Linux Audio

Check our new training course

Loading...
v4.6
   1/*
   2 * Copyright 2014 Advanced Micro Devices, Inc.
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice shall be included in
  12 * all copies or substantial portions of the Software.
  13 *
  14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20 * OTHER DEALINGS IN THE SOFTWARE.
  21 *
  22 */
 
  23#include <linux/firmware.h>
  24#include "drmP.h"
 
 
 
  25#include "amdgpu.h"
  26#include "gmc_v8_0.h"
  27#include "amdgpu_ucode.h"
 
 
  28
  29#include "gmc/gmc_8_1_d.h"
  30#include "gmc/gmc_8_1_sh_mask.h"
  31
  32#include "bif/bif_5_0_d.h"
  33#include "bif/bif_5_0_sh_mask.h"
  34
  35#include "oss/oss_3_0_d.h"
  36#include "oss/oss_3_0_sh_mask.h"
  37
 
 
 
  38#include "vid.h"
  39#include "vi.h"
  40
 
  41
  42static void gmc_v8_0_set_gart_funcs(struct amdgpu_device *adev);
 
 
  43static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
 
  44
  45MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
 
 
 
 
 
 
 
  46
  47static const u32 golden_settings_tonga_a11[] =
  48{
  49	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
  50	mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
  51	mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
  52	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  53	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  54	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  55	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  56};
  57
  58static const u32 tonga_mgcg_cgcg_init[] =
  59{
  60	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  61};
  62
  63static const u32 golden_settings_fiji_a10[] =
  64{
  65	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  66	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  67	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  68	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  69};
  70
  71static const u32 fiji_mgcg_cgcg_init[] =
  72{
  73	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  74};
  75
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  76static const u32 cz_mgcg_cgcg_init[] =
  77{
  78	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  79};
  80
  81static const u32 stoney_mgcg_cgcg_init[] =
  82{
 
  83	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  84};
  85
 
 
 
 
 
  86
  87static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
  88{
  89	switch (adev->asic_type) {
  90	case CHIP_FIJI:
  91		amdgpu_program_register_sequence(adev,
  92						 fiji_mgcg_cgcg_init,
  93						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
  94		amdgpu_program_register_sequence(adev,
  95						 golden_settings_fiji_a10,
  96						 (const u32)ARRAY_SIZE(golden_settings_fiji_a10));
  97		break;
  98	case CHIP_TONGA:
  99		amdgpu_program_register_sequence(adev,
 100						 tonga_mgcg_cgcg_init,
 101						 (const u32)ARRAY_SIZE(tonga_mgcg_cgcg_init));
 102		amdgpu_program_register_sequence(adev,
 103						 golden_settings_tonga_a11,
 104						 (const u32)ARRAY_SIZE(golden_settings_tonga_a11));
 
 
 
 
 
 
 
 
 
 
 
 
 105		break;
 106	case CHIP_CARRIZO:
 107		amdgpu_program_register_sequence(adev,
 108						 cz_mgcg_cgcg_init,
 109						 (const u32)ARRAY_SIZE(cz_mgcg_cgcg_init));
 110		break;
 111	case CHIP_STONEY:
 112		amdgpu_program_register_sequence(adev,
 113						 stoney_mgcg_cgcg_init,
 114						 (const u32)ARRAY_SIZE(stoney_mgcg_cgcg_init));
 
 
 
 115		break;
 116	default:
 117		break;
 118	}
 119}
 120
 121/**
 122 * gmc8_mc_wait_for_idle - wait for MC idle callback.
 123 *
 124 * @adev: amdgpu_device pointer
 125 *
 126 * Wait for the MC (memory controller) to be idle.
 127 * (evergreen+).
 128 * Returns 0 if the MC is idle, -1 if not.
 129 */
 130int gmc_v8_0_mc_wait_for_idle(struct amdgpu_device *adev)
 131{
 132	unsigned i;
 133	u32 tmp;
 134
 135	for (i = 0; i < adev->usec_timeout; i++) {
 136		/* read MC_STATUS */
 137		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__VMC_BUSY_MASK |
 138					       SRBM_STATUS__MCB_BUSY_MASK |
 139					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
 140					       SRBM_STATUS__MCC_BUSY_MASK |
 141					       SRBM_STATUS__MCD_BUSY_MASK |
 142					       SRBM_STATUS__VMC1_BUSY_MASK);
 143		if (!tmp)
 144			return 0;
 145		udelay(1);
 146	}
 147	return -1;
 148}
 149
 150void gmc_v8_0_mc_stop(struct amdgpu_device *adev,
 151		      struct amdgpu_mode_mc_save *save)
 152{
 153	u32 blackout;
 154
 155	if (adev->mode_info.num_crtc)
 156		amdgpu_display_stop_mc_access(adev, save);
 157
 158	amdgpu_asic_wait_for_mc_idle(adev);
 159
 160	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 161	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
 162		/* Block CPU access */
 163		WREG32(mmBIF_FB_EN, 0);
 164		/* blackout the MC */
 165		blackout = REG_SET_FIELD(blackout,
 166					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
 167		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
 168	}
 169	/* wait for the MC to settle */
 170	udelay(100);
 171}
 172
 173void gmc_v8_0_mc_resume(struct amdgpu_device *adev,
 174			struct amdgpu_mode_mc_save *save)
 175{
 176	u32 tmp;
 177
 178	/* unblackout the MC */
 179	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 180	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
 181	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
 182	/* allow CPU access */
 183	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
 184	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
 185	WREG32(mmBIF_FB_EN, tmp);
 186
 187	if (adev->mode_info.num_crtc)
 188		amdgpu_display_resume_mc_access(adev, save);
 189}
 190
 191/**
 192 * gmc_v8_0_init_microcode - load ucode images from disk
 193 *
 194 * @adev: amdgpu_device pointer
 195 *
 196 * Use the firmware interface to load the ucode images into
 197 * the driver (not loaded into hw).
 198 * Returns 0 on success, error on failure.
 199 */
 200static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
 201{
 202	const char *chip_name;
 203	char fw_name[30];
 204	int err;
 205
 206	DRM_DEBUG("\n");
 207
 208	switch (adev->asic_type) {
 209	case CHIP_TONGA:
 210		chip_name = "tonga";
 211		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 212	case CHIP_FIJI:
 213	case CHIP_CARRIZO:
 214	case CHIP_STONEY:
 
 215		return 0;
 216	default: BUG();
 217	}
 218
 219	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
 220	err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
 221	if (err)
 222		goto out;
 223	err = amdgpu_ucode_validate(adev->mc.fw);
 224
 225out:
 226	if (err) {
 227		printk(KERN_ERR
 228		       "mc: Failed to load firmware \"%s\"\n",
 229		       fw_name);
 230		release_firmware(adev->mc.fw);
 231		adev->mc.fw = NULL;
 232	}
 233	return err;
 234}
 235
 236/**
 237 * gmc_v8_0_mc_load_microcode - load MC ucode into the hw
 238 *
 239 * @adev: amdgpu_device pointer
 240 *
 241 * Load the GDDR MC ucode into the hw (CIK).
 242 * Returns 0 on success, error on failure.
 243 */
 244static int gmc_v8_0_mc_load_microcode(struct amdgpu_device *adev)
 245{
 246	const struct mc_firmware_header_v1_0 *hdr;
 247	const __le32 *fw_data = NULL;
 248	const __le32 *io_mc_regs = NULL;
 249	u32 running, blackout = 0;
 250	int i, ucode_size, regs_size;
 251
 252	if (!adev->mc.fw)
 253		return -EINVAL;
 254
 255	/* Skip MC ucode loading on SR-IOV capable boards.
 256	 * vbios does this for us in asic_init in that case.
 
 
 257	 */
 258	if (adev->virtualization.supports_sr_iov)
 259		return 0;
 260
 261	hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
 
 
 
 262	amdgpu_ucode_print_mc_hdr(&hdr->header);
 263
 264	adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 265	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 266	io_mc_regs = (const __le32 *)
 267		(adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 268	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 269	fw_data = (const __le32 *)
 270		(adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 271
 272	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
 273
 274	if (running == 0) {
 275		if (running) {
 276			blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 277			WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
 278		}
 279
 280		/* reset the engine and set to writable */
 281		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 282		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 283
 284		/* load mc io regs */
 285		for (i = 0; i < regs_size; i++) {
 286			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
 287			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
 288		}
 289		/* load the MC ucode */
 290		for (i = 0; i < ucode_size; i++)
 291			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
 292
 293		/* put the engine back into the active state */
 294		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 295		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 296		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 297
 298		/* wait for training to complete */
 299		for (i = 0; i < adev->usec_timeout; i++) {
 300			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 301					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
 302				break;
 303			udelay(1);
 304		}
 305		for (i = 0; i < adev->usec_timeout; i++) {
 306			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 307					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
 308				break;
 309			udelay(1);
 310		}
 
 
 
 
 
 
 
 
 
 
 
 
 311
 312		if (running)
 313			WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 314	}
 315
 316	return 0;
 317}
 318
 319static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
 320				       struct amdgpu_mc *mc)
 321{
 322	if (mc->mc_vram_size > 0xFFC0000000ULL) {
 323		/* leave room for at least 1024M GTT */
 324		dev_warn(adev->dev, "limiting VRAM\n");
 325		mc->real_vram_size = 0xFFC0000000ULL;
 326		mc->mc_vram_size = 0xFFC0000000ULL;
 327	}
 328	amdgpu_vram_location(adev, &adev->mc, 0);
 329	adev->mc.gtt_base_align = 0;
 330	amdgpu_gtt_location(adev, mc);
 331}
 332
 333/**
 334 * gmc_v8_0_mc_program - program the GPU memory controller
 335 *
 336 * @adev: amdgpu_device pointer
 337 *
 338 * Set the location of vram, gart, and AGP in the GPU's
 339 * physical address space (CIK).
 340 */
 341static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
 342{
 343	struct amdgpu_mode_mc_save save;
 344	u32 tmp;
 345	int i, j;
 346
 347	/* Initialize HDP */
 348	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
 349		WREG32((0xb05 + j), 0x00000000);
 350		WREG32((0xb06 + j), 0x00000000);
 351		WREG32((0xb07 + j), 0x00000000);
 352		WREG32((0xb08 + j), 0x00000000);
 353		WREG32((0xb09 + j), 0x00000000);
 354	}
 355	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
 356
 357	if (adev->mode_info.num_crtc)
 358		amdgpu_display_set_vga_render_state(adev, false);
 359
 360	gmc_v8_0_mc_stop(adev, &save);
 361	if (amdgpu_asic_wait_for_mc_idle(adev)) {
 362		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 363	}
 
 
 
 
 
 
 
 
 
 
 
 364	/* Update configuration */
 365	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
 366	       adev->mc.vram_start >> 12);
 367	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 368	       adev->mc.vram_end >> 12);
 369	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
 370	       adev->vram_scratch.gpu_addr >> 12);
 371	tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
 372	tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
 373	WREG32(mmMC_VM_FB_LOCATION, tmp);
 374	/* XXX double check these! */
 375	WREG32(mmHDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
 376	WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
 377	WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
 
 
 
 
 378	WREG32(mmMC_VM_AGP_BASE, 0);
 379	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
 380	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
 381	if (amdgpu_asic_wait_for_mc_idle(adev)) {
 382		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 383	}
 384	gmc_v8_0_mc_resume(adev, &save);
 385
 386	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
 387
 388	tmp = RREG32(mmHDP_MISC_CNTL);
 389	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
 390	WREG32(mmHDP_MISC_CNTL, tmp);
 391
 392	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
 393	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
 394}
 395
 396/**
 397 * gmc_v8_0_mc_init - initialize the memory controller driver params
 398 *
 399 * @adev: amdgpu_device pointer
 400 *
 401 * Look up the amount of vram, vram width, and decide how to place
 402 * vram and gart within the GPU's physical address space (CIK).
 403 * Returns 0 for success.
 404 */
 405static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
 406{
 407	u32 tmp;
 408	int chansize, numchan;
 409
 410	/* Get VRAM informations */
 411	tmp = RREG32(mmMC_ARB_RAMCFG);
 412	if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
 413		chansize = 64;
 414	} else {
 415		chansize = 32;
 416	}
 417	tmp = RREG32(mmMC_SHARED_CHMAP);
 418	switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
 419	case 0:
 420	default:
 421		numchan = 1;
 422		break;
 423	case 1:
 424		numchan = 2;
 425		break;
 426	case 2:
 427		numchan = 4;
 428		break;
 429	case 3:
 430		numchan = 8;
 431		break;
 432	case 4:
 433		numchan = 3;
 434		break;
 435	case 5:
 436		numchan = 6;
 437		break;
 438	case 6:
 439		numchan = 10;
 440		break;
 441	case 7:
 442		numchan = 12;
 443		break;
 444	case 8:
 445		numchan = 16;
 446		break;
 
