amdgpu_device.c 95.1 KB
Newer Older
A
Alex Deucher 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
/*
 * Copyright 2008 Advanced Micro Devices, Inc.
 * Copyright 2008 Red Hat Inc.
 * Copyright 2009 Jerome Glisse.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 *
 * Authors: Dave Airlie
 *          Alex Deucher
 *          Jerome Glisse
 */
28
#include <linux/power_supply.h>
29
#include <linux/kthread.h>
A
Alex Deucher 已提交
30 31 32 33
#include <linux/console.h>
#include <linux/slab.h>
#include <drm/drmP.h>
#include <drm/drm_crtc_helper.h>
34
#include <drm/drm_atomic_helper.h>
A
Alex Deucher 已提交
35 36 37 38 39
#include <drm/amdgpu_drm.h>
#include <linux/vgaarb.h>
#include <linux/vga_switcheroo.h>
#include <linux/efi.h>
#include "amdgpu.h"
40
#include "amdgpu_trace.h"
A
Alex Deucher 已提交
41 42 43
#include "amdgpu_i2c.h"
#include "atom.h"
#include "amdgpu_atombios.h"
44
#include "amdgpu_atomfirmware.h"
45
#include "amd_pcie.h"
K
Ken Wang 已提交
46 47 48
#ifdef CONFIG_DRM_AMDGPU_SI
#include "si.h"
#endif
49 50 51
#ifdef CONFIG_DRM_AMDGPU_CIK
#include "cik.h"
#endif
52
#include "vi.h"
53
#include "soc15.h"
A
Alex Deucher 已提交
54
#include "bif/bif_4_1_d.h"
55
#include <linux/pci.h>
56
#include <linux/firmware.h>
57
#include "amdgpu_vf_error.h"
A
Alex Deucher 已提交
58

59
#include "amdgpu_amdkfd.h"
60
#include "amdgpu_pm.h"
A
Alex Deucher 已提交
61

62
MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
63
MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin");
64
MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin");
65

66 67
#define AMDGPU_RESUME_MS		2000

A
Alex Deucher 已提交
68
static const char *amdgpu_asic_name[] = {
69 70 71 72 73
	"TAHITI",
	"PITCAIRN",
	"VERDE",
	"OLAND",
	"HAINAN",
A
Alex Deucher 已提交
74 75 76 77 78 79 80
	"BONAIRE",
	"KAVERI",
	"KABINI",
	"HAWAII",
	"MULLINS",
	"TOPAZ",
	"TONGA",
81
	"FIJI",
A
Alex Deucher 已提交
82
	"CARRIZO",
S
Samuel Li 已提交
83
	"STONEY",
84 85
	"POLARIS10",
	"POLARIS11",
86
	"POLARIS12",
L
Leo Liu 已提交
87
	"VEGAM",
K
Ken Wang 已提交
88
	"VEGA10",
89
	"VEGA12",
90
	"VEGA20",
91
	"RAVEN",
A
Alex Deucher 已提交
92 93 94
	"LAST",
};

95 96
static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev);

97 98 99 100 101 102 103 104
/**
 * amdgpu_device_is_px - Is the device is a dGPU with HG/PX power control
 *
 * @dev: drm_device pointer
 *
 * Returns true if the device is a dGPU with HG/PX power control,
 * otherwise return false.
 */
A
Alex Deucher 已提交
105 106 107 108
bool amdgpu_device_is_px(struct drm_device *dev)
{
	struct amdgpu_device *adev = dev->dev_private;

109
	if (adev->flags & AMD_IS_PX)
A
Alex Deucher 已提交
110 111 112 113 114 115 116
		return true;
	return false;
}

/*
 * MMIO register access helper functions.
 */
117 118 119 120 121 122 123 124 125
/**
 * amdgpu_mm_rreg - read a memory mapped IO register
 *
 * @adev: amdgpu_device pointer
 * @reg: dword aligned register offset
 * @acc_flags: access flags which require special behavior
 *
 * Returns the 32 bit value from the offset specified.
 */
A
Alex Deucher 已提交
126
uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg,
M
Monk Liu 已提交
127
			uint32_t acc_flags)
A
Alex Deucher 已提交
128
{
129 130
	uint32_t ret;

131
	if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev))
132 133
		return amdgpu_virt_kiq_rreg(adev, reg);

M
Monk Liu 已提交
134
	if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
135
		ret = readl(((void __iomem *)adev->rmmio) + (reg * 4));
A
Alex Deucher 已提交
136 137 138 139 140 141 142 143
	else {
		unsigned long flags;

		spin_lock_irqsave(&adev->mmio_idx_lock, flags);
		writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4));
		ret = readl(((void __iomem *)adev->rmmio) + (mmMM_DATA * 4));
		spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
	}
144 145
	trace_amdgpu_mm_rreg(adev->pdev->device, reg, ret);
	return ret;
A
Alex Deucher 已提交
146 147
}

148 149 150 151 152 153
/*
 * MMIO register read with bytes helper functions
 * @offset:bytes offset from MMIO start
 *
*/

154 155 156 157 158 159 160 161
/**
 * amdgpu_mm_rreg8 - read a memory mapped IO register
 *
 * @adev: amdgpu_device pointer
 * @offset: byte aligned register offset
 *
 * Returns the 8 bit value from the offset specified.
 */
162 163 164 165 166 167 168 169 170 171 172 173
uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset) {
	if (offset < adev->rmmio_size)
		return (readb(adev->rmmio + offset));
	BUG();
}

/*
 * MMIO register write with bytes helper functions
 * @offset:bytes offset from MMIO start
 * @value: the value want to be written to the register
 *
*/
174 175 176 177 178 179 180 181 182
/**
 * amdgpu_mm_wreg8 - read a memory mapped IO register
 *
 * @adev: amdgpu_device pointer
 * @offset: byte aligned register offset
 * @value: 8 bit value to write
 *
 * Writes the value specified to the offset specified.
 */
183 184 185 186 187 188 189
void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value) {
	if (offset < adev->rmmio_size)
		writeb(value, adev->rmmio + offset);
	else
		BUG();
}

190 191 192 193 194 195 196 197 198 199
/**
 * amdgpu_mm_wreg - write to a memory mapped IO register
 *
 * @adev: amdgpu_device pointer
 * @reg: dword aligned register offset
 * @v: 32 bit value to write to the register
 * @acc_flags: access flags which require special behavior
 *
 * Writes the value specified to the offset specified.
 */
A
Alex Deucher 已提交
200
void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v,
M
Monk Liu 已提交
201
		    uint32_t acc_flags)
A
Alex Deucher 已提交
202
{
203
	trace_amdgpu_mm_wreg(adev->pdev->device, reg, v);
204

205 206 207 208
	if (adev->asic_type >= CHIP_VEGA10 && reg == 0) {
		adev->last_mm_index = v;
	}

209
	if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev))
210 211
		return amdgpu_virt_kiq_wreg(adev, reg, v);

M
Monk Liu 已提交
212
	if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
A
Alex Deucher 已提交
213 214 215 216 217 218 219 220 221
		writel(v, ((void __iomem *)adev->rmmio) + (reg * 4));
	else {
		unsigned long flags;

		spin_lock_irqsave(&adev->mmio_idx_lock, flags);
		writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4));
		writel(v, ((void __iomem *)adev->rmmio) + (mmMM_DATA * 4));
		spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
	}
222 223 224 225

	if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) {
		udelay(500);
	}
A
Alex Deucher 已提交
226 227
}

228 229 230 231 232 233 234 235
/**
 * amdgpu_io_rreg - read an IO register
 *
 * @adev: amdgpu_device pointer
 * @reg: dword aligned register offset
 *
 * Returns the 32 bit value from the offset specified.
 */
A
Alex Deucher 已提交
236 237 238 239 240 241 242 243 244 245
u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg)
{
	if ((reg * 4) < adev->rio_mem_size)
		return ioread32(adev->rio_mem + (reg * 4));
	else {
		iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
		return ioread32(adev->rio_mem + (mmMM_DATA * 4));
	}
}

246 247 248 249 250 251 252 253 254
/**
 * amdgpu_io_wreg - write to an IO register
 *
 * @adev: amdgpu_device pointer
 * @reg: dword aligned register offset
 * @v: 32 bit value to write to the register
 *
 * Writes the value specified to the offset specified.
 */
A
Alex Deucher 已提交
255 256
void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
{
257 258 259
	if (adev->asic_type >= CHIP_VEGA10 && reg == 0) {
		adev->last_mm_index = v;
	}
A
Alex Deucher 已提交
260 261 262 263 264 265 266

	if ((reg * 4) < adev->rio_mem_size)
		iowrite32(v, adev->rio_mem + (reg * 4));
	else {
		iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
		iowrite32(v, adev->rio_mem + (mmMM_DATA * 4));
	}
267 268 269 270

	if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) {
		udelay(500);
	}
A
Alex Deucher 已提交
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310
}

/**
 * amdgpu_mm_rdoorbell - read a doorbell dword
 *
 * @adev: amdgpu_device pointer
 * @index: doorbell index
 *
 * Returns the value in the doorbell aperture at the
 * requested doorbell index (CIK).
 */
u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index)
{
	if (index < adev->doorbell.num_doorbells) {
		return readl(adev->doorbell.ptr + index);
	} else {
		DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
		return 0;
	}
}

