == CHIP_RS480) /* don't think rs400 */
3337			tcas_ff = memtcas_rs480_ff[data];
3338		else
3339			tcas_ff = memtcas_ff[data];
3340	} else
3341		tcas_ff = memtcas2_ff[data];
3342
3343	if (rdev->family == CHIP_RS400 ||
3344	    rdev->family == CHIP_RS480) {
3345		/* extra cas latency stored in bits 23-25 0-4 clocks */
3346		data = (temp >> 23) & 0x7;
3347		if (data < 5)
3348			tcas_ff.full += dfixed_const(data);
3349	}
3350
3351	if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
3352		/* on the R300, Tcas is included in Trbs.
3353		 */
3354		temp = RREG32(RADEON_MEM_CNTL);
3355		data = (R300_MEM_NUM_CHANNELS_MASK & temp);
3356		if (data == 1) {
3357			if (R300_MEM_USE_CD_CH_ONLY & temp) {
3358				temp = RREG32(R300_MC_IND_INDEX);
3359				temp &= ~R300_MC_IND_ADDR_MASK;
3360				temp |= R300_MC_READ_CNTL_CD_mcind;
3361				WREG32(R300_MC_IND_INDEX, temp);
3362				temp = RREG32(R300_MC_IND_DATA);
3363				data = (R300_MEM_RBS_POSITION_C_MASK & temp);
3364			} else {
3365				temp = RREG32(R300_MC_READ_CNTL_AB);
3366				data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3367			}
3368		} else {
3369			temp = RREG32(R300_MC_READ_CNTL_AB);
3370			data = (R300_MEM_RBS_POSITION_A_MASK & temp);
3371		}
3372		if (rdev->family == CHIP_RV410 ||
3373		    rdev->family == CHIP_R420 ||
3374		    rdev->family == CHIP_R423)
3375			trbs_ff = memtrbs_r4xx[data];
3376		else
3377			trbs_ff = memtrbs[data];
3378		tcas_ff.full += trbs_ff.full;
3379	}
3380
3381	sclk_eff_ff.full = sclk_ff.full;
3382
3383	if (rdev->flags & RADEON_IS_AGP) {
3384		fixed20_12 agpmode_ff;
3385		agpmode_ff.full = dfixed_const(radeon_agpmode);
3386		temp_ff.full = dfixed_const_666(16);
3387		sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
3388	}
3389	/* TODO PCIE lanes may affect this - agpmode == 16?? */
3390
3391	if (ASIC_IS_R300(rdev)) {
3392		sclk_delay_ff.full = dfixed_const(250);
3393	} else {
3394		if ((rdev->family == CHIP_RV100) ||
3395		    rdev->flags & RADEON_IS_IGP) {
3396			if (rdev->mc.vram_is_ddr)
3397				sclk_delay_ff.full = dfixed_const(41);
3398			else
3399				sclk_delay_ff.full = dfixed_const(33);
3400		} else {
3401			if (rdev->mc.vram_width == 128)
3402				sclk_delay_ff.full = dfixed_const(57);
3403			else
3404				sclk_delay_ff.full = dfixed_const(41);
3405		}
3406	}
3407
3408	mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
3409
3410	if (rdev->mc.vram_is_ddr) {
3411		if (rdev->mc.vram_width == 32) {
3412			k1.full = dfixed_const(40);
3413			c  = 3;
3414		} else {
3415			k1.full = dfixed_const(20);
3416			c  = 1;
3417		}
3418	} else {
3419		k1.full = dfixed_const(40);
3420		c  = 3;
3421	}
3422
3423	temp_ff.full = dfixed_const(2);
3424	mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
3425	temp_ff.full = dfixed_const(c);
3426	mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
3427	temp_ff.full = dfixed_const(4);
3428	mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
3429	mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
3430	mc_latency_mclk.full += k1.full;
3431
3432	mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
3433	mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
3434
3435	/*
3436	  HW cursor time assuming worst case of full size colour cursor.
3437	*/
3438	temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
3439	temp_ff.full += trcd_ff.full;
3440	if (temp_ff.full < tras_ff.full)
3441		temp_ff.full = tras_ff.full;
3442	cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
3443
3444	temp_ff.full = dfixed_const(cur_size);