 
 
 
 
 
 
 447	}
 448	adev->mc.vram_width = numchan * chansize;
 449	/* Could aper size report 0 ? */
 450	adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
 451	adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
 452	/* size in MB on si */
 453	adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 454	adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 455	adev->mc.visible_vram_size = adev->mc.aper_size;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 456
 457	/* In case the PCI BAR is larger than the actual amount of vram */
 458	if (adev->mc.visible_vram_size > adev->mc.real_vram_size)
 459		adev->mc.visible_vram_size = adev->mc.real_vram_size;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 460
 461	/* unless the user had overridden it, set the gart
 462	 * size equal to the 1024 or vram, whichever is larger.
 463	 */
 464	if (amdgpu_gart_size == -1)
 465		adev->mc.gtt_size = max((1024ULL << 20), adev->mc.mc_vram_size);
 466	else
 467		adev->mc.gtt_size = (uint64_t)amdgpu_gart_size << 20;
 468
 469	gmc_v8_0_vram_gtt_location(adev, &adev->mc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 470
 471	return 0;
 
 472}
 473
 474/*
 475 * GART
 476 * VMID 0 is the physical GPU addresses as used by the kernel.
 477 * VMIDs 1-15 are used for userspace clients and are handled
 478 * by the amdgpu vm/hsa code.
 479 */
 480
 481/**
 482 * gmc_v8_0_gart_flush_gpu_tlb - gart tlb flush callback
 483 *
 484 * @adev: amdgpu_device pointer
 485 * @vmid: vm instance to flush
 
 
 486 *
 487 * Flush the TLB for the requested page table (CIK).
 488 */
 489static void gmc_v8_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
 490					uint32_t vmid)
 491{
 492	/* flush hdp cache */
 493	WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
 494
 495	/* bits 0-15 are the VM contexts0-15 */
 496	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 497}
 498
 499/**
 500 * gmc_v8_0_gart_set_pte_pde - update the page tables using MMIO
 501 *
 502 * @adev: amdgpu_device pointer
 503 * @cpu_pt_addr: cpu address of the page table
 504 * @gpu_page_idx: entry in the page table to update
 505 * @addr: dst addr to write into pte/pde
 506 * @flags: access flags
 507 *
 508 * Update the page tables using the CPU.
 509 */
 510static int gmc_v8_0_gart_set_pte_pde(struct amdgpu_device *adev,
 511				     void *cpu_pt_addr,
 512				     uint32_t gpu_page_idx,
 513				     uint64_t addr,
 514				     uint32_t flags)
 515{
 516	void __iomem *ptr = (void *)cpu_pt_addr;
 517	uint64_t value;
 518
 519	/*
 520	 * PTE format on VI:
 521	 * 63:40 reserved
 522	 * 39:12 4k physical page base address
 523	 * 11:7 fragment
 524	 * 6 write
 525	 * 5 read
 526	 * 4 exe
 527	 * 3 reserved
 528	 * 2 snooped
 529	 * 1 system
 530	 * 0 valid
 531	 *
 532	 * PDE format on VI:
 533	 * 63:59 block fragment size
 534	 * 58:40 reserved
 535	 * 39:1 physical base address of PTE
 536	 * bits 5:1 must be 0.
 537	 * 0 valid
 538	 */
 539	value = addr & 0x000000FFFFFFF000ULL;
 540	value |= flags;
 541	writeq(value, ptr + (gpu_page_idx * 8));
 542
 543	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 544}
 545
 546/**
 547 * gmc_v8_0_set_fault_enable_default - update VM fault handling
 548 *
 549 * @adev: amdgpu_device pointer
 550 * @value: true redirects VM faults to the default page
 551 */
 552static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
 553					      bool value)
 554{
 555	u32 tmp;
 556
 557	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 558	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 559			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 560	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 561			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 562	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 563			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 564	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 565			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 566	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 567			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 568	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 569			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 570	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 571			    EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 572	WREG32(mmVM_CONTEXT1_CNTL, tmp);
 573}
 574
 575/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 576 * gmc_v8_0_gart_enable - gart enable
 577 *
 578 * @adev: amdgpu_device pointer
 579 *
 580 * This sets up the TLBs, programs the page tables for VMID0,
 581 * sets up the hw for VMIDs 1-15 which are allocated on
 582 * demand, and sets up the global locations for the LDS, GDS,
 583 * and GPUVM for FSA64 clients (CIK).
 584 * Returns 0 for success, errors for failure.
 585 */
 586static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
 587{
 
 588	int r, i;
 589	u32 tmp;
 590
 591	if (adev->gart.robj == NULL) {
 592		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
 593		return -EINVAL;
 594	}
 595	r = amdgpu_gart_table_vram_pin(adev);
 596	if (r)
 597		return r;
 
 
 
 598	/* Setup TLB control */
 599	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 600	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
 601	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
 602	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
 603	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
 604	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
 605	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 606	/* Setup L2 cache */
 607	tmp = RREG32(mmVM_L2_CNTL);
 608	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
 609	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
 610	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
 611	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
 612	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
 613	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
 614	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
 615	WREG32(mmVM_L2_CNTL, tmp);
 616	tmp = RREG32(mmVM_L2_CNTL2);
 617	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
 618	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
 619	WREG32(mmVM_L2_CNTL2, tmp);
 
 
 620	tmp = RREG32(mmVM_L2_CNTL3);
 621	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
 622	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 4);
 623	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, 4);
 624	WREG32(mmVM_L2_CNTL3, tmp);
 625	/* XXX: set to enable PTE/PDE in system memory */
 626	tmp = RREG32(mmVM_L2_CNTL4);
 627	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
 628	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
 629	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
 630	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
 631	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
 632	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
 633	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
 634	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
 635	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
 636	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
 637	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
 638	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
 639	WREG32(mmVM_L2_CNTL4, tmp);
 640	/* setup context0 */
 641	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
 642	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
 643	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
 644	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 645			(u32)(adev->dummy_page.addr >> 12));
 646	WREG32(mmVM_CONTEXT0_CNTL2, 0);
 647	tmp = RREG32(mmVM_CONTEXT0_CNTL);
 648	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
 649	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
 650	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 651	WREG32(mmVM_CONTEXT0_CNTL, tmp);
 652
 653	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
 654	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
 655	WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
 656
 657	/* empty context1-15 */
 658	/* FIXME start with 4G, once using 2 level pt switch to full
 659	 * vm size space
 660	 */
 661	/* set vm size, must be a multiple of 4 */
 662	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
 663	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
 664	for (i = 1; i < 16; i++) {
 665		if (i < 8)
 666			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
 667			       adev->gart.table_addr >> 12);
 668		else
 669			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
 670			       adev->gart.table_addr >> 12);
 671	}
 672
 673	/* enable context1-15 */
 674	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
 675	       (u32)(adev->dummy_page.addr >> 12));
 676	WREG32(mmVM_CONTEXT1_CNTL2, 4);
 677	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 678	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
 679	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
 680	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 681	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 682	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 683	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 684	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 685	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 686	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 687	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
 688			    amdgpu_vm_block_size - 9);
 689	WREG32(mmVM_CONTEXT1_CNTL, tmp);
 690	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 691		gmc_v8_0_set_fault_enable_default(adev, false);
 692	else
 693		gmc_v8_0_set_fault_enable_default(adev, true);
 694
 695	gmc_v8_0_gart_flush_gpu_tlb(adev, 0);
 696	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
 697		 (unsigned)(adev->mc.gtt_size >> 20),
 698		 (unsigned long long)adev->gart.table_addr);
 699	adev->gart.ready = true;
 700	return 0;
 701}
 702
 703static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
 704{
 705	int r;
 706
 707	if (adev->gart.robj) {
 708		WARN(1, "R600 PCIE GART already initialized\n");
 709		return 0;
 710	}
 711	/* Initialize common gart structure */
 712	r = amdgpu_gart_init(adev);
 713	if (r)
 714		return r;
 715	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
 
 716	return amdgpu_gart_table_vram_alloc(adev);
 717}
 718
 719/**
 720 * gmc_v8_0_gart_disable - gart disable
 721 *
 722 * @adev: amdgpu_device pointer
 723 *
 724 * This disables all VM page table (CIK).
 725 */
 726static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
 727{
 728	u32 tmp;
 729
 730	/* Disable all tables */
 731	WREG32(mmVM_CONTEXT0_CNTL, 0);
 732	WREG32(mmVM_CONTEXT1_CNTL, 0);
 733	/* Setup TLB control */
 734	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 735	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
 736	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
 737	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
 738	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 739	/* Setup L2 cache */
 740	tmp = RREG32(mmVM_L2_CNTL);
 741	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
 742	WREG32(mmVM_L2_CNTL, tmp);
 743	WREG32(mmVM_L2_CNTL2, 0);
 744	amdgpu_gart_table_vram_unpin(adev);
 745}
 746
 747/**
 748 * gmc_v8_0_gart_fini - vm fini callback
 749 *
 750 * @adev: amdgpu_device pointer
 751 *
 752 * Tears down the driver GART/VM setup (CIK).
 753 */
 754static void gmc_v8_0_gart_fini(struct amdgpu_device *adev)
 755{
 756	amdgpu_gart_table_vram_free(adev);
 757	amdgpu_gart_fini(adev);
 758}
 759
 760/*
 761 * vm
 762 * VMID 0 is the physical GPU addresses as used by the kernel.
 763 * VMIDs 1-15 are used for userspace clients and are handled
 764 * by the amdgpu vm/hsa code.
 765 */
 766/**
 767 * gmc_v8_0_vm_init - cik vm init callback
 768 *
 769 * @adev: amdgpu_device pointer
 770 *
 771 * Inits cik specific vm parameters (number of VMs, base of vram for
 772 * VMIDs 1-15) (CIK).
 773 * Returns 0 for success.
 774 */
 775static int gmc_v8_0_vm_init(struct amdgpu_device *adev)
 776{
 777	/*
 778	 * number of VMs
 779	 * VMID 0 is reserved for System
 780	 * amdgpu graphics/compute will use VMIDs 1-7
 781	 * amdkfd will use VMIDs 8-15
 782	 */
 783	adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS;
 784	amdgpu_vm_manager_init(adev);
 785
 786	/* base offset of vram pages */
 787	if (adev->flags & AMD_IS_APU) {
 788		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
 789		tmp <<= 22;
 790		adev->vm_manager.vram_base_offset = tmp;
 791	} else
 792		adev->vm_manager.vram_base_offset = 0;
 793
 794	return 0;
 795}
 796
 797/**
 798 * gmc_v8_0_vm_fini - cik vm fini callback
 799 *
 800 * @adev: amdgpu_device pointer
 801 *
 802 * Tear down any asic specific VM setup (CIK).
 803 */
 804static void gmc_v8_0_vm_fini(struct amdgpu_device *adev)
 805{
 806}
 807
 808/**
 809 * gmc_v8_0_vm_decode_fault - print human readable fault info
 810 *
 811 * @adev: amdgpu_device pointer
 812 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
 813 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
 
 
 814 *
 815 * Print human readable fault information (CIK).
 816 */
 817static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev,
 818				     u32 status, u32 addr, u32 mc_client)
 819{
 820	u32 mc_id;
 821	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
 822	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 823					PROTECTIONS);
 824	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
 825		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
 
 826
 827	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 828			      MEMORY_CLIENT_ID);
 829
 830	printk("VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
 831	       protections, vmid, addr,
 832	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 833			     MEMORY_CLIENT_RW) ?
 834	       "write" : "read", block, mc_client, mc_id);
 835}
 836
 837static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
 838{
 839	switch (mc_seq_vram_type) {
 840	case MC_SEQ_MISC0__MT__GDDR1:
 841		return AMDGPU_VRAM_TYPE_GDDR1;
 842	case MC_SEQ_MISC0__MT__DDR2:
 843		return AMDGPU_VRAM_TYPE_DDR2;
 844	case MC_SEQ_MISC0__MT__GDDR3:
 845		return AMDGPU_VRAM_TYPE_GDDR3;
 846	case MC_SEQ_MISC0__MT__GDDR4:
 847		return AMDGPU_VRAM_TYPE_GDDR4;
 848	case MC_SEQ_MISC0__MT__GDDR5:
 849		return AMDGPU_VRAM_TYPE_GDDR5;
 850	case MC_SEQ_MISC0__MT__HBM:
 851		return AMDGPU_VRAM_TYPE_HBM;
 852	case MC_SEQ_MISC0__MT__DDR3:
 853		return AMDGPU_VRAM_TYPE_DDR3;
 854	default:
 855		return AMDGPU_VRAM_TYPE_UNKNOWN;
 856	}
 857}
 858
 859static int gmc_v8_0_early_init(void *handle)
 860{
 861	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 862
 863	gmc_v8_0_set_gart_funcs(adev);
 864	gmc_v8_0_set_irq_funcs(adev);
 865
 