/**
 * amdgpu_mm_wdoorbell - write a doorbell dword
 *
 * @adev: amdgpu_device pointer
 * @index: doorbell index
 * @v: value to write
 *
 * Writes @v to the doorbell aperture at the
 * requested doorbell index (CIK).
 */
void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v)
{
	if (index < adev->doorbell.num_doorbells) {
		writel(v, adev->doorbell.ptr + index);
	} else {
		DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
	}
}

311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
/**
 * amdgpu_mm_rdoorbell64 - read a doorbell Qword
 *
 * @adev: amdgpu_device pointer
 * @index: doorbell index
 *
 * Returns the value in the doorbell aperture at the
 * requested doorbell index (VEGA10+).
 */
u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index)
{
	if (index < adev->doorbell.num_doorbells) {
		return atomic64_read((atomic64_t *)(adev->doorbell.ptr + index));
	} else {
		DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
		return 0;
	}
}

/**
 * amdgpu_mm_wdoorbell64 - write a doorbell Qword
 *
 * @adev: amdgpu_device pointer
 * @index: doorbell index
 * @v: value to write
 *
 * Writes @v to the doorbell aperture at the
 * requested doorbell index (VEGA10+).
 */
void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v)
{
	if (index < adev->doorbell.num_doorbells) {
		atomic64_set((atomic64_t *)(adev->doorbell.ptr + index), v);
	} else {
		DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
	}
}

A
Alex Deucher 已提交
349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422
/**
 * amdgpu_invalid_rreg - dummy reg read function
 *
 * @adev: amdgpu device pointer
 * @reg: offset of register
 *
 * Dummy register read function.  Used for register blocks
 * that certain asics don't have (all asics).
 * Returns the value in the register.
 */
static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg)
{
	DRM_ERROR("Invalid callback to read register 0x%04X\n", reg);
	BUG();
	return 0;
}

/**
 * amdgpu_invalid_wreg - dummy reg write function
 *
 * @adev: amdgpu device pointer
 * @reg: offset of register
 * @v: value to write to the register
 *
 * Dummy register read function.  Used for register blocks
 * that certain asics don't have (all asics).
 */
static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v)
{
	DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n",
		  reg, v);
	BUG();
}

/**
 * amdgpu_block_invalid_rreg - dummy reg read function
 *
 * @adev: amdgpu device pointer
 * @block: offset of instance
 * @reg: offset of register
 *
 * Dummy register read function.  Used for register blocks
 * that certain asics don't have (all asics).
 * Returns the value in the register.
 */
static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev,
					  uint32_t block, uint32_t reg)
{
	DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n",
		  reg, block);
	BUG();
	return 0;
}

/**
 * amdgpu_block_invalid_wreg - dummy reg write function
 *
 * @adev: amdgpu device pointer
 * @block: offset of instance
 * @reg: offset of register
 * @v: value to write to the register
 *
 * Dummy register read function.  Used for register blocks
 * that certain asics don't have (all asics).
 */
static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev,
				      uint32_t block,
				      uint32_t reg, uint32_t v)
{
	DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n",
		  reg, block, v);
	BUG();
}

423 424 425 426 427 428 429 430
/**
 * amdgpu_device_vram_scratch_init - allocate the VRAM scratch page
 *
 * @adev: amdgpu device pointer
 *
 * Allocates a scratch page of VRAM for use by various things in the
 * driver.
 */
431
static int amdgpu_device_vram_scratch_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
432
{
433 434 435 436 437
	return amdgpu_bo_create_kernel(adev, AMDGPU_GPU_PAGE_SIZE,
				       PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM,
				       &adev->vram_scratch.robj,
				       &adev->vram_scratch.gpu_addr,
				       (void **)&adev->vram_scratch.ptr);
A
Alex Deucher 已提交
438 439
}

440 441 442 443 444 445 446
/**
 * amdgpu_device_vram_scratch_fini - Free the VRAM scratch page
 *
 * @adev: amdgpu device pointer
 *
 * Frees the VRAM scratch page.
 */
447
static void amdgpu_device_vram_scratch_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
448
{
449
	amdgpu_bo_free_kernel(&adev->vram_scratch.robj, NULL, NULL);
A
Alex Deucher 已提交
450 451 452
}

/**
453
 * amdgpu_device_program_register_sequence - program an array of registers.
A
Alex Deucher 已提交
454 455 456 457 458 459 460 461
 *
 * @adev: amdgpu_device pointer
 * @registers: pointer to the register array
 * @array_size: size of the register array
 *
 * Programs an array or registers with and and or masks.
 * This is a helper for setting golden registers.
 */
462 463 464
void amdgpu_device_program_register_sequence(struct amdgpu_device *adev,
					     const u32 *registers,
					     const u32 array_size)
A
Alex Deucher 已提交
465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487
{
	u32 tmp, reg, and_mask, or_mask;
	int i;

	if (array_size % 3)
		return;

	for (i = 0; i < array_size; i +=3) {
		reg = registers[i + 0];
		and_mask = registers[i + 1];
		or_mask = registers[i + 2];

		if (and_mask == 0xffffffff) {
			tmp = or_mask;
		} else {
			tmp = RREG32(reg);
			tmp &= ~and_mask;
			tmp |= or_mask;
		}
		WREG32(reg, tmp);
	}
}

488 489 490 491 492 493 494 495
/**
 * amdgpu_device_pci_config_reset - reset the GPU
 *
 * @adev: amdgpu_device pointer
 *
 * Resets the GPU using the pci config reset sequence.
 * Only applicable to asics prior to vega10.
 */
496
void amdgpu_device_pci_config_reset(struct amdgpu_device *adev)
A
Alex Deucher 已提交
497 498 499 500 501 502 503 504
{
	pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA);
}

/*
 * GPU doorbell aperture helpers function.
 */
/**
505
 * amdgpu_device_doorbell_init - Init doorbell driver information.
A
Alex Deucher 已提交
506 507 508 509 510 511
 *
 * @adev: amdgpu_device pointer
 *
 * Init doorbell driver information (CIK)
 * Returns 0 on success, error on failure.
 */
512
static int amdgpu_device_doorbell_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
513
{
514 515 516 517 518 519 520 521 522
	/* No doorbell on SI hardware generation */
	if (adev->asic_type < CHIP_BONAIRE) {
		adev->doorbell.base = 0;
		adev->doorbell.size = 0;
		adev->doorbell.num_doorbells = 0;
		adev->doorbell.ptr = NULL;
		return 0;
	}

523 524 525
	if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET)
		return -EINVAL;

A
Alex Deucher 已提交
526 527 528 529
	/* doorbell bar mapping */
	adev->doorbell.base = pci_resource_start(adev->pdev, 2);
	adev->doorbell.size = pci_resource_len(adev->pdev, 2);

530
	adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32),
A
Alex Deucher 已提交
531 532 533 534
					     AMDGPU_DOORBELL_MAX_ASSIGNMENT+1);
	if (adev->doorbell.num_doorbells == 0)
		return -EINVAL;

535 536 537 538
	adev->doorbell.ptr = ioremap(adev->doorbell.base,
				     adev->doorbell.num_doorbells *
				     sizeof(u32));
	if (adev->doorbell.ptr == NULL)
A
Alex Deucher 已提交
539 540 541 542 543 544
		return -ENOMEM;

	return 0;
}

/**
545
 * amdgpu_device_doorbell_fini - Tear down doorbell driver information.
A
Alex Deucher 已提交
546 547 548 549 550
 *
 * @adev: amdgpu_device pointer
 *
 * Tear down doorbell driver information (CIK)
 */
551
static void amdgpu_device_doorbell_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
552 553 554 555 556
{
	iounmap(adev->doorbell.ptr);
	adev->doorbell.ptr = NULL;
}

557

A
Alex Deucher 已提交
558 559

/*
560
 * amdgpu_device_wb_*()
561
 * Writeback is the method by which the GPU updates special pages in memory
A
Alex Xie 已提交
562
 * with the status of certain GPU events (fences, ring pointers,etc.).
A
Alex Deucher 已提交
563 564 565
 */

/**
566
 * amdgpu_device_wb_fini - Disable Writeback and free memory
A
Alex Deucher 已提交
567 568 569 570 571 572
 *
 * @adev: amdgpu_device pointer
 *
 * Disables Writeback and frees the Writeback memory (all asics).
 * Used at driver shutdown.
 */
573
static void amdgpu_device_wb_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
574 575
{
	if (adev->wb.wb_obj) {
576 577 578
		amdgpu_bo_free_kernel(&adev->wb.wb_obj,
				      &adev->wb.gpu_addr,
				      (void **)&adev->wb.wb);
A
Alex Deucher 已提交
579 580 581 582 583
		adev->wb.wb_obj = NULL;
	}
}

/**
584
 * amdgpu_device_wb_init- Init Writeback driver info and allocate memory
A
Alex Deucher 已提交
585 586 587
 *
 * @adev: amdgpu_device pointer
 *
588
 * Initializes writeback and allocates writeback memory (all asics).
A
Alex Deucher 已提交
589 590 591
 * Used at driver startup.
 * Returns 0 on success or an -error on failure.
 */
592
static int amdgpu_device_wb_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
593 594 595 596
{
	int r;

	if (adev->wb.wb_obj == NULL) {
597 598
		/* AMDGPU_MAX_WB * sizeof(uint32_t) * 8 = AMDGPU_MAX_WB 256bit slots */
		r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t) * 8,
599 600 601
					    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
					    &adev->wb.wb_obj, &adev->wb.gpu_addr,
					    (void **)&adev->wb.wb);
A
Alex Deucher 已提交
602 603 604 605 606 607 608 609 610
		if (r) {
			dev_warn(adev->dev, "(%d) create WB bo failed\n", r);
			return r;
		}

		adev->wb.num_wb = AMDGPU_MAX_WB;
		memset(&adev->wb.used, 0, sizeof(adev->wb.used));