3445	cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
3446	/*
3447	  Find the total latency for the display data.
3448	*/
3449	disp_latency_overhead.full = dfixed_const(8);
3450	disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
3451	mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
3452	mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
3453
3454	if (mc_latency_mclk.full > mc_latency_sclk.full)
3455		disp_latency.full = mc_latency_mclk.full;
3456	else
3457		disp_latency.full = mc_latency_sclk.full;
3458
3459	/* setup Max GRPH_STOP_REQ default value */
3460	if (ASIC_IS_RV100(rdev))
3461		max_stop_req = 0x5c;
3462	else
3463		max_stop_req = 0x7c;
3464
3465	if (mode1) {
3466		/*  CRTC1
3467		    Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
3468		    GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
3469		*/
3470		stop_req = mode1->hdisplay * pixel_bytes1 / 16;
3471
3472		if (stop_req > max_stop_req)
3473			stop_req = max_stop_req;
3474
3475		/*
3476		  Find the drain rate of the display buffer.
3477		*/
3478		temp_ff.full = dfixed_const((16/pixel_bytes1));
3479		disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
3480
3481		/*
3482		  Find the critical point of the display buffer.
3483		*/
3484		crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
3485		crit_point_ff.full += dfixed_const_half(0);
3486
3487		critical_point = dfixed_trunc(crit_point_ff);
3488
3489		if (rdev->disp_priority == 2) {
3490			critical_point = 0;
3491		}
3492
3493		/*
3494		  The critical point should never be above max_stop_req-4.  Setting
3495		  GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3496		*/
3497		if (max_stop_req - critical_point < 4)
3498			critical_point = 0;
3499
3500		if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3501			/* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3502			critical_point = 0x10;
3503		}
3504
3505		temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3506		temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3507		temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3508		temp &= ~(RADEON_GRPH_START_REQ_MASK);
3509		if ((rdev->family == CHIP_R350) &&
3510		    (stop_req > 0x15)) {
3511			stop_req -= 0x10;
3512		}
3513		temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3514		temp |= RADEON_GRPH_BUFFER_SIZE;
3515		temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3516			  RADEON_GRPH_CRITICAL_AT_SOF |
3517			  RADEON_GRPH_STOP_CNTL);
3518		/*
3519		  Write the result into the register.
3520		*/
3521		WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3522						       (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3523
3524#if 0
3525		if ((rdev->family == CHIP_RS400) ||
3526		    (rdev->family == CHIP_RS480)) {
3527			/* attempt to program RS400 disp regs correctly ??? */
3528			temp = RREG32(RS400_DISP1_REG_CNTL);
3529			temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3530				  RS400_DISP1_STOP_REQ_LEVEL_MASK);
3531			WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3532						       (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3533						       (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3534			temp = RREG32(RS400_DMIF_MEM_CNTL1);
3535			temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3536				  RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3537			WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3538						      (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3539						      (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3540		}
3541#endif
3542
3543		DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3544			  /* 	  (unsigned int)info->SavedReg->grph_buffer_cntl, */
3545			  (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3546	}
3547
3548	if (mode2) {
3549		u32 grph2_cntl;
3550		stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3551
3552		if (stop_req > max_stop_req)
3553			stop_req = max_stop_req;