 
 
 
 
 
 
 
 866	return 0;
 867}
 868
 869static int gmc_v8_0_late_init(void *handle)
 870{
 871	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 872
 873	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
 874		return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
 875	else
 876		return 0;
 877}
 878
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 879#define mmMC_SEQ_MISC0_FIJI 0xA71
 880
 881static int gmc_v8_0_sw_init(void *handle)
 882{
 883	int r;
 884	int dma_bits;
 885	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 886
 
 
 887	if (adev->flags & AMD_IS_APU) {
 888		adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
 889	} else {
 890		u32 tmp;
 891
 892		if (adev->asic_type == CHIP_FIJI)
 
 893			tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
 894		else
 895			tmp = RREG32(mmMC_SEQ_MISC0);
 896		tmp &= MC_SEQ_MISC0__MT__MASK;
 897		adev->mc.vram_type = gmc_v8_0_convert_vram_type(tmp);
 898	}
 899
 900	r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault);
 901	if (r)
 902		return r;
 903
 904	r = amdgpu_irq_add_id(adev, 147, &adev->mc.vm_fault);
 905	if (r)
 906		return r;
 907
 908	/* Adjust VM size here.
 909	 * Currently set to 4GB ((1 << 20) 4k pages).
 910	 * Max GPUVM size for cayman and SI is 40 bits.
 911	 */
 912	adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
 913
 914	/* Set the internal MC address mask
 915	 * This is the max address of the GPU's
 916	 * internal address space.
 917	 */
 918	adev->mc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
 919
 920	/* set DMA mask + need_dma32 flags.
 921	 * PCIE - can handle 40-bits.
 922	 * IGP - can handle 40-bits
 923	 * PCI - dma32 for legacy pci gart, 40 bits on newer asics
 924	 */
 925	adev->need_dma32 = false;
 926	dma_bits = adev->need_dma32 ? 32 : 40;
 927	r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
 928	if (r) {
 929		adev->need_dma32 = true;
 930		dma_bits = 32;
 931		printk(KERN_WARNING "amdgpu: No suitable DMA available.\n");
 932	}
 933	r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
 934	if (r) {
 935		pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
 936		printk(KERN_WARNING "amdgpu: No coherent DMA available.\n");
 937	}
 
 938
 939	r = gmc_v8_0_init_microcode(adev);
 940	if (r) {
 941		DRM_ERROR("Failed to load mc firmware!\n");
 942		return r;
 943	}
 944
 945	r = gmc_v8_0_mc_init(adev);
 946	if (r)
 947		return r;
 948
 
 
 949	/* Memory manager */
 950	r = amdgpu_bo_init(adev);
 951	if (r)
 952		return r;
 953
 954	r = gmc_v8_0_gart_init(adev);
 955	if (r)
 956		return r;
 957
 958	if (!adev->vm_manager.enabled) {
 959		r = gmc_v8_0_vm_init(adev);
 960		if (r) {
 961			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
 962			return r;
 963		}
 964		adev->vm_manager.enabled = true;
 
 
 
 
 
 
 
 
 
 
 965	}
 966
 967	return r;
 
 
 
 
 
 
 968}
 969
 970static int gmc_v8_0_sw_fini(void *handle)
 971{
 972	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 973
 974	if (adev->vm_manager.enabled) {
 975		amdgpu_vm_manager_fini(adev);
 976		gmc_v8_0_vm_fini(adev);
 977		adev->vm_manager.enabled = false;
 978	}
 979	gmc_v8_0_gart_fini(adev);
 980	amdgpu_gem_force_release(adev);
 
 
 
 981	amdgpu_bo_fini(adev);
 
 
 982
 983	return 0;
 984}
 985
 986static int gmc_v8_0_hw_init(void *handle)
 987{
 988	int r;
 989	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 990
 991	gmc_v8_0_init_golden_registers(adev);
 992
 993	gmc_v8_0_mc_program(adev);
 994
 995	if (adev->asic_type == CHIP_TONGA) {
 996		r = gmc_v8_0_mc_load_microcode(adev);
 
 
 
 
 
 
 
 
 997		if (r) {
 998			DRM_ERROR("Failed to load MC firmware!\n");
 999			return r;
1000		}
1001	}
1002
1003	r = gmc_v8_0_gart_enable(adev);
1004	if (r)
1005		return r;
1006
1007	return r;
1008}
1009
1010static int gmc_v8_0_hw_fini(void *handle)
1011{
1012	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1013
1014	amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
1015	gmc_v8_0_gart_disable(adev);
1016
1017	return 0;
1018}
1019
1020static int gmc_v8_0_suspend(void *handle)
1021{
1022	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1023
1024	if (adev->vm_manager.enabled) {
1025		gmc_v8_0_vm_fini(adev);
1026		adev->vm_manager.enabled = false;
1027	}
1028	gmc_v8_0_hw_fini(adev);
1029
1030	return 0;
1031}
1032
1033static int gmc_v8_0_resume(void *handle)
1034{
1035	int r;
1036	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1037
1038	r = gmc_v8_0_hw_init(adev);
1039	if (r)
1040		return r;
1041
1042	if (!adev->vm_manager.enabled) {
1043		r = gmc_v8_0_vm_init(adev);
1044		if (r) {
1045			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
1046			return r;
1047		}
1048		adev->vm_manager.enabled = true;
1049	}
1050
1051	return r;
1052}
1053
1054static bool gmc_v8_0_is_idle(void *handle)
1055{
1056	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1057	u32 tmp = RREG32(mmSRBM_STATUS);
1058
1059	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1060		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1061		return false;
1062
1063	return true;
1064}
1065
1066static int gmc_v8_0_wait_for_idle(void *handle)
1067{
1068	unsigned i;
1069	u32 tmp;
1070	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1071
1072	for (i = 0; i < adev->usec_timeout; i++) {
1073		/* read MC_STATUS */
1074		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1075					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1076					       SRBM_STATUS__MCC_BUSY_MASK |
1077					       SRBM_STATUS__MCD_BUSY_MASK |
1078					       SRBM_STATUS__VMC_BUSY_MASK |
1079					       SRBM_STATUS__VMC1_BUSY_MASK);
1080		if (!tmp)
1081			return 0;
1082		udelay(1);
1083	}
1084	return -ETIMEDOUT;
1085
1086}
1087
1088static void gmc_v8_0_print_status(void *handle)
1089{
1090	int i, j;
1091	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1092
1093	dev_info(adev->dev, "GMC 8.x registers\n");
1094	dev_info(adev->dev, "  SRBM_STATUS=0x%08X\n",
1095		RREG32(mmSRBM_STATUS));
1096	dev_info(adev->dev, "  SRBM_STATUS2=0x%08X\n",
1097		RREG32(mmSRBM_STATUS2));
1098
1099	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1100		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR));
1101	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1102		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS));
1103	dev_info(adev->dev, "  MC_VM_MX_L1_TLB_CNTL=0x%08X\n",
1104		 RREG32(mmMC_VM_MX_L1_TLB_CNTL));
1105	dev_info(adev->dev, "  VM_L2_CNTL=0x%08X\n",
1106		 RREG32(mmVM_L2_CNTL));
1107	dev_info(adev->dev, "  VM_L2_CNTL2=0x%08X\n",
1108		 RREG32(mmVM_L2_CNTL2));
1109	dev_info(adev->dev, "  VM_L2_CNTL3=0x%08X\n",
1110		 RREG32(mmVM_L2_CNTL3));
1111	dev_info(adev->dev, "  VM_L2_CNTL4=0x%08X\n",
1112		 RREG32(mmVM_L2_CNTL4));
1113	dev_info(adev->dev, "  VM_CONTEXT0_PAGE_TABLE_START_ADDR=0x%08X\n",
1114		 RREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR));
1115	dev_info(adev->dev, "  VM_CONTEXT0_PAGE_TABLE_END_ADDR=0x%08X\n",
1116		 RREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR));
1117	dev_info(adev->dev, "  VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR=0x%08X\n",
1118		 RREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR));
1119	dev_info(adev->dev, "  VM_CONTEXT0_CNTL2=0x%08X\n",
1120		 RREG32(mmVM_CONTEXT0_CNTL2));
1121	dev_info(adev->dev, "  VM_CONTEXT0_CNTL=0x%08X\n",
1122		 RREG32(mmVM_CONTEXT0_CNTL));
1123	dev_info(adev->dev, "  VM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR=0x%08X\n",
1124		 RREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR));
1125	dev_info(adev->dev, "  VM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR=0x%08X\n",
1126		 RREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR));
1127	dev_info(adev->dev, "  mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET=0x%08X\n",
1128		 RREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET));
1129	dev_info(adev->dev, "  VM_CONTEXT1_PAGE_TABLE_START_ADDR=0x%08X\n",
1130		 RREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR));
1131	dev_info(adev->dev, "  VM_CONTEXT1_PAGE_TABLE_END_ADDR=0x%08X\n",
1132		 RREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR));
1133	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR=0x%08X\n",
1134		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR));
1135	dev_info(adev->dev, "  VM_CONTEXT1_CNTL2=0x%08X\n",
1136		 RREG32(mmVM_CONTEXT1_CNTL2));
1137	dev_info(adev->dev, "  VM_CONTEXT1_CNTL=0x%08X\n",
1138		 RREG32(mmVM_CONTEXT1_CNTL));
1139	for (i = 0; i < 16; i++) {
1140		if (i < 8)
1141			dev_info(adev->dev, "  VM_CONTEXT%d_PAGE_TABLE_BASE_ADDR=0x%08X\n",
1142				 i, RREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i));
1143		else
1144			dev_info(adev->dev, "  VM_CONTEXT%d_PAGE_TABLE_BASE_ADDR=0x%08X\n",
1145				 i, RREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8));
1146	}
1147	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_LOW_ADDR=0x%08X\n",
1148		 RREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR));
1149	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_HIGH_ADDR=0x%08X\n",
1150		 RREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR));
1151	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR=0x%08X\n",
1152		 RREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR));
1153	dev_info(adev->dev, "  MC_VM_FB_LOCATION=0x%08X\n",
1154		 RREG32(mmMC_VM_FB_LOCATION));
1155	dev_info(adev->dev, "  MC_VM_AGP_BASE=0x%08X\n",
1156		 RREG32(mmMC_VM_AGP_BASE));
1157	dev_info(adev->dev, "  MC_VM_AGP_TOP=0x%08X\n",
1158		 RREG32(mmMC_VM_AGP_TOP));
1159	dev_info(adev->dev, "  MC_VM_AGP_BOT=0x%08X\n",
1160		 RREG32(mmMC_VM_AGP_BOT));
1161
1162	dev_info(adev->dev, "  HDP_REG_COHERENCY_FLUSH_CNTL=0x%08X\n",
1163		 RREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL));
1164	dev_info(adev->dev, "  HDP_NONSURFACE_BASE=0x%08X\n",
1165		 RREG32(mmHDP_NONSURFACE_BASE));
1166	dev_info(adev->dev, "  HDP_NONSURFACE_INFO=0x%08X\n",
1167		 RREG32(mmHDP_NONSURFACE_INFO));
1168	dev_info(adev->dev, "  HDP_NONSURFACE_SIZE=0x%08X\n",
1169		 RREG32(mmHDP_NONSURFACE_SIZE));
1170	dev_info(adev->dev, "  HDP_MISC_CNTL=0x%08X\n",
1171		 RREG32(mmHDP_MISC_CNTL));
1172	dev_info(adev->dev, "  HDP_HOST_PATH_CNTL=0x%08X\n",
1173		 RREG32(mmHDP_HOST_PATH_CNTL));
1174
1175	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
1176		dev_info(adev->dev, "  %d:\n", i);
1177		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1178			 0xb05 + j, RREG32(0xb05 + j));
1179		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1180			 0xb06 + j, RREG32(0xb06 + j));
1181		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1182			 0xb07 + j, RREG32(0xb07 + j));
1183		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1184			 0xb08 + j, RREG32(0xb08 + j));
1185		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1186			 0xb09 + j, RREG32(0xb09 + j));
1187	}
1188
1189	dev_info(adev->dev, "  BIF_FB_EN=0x%08X\n",
1190		 RREG32(mmBIF_FB_EN));
1191}
1192
1193static int gmc_v8_0_soft_reset(void *handle)
1194{
1195	struct amdgpu_mode_mc_save save;
1196	u32 srbm_soft_reset = 0;
1197	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1198	u32 tmp = RREG32(mmSRBM_STATUS);
1199
1200	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1201		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1202						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1203
1204	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1205		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1206		if (!(adev->flags & AMD_IS_APU))
1207			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1208							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1209	}
1210
1211	if (srbm_soft_reset) {
1212		gmc_v8_0_print_status((void *)adev);
 
 
 
 
 