		/* clear wb memory */
M
Monk Liu 已提交
611
		memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t) * 8);
A
Alex Deucher 已提交
612 613 614 615 616 617
	}

	return 0;
}

/**
618
 * amdgpu_device_wb_get - Allocate a wb entry
A
Alex Deucher 已提交
619 620 621 622 623 624 625
 *
 * @adev: amdgpu_device pointer
 * @wb: wb index
 *
 * Allocate a wb slot for use by the driver (all asics).
 * Returns 0 on success or -EINVAL on failure.
 */
626
int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb)
A
Alex Deucher 已提交
627 628 629
{
	unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb);

630
	if (offset < adev->wb.num_wb) {
K
Ken Wang 已提交
631
		__set_bit(offset, adev->wb.used);
M
Monk Liu 已提交
632
		*wb = offset << 3; /* convert to dw offset */
633 634 635 636 637 638
		return 0;
	} else {
		return -EINVAL;
	}
}

A
Alex Deucher 已提交
639
/**
640
 * amdgpu_device_wb_free - Free a wb entry
A
Alex Deucher 已提交
641 642 643 644 645 646
 *
 * @adev: amdgpu_device pointer
 * @wb: wb index
 *
 * Free a wb slot allocated for use by the driver (all asics)
 */
647
void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb)
A
Alex Deucher 已提交
648
{
M
Monk Liu 已提交
649
	wb >>= 3;
A
Alex Deucher 已提交
650
	if (wb < adev->wb.num_wb)
M
Monk Liu 已提交
651
		__clear_bit(wb, adev->wb.used);
A
Alex Deucher 已提交
652 653 654
}

/**
655
 * amdgpu_device_vram_location - try to find VRAM location
656
 *
A
Alex Deucher 已提交
657 658 659 660
 * @adev: amdgpu device structure holding all necessary informations
 * @mc: memory controller structure holding memory informations
 * @base: base address at which to put VRAM
 *
661
 * Function will try to place VRAM at base address provided
662
 * as parameter.
A
Alex Deucher 已提交
663
 */
664
void amdgpu_device_vram_location(struct amdgpu_device *adev,
665
				 struct amdgpu_gmc *mc, u64 base)
A
Alex Deucher 已提交
666 667 668 669 670 671 672 673 674 675 676 677 678
{
	uint64_t limit = (uint64_t)amdgpu_vram_limit << 20;

	mc->vram_start = base;
	mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
	if (limit && limit < mc->real_vram_size)
		mc->real_vram_size = limit;
	dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
			mc->mc_vram_size >> 20, mc->vram_start,
			mc->vram_end, mc->real_vram_size >> 20);
}

/**
679
 * amdgpu_device_gart_location - try to find GART location
680
 *
A
Alex Deucher 已提交
681 682 683
 * @adev: amdgpu device structure holding all necessary informations
 * @mc: memory controller structure holding memory informations
 *
684
 * Function will place try to place GART before or after VRAM.
A
Alex Deucher 已提交
685
 *
686
 * If GART size is bigger than space left then we ajust GART size.
A
Alex Deucher 已提交
687 688
 * Thus function will never fails.
 */
689
void amdgpu_device_gart_location(struct amdgpu_device *adev,
690
				 struct amdgpu_gmc *mc)
A
Alex Deucher 已提交
691 692 693
{
	u64 size_af, size_bf;

694 695
	mc->gart_size += adev->pm.smu_prv_buffer_size;

696
	size_af = adev->gmc.mc_mask - mc->vram_end;
697
	size_bf = mc->vram_start;
A
Alex Deucher 已提交
698
	if (size_bf > size_af) {
699
		if (mc->gart_size > size_bf) {
700
			dev_warn(adev->dev, "limiting GART\n");
701
			mc->gart_size = size_bf;
A
Alex Deucher 已提交
702
		}
703
		mc->gart_start = 0;
A
Alex Deucher 已提交
704
	} else {
705
		if (mc->gart_size > size_af) {
706
			dev_warn(adev->dev, "limiting GART\n");
707
			mc->gart_size = size_af;
A
Alex Deucher 已提交
708
		}
709 710 711 712
		/* VCE doesn't like it when BOs cross a 4GB segment, so align
		 * the GART base on a 4GB boundary as well.
		 */
		mc->gart_start = ALIGN(mc->vram_end + 1, 0x100000000ULL);
A
Alex Deucher 已提交
713
	}
714
	mc->gart_end = mc->gart_start + mc->gart_size - 1;
715
	dev_info(adev->dev, "GART: %lluM 0x%016llX - 0x%016llX\n",
716
			mc->gart_size >> 20, mc->gart_start, mc->gart_end);
A
Alex Deucher 已提交
717 718
}

719 720 721 722 723 724 725 726 727 728 729
/**
 * amdgpu_device_resize_fb_bar - try to resize FB BAR
 *
 * @adev: amdgpu_device pointer
 *
 * Try to resize FB BAR to make all VRAM CPU accessible. We try very hard not
 * to fail, but if any of the BARs is not accessible after the size we abort
 * driver loading by returning -ENODEV.
 */
int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev)
{
730
	u64 space_needed = roundup_pow_of_two(adev->gmc.real_vram_size);
731
	u32 rbar_size = order_base_2(((space_needed >> 20) | 1)) - 1;
732 733 734
	struct pci_bus *root;
	struct resource *res;
	unsigned i;
735 736 737
	u16 cmd;
	int r;

738 739 740 741
	/* Bypass for VF */
	if (amdgpu_sriov_vf(adev))
		return 0;

742 743 744 745 746 747
	/* Check if the root BUS has 64bit memory resources */
	root = adev->pdev->bus;
	while (root->parent)
		root = root->parent;

	pci_bus_for_each_resource(root, res, i) {
748
		if (res && res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) &&
749 750 751 752 753 754 755 756
		    res->start > 0x100000000ull)
			break;
	}

	/* Trying to resize is pointless without a root hub window above 4GB */
	if (!res)
		return 0;

757 758 759 760 761 762
	/* Disable memory decoding while we change the BAR addresses and size */
	pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd);
	pci_write_config_word(adev->pdev, PCI_COMMAND,
			      cmd & ~PCI_COMMAND_MEMORY);

	/* Free the VRAM and doorbell BAR, we most likely need to move both. */
763
	amdgpu_device_doorbell_fini(adev);
764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779
	if (adev->asic_type >= CHIP_BONAIRE)
		pci_release_resource(adev->pdev, 2);

	pci_release_resource(adev->pdev, 0);

	r = pci_resize_resource(adev->pdev, 0, rbar_size);
	if (r == -ENOSPC)
		DRM_INFO("Not enough PCI address space for a large BAR.");
	else if (r && r != -ENOTSUPP)
		DRM_ERROR("Problem resizing BAR0 (%d).", r);

	pci_assign_unassigned_bus_resources(adev->pdev->bus);

	/* When the doorbell or fb BAR isn't available we have no chance of
	 * using the device.
	 */
780
	r = amdgpu_device_doorbell_init(adev);
781 782 783 784 785 786 787
	if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET))
		return -ENODEV;

	pci_write_config_word(adev->pdev, PCI_COMMAND, cmd);

	return 0;
}
788

A
Alex Deucher 已提交
789 790 791 792
/*
 * GPU helpers function.
 */
/**
A
Alex Deucher 已提交
793
 * amdgpu_device_need_post - check if the hw need post or not
A
Alex Deucher 已提交
794 795 796
 *
 * @adev: amdgpu_device pointer
 *
797 798 799
 * Check if the asic has been initialized (all asics) at driver startup
 * or post is needed if  hw reset is performed.
 * Returns true if need or false if not.
A
Alex Deucher 已提交
800
 */
A
Alex Deucher 已提交
801
bool amdgpu_device_need_post(struct amdgpu_device *adev)
A
Alex Deucher 已提交
802 803 804
{
	uint32_t reg;

805 806 807 808
	if (amdgpu_sriov_vf(adev))
		return false;

	if (amdgpu_passthrough(adev)) {
M
Monk Liu 已提交
809 810 811 812
		/* for FIJI: In whole GPU pass-through virtualization case, after VM reboot
		 * some old smc fw still need driver do vPost otherwise gpu hang, while
		 * those smc fw version above 22.15 doesn't have this flaw, so we force
		 * vpost executed for smc version below 22.15
813 814 815 816 817 818 819 820 821 822
		 */
		if (adev->asic_type == CHIP_FIJI) {
			int err;
			uint32_t fw_ver;
			err = request_firmware(&adev->pm.fw, "amdgpu/fiji_smc.bin", adev->dev);
			/* force vPost if error occured */
			if (err)
				return true;

			fw_ver = *((uint32_t *)adev->pm.fw->data + 69);
M
Monk Liu 已提交
823 824
			if (fw_ver < 0x00160e00)
				return true;
825 826
		}
	}
827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843

	if (adev->has_hw_reset) {
		adev->has_hw_reset = false;
		return true;
	}

	/* bios scratch used on CIK+ */
	if (adev->asic_type >= CHIP_BONAIRE)
		return amdgpu_atombios_scratch_need_asic_init(adev);