3554
3555		/*
3556		  Find the drain rate of the display buffer.
3557		*/
3558		temp_ff.full = dfixed_const((16/pixel_bytes2));
3559		disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3560
3561		grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3562		grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3563		grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3564		grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3565		if ((rdev->family == CHIP_R350) &&
3566		    (stop_req > 0x15)) {
3567			stop_req -= 0x10;
3568		}
3569		grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3570		grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3571		grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
3572			  RADEON_GRPH_CRITICAL_AT_SOF |
3573			  RADEON_GRPH_STOP_CNTL);
3574
3575		if ((rdev->family == CHIP_RS100) ||
3576		    (rdev->family == CHIP_RS200))
3577			critical_point2 = 0;
3578		else {
3579			temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3580			temp_ff.full = dfixed_const(temp);
3581			temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3582			if (sclk_ff.full < temp_ff.full)
3583				temp_ff.full = sclk_ff.full;
3584
3585			read_return_rate.full = temp_ff.full;
3586
3587			if (mode1) {
3588				temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3589				time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3590			} else {
3591				time_disp1_drop_priority.full = 0;
3592			}
3593			crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3594			crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3595			crit_point_ff.full += dfixed_const_half(0);
3596
3597			critical_point2 = dfixed_trunc(crit_point_ff);
3598
3599			if (rdev->disp_priority == 2) {
3600				critical_point2 = 0;
3601			}
3602
3603			if (max_stop_req - critical_point2 < 4)
3604				critical_point2 = 0;
3605
3606		}
3607
3608		if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3609			/* some R300 cards have problem with this set to 0 */
3610			critical_point2 = 0x10;
3611		}
3612
3613		WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3614						  (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3615
3616		if ((rdev->family == CHIP_RS400) ||
3617		    (rdev->family == CHIP_RS480)) {
3618#if 0
3619			/* attempt to program RS400 disp2 regs correctly ??? */
3620			temp = RREG32(RS400_DISP2_REQ_CNTL1);
3621			temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3622				  RS400_DISP2_STOP_REQ_LEVEL_MASK);
3623			WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3624						       (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3625						       (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3626			temp = RREG32(RS400_DISP2_REQ_CNTL2);
3627			temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3628				  RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3629			WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3630						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3631						       (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3632#endif
3633			WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3634			WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3635			WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
3636			WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3637		}
3638
3639		DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3640			  (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3641	}
3642
3643	/* Save number of lines the linebuffer leads before the scanout */
3644	if (mode1)
3645	    rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
3646
3647	if (mode2)
3648	    rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
3649}
3650
3651int r100_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
3652{
3653	uint32_t scratch;
3654	uint32_t tmp = 0;
3655	unsigned i;
3656	int r;
3657
3658	r = radeon_scratch_get(rdev, &scratch);
3659	if (r) {
3660		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3661		return r;
3662	}
3663	WREG32(scratch, 0xCAFEDEAD);
3664	r = radeon_ring_lock(rdev, ring, 2);
3665	if (r) {
3666		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3667		radeon_scratch_free(rdev, scratch);
3668		return r;
3669	}
3670	radeon_ring_write(ring, PACKET0(scratch, 0));
3671	radeon_ring_write(ring, 0xDEADBEEF);
3672	radeon_ring_unlock_commit(rdev, ring, false);
3673	for (i = 0; i < rdev->usec_timeout; i++) {
3674		tmp = RREG32(scratch);
3675		if (tmp == 0xDEADBEEF) {
3676			break;
3677		}
3678		udelay(1);
3679	}
3680	if (i < rdev->usec_timeout) {
3681		DRM_INFO("ring test succeeded in %d usecs\n", i);
3682	} else {
3683		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
3684			  scratch, tmp);
3685		r = -EINVAL;
3686	}
3687	radeon_scratch_free(rdev, scratch);
3688	return r;
3689}
3690
3691void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3692{
3693	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
3694
3695	if (ring->rptr_save_reg) {