 
1213
1214		gmc_v8_0_mc_stop(adev, &save);
1215		if (gmc_v8_0_wait_for_idle(adev)) {
1216			dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1217		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1218
 
 
1219
1220		tmp = RREG32(mmSRBM_SOFT_RESET);
1221		tmp |= srbm_soft_reset;
1222		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1223		WREG32(mmSRBM_SOFT_RESET, tmp);
1224		tmp = RREG32(mmSRBM_SOFT_RESET);
1225
1226		udelay(50);
1227
1228		tmp &= ~srbm_soft_reset;
1229		WREG32(mmSRBM_SOFT_RESET, tmp);
1230		tmp = RREG32(mmSRBM_SOFT_RESET);
1231
1232		/* Wait a little for things to settle down */
1233		udelay(50);
 
1234
1235		gmc_v8_0_mc_resume(adev, &save);
1236		udelay(50);
1237
1238		gmc_v8_0_print_status((void *)adev);
1239	}
 
 
 
 
1240
 
1241	return 0;
1242}
1243
1244static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1245					     struct amdgpu_irq_src *src,
1246					     unsigned type,
1247					     enum amdgpu_interrupt_state state)
1248{
1249	u32 tmp;
1250	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1251		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1252		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1253		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1254		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1255		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1256		    VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1257
1258	switch (state) {
1259	case AMDGPU_IRQ_STATE_DISABLE:
1260		/* system context */
1261		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1262		tmp &= ~bits;
1263		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1264		/* VMs */
1265		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1266		tmp &= ~bits;
1267		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1268		break;
1269	case AMDGPU_IRQ_STATE_ENABLE:
1270		/* system context */
1271		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1272		tmp |= bits;
1273		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1274		/* VMs */
1275		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1276		tmp |= bits;
1277		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1278		break;
1279	default:
1280		break;
1281	}
1282
1283	return 0;
1284}
1285
1286static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1287				      struct amdgpu_irq_src *source,
1288				      struct amdgpu_iv_entry *entry)
1289{
1290	u32 addr, status, mc_client;
 
 
 
 
 
 
 
1291
1292	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1293	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1294	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1295	/* reset addr and status */
1296	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1297
1298	if (!addr && !status)
1299		return 0;
1300
1301	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1302		gmc_v8_0_set_fault_enable_default(adev, false);
1303
1304	dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1305		entry->src_id, entry->src_data);
1306	dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1307		addr);
1308	dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1309		status);
1310	gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1311
1312	return 0;
1313}
1314
1315static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1316		bool enable)
1317{
1318	uint32_t data;
1319
1320	if (enable) {
1321		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1322		data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1323		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1324
1325		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1326		data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1327		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1328
1329		data = RREG32(mmMC_HUB_MISC_VM_CG);
1330		data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1331		WREG32(mmMC_HUB_MISC_VM_CG, data);
1332
1333		data = RREG32(mmMC_XPB_CLK_GAT);
1334		data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1335		WREG32(mmMC_XPB_CLK_GAT, data);
1336
1337		data = RREG32(mmATC_MISC_CG);
1338		data |= ATC_MISC_CG__ENABLE_MASK;
1339		WREG32(mmATC_MISC_CG, data);
1340
1341		data = RREG32(mmMC_CITF_MISC_WR_CG);
1342		data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1343		WREG32(mmMC_CITF_MISC_WR_CG, data);
1344
1345		data = RREG32(mmMC_CITF_MISC_RD_CG);
1346		data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1347		WREG32(mmMC_CITF_MISC_RD_CG, data);
1348
1349		data = RREG32(mmMC_CITF_MISC_VM_CG);
1350		data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1351		WREG32(mmMC_CITF_MISC_VM_CG, data);
1352
1353		data = RREG32(mmVM_L2_CG);
1354		data |= VM_L2_CG__ENABLE_MASK;
1355		WREG32(mmVM_L2_CG, data);
1356	} else {
1357		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1358		data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1359		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1360
1361		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1362		data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1363		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1364
1365		data = RREG32(mmMC_HUB_MISC_VM_CG);
1366		data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1367		WREG32(mmMC_HUB_MISC_VM_CG, data);
1368
1369		data = RREG32(mmMC_XPB_CLK_GAT);
1370		data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1371		WREG32(mmMC_XPB_CLK_GAT, data);
1372
1373		data = RREG32(mmATC_MISC_CG);
1374		data &= ~ATC_MISC_CG__ENABLE_MASK;
1375		WREG32(mmATC_MISC_CG, data);
1376
1377		data = RREG32(mmMC_CITF_MISC_WR_CG);
1378		data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1379		WREG32(mmMC_CITF_MISC_WR_CG, data);
1380
1381		data = RREG32(mmMC_CITF_MISC_RD_CG);
1382		data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1383		WREG32(mmMC_CITF_MISC_RD_CG, data);
1384
1385		data = RREG32(mmMC_CITF_MISC_VM_CG);
1386		data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1387		WREG32(mmMC_CITF_MISC_VM_CG, data);
1388
1389		data = RREG32(mmVM_L2_CG);
1390		data &= ~VM_L2_CG__ENABLE_MASK;
1391		WREG32(mmVM_L2_CG, data);
1392	}
1393}
1394
1395static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1396		bool enable)
1397{
1398	uint32_t data;
1399
1400	if (enable) {
1401		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1402		data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1403		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1404
1405		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1406		data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1407		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1408
1409		data = RREG32(mmMC_HUB_MISC_VM_CG);
1410		data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1411		WREG32(mmMC_HUB_MISC_VM_CG, data);
1412
1413		data = RREG32(mmMC_XPB_CLK_GAT);
1414		data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1415		WREG32(mmMC_XPB_CLK_GAT, data);
1416
1417		data = RREG32(mmATC_MISC_CG);
1418		data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1419		WREG32(mmATC_MISC_CG, data);
1420
1421		data = RREG32(mmMC_CITF_MISC_WR_CG);
1422		data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1423		WREG32(mmMC_CITF_MISC_WR_CG, data);
1424
1425		data = RREG32(mmMC_CITF_MISC_RD_CG);
1426		data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1427		WREG32(mmMC_CITF_MISC_RD_CG, data);
1428
1429		data = RREG32(mmMC_CITF_MISC_VM_CG);
1430		data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1431		WREG32(mmMC_CITF_MISC_VM_CG, data);
1432
1433		data = RREG32(mmVM_L2_CG);
1434		data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1435		WREG32(mmVM_L2_CG, data);
1436	} else {
1437		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1438		data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1439		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1440
1441		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1442		data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1443		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1444
1445		data = RREG32(mmMC_HUB_MISC_VM_CG);
1446		data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1447		WREG32(mmMC_HUB_MISC_VM_CG, data);
1448
1449		data = RREG32(mmMC_XPB_CLK_GAT);
1450		data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1451		WREG32(mmMC_XPB_CLK_GAT, data);
1452
1453		data = RREG32(mmATC_MISC_CG);
1454		data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1455		WREG32(mmATC_MISC_CG, data);
1456
1457		data = RREG32(mmMC_CITF_MISC_WR_CG);
1458		data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1459		WREG32(mmMC_CITF_MISC_WR_CG, data);
1460
1461		data = RREG32(mmMC_CITF_MISC_RD_CG);
1462		data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1463		WREG32(mmMC_CITF_MISC_RD_CG, data);
1464
1465		data = RREG32(mmMC_CITF_MISC_VM_CG);
1466		data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1467		WREG32(mmMC_CITF_MISC_VM_CG, data);
1468
1469		data = RREG32(mmVM_L2_CG);
1470		data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1471		WREG32(mmVM_L2_CG, data);
1472	}
1473}
1474
1475static int gmc_v8_0_set_clockgating_state(void *handle,
1476					  enum amd_clockgating_state state)
1477{
1478	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1479
 
 
 
1480	switch (adev->asic_type) {
1481	case CHIP_FIJI:
1482		fiji_update_mc_medium_grain_clock_gating(adev,
1483				state == AMD_CG_STATE_GATE ? true : false);
1484		fiji_update_mc_light_sleep(adev,
1485				state == AMD_CG_STATE_GATE ? true : false);
1486		break;
1487	default:
1488		break;
1489	}
1490	return 0;
1491}
1492
1493static int gmc_v8_0_set_powergating_state(void *handle,
1494					  enum amd_powergating_state state)
1495{
1496	return 0;
1497}
1498
1499const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1500	.early_init = gmc_v8_0_early_init,
1501	.late_init = gmc_v8_0_late_init,
1502	.sw_init = gmc_v8_0_sw_init,
1503	.sw_fini = gmc_v8_0_sw_fini,
1504	.hw_init = gmc_v8_0_hw_init,
1505	.hw_fini = gmc_v8_0_hw_fini,
1506	.suspend = gmc_v8_0_suspend,
1507	.resume = gmc_v8_0_resume,
1508	.is_idle = gmc_v8_0_is_idle,
1509	.wait_for_idle = gmc_v8_0_wait_for_idle,
 
 
1510	.soft_reset = gmc_v8_0_soft_reset,
1511	.print_status = gmc_v8_0_print_status,
1512	.set_clockgating_state = gmc_v8_0_set_clockgating_state,
1513	.set_powergating_state = gmc_v8_0_set_powergating_state,
 
1514};
1515
1516static const struct amdgpu_gart_funcs gmc_v8_0_gart_funcs = {
1517	.flush_gpu_tlb = gmc_v8_0_gart_flush_gpu_tlb,
1518	.set_pte_pde = gmc_v8_0_gart_set_pte_pde,
 
 
 
 
 
 
1519};
1520
1521static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1522	.set = gmc_v8_0_vm_fault_interrupt_state,
1523	.process = gmc_v8_0_process_interrupt,
1524};
1525
1526static void gmc_v8_0_set_gart_funcs(struct amdgpu_device *adev)
1527{
1528	if (adev->gart.gart_funcs == NULL)
1529		adev->gart.gart_funcs = &gmc_v8_0_gart_funcs;
1530}
1531
1532static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1533{
1534	adev->mc.vm_fault.num_types = 1;
1535	adev->mc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1536}
v5.14.15
   1/*
   2 * Copyright 2014 Advanced Micro Devices, Inc.
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice shall be included in
  12 * all copies or substantial portions of the Software.
  13 *
  14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20 * OTHER DEALINGS IN THE SOFTWARE.
  21 *
  22 */
  23
  24#include <linux/firmware.h>
  25#include <linux/module.h>
  26#include <linux/pci.h>
  27
  28#include <drm/drm_cache.h>
  29#include "amdgpu.h"
  30#include "gmc_v8_0.h"
  31#include "amdgpu_ucode.h"
  32#include "amdgpu_amdkfd.h"
  33#include "amdgpu_gem.h"
  34
  35#include "gmc/gmc_8_1_d.h"
  36#include "gmc/gmc_8_1_sh_mask.h"
  37
  38#include "bif/bif_5_0_d.h"
  39#include "bif/bif_5_0_sh_mask.h"
  40
  41#include "oss/oss_3_0_d.h"
  42#include "oss/oss_3_0_sh_mask.h"
  43
  44#include "dce/dce_10_0_d.h"
  45#include "dce/dce_10_0_sh_mask.h"
  46
  47#include "vid.h"
  48#include "vi.h"
  49
  50#include "amdgpu_atombios.h"
  51
  52#include "ivsrcid/ivsrcid_vislands30.h"
  53
  54static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev);
  55static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
  56static int gmc_v8_0_wait_for_idle(void *handle);
  57
  58MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
  59MODULE_FIRMWARE("amdgpu/polaris11_mc.bin");
  60MODULE_FIRMWARE("amdgpu/polaris10_mc.bin");
  61MODULE_FIRMWARE("amdgpu/polaris12_mc.bin");
  62MODULE_FIRMWARE("amdgpu/polaris12_32_mc.bin");
  63MODULE_FIRMWARE("amdgpu/polaris11_k_mc.bin");
  64MODULE_FIRMWARE("amdgpu/polaris10_k_mc.bin");
  65MODULE_FIRMWARE("amdgpu/polaris12_k_mc.bin");
  66
  67static const u32 golden_settings_tonga_a11[] =
  68{
  69	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
  70	mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
  71	mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
  72	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  73	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  74	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  75	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  76};
  77
  78static const u32 tonga_mgcg_cgcg_init[] =
  79{
  80	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  81};
  82
  83static const u32 golden_settings_fiji_a10[] =
  84{
  85	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  86	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  87	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  88	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  89};
  90
  91static const u32 fiji_mgcg_cgcg_init[] =
  92{
  93	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  94};
  95
  96static const u32 golden_settings_polaris11_a11[] =
  97{
  98	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  99	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 100	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 101	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
 102};
 103
 104static const u32 golden_settings_polaris10_a11[] =
 105{
 106	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
 107	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 108	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 109	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 110	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
 111};
 112
 113static const u32 cz_mgcg_cgcg_init[] =
 114{
 115	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
 116};
 117
 118static const u32 stoney_mgcg_cgcg_init[] =
 119{
 120	mmATC_MISC_CG, 0xffffffff, 0x000c0200,
 121	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
 122};
 123
 124static const u32 golden_settings_stoney_common[] =
 125{
 126	mmMC_HUB_RDREQ_UVD, MC_HUB_RDREQ_UVD__PRESCALE_MASK, 0x00000004,
 127	mmMC_RD_GRP_OTH, MC_RD_GRP_OTH__UVD_MASK, 0x00600000
 128};
 129
 130static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
 131{
 132	switch (adev->asic_type) {
 133	case CHIP_FIJI:
 134		amdgpu_device_program_register_sequence(adev,
 135							fiji_mgcg_cgcg_init,
 136							ARRAY_SIZE(fiji_mgcg_cgcg_init));
 137		amdgpu_device_program_register_sequence(adev,
 138							golden_settings_fiji_a10,
 139							ARRAY_SIZE(golden_settings_fiji_a10));
 140		break;
 141	case CHIP_TONGA:
 142		amdgpu_device_program_register_sequence(adev,
 143							tonga_mgcg_cgcg_init,
 144							ARRAY_SIZE(tonga_mgcg_cgcg_init));
 145		amdgpu_device_program_register_sequence(adev,
 146							golden_settings_tonga_a11,
 147							ARRAY_SIZE(golden_settings_tonga_a11));
 148		break;
 149	case CHIP_POLARIS11:
 150	case CHIP_POLARIS12:
 151	case CHIP_VEGAM:
 152		amdgpu_device_program_register_sequence(adev,
 153							golden_settings_polaris11_a11,
 154							ARRAY_SIZE(golden_settings_polaris11_a11));
 155		break;
 156	case CHIP_POLARIS10:
 157		amdgpu_device_program_register_sequence(adev,
 158							golden_settings_polaris10_a11,
 159							ARRAY_SIZE(golden_settings_polaris10_a11));
 160		break;
 161	case CHIP_CARRIZO:
 162		amdgpu_device_program_register_sequence(adev,
 163							cz_mgcg_cgcg_init,
 164							ARRAY_SIZE(cz_mgcg_cgcg_init));
 165		break;
 166	case CHIP_STONEY:
 167		amdgpu_device_program_register_sequence(adev,
 168							stoney_mgcg_cgcg_init,
 169							ARRAY_SIZE(stoney_mgcg_cgcg_init));
 170		amdgpu_device_program_register_sequence(adev,
 171							golden_settings_stoney_common,
 172							ARRAY_SIZE(golden_settings_stoney_common));
 173		break;
 174	default:
 175		break;
 176	}
 177}
 178
 179static void gmc_v8_0_mc_stop(struct amdgpu_device *adev)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 180{
 181	u32 blackout;
 182
 183	gmc_v8_0_wait_for_idle(adev);
 