	/* check MEM_SIZE for older asics */
	reg = amdgpu_asic_get_config_memsize(adev);

	if ((reg != 0) && (reg != 0xffffffff))
		return false;

	return true;
844 845
}

A
Alex Deucher 已提交
846 847
/* if we get transitioned to only one device, take VGA back */
/**
848
 * amdgpu_device_vga_set_decode - enable/disable vga decode
A
Alex Deucher 已提交
849 850 851 852 853 854 855
 *
 * @cookie: amdgpu_device pointer
 * @state: enable/disable vga decode
 *
 * Enable/disable vga decode (all asics).
 * Returns VGA resource flags.
 */
856
static unsigned int amdgpu_device_vga_set_decode(void *cookie, bool state)
A
Alex Deucher 已提交
857 858 859 860 861 862 863 864 865 866
{
	struct amdgpu_device *adev = cookie;
	amdgpu_asic_set_vga_state(adev, state);
	if (state)
		return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM |
		       VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
	else
		return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
}

867 868 869 870 871 872 873 874 875 876
/**
 * amdgpu_device_check_block_size - validate the vm block size
 *
 * @adev: amdgpu_device pointer
 *
 * Validates the vm block size specified via module parameter.
 * The vm block size defines number of bits in page table versus page directory,
 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
 * page table and the remaining bits are in the page directory.
 */
877
static void amdgpu_device_check_block_size(struct amdgpu_device *adev)
878 879 880 881
{
	/* defines number of bits in page table versus page directory,
	 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
	 * page table and the remaining bits are in the page directory */
882 883
	if (amdgpu_vm_block_size == -1)
		return;
884

885
	if (amdgpu_vm_block_size < 9) {
886 887
		dev_warn(adev->dev, "VM page table size (%d) too small\n",
			 amdgpu_vm_block_size);
888
		amdgpu_vm_block_size = -1;
889 890 891
	}
}

892 893 894 895 896 897 898 899
/**
 * amdgpu_device_check_vm_size - validate the vm size
 *
 * @adev: amdgpu_device pointer
 *
 * Validates the vm size in GB specified via module parameter.
 * The VM size is the size of the GPU virtual memory space in GB.
 */
900
static void amdgpu_device_check_vm_size(struct amdgpu_device *adev)
901
{
902 903 904 905
	/* no need to check the default value */
	if (amdgpu_vm_size == -1)
		return;

906 907 908
	if (amdgpu_vm_size < 1) {
		dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n",
			 amdgpu_vm_size);
909
		amdgpu_vm_size = -1;
910 911 912
	}
}

913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952
static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev)
{
	struct sysinfo si;
	bool is_os_64 = (sizeof(void *) == 8) ? true : false;
	uint64_t total_memory;
	uint64_t dram_size_seven_GB = 0x1B8000000;
	uint64_t dram_size_three_GB = 0xB8000000;

	if (amdgpu_smu_memory_pool_size == 0)
		return;

	if (!is_os_64) {
		DRM_WARN("Not 64-bit OS, feature not supported\n");
		goto def_value;
	}
	si_meminfo(&si);
	total_memory = (uint64_t)si.totalram * si.mem_unit;

	if ((amdgpu_smu_memory_pool_size == 1) ||
		(amdgpu_smu_memory_pool_size == 2)) {
		if (total_memory < dram_size_three_GB)
			goto def_value1;
	} else if ((amdgpu_smu_memory_pool_size == 4) ||
		(amdgpu_smu_memory_pool_size == 8)) {
		if (total_memory < dram_size_seven_GB)
			goto def_value1;
	} else {
		DRM_WARN("Smu memory pool size not supported\n");
		goto def_value;
	}
	adev->pm.smu_prv_buffer_size = amdgpu_smu_memory_pool_size << 28;

	return;

def_value1:
	DRM_WARN("No enough system memory\n");
def_value:
	adev->pm.smu_prv_buffer_size = 0;
}

A
Alex Deucher 已提交
953
/**
954
 * amdgpu_device_check_arguments - validate module params
A
Alex Deucher 已提交
955 956 957 958 959 960
 *
 * @adev: amdgpu_device pointer
 *
 * Validates certain module parameters and updates
 * the associated values used by the driver (all asics).
 */
961
static void amdgpu_device_check_arguments(struct amdgpu_device *adev)
A
Alex Deucher 已提交
962
{
963 964 965 966
	if (amdgpu_sched_jobs < 4) {
		dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n",
			 amdgpu_sched_jobs);
		amdgpu_sched_jobs = 4;
967
	} else if (!is_power_of_2(amdgpu_sched_jobs)){
968 969 970 971
		dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n",
			 amdgpu_sched_jobs);
		amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs);
	}
A
Alex Deucher 已提交
972

973
	if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) {
974 975 976
		/* gart size must be greater or equal to 32M */
		dev_warn(adev->dev, "gart size (%d) too small\n",
			 amdgpu_gart_size);
977
		amdgpu_gart_size = -1;
A
Alex Deucher 已提交
978 979
	}

980
	if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) {
981
		/* gtt size must be greater or equal to 32M */
982 983 984
		dev_warn(adev->dev, "gtt size (%d) too small\n",
				 amdgpu_gtt_size);
		amdgpu_gtt_size = -1;
A
Alex Deucher 已提交
985 986
	}

987 988 989 990 991 992 993
	/* valid range is between 4 and 9 inclusive */
	if (amdgpu_vm_fragment_size != -1 &&
	    (amdgpu_vm_fragment_size > 9 || amdgpu_vm_fragment_size < 4)) {
		dev_warn(adev->dev, "valid range is between 4 and 9\n");
		amdgpu_vm_fragment_size = -1;
	}

994 995
	amdgpu_device_check_smu_prv_buffer_size(adev);

996
	amdgpu_device_check_vm_size(adev);
A
Alex Deucher 已提交
997

998
	amdgpu_device_check_block_size(adev);
C
Christian König 已提交
999

1000
	if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 ||
1001
	    !is_power_of_2(amdgpu_vram_page_split))) {
C
Christian König 已提交
1002 1003 1004 1005
		dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
			 amdgpu_vram_page_split);
		amdgpu_vram_page_split = 1024;
	}
1006 1007 1008 1009 1010

	if (amdgpu_lockup_timeout == 0) {
		dev_warn(adev->dev, "lockup_timeout msut be > 0, adjusting to 10000\n");
		amdgpu_lockup_timeout = 10000;
	}
1011 1012

	adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type);
A
Alex Deucher 已提交
1013 1014 1015 1016 1017 1018
}

/**
 * amdgpu_switcheroo_set_state - set switcheroo state
 *
 * @pdev: pci dev pointer
1019
 * @state: vga_switcheroo state
A
Alex Deucher 已提交
1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031
 *
 * Callback for the switcheroo driver.  Suspends or resumes the
 * the asics before or after it is powered up using ACPI methods.
 */
static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_state state)
{
	struct drm_device *dev = pci_get_drvdata(pdev);

	if (amdgpu_device_is_px(dev) && state == VGA_SWITCHEROO_OFF)
		return;

	if (state == VGA_SWITCHEROO_ON) {
1032
		pr_info("amdgpu: switched on\n");
A
Alex Deucher 已提交
1033 1034 1035
		/* don't suspend or resume card normally */
		dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;

1036
		amdgpu_device_resume(dev, true, true);
A
Alex Deucher 已提交
1037 1038 1039 1040

		dev->switch_power_state = DRM_SWITCH_POWER_ON;
		drm_kms_helper_poll_enable(dev);
	} else {
1041
		pr_info("amdgpu: switched off\n");
A
Alex Deucher 已提交
1042 1043
		drm_kms_helper_poll_disable(dev);
		dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
1044
		amdgpu_device_suspend(dev, true, true);
A
Alex Deucher 已提交
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
		dev->switch_power_state = DRM_SWITCH_POWER_OFF;
	}
}

/**
 * amdgpu_switcheroo_can_switch - see if switcheroo state can change
 *
 * @pdev: pci dev pointer
 *
 * Callback for the switcheroo driver.  Check of the switcheroo
 * state can be changed.
 * Returns true if the state can be changed, false if not.
 */
static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev)
{
	struct drm_device *dev = pci_get_drvdata(pdev);

	/*
	* FIXME: open_count is protected by drm_global_mutex but that would lead to
	* locking inversion with the driver load path. And the access here is
	* completely racy anyway. So don't bother with locking for now.
	*/
	return dev->open_count == 0;
}

static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = {
	.set_gpu_state = amdgpu_switcheroo_set_state,
	.reprobe = NULL,
	.can_switch = amdgpu_switcheroo_can_switch,
};

1076 1077 1078
/**
 * amdgpu_device_ip_set_clockgating_state - set the CG state
 *
1079
 * @dev: amdgpu_device pointer
1080 1081 1082 1083 1084 1085 1086
 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
 * @state: clockgating state (gate or ungate)
 *
 * Sets the requested clockgating state for all instances of
 * the hardware IP specified.
 * Returns the error code from the last instance.
 */
1087
int amdgpu_device_ip_set_clockgating_state(void *dev,
1088 1089
					   enum amd_ip_block_type block_type,
					   enum amd_clockgating_state state)
A
Alex Deucher 已提交
1090
{
1091
	struct amdgpu_device *adev = dev;
A
Alex Deucher 已提交
1092 1093 1094
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1095
		if (!adev->ip_blocks[i].status.valid)
1096
			continue;
1097 1098 1099 1100 1101 1102 1103 1104 1105
		if (adev->ip_blocks[i].version->type != block_type)
			continue;
		if (!adev->ip_blocks[i].version->funcs->set_clockgating_state)
			continue;
		r = adev->ip_blocks[i].version->funcs->set_clockgating_state(
			(void *)adev, state);
		if (r)
			DRM_ERROR("set_clockgating_state of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
A
Alex Deucher 已提交
1106 1107 1108 1109
	}
	return r;
}