3696		u32 next_rptr = ring->wptr + 2 + 3;
3697		radeon_ring_write(ring, PACKET0(ring->rptr_save_reg, 0));
3698		radeon_ring_write(ring, next_rptr);
3699	}
3700
3701	radeon_ring_write(ring, PACKET0(RADEON_CP_IB_BASE, 1));
3702	radeon_ring_write(ring, ib->gpu_addr);
3703	radeon_ring_write(ring, ib->length_dw);
3704}
3705
3706int r100_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
3707{
3708	struct radeon_ib ib;
3709	uint32_t scratch;
3710	uint32_t tmp = 0;
3711	unsigned i;
3712	int r;
3713
3714	r = radeon_scratch_get(rdev, &scratch);
3715	if (r) {
3716		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3717		return r;
3718	}
3719	WREG32(scratch, 0xCAFEDEAD);
3720	r = radeon_ib_get(rdev, RADEON_RING_TYPE_GFX_INDEX, &ib, NULL, 256);
3721	if (r) {
3722		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
3723		goto free_scratch;
3724	}
3725	ib.ptr[0] = PACKET0(scratch, 0);
3726	ib.ptr[1] = 0xDEADBEEF;
3727	ib.ptr[2] = PACKET2(0);
3728	ib.ptr[3] = PACKET2(0);
3729	ib.ptr[4] = PACKET2(0);
3730	ib.ptr[5] = PACKET2(0);
3731	ib.ptr[6] = PACKET2(0);
3732	ib.ptr[7] = PACKET2(0);
3733	ib.length_dw = 8;
3734	r = radeon_ib_schedule(rdev, &ib, NULL, false);
3735	if (r) {
3736		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
3737		goto free_ib;
3738	}
3739	r = radeon_fence_wait_timeout(ib.fence, false, usecs_to_jiffies(
3740		RADEON_USEC_IB_TEST_TIMEOUT));
3741	if (r < 0) {
3742		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
3743		goto free_ib;
3744	} else if (r == 0) {
3745		DRM_ERROR("radeon: fence wait timed out.\n");
3746		r = -ETIMEDOUT;
3747		goto free_ib;
3748	}
3749	r = 0;
3750	for (i = 0; i < rdev->usec_timeout; i++) {
3751		tmp = RREG32(scratch);
3752		if (tmp == 0xDEADBEEF) {
3753			break;
3754		}
3755		udelay(1);
3756	}
3757	if (i < rdev->usec_timeout) {
3758		DRM_INFO("ib test succeeded in %u usecs\n", i);
3759	} else {
3760		DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n",
3761			  scratch, tmp);
3762		r = -EINVAL;
3763	}
3764free_ib:
3765	radeon_ib_free(rdev, &ib);
3766free_scratch:
3767	radeon_scratch_free(rdev, scratch);
3768	return r;
3769}
3770
3771void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3772{
3773	/* Shutdown CP we shouldn't need to do that but better be safe than
3774	 * sorry
3775	 */
3776	rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
3777	WREG32(R_000740_CP_CSQ_CNTL, 0);
3778
3779	/* Save few CRTC registers */
3780	save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3781	save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3782	save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3783	save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3784	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3785		save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3786		save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3787	}
3788
3789	/* Disable VGA aperture access */
3790	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3791	/* Disable cursor, overlay, crtc */
3792	WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3793	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3794					S_000054_CRTC_DISPLAY_DIS(1));
3795	WREG32(R_000050_CRTC_GEN_CNTL,
3796			(C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3797			S_000050_CRTC_DISP_REQ_EN_B(1));
3798	WREG32(R_000420_OV0_SCALE_CNTL,
3799		C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3800	WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3801	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3802		WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3803						S_000360_CUR2_LOCK(1));
3804		WREG32(R_0003F8_CRTC2_GEN_CNTL,
3805			(C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3806			S_0003F8_CRTC2_DISPLAY_DIS(1) |
3807			S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3808		WREG32(R_000360_CUR2_OFFSET,
3809			C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3810	}
3811}
3812
3813void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3814{
3815	/* Update base address for crtc */
3816	WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3817	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3818		WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3819	}
3820	/* Restore CRTC registers */
3821	WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3822	WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3823	WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3824	if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3825		WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3826	}
3827}
3828
3829void r100_vga_render_disable(struct radeon_device *rdev)
3830{
3831	u32 tmp;
3832
3833	tmp = RREG8(R_0003C2_GENMO_WT);
3834	WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3835}
3836
3837static void r100_debugfs(struct radeon_device *rdev)
3838{
3839	int r;
3840
3841	r = r100_debugfs_mc_info_init(rdev);
3842	if (r)