 
 
 184
 185	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 186	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
 187		/* Block CPU access */
 188		WREG32(mmBIF_FB_EN, 0);
 189		/* blackout the MC */
 190		blackout = REG_SET_FIELD(blackout,
 191					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
 192		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
 193	}
 194	/* wait for the MC to settle */
 195	udelay(100);
 196}
 197
 198static void gmc_v8_0_mc_resume(struct amdgpu_device *adev)
 
 199{
 200	u32 tmp;
 201
 202	/* unblackout the MC */
 203	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 204	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
 205	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
 206	/* allow CPU access */
 207	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
 208	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
 209	WREG32(mmBIF_FB_EN, tmp);
 
 
 
 210}
 211
 212/**
 213 * gmc_v8_0_init_microcode - load ucode images from disk
 214 *
 215 * @adev: amdgpu_device pointer
 216 *
 217 * Use the firmware interface to load the ucode images into
 218 * the driver (not loaded into hw).
 219 * Returns 0 on success, error on failure.
 220 */
 221static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
 222{
 223	const char *chip_name;
 224	char fw_name[30];
 225	int err;
 226
 227	DRM_DEBUG("\n");
 228
 229	switch (adev->asic_type) {
 230	case CHIP_TONGA:
 231		chip_name = "tonga";
 232		break;
 233	case CHIP_POLARIS11:
 234		if (ASICID_IS_P21(adev->pdev->device, adev->pdev->revision) ||
 235		    ASICID_IS_P31(adev->pdev->device, adev->pdev->revision))
 236			chip_name = "polaris11_k";
 237		else
 238			chip_name = "polaris11";
 239		break;
 240	case CHIP_POLARIS10:
 241		if (ASICID_IS_P30(adev->pdev->device, adev->pdev->revision))
 242			chip_name = "polaris10_k";
 243		else
 244			chip_name = "polaris10";
 245		break;
 246	case CHIP_POLARIS12:
 247		if (ASICID_IS_P23(adev->pdev->device, adev->pdev->revision)) {
 248			chip_name = "polaris12_k";
 249		} else {
 250			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, ixMC_IO_DEBUG_UP_159);
 251			/* Polaris12 32bit ASIC needs a special MC firmware */
 252			if (RREG32(mmMC_SEQ_IO_DEBUG_DATA) == 0x05b4dc40)
 253				chip_name = "polaris12_32";
 254			else
 255				chip_name = "polaris12";
 256		}
 257		break;
 258	case CHIP_FIJI:
 259	case CHIP_CARRIZO:
 260	case CHIP_STONEY:
 261	case CHIP_VEGAM:
 262		return 0;
 263	default: BUG();
 264	}
 265
 266	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
 267	err = request_firmware(&adev->gmc.fw, fw_name, adev->dev);
 268	if (err)
 269		goto out;
 270	err = amdgpu_ucode_validate(adev->gmc.fw);
 271
 272out:
 273	if (err) {
 274		pr_err("mc: Failed to load firmware \"%s\"\n", fw_name);
 275		release_firmware(adev->gmc.fw);
 276		adev->gmc.fw = NULL;
 
 
 277	}
 278	return err;
 279}
 280
 281/**
 282 * gmc_v8_0_tonga_mc_load_microcode - load tonga MC ucode into the hw
 283 *
 284 * @adev: amdgpu_device pointer
 285 *
 286 * Load the GDDR MC ucode into the hw (VI).
 287 * Returns 0 on success, error on failure.
 288 */
 289static int gmc_v8_0_tonga_mc_load_microcode(struct amdgpu_device *adev)
 290{
 291	const struct mc_firmware_header_v1_0 *hdr;
 292	const __le32 *fw_data = NULL;
 293	const __le32 *io_mc_regs = NULL;
 294	u32 running;
 295	int i, ucode_size, regs_size;
 296
 
 
 
 297	/* Skip MC ucode loading on SR-IOV capable boards.
 298	 * vbios does this for us in asic_init in that case.
 299	 * Skip MC ucode loading on VF, because hypervisor will do that
 300	 * for this adaptor.
 301	 */
 302	if (amdgpu_sriov_bios(adev))
 303		return 0;
 304
 305	if (!adev->gmc.fw)
 306		return -EINVAL;
 307
 308	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
 309	amdgpu_ucode_print_mc_hdr(&hdr->header);
 310
 311	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 312	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 313	io_mc_regs = (const __le32 *)
 314		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 315	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 316	fw_data = (const __le32 *)
 317		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 318
 319	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
 320
 321	if (running == 0) {
 
 
 
 
 
 322		/* reset the engine and set to writable */
 323		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 324		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 325
 326		/* load mc io regs */
 327		for (i = 0; i < regs_size; i++) {
 328			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
 329			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
 330		}
 331		/* load the MC ucode */
 332		for (i = 0; i < ucode_size; i++)
 333			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
 334
 335		/* put the engine back into the active state */
 336		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 337		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 338		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 339
 340		/* wait for training to complete */
 341		for (i = 0; i < adev->usec_timeout; i++) {
 342			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 343					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
 344				break;
 345			udelay(1);
 346		}
 347		for (i = 0; i < adev->usec_timeout; i++) {
 348			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 349					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
 350				break;
 351			udelay(1);
 352		}
 353	}
 354
 355	return 0;
 356}
 357
 358static int gmc_v8_0_polaris_mc_load_microcode(struct amdgpu_device *adev)
 359{
 360	const struct mc_firmware_header_v1_0 *hdr;
 361	const __le32 *fw_data = NULL;
 362	const __le32 *io_mc_regs = NULL;
 363	u32 data;
 364	int i, ucode_size, regs_size;
 365
 366	/* Skip MC ucode loading on SR-IOV capable boards.
 367	 * vbios does this for us in asic_init in that case.
 368	 * Skip MC ucode loading on VF, because hypervisor will do that
 369	 * for this adaptor.
 370	 */
 371	if (amdgpu_sriov_bios(adev))
 372		return 0;
 373
 374	if (!adev->gmc.fw)
 375		return -EINVAL;
 376
 377	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
 378	amdgpu_ucode_print_mc_hdr(&hdr->header);
 379
 380	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 381	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 382	io_mc_regs = (const __le32 *)
 383		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 384	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 385	fw_data = (const __le32 *)
 386		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 387
 388	data = RREG32(mmMC_SEQ_MISC0);
 389	data &= ~(0x40);
 390	WREG32(mmMC_SEQ_MISC0, data);
 391
 392	/* load mc io regs */
 393	for (i = 0; i < regs_size; i++) {
 394		WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
 395		WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
 396	}
 397
 398	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 399	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 400
 401	/* load the MC ucode */
 402	for (i = 0; i < ucode_size; i++)
 403		WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
 404
 405	/* put the engine back into the active state */
 406	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 407	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 408	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 409
 410	/* wait for training to complete */
 411	for (i = 0; i < adev->usec_timeout; i++) {
 412		data = RREG32(mmMC_SEQ_MISC0);
 413		if (data & 0x80)
 414			break;
 415		udelay(1);
 416	}
 417
 418	return 0;
 419}
 420
 421static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
 422				       struct amdgpu_gmc *mc)
 423{
 424	u64 base = 0;
 425
 426	if (!amdgpu_sriov_vf(adev))
 427		base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
 428	base <<= 24;
 429
 430	amdgpu_gmc_vram_location(adev, mc, base);
 431	amdgpu_gmc_gart_location(adev, mc);
 
 432}
 433
 434/**
 435 * gmc_v8_0_mc_program - program the GPU memory controller
 436 *
 437 * @adev: amdgpu_device pointer
 438 *
 439 * Set the location of vram, gart, and AGP in the GPU's
 440 * physical address space (VI).
 441 */
 442static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
 443{
 
 444	u32 tmp;
 445	int i, j;
 446
 447	/* Initialize HDP */
 448	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
 449		WREG32((0xb05 + j), 0x00000000);
 450		WREG32((0xb06 + j), 0x00000000);
 451		WREG32((0xb07 + j), 0x00000000);
 452		WREG32((0xb08 + j), 0x00000000);
 453		WREG32((0xb09 + j), 0x00000000);
 454	}
 455	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
 456
 457	if (gmc_v8_0_wait_for_idle((void *)adev)) {
 
 
 
 
 458		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 459	}
 460	if (adev->mode_info.num_crtc) {
 461		/* Lockout access through VGA aperture*/
 462		tmp = RREG32(mmVGA_HDP_CONTROL);
 463		tmp = REG_SET_FIELD(tmp, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
 464		WREG32(mmVGA_HDP_CONTROL, tmp);
 465
 466		/* disable VGA render */
 467		tmp = RREG32(mmVGA_RENDER_CONTROL);
 468		tmp = REG_SET_FIELD(tmp, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
 469		WREG32(mmVGA_RENDER_CONTROL, tmp);
 470	}
 471	/* Update configuration */
 472	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
 473	       adev->gmc.vram_start >> 12);
 474	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 475	       adev->gmc.vram_end >> 12);
 476	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
 477	       adev->vram_scratch.gpu_addr >> 12);
 478
 479	if (amdgpu_sriov_vf(adev)) {
 480		tmp = ((adev->gmc.vram_end >> 24) & 0xFFFF) << 16;
 481		tmp |= ((adev->gmc.vram_start >> 24) & 0xFFFF);
 482		WREG32(mmMC_VM_FB_LOCATION, tmp);
 483		/* XXX double check these! */
 484		WREG32(mmHDP_NONSURFACE_BASE, (adev->gmc.vram_start >> 8));
 485		WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
 486		WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
 487	}
 488
 489	WREG32(mmMC_VM_AGP_BASE, 0);
 490	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
 491	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
 492	if (gmc_v8_0_wait_for_idle((void *)adev)) {
 493		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 494	}
 