1110 1111 1112
/**
 * amdgpu_device_ip_set_powergating_state - set the PG state
 *
1113
 * @dev: amdgpu_device pointer
1114 1115 1116 1117 1118 1119 1120
 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
 * @state: powergating state (gate or ungate)
 *
 * Sets the requested powergating state for all instances of
 * the hardware IP specified.
 * Returns the error code from the last instance.
 */
1121
int amdgpu_device_ip_set_powergating_state(void *dev,
1122 1123
					   enum amd_ip_block_type block_type,
					   enum amd_powergating_state state)
A
Alex Deucher 已提交
1124
{
1125
	struct amdgpu_device *adev = dev;
A
Alex Deucher 已提交
1126 1127 1128
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1129
		if (!adev->ip_blocks[i].status.valid)
1130
			continue;
1131 1132 1133 1134 1135 1136 1137 1138 1139
		if (adev->ip_blocks[i].version->type != block_type)
			continue;
		if (!adev->ip_blocks[i].version->funcs->set_powergating_state)
			continue;
		r = adev->ip_blocks[i].version->funcs->set_powergating_state(
			(void *)adev, state);
		if (r)
			DRM_ERROR("set_powergating_state of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
A
Alex Deucher 已提交
1140 1141 1142 1143
	}
	return r;
}

1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
/**
 * amdgpu_device_ip_get_clockgating_state - get the CG state
 *
 * @adev: amdgpu_device pointer
 * @flags: clockgating feature flags
 *
 * Walks the list of IPs on the device and updates the clockgating
 * flags for each IP.
 * Updates @flags with the feature flags for each hardware IP where
 * clockgating is enabled.
 */
1155 1156
void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev,
					    u32 *flags)
1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
{
	int i;

	for (i = 0; i < adev->num_ip_blocks; i++) {
		if (!adev->ip_blocks[i].status.valid)
			continue;
		if (adev->ip_blocks[i].version->funcs->get_clockgating_state)
			adev->ip_blocks[i].version->funcs->get_clockgating_state((void *)adev, flags);
	}
}

1168 1169 1170 1171 1172 1173 1174 1175 1176
/**
 * amdgpu_device_ip_wait_for_idle - wait for idle
 *
 * @adev: amdgpu_device pointer
 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
 *
 * Waits for the request hardware IP to be idle.
 * Returns 0 for success or a negative error code on failure.
 */
1177 1178
int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev,
				   enum amd_ip_block_type block_type)
1179 1180 1181 1182
{
	int i, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1183
		if (!adev->ip_blocks[i].status.valid)
1184
			continue;
1185 1186
		if (adev->ip_blocks[i].version->type == block_type) {
			r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev);
1187 1188 1189 1190 1191 1192 1193 1194 1195
			if (r)
				return r;
			break;
		}
	}
	return 0;

}

1196 1197 1198 1199 1200 1201 1202 1203 1204
/**
 * amdgpu_device_ip_is_idle - is the hardware IP idle
 *
 * @adev: amdgpu_device pointer
 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
 *
 * Check if the hardware IP is idle or not.
 * Returns true if it the IP is idle, false if not.
 */
1205 1206
bool amdgpu_device_ip_is_idle(struct amdgpu_device *adev,
			      enum amd_ip_block_type block_type)
1207 1208 1209 1210
{
	int i;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1211
		if (!adev->ip_blocks[i].status.valid)
1212
			continue;
1213 1214
		if (adev->ip_blocks[i].version->type == block_type)
			return adev->ip_blocks[i].version->funcs->is_idle((void *)adev);
1215 1216 1217 1218 1219
	}
	return true;

}

1220 1221 1222 1223
/**
 * amdgpu_device_ip_get_ip_block - get a hw IP pointer
 *
 * @adev: amdgpu_device pointer
1224
 * @type: Type of hardware IP (SMU, GFX, UVD, etc.)
1225 1226 1227 1228
 *
 * Returns a pointer to the hardware IP block structure
 * if it exists for the asic, otherwise NULL.
 */
1229 1230 1231
struct amdgpu_ip_block *
amdgpu_device_ip_get_ip_block(struct amdgpu_device *adev,
			      enum amd_ip_block_type type)
A
Alex Deucher 已提交
1232 1233 1234 1235
{
	int i;

	for (i = 0; i < adev->num_ip_blocks; i++)
1236
		if (adev->ip_blocks[i].version->type == type)
A
Alex Deucher 已提交
1237 1238 1239 1240 1241 1242
			return &adev->ip_blocks[i];

	return NULL;
}

/**
1243
 * amdgpu_device_ip_block_version_cmp
A
Alex Deucher 已提交
1244 1245
 *
 * @adev: amdgpu_device pointer
1246
 * @type: enum amd_ip_block_type
A
Alex Deucher 已提交
1247 1248 1249 1250 1251 1252
 * @major: major version
 * @minor: minor version
 *
 * return 0 if equal or greater
 * return 1 if smaller or the ip_block doesn't exist
 */
1253 1254 1255
int amdgpu_device_ip_block_version_cmp(struct amdgpu_device *adev,
				       enum amd_ip_block_type type,
				       u32 major, u32 minor)
A
Alex Deucher 已提交
1256
{
1257
	struct amdgpu_ip_block *ip_block = amdgpu_device_ip_get_ip_block(adev, type);
A
Alex Deucher 已提交
1258

1259 1260 1261
	if (ip_block && ((ip_block->version->major > major) ||
			((ip_block->version->major == major) &&
			(ip_block->version->minor >= minor))))
A
Alex Deucher 已提交
1262 1263 1264 1265 1266
		return 0;

	return 1;
}

1267
/**
1268
 * amdgpu_device_ip_block_add
1269 1270 1271 1272 1273 1274 1275
 *
 * @adev: amdgpu_device pointer
 * @ip_block_version: pointer to the IP to add
 *
 * Adds the IP block driver information to the collection of IPs
 * on the asic.
 */
1276 1277
int amdgpu_device_ip_block_add(struct amdgpu_device *adev,
			       const struct amdgpu_ip_block_version *ip_block_version)
1278 1279 1280 1281
{
	if (!ip_block_version)
		return -EINVAL;

1282
	DRM_INFO("add ip block number %d <%s>\n", adev->num_ip_blocks,
1283 1284
		  ip_block_version->funcs->name);

1285 1286 1287 1288 1289
	adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version;

	return 0;
}

1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301
/**
 * amdgpu_device_enable_virtual_display - enable virtual display feature
 *
 * @adev: amdgpu_device pointer
 *
 * Enabled the virtual display feature if the user has enabled it via
 * the module parameter virtual_display.  This feature provides a virtual
 * display hardware on headless boards or in virtualized environments.
 * This function parses and validates the configuration string specified by
 * the user and configues the virtual display configuration (number of
 * virtual connectors, crtcs, etc.) specified.
 */
1302
static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev)
1303 1304 1305 1306 1307 1308
{
	adev->enable_virtual_display = false;

	if (amdgpu_virtual_display) {
		struct drm_device *ddev = adev->ddev;
		const char *pci_address_name = pci_name(ddev->pdev);
1309
		char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname;
1310 1311 1312

		pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL);
		pciaddstr_tmp = pciaddstr;
1313 1314
		while ((pciaddname_tmp = strsep(&pciaddstr_tmp, ";"))) {
			pciaddname = strsep(&pciaddname_tmp, ",");
1315 1316
			if (!strcmp("all", pciaddname)
			    || !strcmp(pci_address_name, pciaddname)) {
1317 1318 1319
				long num_crtc;
				int res = -1;

1320
				adev->enable_virtual_display = true;
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334

				if (pciaddname_tmp)
					res = kstrtol(pciaddname_tmp, 10,
						      &num_crtc);

				if (!res) {
					if (num_crtc < 1)
						num_crtc = 1;
					if (num_crtc > 6)
						num_crtc = 6;
					adev->mode_info.num_crtc = num_crtc;
				} else {
					adev->mode_info.num_crtc = 1;
				}
1335 1336 1337 1338
				break;
			}
		}

1339 1340 1341
		DRM_INFO("virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n",
			 amdgpu_virtual_display, pci_address_name,
			 adev->enable_virtual_display, adev->mode_info.num_crtc);
1342 1343 1344 1345 1346

		kfree(pciaddstr);
	}
}

1347 1348 1349 1350 1351 1352 1353 1354 1355 1356
/**
 * amdgpu_device_parse_gpu_info_fw - parse gpu info firmware
 *
 * @adev: amdgpu_device pointer
 *
 * Parses the asic configuration parameters specified in the gpu info
 * firmware and makes them availale to the driver for use in configuring
 * the asic.
 * Returns 0 on success, -EINVAL on failure.
 */
1357 1358 1359 1360 1361 1362 1363
static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev)
{
	const char *chip_name;
	char fw_name[30];
	int err;
	const struct gpu_info_firmware_header_v1_0 *hdr;

1364 1365
	adev->firmware.gpu_info_fw = NULL;