3843		dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3844}
3845
3846static void r100_mc_program(struct radeon_device *rdev)
3847{
3848	struct r100_mc_save save;
3849
3850	/* Stops all mc clients */
3851	r100_mc_stop(rdev, &save);
3852	if (rdev->flags & RADEON_IS_AGP) {
3853		WREG32(R_00014C_MC_AGP_LOCATION,
3854			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3855			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3856		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3857		if (rdev->family > CHIP_RV200)
3858			WREG32(R_00015C_AGP_BASE_2,
3859				upper_32_bits(rdev->mc.agp_base) & 0xff);
3860	} else {
3861		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3862		WREG32(R_000170_AGP_BASE, 0);
3863		if (rdev->family > CHIP_RV200)
3864			WREG32(R_00015C_AGP_BASE_2, 0);
3865	}
3866	/* Wait for mc idle */
3867	if (r100_mc_wait_for_idle(rdev))
3868		dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3869	/* Program MC, should be a 32bits limited address space */
3870	WREG32(R_000148_MC_FB_LOCATION,
3871		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3872		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3873	r100_mc_resume(rdev, &save);
3874}
3875
3876static void r100_clock_startup(struct radeon_device *rdev)
3877{
3878	u32 tmp;
3879
3880	if (radeon_dynclks != -1 && radeon_dynclks)
3881		radeon_legacy_set_clock_gating(rdev, 1);
3882	/* We need to force on some of the block */
3883	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3884	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3885	if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3886		tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3887	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3888}
3889
3890static int r100_startup(struct radeon_device *rdev)
3891{
3892	int r;
3893
3894	/* set common regs */
3895	r100_set_common_regs(rdev);
3896	/* program mc */
3897	r100_mc_program(rdev);
3898	/* Resume clock */
3899	r100_clock_startup(rdev);
3900	/* Initialize GART (initialize after TTM so we can allocate
3901	 * memory through TTM but finalize after TTM) */
3902	r100_enable_bm(rdev);
3903	if (rdev->flags & RADEON_IS_PCI) {
3904		r = r100_pci_gart_enable(rdev);
3905		if (r)
3906			return r;
3907	}
3908
3909	/* allocate wb buffer */
3910	r = radeon_wb_init(rdev);
3911	if (r)
3912		return r;
3913
3914	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
3915	if (r) {
3916		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
3917		return r;
3918	}
3919
3920	/* Enable IRQ */
3921	if (!rdev->irq.installed) {
3922		r = radeon_irq_kms_init(rdev);
3923		if (r)
3924			return r;
3925	}
3926
3927	r100_irq_set(rdev);
3928	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3929	/* 1M ring buffer */
3930	r = r100_cp_init(rdev, 1024 * 1024);
3931	if (r) {
3932		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
3933		return r;
3934	}
3935
3936	r = radeon_ib_pool_init(rdev);
3937	if (r) {
3938		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
3939		return r;
3940	}
3941
3942	return 0;
3943}
3944
3945int r100_resume(struct radeon_device *rdev)
3946{
3947	int r;
3948
3949	/* Make sur GART are not working */
3950	if (rdev->flags & RADEON_IS_PCI)
3951		r100_pci_gart_disable(rdev);
3952	/* Resume clock before doing reset */
3953	r100_clock_startup(rdev);
3954	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
3955	if (radeon_asic_reset(rdev)) {
3956		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3957			RREG32(R_000E40_RBBM_STATUS),
3958			RREG32(R_0007C0_CP_STAT));
3959	}
3960	/* post */
3961	radeon_combios_asic_init(rdev->ddev);
3962	/* Resume clock after posting */
3963	r100_clock_startup(rdev);
3964	/* Initialize surface registers */
3965	radeon_surface_init(rdev);
3966
3967	rdev->accel_working = true;
3968	r = r100_startup(rdev);
3969	if (r) {
3970		rdev->accel_working = false;
3971	}
3972	return r;
3973}
3974
3975int r100_suspend(struct radeon_device *rdev)
3976{
3977	radeon_pm_suspend(rdev);
3978	r100_cp_disable(rdev);
3979	radeon_wb_disable(rdev);
3980	r100_irq_disable(rdev);
3981	if (rdev->flags & RADEON_IS_PCI)
3982		r100_pci_gart_disable(rdev);
3983	return 0;
3984}
3985
3986void r100_fini(struct radeon_device *rdev)
3987{
3988	radeon_pm_fini(rdev);
3989	r100_cp_fini(rdev);
3990	radeon_wb_fini(rdev);
3991	radeon_ib_pool_fini(rdev);
3992	radeon_gem_fini(rdev);
3993	if (rdev->flags & RADEON_IS_PCI)
3994		r100_pci_gart_fini(rdev);
3995	radeon_agp_fini(rdev);
3996	radeon_irq_kms_fini(rdev);
3997	radeon_fence_driver_fini(rdev);
3998	radeon_bo_fini(rdev);
3999	radeon_atombios_fini(rdev);
4000	kfree(rdev->bios);
4001	rdev->bios = NULL;
4002}
4003
4004/*
4005 * Due to how kexec works, it can leave the hw fully initialised when it
4006 * boots the new kernel. However doing our init sequence with the CP and
4007 * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
4008 * do some quick sanity checks and restore sane values to avoid this
4009 * problem.