 495
 496	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
 497
 498	tmp = RREG32(mmHDP_MISC_CNTL);
 499	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
 500	WREG32(mmHDP_MISC_CNTL, tmp);
 501
 502	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
 503	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
 504}
 505
 506/**
 507 * gmc_v8_0_mc_init - initialize the memory controller driver params
 508 *
 509 * @adev: amdgpu_device pointer
 510 *
 511 * Look up the amount of vram, vram width, and decide how to place
 512 * vram and gart within the GPU's physical address space (VI).
 513 * Returns 0 for success.
 514 */
 515static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
 516{
 517	int r;
 
 518
 519	adev->gmc.vram_width = amdgpu_atombios_get_vram_width(adev);
 520	if (!adev->gmc.vram_width) {
 521		u32 tmp;
 522		int chansize, numchan;
 523
 524		/* Get VRAM informations */
 525		tmp = RREG32(mmMC_ARB_RAMCFG);
 526		if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
 527			chansize = 64;
 528		} else {
 529			chansize = 32;
 530		}
 531		tmp = RREG32(mmMC_SHARED_CHMAP);
 532		switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
 533		case 0:
 534		default:
 535			numchan = 1;
 536			break;
 537		case 1:
 538			numchan = 2;
 539			break;
 540		case 2:
 541			numchan = 4;
 542			break;
 543		case 3:
 544			numchan = 8;
 545			break;
 546		case 4:
 547			numchan = 3;
 548			break;
 549		case 5:
 550			numchan = 6;
 551			break;
 552		case 6:
 553			numchan = 10;
 554			break;
 555		case 7:
 556			numchan = 12;
 557			break;
 558		case 8:
 559			numchan = 16;
 560			break;
 561		}
 562		adev->gmc.vram_width = numchan * chansize;
 563	}
 
 
 
 
 564	/* size in MB on si */
 565	adev->gmc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 566	adev->gmc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 567
 568	if (!(adev->flags & AMD_IS_APU)) {
 569		r = amdgpu_device_resize_fb_bar(adev);
 570		if (r)
 571			return r;
 572	}
 573	adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
 574	adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
 575
 576#ifdef CONFIG_X86_64
 577	if (adev->flags & AMD_IS_APU) {
 578		adev->gmc.aper_base = ((u64)RREG32(mmMC_VM_FB_OFFSET)) << 22;
 579		adev->gmc.aper_size = adev->gmc.real_vram_size;
 580	}
 581#endif
 582
 583	/* In case the PCI BAR is larger than the actual amount of vram */
 584	adev->gmc.visible_vram_size = adev->gmc.aper_size;
 585	if (adev->gmc.visible_vram_size > adev->gmc.real_vram_size)
 586		adev->gmc.visible_vram_size = adev->gmc.real_vram_size;
 587
 588	/* set the gart size */
 589	if (amdgpu_gart_size == -1) {
 590		switch (adev->asic_type) {
 591		case CHIP_POLARIS10: /* all engines support GPUVM */
 592		case CHIP_POLARIS11: /* all engines support GPUVM */
 593		case CHIP_POLARIS12: /* all engines support GPUVM */
 594		case CHIP_VEGAM:     /* all engines support GPUVM */
 595		default:
 596			adev->gmc.gart_size = 256ULL << 20;
 597			break;
 598		case CHIP_TONGA:   /* UVD, VCE do not support GPUVM */
 599		case CHIP_FIJI:    /* UVD, VCE do not support GPUVM */
 600		case CHIP_CARRIZO: /* UVD, VCE do not support GPUVM, DCE SG support */
 601		case CHIP_STONEY:  /* UVD does not support GPUVM, DCE SG support */
 602			adev->gmc.gart_size = 1024ULL << 20;
 603			break;
 604		}
 605	} else {
 606		adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
 607	}
 608
 609	adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
 610	gmc_v8_0_vram_gtt_location(adev, &adev->gmc);
 
 
 
 
 
 611
 612	return 0;
 613}
 614
 615/**
 616 * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
 617 *
 618 * @adev: amdgpu_device pointer
 619 * @pasid: pasid to be flush
 620 * @flush_type: type of flush
 621 * @all_hub: flush all hubs
 622 *
 623 * Flush the TLB for the requested pasid.
 624 */
 625static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 626					uint16_t pasid, uint32_t flush_type,
 627					bool all_hub)
 628{
 629	int vmid;
 630	unsigned int tmp;
 631
 632	if (amdgpu_in_reset(adev))
 633		return -EIO;
 634
 635	for (vmid = 1; vmid < 16; vmid++) {
 636
 637		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
 638		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
 639			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
 640			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 641			RREG32(mmVM_INVALIDATE_RESPONSE);
 642			break;
 643		}
 644	}
 645
 646	return 0;
 647
 648}
 649
 650/*
 651 * GART
 652 * VMID 0 is the physical GPU addresses as used by the kernel.
 653 * VMIDs 1-15 are used for userspace clients and are handled
 654 * by the amdgpu vm/hsa code.
 655 */
 656
 657/**
 658 * gmc_v8_0_flush_gpu_tlb - gart tlb flush callback
 659 *
 660 * @adev: amdgpu_device pointer
 661 * @vmid: vm instance to flush
 662 * @vmhub: which hub to flush
 663 * @flush_type: type of flush
 664 *
 665 * Flush the TLB for the requested page table (VI).
 666 */
 667static void gmc_v8_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 668					uint32_t vmhub, uint32_t flush_type)
 669{
 
 
 
 670	/* bits 0-15 are the VM contexts0-15 */
 671	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 672}
 673
 674static uint64_t gmc_v8_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 675					    unsigned vmid, uint64_t pd_addr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 676{
 677	uint32_t reg;
 
 678
 679	if (vmid < 8)
 680		reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
 681	else
 682		reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8;
 683	amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 684
 685	/* bits 0-15 are the VM contexts0-15 */
 686	amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
 687
 688	return pd_addr;
 689}
 690
 691static void gmc_v8_0_emit_pasid_mapping(struct amdgpu_ring *ring, unsigned vmid,
 692					unsigned pasid)
 693{
 694	amdgpu_ring_emit_wreg(ring, mmIH_VMID_0_LUT + vmid, pasid);
 695}
 696
 697/*
 698 * PTE format on VI:
 699 * 63:40 reserved
 700 * 39:12 4k physical page base address
 701 * 11:7 fragment
 702 * 6 write
 703 * 5 read
 704 * 4 exe
 705 * 3 reserved
 706 * 2 snooped
 707 * 1 system
 708 * 0 valid
 709 *
 710 * PDE format on VI:
 711 * 63:59 block fragment size
 712 * 58:40 reserved
 713 * 39:1 physical base address of PTE
 714 * bits 5:1 must be 0.
 715 * 0 valid
 716 */
 717
 718static void gmc_v8_0_get_vm_pde(struct amdgpu_device *adev, int level,
 719				uint64_t *addr, uint64_t *flags)
 720{
 721	BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
 722}
 723
 724static void gmc_v8_0_get_vm_pte(struct amdgpu_device *adev,
 725				struct amdgpu_bo_va_mapping *mapping,
 726				uint64_t *flags)
 727{
 728	*flags &= ~AMDGPU_PTE_EXECUTABLE;
 729	*flags |= mapping->flags & AMDGPU_PTE_EXECUTABLE;
 730	*flags &= ~AMDGPU_PTE_PRT;
 731}
 732
 733/**
 734 * gmc_v8_0_set_fault_enable_default - update VM fault handling
 735 *
 736 * @adev: amdgpu_device pointer
 737 * @value: true redirects VM faults to the default page
 738 */
 739static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
 740					      bool value)
 741{
 742	u32 tmp;
 743
 744	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 745	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 746			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 747	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 748			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 749	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 750			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 751	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 752			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 753	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 754			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 755	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 756			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 757	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 758			    EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 759	WREG32(mmVM_CONTEXT1_CNTL, tmp);
 760}
 761
 762/**
 763 * gmc_v8_0_set_prt - set PRT VM fault
 764 *
 765 * @adev: amdgpu_device pointer
 766 * @enable: enable/disable VM fault handling for PRT
 767*/
 768static void gmc_v8_0_set_prt(struct amdgpu_device *adev, bool enable)
 769{
 770	u32 tmp;
 771
 772	if (enable && !adev->gmc.prt_warning) {
 773		dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
 774		adev->gmc.prt_warning = true;
 775	}
 776
 777	tmp = RREG32(mmVM_PRT_CNTL);
 778	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 779			    CB_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
 780	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 781			    CB_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
 782	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 783			    TC_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
 784	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 785			    TC_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
 786	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 787			    L2_CACHE_STORE_INVALID_ENTRIES, enable);
 788	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 789			    L1_TLB_STORE_INVALID_ENTRIES, enable);
 790	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 791			    MASK_PDE0_FAULT, enable);
 792	WREG32(mmVM_PRT_CNTL, tmp);
 793
 794	if (enable) {
 795		uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
 796		uint32_t high = adev->vm_manager.max_pfn -
 797			(AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
 798
 799		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
 800		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
 801		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
 802		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
 803		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
 804		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
 805		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
 806		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
 807	} else {
 808		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
 809		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
 810		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
 811		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
 812		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
 813		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
 814		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
 815		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
 816	}
 817}
 818
 819/**
 820 * gmc_v8_0_gart_enable - gart enable
 821 *
 822 * @adev: amdgpu_device pointer
 823 *
 824 * This sets up the TLBs, programs the page tables for VMID0,
 825 * sets up the hw for VMIDs 1-15 which are allocated on
 826 * demand, and sets up the global locations for the LDS, GDS,
 827 * and GPUVM for FSA64 clients (VI).
 828 * Returns 0 for success, errors for failure.
 829 */
 830static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
 831{
 832	uint64_t table_addr;
 833	int r, i;
 834	u32 tmp, field;
 835
 836	if (adev->gart.bo == NULL) {
 837		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
 838		return -EINVAL;
 839	}
 840	r = amdgpu_gart_table_vram_pin(adev);
 841	if (r)
 842		return r;
 843
 844	table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
 845
 846	/* Setup TLB control */
 847	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 848	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
 849	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
 850	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
 851	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
 852	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
 853	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 854	/* Setup L2 cache */
 855	tmp = RREG32(mmVM_L2_CNTL);
 856	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
 857	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
 858	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
 859	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
 860	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
 861	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
 862	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
 863	WREG32(mmVM_L2_CNTL, tmp);
 864	tmp = RREG32(mmVM_L2_CNTL2);
 865	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
 866	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
 867	WREG32(mmVM_L2_CNTL2, tmp);
 868
 869	field = adev->vm_manager.fragment_size;
 870	tmp = RREG32(mmVM_L2_CNTL3);
 871	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
 872	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, field);
 873	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, field);
 874	WREG32(mmVM_L2_CNTL3, tmp);
 875	/* XXX: set to enable PTE/PDE in system memory */
 876	tmp = RREG32(mmVM_L2_CNTL4);
 877	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
 878	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
 879	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
 880	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
 881	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
 882	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
 883	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
 884	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
 885	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
 886	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
 887	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
 888	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
 889	WREG32(mmVM_L2_CNTL4, tmp);
 890	/* setup context0 */
 891	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
 892	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
 893	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
 894	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 895			(u32)(adev->dummy_page_addr >> 12));
 896	WREG32(mmVM_CONTEXT0_CNTL2, 0);
 897	tmp = RREG32(mmVM_CONTEXT0_CNTL);
 898	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
 899	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
 900	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 901	WREG32(mmVM_CONTEXT0_CNTL, tmp);
 902
 903	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
 904	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
 905	WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
 906
 907	/* empty context1-15 */
 908	/* FIXME start with 4G, once using 2 level pt switch to full
 909	 * vm size space
 910	 */
 911	/* set vm size, must be a multiple of 4 */
 912	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
 913	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
 914	for (i = 1; i < AMDGPU_NUM_VMID; i++) {
 915		if (i < 8)
 916			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
 917			       table_addr >> 12);
 918		else
 919			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
 920			       table_addr >> 12);
 921	}
 922
 923	/* enable context1-15 */
 924	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
 925	       (u32)(adev->dummy_page_addr >> 12));
 926	WREG32(mmVM_CONTEXT1_CNTL2, 4);
 927	tmp = RREG32(mmVM_CONTEXT1_CNTL);
 928	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
 929	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
 930	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 931	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 932	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 933	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 934	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 935	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 936	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 937	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
 938			    adev->vm_manager.block_size - 9);
 939	WREG32(mmVM_CONTEXT1_CNTL, tmp);
 940	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 941		gmc_v8_0_set_fault_enable_default(adev, false);
 942	else
 943		gmc_v8_0_set_fault_enable_default(adev, true);
 944
 945	gmc_v8_0_flush_gpu_tlb(adev, 0, 0, 0);
 946	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
 947		 (unsigned)(adev->gmc.gart_size >> 20),
 948		 (unsigned long long)table_addr);
 949	adev->gart.ready = true;
 950	return 0;
 951}
 952
 953static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
 954{
 955	int r;
 956
 957	if (adev->gart.bo) {
 958		WARN(1, "R600 PCIE GART already initialized\n");
 959		return 0;
 960	}
 961	/* Initialize common gart structure */
 962	r = amdgpu_gart_init(adev);
 963	if (r)
 964		return r;
 965	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
 966	adev->gart.gart_pte_flags = AMDGPU_PTE_EXECUTABLE;
 967	return amdgpu_gart_table_vram_alloc(adev);
 968}
 969
 970/**
 971 * gmc_v8_0_gart_disable - gart disable
 972 *
 973 * @adev: amdgpu_device pointer
 974 *
 975 * This disables all VM page table (VI).
 976 */
 977static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
 978{
 979	u32 tmp;
 980
 981	/* Disable all tables */
 982	WREG32(mmVM_CONTEXT0_CNTL, 0);
 983	WREG32(mmVM_CONTEXT1_CNTL, 0);
 984	/* Setup TLB control */
 985	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 986	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
 987	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
 988	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
 989	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 990	/* Setup L2 cache */
 991	tmp = RREG32(mmVM_L2_CNTL);
 992	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
 993	WREG32(mmVM_L2_CNTL, tmp);
 994	WREG32(mmVM_L2_CNTL2, 0);
 995	amdgpu_gart_table_vram_unpin(adev);
 996}
 997
 998/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 999 * gmc_v8_0_vm_decode_fault - print human readable fault info
1000 *
1001 * @adev: amdgpu_device pointer
1002 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
1003 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
1004 * @mc_client: VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT register value
1005 * @pasid: debug logging only - no functional use
1006 *
1007 * Print human readable fault information (VI).
1008 */
1009static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev, u32 status,
1010				     u32 addr, u32 mc_client, unsigned pasid)
1011{
 