1366 1367 1368 1369 1370
	switch (adev->asic_type) {
	case CHIP_TOPAZ:
	case CHIP_TONGA:
	case CHIP_FIJI:
	case CHIP_POLARIS10:
1371
	case CHIP_POLARIS11:
1372
	case CHIP_POLARIS12:
1373
	case CHIP_VEGAM:
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389
	case CHIP_CARRIZO:
	case CHIP_STONEY:
#ifdef CONFIG_DRM_AMDGPU_SI
	case CHIP_VERDE:
	case CHIP_TAHITI:
	case CHIP_PITCAIRN:
	case CHIP_OLAND:
	case CHIP_HAINAN:
#endif
#ifdef CONFIG_DRM_AMDGPU_CIK
	case CHIP_BONAIRE:
	case CHIP_HAWAII:
	case CHIP_KAVERI:
	case CHIP_KABINI:
	case CHIP_MULLINS:
#endif
1390
	case CHIP_VEGA20:
1391 1392 1393 1394 1395
	default:
		return 0;
	case CHIP_VEGA10:
		chip_name = "vega10";
		break;
1396 1397 1398
	case CHIP_VEGA12:
		chip_name = "vega12";
		break;
1399 1400 1401
	case CHIP_RAVEN:
		chip_name = "raven";
		break;
1402 1403 1404
	}

	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name);
1405
	err = request_firmware(&adev->firmware.gpu_info_fw, fw_name, adev->dev);
1406 1407 1408 1409 1410 1411
	if (err) {
		dev_err(adev->dev,
			"Failed to load gpu_info firmware \"%s\"\n",
			fw_name);
		goto out;
	}
1412
	err = amdgpu_ucode_validate(adev->firmware.gpu_info_fw);
1413 1414 1415 1416 1417 1418 1419
	if (err) {
		dev_err(adev->dev,
			"Failed to validate gpu_info firmware \"%s\"\n",
			fw_name);
		goto out;
	}

1420
	hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data;
1421 1422 1423 1424 1425 1426
	amdgpu_ucode_print_gpu_info_hdr(&hdr->header);

	switch (hdr->version_major) {
	case 1:
	{
		const struct gpu_info_firmware_v1_0 *gpu_info_fw =
1427
			(const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data +
1428 1429
								le32_to_cpu(hdr->header.ucode_array_offset_bytes));

1430 1431 1432 1433
		adev->gfx.config.max_shader_engines = le32_to_cpu(gpu_info_fw->gc_num_se);
		adev->gfx.config.max_cu_per_sh = le32_to_cpu(gpu_info_fw->gc_num_cu_per_sh);
		adev->gfx.config.max_sh_per_se = le32_to_cpu(gpu_info_fw->gc_num_sh_per_se);
		adev->gfx.config.max_backends_per_se = le32_to_cpu(gpu_info_fw->gc_num_rb_per_se);
1434
		adev->gfx.config.max_texture_channel_caches =
1435 1436 1437 1438 1439
			le32_to_cpu(gpu_info_fw->gc_num_tccs);
		adev->gfx.config.max_gprs = le32_to_cpu(gpu_info_fw->gc_num_gprs);
		adev->gfx.config.max_gs_threads = le32_to_cpu(gpu_info_fw->gc_num_max_gs_thds);
		adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gpu_info_fw->gc_gs_table_depth);
		adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gpu_info_fw->gc_gsprim_buff_depth);
1440
		adev->gfx.config.double_offchip_lds_buf =
1441 1442
			le32_to_cpu(gpu_info_fw->gc_double_offchip_lds_buffer);
		adev->gfx.cu_info.wave_front_size = le32_to_cpu(gpu_info_fw->gc_wave_size);
1443 1444 1445 1446 1447
		adev->gfx.cu_info.max_waves_per_simd =
			le32_to_cpu(gpu_info_fw->gc_max_waves_per_simd);
		adev->gfx.cu_info.max_scratch_slots_per_cu =
			le32_to_cpu(gpu_info_fw->gc_max_scratch_slots_per_cu);
		adev->gfx.cu_info.lds_size = le32_to_cpu(gpu_info_fw->gc_lds_size);
1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459
		break;
	}
	default:
		dev_err(adev->dev,
			"Unsupported gpu_info table %d\n", hdr->header.ucode_version);
		err = -EINVAL;
		goto out;
	}
out:
	return err;
}

1460 1461 1462 1463 1464 1465 1466 1467 1468 1469
/**
 * amdgpu_device_ip_early_init - run early init for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * Early initialization pass for hardware IPs.  The hardware IPs that make
 * up each asic are discovered each IP's early_init callback is run.  This
 * is the first stage in initializing the asic.
 * Returns 0 on success, negative error code on failure.
 */
1470
static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1471
{
1472
	int i, r;
A
Alex Deucher 已提交
1473

1474
	amdgpu_device_enable_virtual_display(adev);
1475

A
Alex Deucher 已提交
1476
	switch (adev->asic_type) {
1477 1478
	case CHIP_TOPAZ:
	case CHIP_TONGA:
1479
	case CHIP_FIJI:
1480
	case CHIP_POLARIS10:
1481
	case CHIP_POLARIS11:
1482
	case CHIP_POLARIS12:
1483
	case CHIP_VEGAM:
1484
	case CHIP_CARRIZO:
1485 1486
	case CHIP_STONEY:
		if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY)
1487 1488 1489 1490 1491 1492 1493 1494
			adev->family = AMDGPU_FAMILY_CZ;
		else
			adev->family = AMDGPU_FAMILY_VI;

		r = vi_set_ip_blocks(adev);
		if (r)
			return r;
		break;
K
Ken Wang 已提交
1495 1496 1497 1498 1499 1500
#ifdef CONFIG_DRM_AMDGPU_SI
	case CHIP_VERDE:
	case CHIP_TAHITI:
	case CHIP_PITCAIRN:
	case CHIP_OLAND:
	case CHIP_HAINAN:
K
Ken Wang 已提交
1501
		adev->family = AMDGPU_FAMILY_SI;
K
Ken Wang 已提交
1502 1503 1504 1505 1506
		r = si_set_ip_blocks(adev);
		if (r)
			return r;
		break;
#endif
1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522
#ifdef CONFIG_DRM_AMDGPU_CIK
	case CHIP_BONAIRE:
	case CHIP_HAWAII:
	case CHIP_KAVERI:
	case CHIP_KABINI:
	case CHIP_MULLINS:
		if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII))
			adev->family = AMDGPU_FAMILY_CI;
		else
			adev->family = AMDGPU_FAMILY_KV;

		r = cik_set_ip_blocks(adev);
		if (r)
			return r;
		break;
#endif
1523 1524
	case CHIP_VEGA10:
	case CHIP_VEGA12:
1525
	case CHIP_VEGA20:
1526
	case CHIP_RAVEN:
1527 1528 1529 1530
		if (adev->asic_type == CHIP_RAVEN)
			adev->family = AMDGPU_FAMILY_RV;
		else
			adev->family = AMDGPU_FAMILY_AI;
1531 1532 1533 1534 1535

		r = soc15_set_ip_blocks(adev);
		if (r)
			return r;
		break;
A
Alex Deucher 已提交
1536 1537 1538 1539 1540
	default:
		/* FIXME: not supported yet */
		return -EINVAL;
	}

1541 1542 1543 1544
	r = amdgpu_device_parse_gpu_info_fw(adev);
	if (r)
		return r;

1545 1546
	amdgpu_amdkfd_device_probe(adev);

1547 1548 1549
	if (amdgpu_sriov_vf(adev)) {
		r = amdgpu_virt_request_full_gpu(adev, true);
		if (r)
1550
			return -EAGAIN;
1551 1552
	}

1553 1554
	adev->powerplay.pp_feature = amdgpu_pp_feature_mask;

A
Alex Deucher 已提交
1555 1556
	for (i = 0; i < adev->num_ip_blocks; i++) {
		if ((amdgpu_ip_block_mask & (1 << i)) == 0) {
1557 1558
			DRM_ERROR("disabled ip block: %d <%s>\n",
				  i, adev->ip_blocks[i].version->funcs->name);
1559
			adev->ip_blocks[i].status.valid = false;
A
Alex Deucher 已提交
1560
		} else {
1561 1562
			if (adev->ip_blocks[i].version->funcs->early_init) {
				r = adev->ip_blocks[i].version->funcs->early_init((void *)adev);
1563
				if (r == -ENOENT) {
1564
					adev->ip_blocks[i].status.valid = false;
1565
				} else if (r) {
1566 1567
					DRM_ERROR("early_init of IP block <%s> failed %d\n",
						  adev->ip_blocks[i].version->funcs->name, r);
A
Alex Deucher 已提交
1568
					return r;
1569
				} else {
1570
					adev->ip_blocks[i].status.valid = true;
1571
				}
1572
			} else {
1573
				adev->ip_blocks[i].status.valid = true;
A
Alex Deucher 已提交
1574 1575 1576 1577
			}
		}
	}

1578 1579 1580
	adev->cg_flags &= amdgpu_cg_mask;
	adev->pg_flags &= amdgpu_pg_mask;