4010 */
4011void r100_restore_sanity(struct radeon_device *rdev)
4012{
4013	u32 tmp;
4014
4015	tmp = RREG32(RADEON_CP_CSQ_CNTL);
4016	if (tmp) {
4017		WREG32(RADEON_CP_CSQ_CNTL, 0);
4018	}
4019	tmp = RREG32(RADEON_CP_RB_CNTL);
4020	if (tmp) {
4021		WREG32(RADEON_CP_RB_CNTL, 0);
4022	}
4023	tmp = RREG32(RADEON_SCRATCH_UMSK);
4024	if (tmp) {
4025		WREG32(RADEON_SCRATCH_UMSK, 0);
4026	}
4027}
4028
4029int r100_init(struct radeon_device *rdev)
4030{
4031	int r;
4032
4033	/* Register debugfs file specific to this group of asics */
4034	r100_debugfs(rdev);
4035	/* Disable VGA */
4036	r100_vga_render_disable(rdev);
4037	/* Initialize scratch registers */
4038	radeon_scratch_init(rdev);
4039	/* Initialize surface registers */
4040	radeon_surface_init(rdev);
4041	/* sanity check some register to avoid hangs like after kexec */
4042	r100_restore_sanity(rdev);
4043	/* TODO: disable VGA need to use VGA request */
4044	/* BIOS*/
4045	if (!radeon_get_bios(rdev)) {
4046		if (ASIC_IS_AVIVO(rdev))
4047			return -EINVAL;
4048	}
4049	if (rdev->is_atom_bios) {
4050		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
4051		return -EINVAL;
4052	} else {
4053		r = radeon_combios_init(rdev);
4054		if (r)
4055			return r;
4056	}
4057	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
4058	if (radeon_asic_reset(rdev)) {
4059		dev_warn(rdev->dev,
4060			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
4061			RREG32(R_000E40_RBBM_STATUS),
4062			RREG32(R_0007C0_CP_STAT));
4063	}
4064	/* check if cards are posted or not */
4065	if (radeon_boot_test_post_card(rdev) == false)
4066		return -EINVAL;
4067	/* Set asic errata */
4068	r100_errata(rdev);
4069	/* Initialize clocks */
4070	radeon_get_clock_info(rdev->ddev);
4071	/* initialize AGP */
4072	if (rdev->flags & RADEON_IS_AGP) {
4073		r = radeon_agp_init(rdev);
4074		if (r) {
4075			radeon_agp_disable(rdev);
4076		}
4077	}
4078	/* initialize VRAM */
4079	r100_mc_init(rdev);
4080	/* Fence driver */
4081	r = radeon_fence_driver_init(rdev);
4082	if (r)
4083		return r;
4084	/* Memory manager */
4085	r = radeon_bo_init(rdev);
4086	if (r)
4087		return r;
4088	if (rdev->flags & RADEON_IS_PCI) {
4089		r = r100_pci_gart_init(rdev);
4090		if (r)
4091			return r;
4092	}
4093	r100_set_safe_registers(rdev);
4094
4095	/* Initialize power management */
4096	radeon_pm_init(rdev);
4097
4098	rdev->accel_working = true;
4099	r = r100_startup(rdev);
4100	if (r) {
4101		/* Somethings want wront with the accel init stop accel */
4102		dev_err(rdev->dev, "Disabling GPU acceleration\n");
4103		r100_cp_fini(rdev);
4104		radeon_wb_fini(rdev);
4105		radeon_ib_pool_fini(rdev);
4106		radeon_irq_kms_fini(rdev);
4107		if (rdev->flags & RADEON_IS_PCI)
4108			r100_pci_gart_fini(rdev);
4109		rdev->accel_working = false;
4110	}
4111	return 0;
4112}
4113
4114uint32_t r100_mm_rreg_slow(struct radeon_device *rdev, uint32_t reg)
4115{
4116	unsigned long flags;
4117	uint32_t ret;
4118
4119	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4120	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4121	ret = readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4122	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4123	return ret;
4124}
4125
4126void r100_mm_wreg_slow(struct radeon_device *rdev, uint32_t reg, uint32_t v)
4127{
4128	unsigned long flags;
4129
4130	spin_lock_irqsave(&rdev->mmio_idx_lock, flags);
4131	writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
4132	writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
4133	spin_unlock_irqrestore(&rdev->mmio_idx_lock, flags);
4134}
4135
4136u32 r100_io_rreg(struct radeon_device *rdev, u32 reg)
4137{
4138	if (reg < rdev->rio_mem_size)
4139		return ioread32(rdev->rio_mem + reg);
4140	else {
4141		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4142		return ioread32(rdev->rio_mem + RADEON_MM_DATA);
4143	}
4144}
4145
4146void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v)
4147{
4148	if (reg < rdev->rio_mem_size)
4149		iowrite32(v, rdev->rio_mem + reg);
4150	else {
4151		iowrite32(reg, rdev->rio_mem + RADEON_MM_INDEX);
4152		iowrite32(v, rdev->rio_mem + RADEON_MM_DATA);
4153	}
4154}