1012	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
1013	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1014					PROTECTIONS);
1015	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
1016		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
1017	u32 mc_id;
1018
1019	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1020			      MEMORY_CLIENT_ID);
1021
1022	dev_err(adev->dev, "VM fault (0x%02x, vmid %d, pasid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
1023	       protections, vmid, pasid, addr,
1024	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1025			     MEMORY_CLIENT_RW) ?
1026	       "write" : "read", block, mc_client, mc_id);
1027}
1028
1029static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
1030{
1031	switch (mc_seq_vram_type) {
1032	case MC_SEQ_MISC0__MT__GDDR1:
1033		return AMDGPU_VRAM_TYPE_GDDR1;
1034	case MC_SEQ_MISC0__MT__DDR2:
1035		return AMDGPU_VRAM_TYPE_DDR2;
1036	case MC_SEQ_MISC0__MT__GDDR3:
1037		return AMDGPU_VRAM_TYPE_GDDR3;
1038	case MC_SEQ_MISC0__MT__GDDR4:
1039		return AMDGPU_VRAM_TYPE_GDDR4;
1040	case MC_SEQ_MISC0__MT__GDDR5:
1041		return AMDGPU_VRAM_TYPE_GDDR5;
1042	case MC_SEQ_MISC0__MT__HBM:
1043		return AMDGPU_VRAM_TYPE_HBM;
1044	case MC_SEQ_MISC0__MT__DDR3:
1045		return AMDGPU_VRAM_TYPE_DDR3;
1046	default:
1047		return AMDGPU_VRAM_TYPE_UNKNOWN;
1048	}
1049}
1050
1051static int gmc_v8_0_early_init(void *handle)
1052{
1053	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1054
1055	gmc_v8_0_set_gmc_funcs(adev);
1056	gmc_v8_0_set_irq_funcs(adev);
1057
1058	adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
1059	adev->gmc.shared_aperture_end =
1060		adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
1061	adev->gmc.private_aperture_start =
1062		adev->gmc.shared_aperture_end + 1;
1063	adev->gmc.private_aperture_end =
1064		adev->gmc.private_aperture_start + (4ULL << 30) - 1;
1065
1066	return 0;
1067}
1068
1069static int gmc_v8_0_late_init(void *handle)
1070{
1071	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1072
1073	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
1074		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
1075	else
1076		return 0;
1077}
1078
1079static unsigned gmc_v8_0_get_vbios_fb_size(struct amdgpu_device *adev)
1080{
1081	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
1082	unsigned size;
1083
1084	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
1085		size = AMDGPU_VBIOS_VGA_ALLOCATION;
1086	} else {
1087		u32 viewport = RREG32(mmVIEWPORT_SIZE);
1088		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
1089			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
1090			4);
1091	}
1092
1093	return size;
1094}
1095
1096#define mmMC_SEQ_MISC0_FIJI 0xA71
1097
1098static int gmc_v8_0_sw_init(void *handle)
1099{
1100	int r;
 
1101	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1102
1103	adev->num_vmhubs = 1;
1104
1105	if (adev->flags & AMD_IS_APU) {
1106		adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
1107	} else {
1108		u32 tmp;
1109
1110		if ((adev->asic_type == CHIP_FIJI) ||
1111		    (adev->asic_type == CHIP_VEGAM))
1112			tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
1113		else
1114			tmp = RREG32(mmMC_SEQ_MISC0);
1115		tmp &= MC_SEQ_MISC0__MT__MASK;
1116		adev->gmc.vram_type = gmc_v8_0_convert_vram_type(tmp);
1117	}
1118
1119	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_PAGE_INV_FAULT, &adev->gmc.vm_fault);
1120	if (r)
1121		return r;
1122
1123	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_MEM_PROT_FAULT, &adev->gmc.vm_fault);
1124	if (r)
1125		return r;
1126
1127	/* Adjust VM size here.
1128	 * Currently set to 4GB ((1 << 20) 4k pages).
1129	 * Max GPUVM size for cayman and SI is 40 bits.
1130	 */
1131	amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
1132
1133	/* Set the internal MC address mask
1134	 * This is the max address of the GPU's
1135	 * internal address space.
1136	 */
1137	adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
1138
1139	r = dma_set_mask_and_coherent(adev->dev, DMA_BIT_MASK(40));
 
 
 
 
 
 
 
1140	if (r) {
1141		pr_warn("No suitable DMA available\n");
1142		return r;
 
 
 
 
 
 
1143	}
1144	adev->need_swiotlb = drm_need_swiotlb(40);
1145
1146	r = gmc_v8_0_init_microcode(adev);
1147	if (r) {
1148		DRM_ERROR("Failed to load mc firmware!\n");
1149		return r;
1150	}
1151
1152	r = gmc_v8_0_mc_init(adev);
1153	if (r)
1154		return r;
1155
1156	amdgpu_gmc_get_vbios_allocations(adev);
1157
1158	/* Memory manager */
1159	r = amdgpu_bo_init(adev);
1160	if (r)
1161		return r;
1162
1163	r = gmc_v8_0_gart_init(adev);
1164	if (r)
1165		return r;
1166
1167	/*
1168	 * number of VMs
1169	 * VMID 0 is reserved for System
1170	 * amdgpu graphics/compute will use VMIDs 1-7
1171	 * amdkfd will use VMIDs 8-15
1172	 */
1173	adev->vm_manager.first_kfd_vmid = 8;
1174	amdgpu_vm_manager_init(adev);
1175
1176	/* base offset of vram pages */
1177	if (adev->flags & AMD_IS_APU) {
1178		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
1179
1180		tmp <<= 22;
1181		adev->vm_manager.vram_base_offset = tmp;
1182	} else {
1183		adev->vm_manager.vram_base_offset = 0;
1184	}
1185
1186	adev->gmc.vm_fault_info = kmalloc(sizeof(struct kfd_vm_fault_info),
1187					GFP_KERNEL);
1188	if (!adev->gmc.vm_fault_info)
1189		return -ENOMEM;
1190	atomic_set(&adev->gmc.vm_fault_info_updated, 0);
1191
1192	return 0;
1193}
1194
1195static int gmc_v8_0_sw_fini(void *handle)
1196{
1197	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1198
 
 
 
 
 
 
1199	amdgpu_gem_force_release(adev);
1200	amdgpu_vm_manager_fini(adev);
1201	kfree(adev->gmc.vm_fault_info);
1202	amdgpu_gart_table_vram_free(adev);
1203	amdgpu_bo_fini(adev);
1204	release_firmware(adev->gmc.fw);
1205	adev->gmc.fw = NULL;
1206
1207	return 0;
1208}
1209
1210static int gmc_v8_0_hw_init(void *handle)
1211{
1212	int r;
1213	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1214
1215	gmc_v8_0_init_golden_registers(adev);
1216
1217	gmc_v8_0_mc_program(adev);
1218
1219	if (adev->asic_type == CHIP_TONGA) {
1220		r = gmc_v8_0_tonga_mc_load_microcode(adev);
1221		if (r) {
1222			DRM_ERROR("Failed to load MC firmware!\n");
1223			return r;
1224		}
1225	} else if (adev->asic_type == CHIP_POLARIS11 ||
1226			adev->asic_type == CHIP_POLARIS10 ||
1227			adev->asic_type == CHIP_POLARIS12) {
1228		r = gmc_v8_0_polaris_mc_load_microcode(adev);
1229		if (r) {
1230			DRM_ERROR("Failed to load MC firmware!\n");
1231			return r;
1232		}
1233	}
1234
1235	r = gmc_v8_0_gart_enable(adev);
1236	if (r)
1237		return r;
1238
1239	return r;
1240}
1241
1242static int gmc_v8_0_hw_fini(void *handle)
1243{
1244	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1245
1246	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
1247	gmc_v8_0_gart_disable(adev);
1248
1249	return 0;
1250}
1251
1252static int gmc_v8_0_suspend(void *handle)
1253{
1254	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1255
 
 
 
 
1256	gmc_v8_0_hw_fini(adev);
1257
1258	return 0;
1259}
1260
1261static int gmc_v8_0_resume(void *handle)
1262{
1263	int r;
1264	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1265
1266	r = gmc_v8_0_hw_init(adev);
1267	if (r)
1268		return r;
1269
1270	amdgpu_vmid_reset_all(adev);
 
 
 
 
 
 
 
1271
1272	return 0;
1273}
1274
1275static bool gmc_v8_0_is_idle(void *handle)
1276{
1277	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1278	u32 tmp = RREG32(mmSRBM_STATUS);
1279
1280	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1281		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1282		return false;
1283
1284	return true;
1285}
1286
1287static int gmc_v8_0_wait_for_idle(void *handle)
1288{
1289	unsigned i;
1290	u32 tmp;
1291	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1292
1293	for (i = 0; i < adev->usec_timeout; i++) {
1294		/* read MC_STATUS */
1295		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1296					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1297					       SRBM_STATUS__MCC_BUSY_MASK |
1298					       SRBM_STATUS__MCD_BUSY_MASK |
1299					       SRBM_STATUS__VMC_BUSY_MASK |
1300					       SRBM_STATUS__VMC1_BUSY_MASK);
1301		if (!tmp)
1302			return 0;
1303		udelay(1);
1304	}
1305	return -ETIMEDOUT;
1306
1307}
1308
1309static bool gmc_v8_0_check_soft_reset(void *handle)
1310{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1311	u32 srbm_soft_reset = 0;
1312	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1313	u32 tmp = RREG32(mmSRBM_STATUS);
1314
1315	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1316		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1317						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1318
1319	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1320		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1321		if (!(adev->flags & AMD_IS_APU))
1322			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1323							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1324	}
 