A
Alex Deucher 已提交
1581 1582 1583
	return 0;
}

1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594
/**
 * amdgpu_device_ip_init - run init for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * Main initialization pass for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked and the sw_init and hw_init callbacks
 * are run.  sw_init initializes the software state associated with each IP
 * and hw_init initializes the hardware associated with each IP.
 * Returns 0 on success, negative error code on failure.
 */
1595
static int amdgpu_device_ip_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1596 1597 1598 1599
{
	int i, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1600
		if (!adev->ip_blocks[i].status.valid)
A
Alex Deucher 已提交
1601
			continue;
1602
		r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev);
1603
		if (r) {
1604 1605
			DRM_ERROR("sw_init of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
A
Alex Deucher 已提交
1606
			return r;
1607
		}
1608
		adev->ip_blocks[i].status.sw = true;
1609

A
Alex Deucher 已提交
1610
		/* need to do gmc hw init early so we can allocate gpu mem */
1611
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
1612
			r = amdgpu_device_vram_scratch_init(adev);
1613 1614
			if (r) {
				DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r);
A
Alex Deucher 已提交
1615
				return r;
1616
			}
1617
			r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev);
1618 1619
			if (r) {
				DRM_ERROR("hw_init %d failed %d\n", i, r);
A
Alex Deucher 已提交
1620
				return r;
1621
			}
1622
			r = amdgpu_device_wb_init(adev);
1623
			if (r) {
1624
				DRM_ERROR("amdgpu_device_wb_init failed %d\n", r);
A
Alex Deucher 已提交
1625
				return r;
1626
			}
1627
			adev->ip_blocks[i].status.hw = true;
M
Monk Liu 已提交
1628 1629 1630 1631 1632 1633 1634 1635 1636

			/* right after GMC hw init, we create CSA */
			if (amdgpu_sriov_vf(adev)) {
				r = amdgpu_allocate_static_csa(adev);
				if (r) {
					DRM_ERROR("allocate CSA failed %d\n", r);
					return r;
				}
			}
A
Alex Deucher 已提交
1637 1638 1639 1640
		}
	}

	for (i = 0; i < adev->num_ip_blocks; i++) {
1641
		if (!adev->ip_blocks[i].status.sw)
A
Alex Deucher 已提交
1642
			continue;
1643
		if (adev->ip_blocks[i].status.hw)
A
Alex Deucher 已提交
1644
			continue;
1645
		r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev);
1646
		if (r) {
1647 1648
			DRM_ERROR("hw_init of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
A
Alex Deucher 已提交
1649
			return r;
1650
		}
1651
		adev->ip_blocks[i].status.hw = true;
A
Alex Deucher 已提交
1652 1653
	}

1654
	amdgpu_amdkfd_device_init(adev);
1655 1656 1657 1658

	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_release_full_gpu(adev, true);

A
Alex Deucher 已提交
1659 1660 1661
	return 0;
}

1662 1663 1664 1665 1666 1667 1668 1669 1670
/**
 * amdgpu_device_fill_reset_magic - writes reset magic to gart pointer
 *
 * @adev: amdgpu_device pointer
 *
 * Writes a reset magic value to the gart pointer in VRAM.  The driver calls
 * this function before a GPU reset.  If the value is retained after a
 * GPU reset, VRAM has not been lost.  Some GPU resets may destry VRAM contents.
 */
1671
static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev)
1672 1673 1674 1675
{
	memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM);
}

1676 1677 1678 1679 1680 1681 1682 1683 1684 1685
/**
 * amdgpu_device_check_vram_lost - check if vram is valid
 *
 * @adev: amdgpu_device pointer
 *
 * Checks the reset magic value written to the gart pointer in VRAM.
 * The driver calls this after a GPU reset to see if the contents of
 * VRAM is lost or now.
 * returns true if vram is lost, false if not.
 */
1686
static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev)
1687 1688 1689 1690 1691
{
	return !!memcmp(adev->gart.ptr, adev->reset_magic,
			AMDGPU_RESET_MAGIC_NUM);
}

1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702
/**
 * amdgpu_device_ip_late_set_cg_state - late init for clockgating
 *
 * @adev: amdgpu_device pointer
 *
 * Late initialization pass enabling clockgating for hardware IPs.
 * The list of all the hardware IPs that make up the asic is walked and the
 * set_clockgating_state callbacks are run.  This stage is run late
 * in the init process.
 * Returns 0 on success, negative error code on failure.
 */
1703
static int amdgpu_device_ip_late_set_cg_state(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1704 1705 1706
{
	int i = 0, r;

1707 1708 1709
	if (amdgpu_emu_mode == 1)
		return 0;

A
Alex Deucher 已提交
1710
	for (i = 0; i < adev->num_ip_blocks; i++) {
1711
		if (!adev->ip_blocks[i].status.valid)
A
Alex Deucher 已提交
1712
			continue;
1713
		/* skip CG for VCE/UVD, it's handled specially */
1714
		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
1715
		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
1716
		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
1717
		    adev->ip_blocks[i].version->funcs->set_clockgating_state) {
1718
			/* enable clockgating to save power */
1719 1720
			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
										     AMD_CG_STATE_GATE);
1721 1722
			if (r) {
				DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n",
1723
					  adev->ip_blocks[i].version->funcs->name, r);
1724 1725
				return r;
			}
1726
		}
A
Alex Deucher 已提交
1727
	}
1728

1729 1730 1731 1732 1733 1734
	return 0;
}

static int amdgpu_device_ip_late_set_pg_state(struct amdgpu_device *adev)
{
	int i = 0, r;
1735

1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756
	if (amdgpu_emu_mode == 1)
		return 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
		if (!adev->ip_blocks[i].status.valid)
			continue;
		/* skip CG for VCE/UVD, it's handled specially */
		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
		    adev->ip_blocks[i].version->funcs->set_powergating_state) {
			/* enable powergating to save power */
			r = adev->ip_blocks[i].version->funcs->set_powergating_state((void *)adev,
										     AMD_PG_STATE_GATE);
			if (r) {
				DRM_ERROR("set_powergating_state(gate) of IP block <%s> failed %d\n",
					  adev->ip_blocks[i].version->funcs->name, r);
				return r;
			}
		}
	}
1757 1758 1759
	return 0;
}

1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771
/**
 * amdgpu_device_ip_late_init - run late init for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * Late initialization pass for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked and the late_init callbacks are run.
 * late_init covers any special initialization that an IP requires
 * after all of the have been initialized or something that needs to happen
 * late in the init process.
 * Returns 0 on success, negative error code on failure.
 */
1772
static int amdgpu_device_ip_late_init(struct amdgpu_device *adev)
1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789
{
	int i = 0, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
		if (!adev->ip_blocks[i].status.valid)
			continue;
		if (adev->ip_blocks[i].version->funcs->late_init) {
			r = adev->ip_blocks[i].version->funcs->late_init((void *)adev);
			if (r) {
				DRM_ERROR("late_init of IP block <%s> failed %d\n",
					  adev->ip_blocks[i].version->funcs->name, r);
				return r;
			}
			adev->ip_blocks[i].status.late_initialized = true;
		}
	}

1790 1791 1792
	amdgpu_device_ip_late_set_cg_state(adev);
	amdgpu_device_ip_late_set_pg_state(adev);

1793 1794
	queue_delayed_work(system_wq, &adev->late_init_work,
			   msecs_to_jiffies(AMDGPU_RESUME_MS));
A
Alex Deucher 已提交
1795

1796
	amdgpu_device_fill_reset_magic(adev);
A
Alex Deucher 已提交
1797 1798 1799 1800

	return 0;
}

1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811
/**
 * amdgpu_device_ip_fini - run fini for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * Main teardown pass for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked and the hw_fini and sw_fini callbacks
 * are run.  hw_fini tears down the hardware associated with each IP
 * and sw_fini tears down any software state associated with each IP.
 * Returns 0 on success, negative error code on failure.
 */
1812
static int amdgpu_device_ip_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1813 1814 1815
{
	int i, r;

1816
	amdgpu_amdkfd_device_fini(adev);
1817 1818
	/* need to disable SMC first */
	for (i = 0; i < adev->num_ip_blocks; i++) {
1819
		if (!adev->ip_blocks[i].status.hw)
1820
			continue;
1821 1822
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC &&
			adev->ip_blocks[i].version->funcs->set_clockgating_state) {
1823
			/* ungate blocks before hw fini so that we can shutdown the blocks safely */
1824 1825
			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
										     AMD_CG_STATE_UNGATE);
1826 1827
			if (r) {
				DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n",
1828
					  adev->ip_blocks[i].version->funcs->name, r);
1829 1830
				return r;
			}
1831 1832
			if (adev->powerplay.pp_funcs->set_powergating_by_smu)
				amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, false);
1833
			r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
1834 1835 1836
			/* XXX handle errors */
			if (r) {
				DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
1837
					  adev->ip_blocks[i].version->funcs->name, r);
1838
			}
1839
			adev->ip_blocks[i].status.hw = false;
1840 1841 1842 1843
			break;
		}
	}

A
Alex Deucher 已提交
1844
	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
1845
		if (!adev->ip_blocks[i].status.hw)
A
Alex Deucher 已提交
1846
			continue;
1847 1848

		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
1849
			adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
1850
			adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
1851
			adev->ip_blocks[i].version->funcs->set_clockgating_state) {
1852 1853 1854 1855 1856 1857 1858 1859
			/* ungate blocks before hw fini so that we can shutdown the blocks safely */
			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
										     AMD_CG_STATE_UNGATE);
			if (r) {
				DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n",
					  adev->ip_blocks[i].version->funcs->name, r);
				return r;
			}
1860
		}
1861