1325	if (srbm_soft_reset) {
1326		adev->gmc.srbm_soft_reset = srbm_soft_reset;
1327		return true;
1328	} else {
1329		adev->gmc.srbm_soft_reset = 0;
1330		return false;
1331	}
1332}
1333
1334static int gmc_v8_0_pre_soft_reset(void *handle)
1335{
1336	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1337
1338	if (!adev->gmc.srbm_soft_reset)
1339		return 0;
1340
1341	gmc_v8_0_mc_stop(adev);
1342	if (gmc_v8_0_wait_for_idle(adev)) {
1343		dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1344	}
1345
1346	return 0;
1347}
1348
1349static int gmc_v8_0_soft_reset(void *handle)
1350{
1351	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1352	u32 srbm_soft_reset;
1353
1354	if (!adev->gmc.srbm_soft_reset)
1355		return 0;
1356	srbm_soft_reset = adev->gmc.srbm_soft_reset;
1357
1358	if (srbm_soft_reset) {
1359		u32 tmp;
1360
1361		tmp = RREG32(mmSRBM_SOFT_RESET);
1362		tmp |= srbm_soft_reset;
1363		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1364		WREG32(mmSRBM_SOFT_RESET, tmp);
1365		tmp = RREG32(mmSRBM_SOFT_RESET);
1366
1367		udelay(50);
1368
1369		tmp &= ~srbm_soft_reset;
1370		WREG32(mmSRBM_SOFT_RESET, tmp);
1371		tmp = RREG32(mmSRBM_SOFT_RESET);
1372
1373		/* Wait a little for things to settle down */
1374		udelay(50);
1375	}
1376
1377	return 0;
1378}
1379
1380static int gmc_v8_0_post_soft_reset(void *handle)
1381{
1382	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1383
1384	if (!adev->gmc.srbm_soft_reset)
1385		return 0;
1386
1387	gmc_v8_0_mc_resume(adev);
1388	return 0;
1389}
1390
1391static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1392					     struct amdgpu_irq_src *src,
1393					     unsigned type,
1394					     enum amdgpu_interrupt_state state)
1395{
1396	u32 tmp;
1397	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1398		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1399		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1400		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1401		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1402		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1403		    VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1404
1405	switch (state) {
1406	case AMDGPU_IRQ_STATE_DISABLE:
1407		/* system context */
1408		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1409		tmp &= ~bits;
1410		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1411		/* VMs */
1412		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1413		tmp &= ~bits;
1414		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1415		break;
1416	case AMDGPU_IRQ_STATE_ENABLE:
1417		/* system context */
1418		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1419		tmp |= bits;
1420		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1421		/* VMs */
1422		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1423		tmp |= bits;
1424		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1425		break;
1426	default:
1427		break;
1428	}
1429
1430	return 0;
1431}
1432
1433static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1434				      struct amdgpu_irq_src *source,
1435				      struct amdgpu_iv_entry *entry)
1436{
1437	u32 addr, status, mc_client, vmid;
1438
1439	if (amdgpu_sriov_vf(adev)) {
1440		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1441			entry->src_id, entry->src_data[0]);
1442		dev_err(adev->dev, " Can't decode VM fault info here on SRIOV VF\n");
1443		return 0;
1444	}
1445
1446	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1447	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1448	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1449	/* reset addr and status */
1450	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1451
1452	if (!addr && !status)
1453		return 0;
1454
1455	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1456		gmc_v8_0_set_fault_enable_default(adev, false);
1457
1458	if (printk_ratelimit()) {
1459		struct amdgpu_task_info task_info;
1460
1461		memset(&task_info, 0, sizeof(struct amdgpu_task_info));
1462		amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
1463
1464		dev_err(adev->dev, "GPU fault detected: %d 0x%08x for process %s pid %d thread %s pid %d\n",
1465			entry->src_id, entry->src_data[0], task_info.process_name,
1466			task_info.tgid, task_info.task_name, task_info.pid);
1467		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1468			addr);
1469		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1470			status);
1471		gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client,
1472					 entry->pasid);
1473	}
1474
1475	vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1476			     VMID);
1477	if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
1478		&& !atomic_read(&adev->gmc.vm_fault_info_updated)) {
1479		struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
1480		u32 protections = REG_GET_FIELD(status,
1481					VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1482					PROTECTIONS);
1483
1484		info->vmid = vmid;
1485		info->mc_id = REG_GET_FIELD(status,
1486					    VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1487					    MEMORY_CLIENT_ID);
1488		info->status = status;
1489		info->page_addr = addr;
1490		info->prot_valid = protections & 0x7 ? true : false;
1491		info->prot_read = protections & 0x8 ? true : false;
1492		info->prot_write = protections & 0x10 ? true : false;
1493		info->prot_exec = protections & 0x20 ? true : false;
1494		mb();
1495		atomic_set(&adev->gmc.vm_fault_info_updated, 1);
1496	}
1497
1498	return 0;
1499}
1500
1501static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1502						     bool enable)
1503{
1504	uint32_t data;
1505
1506	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
1507		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1508		data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1509		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1510
1511		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1512		data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1513		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1514
1515		data = RREG32(mmMC_HUB_MISC_VM_CG);
1516		data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1517		WREG32(mmMC_HUB_MISC_VM_CG, data);
1518
1519		data = RREG32(mmMC_XPB_CLK_GAT);
1520		data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1521		WREG32(mmMC_XPB_CLK_GAT, data);
1522
1523		data = RREG32(mmATC_MISC_CG);
1524		data |= ATC_MISC_CG__ENABLE_MASK;
1525		WREG32(mmATC_MISC_CG, data);
1526
1527		data = RREG32(mmMC_CITF_MISC_WR_CG);
1528		data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1529		WREG32(mmMC_CITF_MISC_WR_CG, data);
1530
1531		data = RREG32(mmMC_CITF_MISC_RD_CG);
1532		data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1533		WREG32(mmMC_CITF_MISC_RD_CG, data);
1534
1535		data = RREG32(mmMC_CITF_MISC_VM_CG);
1536		data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1537		WREG32(mmMC_CITF_MISC_VM_CG, data);
1538
1539		data = RREG32(mmVM_L2_CG);
1540		data |= VM_L2_CG__ENABLE_MASK;
1541		WREG32(mmVM_L2_CG, data);
1542	} else {
1543		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1544		data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1545		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1546
1547		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1548		data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1549		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1550
1551		data = RREG32(mmMC_HUB_MISC_VM_CG);
1552		data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1553		WREG32(mmMC_HUB_MISC_VM_CG, data);
1554
1555		data = RREG32(mmMC_XPB_CLK_GAT);
1556		data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1557		WREG32(mmMC_XPB_CLK_GAT, data);
1558
1559		data = RREG32(mmATC_MISC_CG);
1560		data &= ~ATC_MISC_CG__ENABLE_MASK;
1561		WREG32(mmATC_MISC_CG, data);
1562
1563		data = RREG32(mmMC_CITF_MISC_WR_CG);
1564		data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1565		WREG32(mmMC_CITF_MISC_WR_CG, data);
1566
1567		data = RREG32(mmMC_CITF_MISC_RD_CG);
1568		data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1569		WREG32(mmMC_CITF_MISC_RD_CG, data);
1570
1571		data = RREG32(mmMC_CITF_MISC_VM_CG);
1572		data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1573		WREG32(mmMC_CITF_MISC_VM_CG, data);
1574
1575		data = RREG32(mmVM_L2_CG);
1576		data &= ~VM_L2_CG__ENABLE_MASK;
1577		WREG32(mmVM_L2_CG, data);
1578	}
1579}
1580
1581static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1582				       bool enable)
1583{
1584	uint32_t data;
1585
1586	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) {
1587		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1588		data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1589		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1590
1591		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1592		data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1593		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1594
1595		data = RREG32(mmMC_HUB_MISC_VM_CG);
1596		data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1597		WREG32(mmMC_HUB_MISC_VM_CG, data);
1598
1599		data = RREG32(mmMC_XPB_CLK_GAT);
1600		data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1601		WREG32(mmMC_XPB_CLK_GAT, data);
1602
1603		data = RREG32(mmATC_MISC_CG);
1604		data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1605		WREG32(mmATC_MISC_CG, data);
1606
1607		data = RREG32(mmMC_CITF_MISC_WR_CG);
1608		data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1609		WREG32(mmMC_CITF_MISC_WR_CG, data);
1610
1611		data = RREG32(mmMC_CITF_MISC_RD_CG);
1612		data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1613		WREG32(mmMC_CITF_MISC_RD_CG, data);
1614
1615		data = RREG32(mmMC_CITF_MISC_VM_CG);
1616		data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1617		WREG32(mmMC_CITF_MISC_VM_CG, data);
1618
1619		data = RREG32(mmVM_L2_CG);
1620		data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1621		WREG32(mmVM_L2_CG, data);
1622	} else {
1623		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1624		data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1625		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1626
1627		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1628		data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1629		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1630
1631		data = RREG32(mmMC_HUB_MISC_VM_CG);
1632		data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1633		WREG32(mmMC_HUB_MISC_VM_CG, data);
1634
1635		data = RREG32(mmMC_XPB_CLK_GAT);
1636		data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1637		WREG32(mmMC_XPB_CLK_GAT, data);
1638
1639		data = RREG32(mmATC_MISC_CG);
1640		data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1641		WREG32(mmATC_MISC_CG, data);
1642
1643		data = RREG32(mmMC_CITF_MISC_WR_CG);
1644		data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1645		WREG32(mmMC_CITF_MISC_WR_CG, data);
1646
1647		data = RREG32(mmMC_CITF_MISC_RD_CG);
1648		data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1649		WREG32(mmMC_CITF_MISC_RD_CG, data);
1650
1651		data = RREG32(mmMC_CITF_MISC_VM_CG);
1652		data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1653		WREG32(mmMC_CITF_MISC_VM_CG, data);
1654
1655		data = RREG32(mmVM_L2_CG);
1656		data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1657		WREG32(mmVM_L2_CG, data);
1658	}
1659}
1660
1661static int gmc_v8_0_set_clockgating_state(void *handle,
1662					  enum amd_clockgating_state state)
1663{
1664	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1665
1666	if (amdgpu_sriov_vf(adev))
1667		return 0;
1668
1669	switch (adev->asic_type) {
1670	case CHIP_FIJI:
1671		fiji_update_mc_medium_grain_clock_gating(adev,
1672				state == AMD_CG_STATE_GATE);
1673		fiji_update_mc_light_sleep(adev,
1674				state == AMD_CG_STATE_GATE);
1675		break;
1676	default:
1677		break;
1678	}
1679	return 0;
1680}
1681
1682static int gmc_v8_0_set_powergating_state(void *handle,
1683					  enum amd_powergating_state state)
1684{
1685	return 0;
1686}
1687
1688static void gmc_v8_0_get_clockgating_state(void *handle, u32 *flags)
1689{
1690	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1691	int data;
1692
1693	if (amdgpu_sriov_vf(adev))
1694		*flags = 0;
1695
1696	/* AMD_CG_SUPPORT_MC_MGCG */
1697	data = RREG32(mmMC_HUB_MISC_HUB_CG);
1698	if (data & MC_HUB_MISC_HUB_CG__ENABLE_MASK)
1699		*flags |= AMD_CG_SUPPORT_MC_MGCG;
1700
1701	/* AMD_CG_SUPPORT_MC_LS */
1702	if (data & MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK)
1703		*flags |= AMD_CG_SUPPORT_MC_LS;
1704}
1705
1706static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1707	.name = "gmc_v8_0",
1708	.early_init = gmc_v8_0_early_init,
1709	.late_init = gmc_v8_0_late_init,
1710	.sw_init = gmc_v8_0_sw_init,
1711	.sw_fini = gmc_v8_0_sw_fini,
1712	.hw_init = gmc_v8_0_hw_init,
1713	.hw_fini = gmc_v8_0_hw_fini,
1714	.suspend = gmc_v8_0_suspend,
1715	.resume = gmc_v8_0_resume,
1716	.is_idle = gmc_v8_0_is_idle,
1717	.wait_for_idle = gmc_v8_0_wait_for_idle,
1718	.check_soft_reset = gmc_v8_0_check_soft_reset,
1719	.pre_soft_reset = gmc_v8_0_pre_soft_reset,
1720	.soft_reset = gmc_v8_0_soft_reset,
1721	.post_soft_reset = gmc_v8_0_post_soft_reset,
1722	.set_clockgating_state = gmc_v8_0_set_clockgating_state,
1723	.set_powergating_state = gmc_v8_0_set_powergating_state,
1724	.get_clockgating_state = gmc_v8_0_get_clockgating_state,
1725};
1726
1727static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
1728	.flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
1729	.flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
1730	.emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
1731	.emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
1732	.set_prt = gmc_v8_0_set_prt,
1733	.get_vm_pde = gmc_v8_0_get_vm_pde,
1734	.get_vm_pte = gmc_v8_0_get_vm_pte,
1735	.get_vbios_fb_size = gmc_v8_0_get_vbios_fb_size,
1736};
1737
1738static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1739	.set = gmc_v8_0_vm_fault_interrupt_state,
1740	.process = gmc_v8_0_process_interrupt,
1741};
1742
1743static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev)
1744{
1745	adev->gmc.gmc_funcs = &gmc_v8_0_gmc_funcs;
 
1746}
1747
1748static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1749{
1750	adev->gmc.vm_fault.num_types = 1;
1751	adev->gmc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1752}
1753
1754const struct amdgpu_ip_block_version gmc_v8_0_ip_block =
1755{
1756	.type = AMD_IP_BLOCK_TYPE_GMC,
1757	.major = 8,
1758	.minor = 0,
1759	.rev = 0,
1760	.funcs = &gmc_v8_0_ip_funcs,
1761};
1762
1763const struct amdgpu_ip_block_version gmc_v8_1_ip_block =
1764{
1765	.type = AMD_IP_BLOCK_TYPE_GMC,
1766	.major = 8,
1767	.minor = 1,
1768	.rev = 0,
1769	.funcs = &gmc_v8_0_ip_funcs,
1770};
1771
1772const struct amdgpu_ip_block_version gmc_v8_5_ip_block =
1773{
1774	.type = AMD_IP_BLOCK_TYPE_GMC,
1775	.major = 8,
1776	.minor = 5,
1777	.rev = 0,
1778	.funcs = &gmc_v8_0_ip_funcs,
1779};