1862
		r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
A
Alex Deucher 已提交
1863
		/* XXX handle errors */
1864
		if (r) {
1865 1866
			DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
1867
		}
1868

1869
		adev->ip_blocks[i].status.hw = false;
A
Alex Deucher 已提交
1870 1871
	}

1872

A
Alex Deucher 已提交
1873
	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
1874
		if (!adev->ip_blocks[i].status.sw)
A
Alex Deucher 已提交
1875
			continue;
1876 1877 1878 1879 1880 1881 1882

		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
			amdgpu_free_static_csa(adev);
			amdgpu_device_wb_fini(adev);
			amdgpu_device_vram_scratch_fini(adev);
		}

1883
		r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev);
A
Alex Deucher 已提交
1884
		/* XXX handle errors */
1885
		if (r) {
1886 1887
			DRM_DEBUG("sw_fini of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
1888
		}
1889 1890
		adev->ip_blocks[i].status.sw = false;
		adev->ip_blocks[i].status.valid = false;
A
Alex Deucher 已提交
1891 1892
	}

M
Monk Liu 已提交
1893
	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
1894
		if (!adev->ip_blocks[i].status.late_initialized)
1895
			continue;
1896 1897 1898
		if (adev->ip_blocks[i].version->funcs->late_fini)
			adev->ip_blocks[i].version->funcs->late_fini((void *)adev);
		adev->ip_blocks[i].status.late_initialized = false;
M
Monk Liu 已提交
1899 1900
	}

1901
	if (amdgpu_sriov_vf(adev))
1902 1903
		if (amdgpu_virt_release_full_gpu(adev, false))
			DRM_ERROR("failed to release exclusive mode on fini\n");
M
Monk Liu 已提交
1904

A
Alex Deucher 已提交
1905 1906 1907
	return 0;
}

1908 1909 1910 1911 1912 1913 1914 1915 1916
/**
 * amdgpu_device_ip_late_init_func_handler - work handler for clockgating
 *
 * @work: work_struct
 *
 * Work handler for amdgpu_device_ip_late_set_cg_state.  We put the
 * clockgating setup into a worker thread to speed up driver init and
 * resume from suspend.
 */
1917
static void amdgpu_device_ip_late_init_func_handler(struct work_struct *work)
1918 1919 1920
{
	struct amdgpu_device *adev =
		container_of(work, struct amdgpu_device, late_init_work.work);
1921 1922 1923 1924 1925
	int r;

	r = amdgpu_ib_ring_tests(adev);
	if (r)
		DRM_ERROR("ib ring test failed (%d).\n", r);
1926 1927
}

1928
/**
1929
 * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1)
1930 1931 1932 1933 1934 1935 1936 1937 1938
 *
 * @adev: amdgpu_device pointer
 *
 * Main suspend function for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked, clockgating is disabled and the
 * suspend callbacks are run.  suspend puts the hardware and software state
 * in each IP into a state suitable for suspend.
 * Returns 0 on success, negative error code on failure.
 */
1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987
static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev)
{
	int i, r;

	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_request_full_gpu(adev, false);

	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
		if (!adev->ip_blocks[i].status.valid)
			continue;
		/* displays are handled separately */
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) {
			/* ungate blocks so that suspend can properly shut them down */
			if (adev->ip_blocks[i].version->funcs->set_clockgating_state) {
				r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
											     AMD_CG_STATE_UNGATE);
				if (r) {
					DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n",
						  adev->ip_blocks[i].version->funcs->name, r);
				}
			}
			/* XXX handle errors */
			r = adev->ip_blocks[i].version->funcs->suspend(adev);
			/* XXX handle errors */
			if (r) {
				DRM_ERROR("suspend of IP block <%s> failed %d\n",
					  adev->ip_blocks[i].version->funcs->name, r);
			}
		}
	}

	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_release_full_gpu(adev, false);

	return 0;
}

/**
 * amdgpu_device_ip_suspend_phase2 - run suspend for hardware IPs (phase 2)
 *
 * @adev: amdgpu_device pointer
 *
 * Main suspend function for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked, clockgating is disabled and the
 * suspend callbacks are run.  suspend puts the hardware and software state
 * in each IP into a state suitable for suspend.
 * Returns 0 on success, negative error code on failure.
 */
static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1988 1989 1990
{
	int i, r;

1991 1992 1993
	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_request_full_gpu(adev, false);

1994
	/* ungate SMC block first */
1995 1996
	r = amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_SMC,
						   AMD_CG_STATE_UNGATE);
1997
	if (r) {
1998
		DRM_ERROR("set_clockgating_state(ungate) SMC failed %d\n", r);
1999 2000
	}

2001 2002 2003 2004
	/* call smu to disable gfx off feature first when suspend */
	if (adev->powerplay.pp_funcs->set_powergating_by_smu)
		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, false);

A
Alex Deucher 已提交
2005
	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
2006
		if (!adev->ip_blocks[i].status.valid)
A
Alex Deucher 已提交
2007
			continue;
2008 2009 2010
		/* displays are handled in phase1 */
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE)
			continue;
A
Alex Deucher 已提交
2011
		/* ungate blocks so that suspend can properly shut them down */
2012
		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_SMC &&
2013
			adev->ip_blocks[i].version->funcs->set_clockgating_state) {
2014 2015
			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
										     AMD_CG_STATE_UNGATE);
2016
			if (r) {
2017 2018
				DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n",
					  adev->ip_blocks[i].version->funcs->name, r);
2019
			}
2020
		}
A
Alex Deucher 已提交
2021
		/* XXX handle errors */
2022
		r = adev->ip_blocks[i].version->funcs->suspend(adev);
A
Alex Deucher 已提交
2023
		/* XXX handle errors */
2024
		if (r) {
2025 2026
			DRM_ERROR("suspend of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
2027
		}
A
Alex Deucher 已提交
2028 2029
	}

2030 2031 2032
	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_release_full_gpu(adev, false);

A
Alex Deucher 已提交
2033 2034 2035
	return 0;
}

2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058
/**
 * amdgpu_device_ip_suspend - run suspend for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * Main suspend function for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked, clockgating is disabled and the
 * suspend callbacks are run.  suspend puts the hardware and software state
 * in each IP into a state suitable for suspend.
 * Returns 0 on success, negative error code on failure.
 */
int amdgpu_device_ip_suspend(struct amdgpu_device *adev)
{
	int r;

	r = amdgpu_device_ip_suspend_phase1(adev);
	if (r)
		return r;
	r = amdgpu_device_ip_suspend_phase2(adev);

	return r;
}

2059
static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev)
2060 2061 2062
{
	int i, r;

2063 2064 2065 2066 2067
	static enum amd_ip_block_type ip_order[] = {
		AMD_IP_BLOCK_TYPE_GMC,
		AMD_IP_BLOCK_TYPE_COMMON,
		AMD_IP_BLOCK_TYPE_IH,
	};
2068

2069 2070 2071
	for (i = 0; i < ARRAY_SIZE(ip_order); i++) {
		int j;
		struct amdgpu_ip_block *block;
2072

2073 2074 2075 2076 2077 2078 2079 2080
		for (j = 0; j < adev->num_ip_blocks; j++) {
			block = &adev->ip_blocks[j];

			if (block->version->type != ip_order[i] ||
				!block->status.valid)
				continue;

			r = block->version->funcs->hw_init(adev);
2081
			DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded");
2082 2083
			if (r)
				return r;
2084 2085 2086 2087 2088 2089
		}
	}

	return 0;
}

2090
static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev)
2091 2092 2093
{
	int i, r;

2094 2095
	static enum amd_ip_block_type ip_order[] = {
		AMD_IP_BLOCK_TYPE_SMC,
M
Monk Liu 已提交
2096
		AMD_IP_BLOCK_TYPE_PSP,
2097 2098 2099
		AMD_IP_BLOCK_TYPE_DCE,
		AMD_IP_BLOCK_TYPE_GFX,
		AMD_IP_BLOCK_TYPE_SDMA,
2100 2101
		AMD_IP_BLOCK_TYPE_UVD,
		AMD_IP_BLOCK_TYPE_VCE
2102
	};
2103

2104 2105 2106
	for (i = 0; i < ARRAY_SIZE(ip_order); i++) {
		int j;
		struct amdgpu_ip_block *block;
2107

2108 2109 2110 2111 2112 2113 2114 2115
		for (j = 0; j < adev->num_ip_blocks; j++) {
			block = &adev->ip_blocks[j];

			if (block->version->type != ip_order[i] ||
				!block->status.valid)
				continue;

			r = block->version->funcs->hw_init(adev);
2116
			DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded");
2117 2118
			if (r)
				return r;
2119 2120 2121 2122 2123 2124
		}
	}

	return 0;
}

2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136
/**
 * amdgpu_device_ip_resume_phase1 - run resume for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * First resume function for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked and the resume callbacks are run for
 * COMMON, GMC, and IH.  resume puts the hardware into a functional state
 * after a suspend and updates the software state as necessary.  This
 * function is also used for restoring the GPU after a GPU reset.
 * Returns 0 on success, negative error code on failure.
 */
2137
static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev)
A
Alex Deucher 已提交
2138 2139 2140
{
	int i, r;

2141 2142 2143 2144
	for (i = 0; i < adev->num_ip_blocks; i++) {
		if (!adev->ip_blocks[i].status.valid)
			continue;
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
2145 2146
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) {
2147 2148 2149 2150 2151 2152
			r = adev->ip_blocks[i].version->funcs->resume(adev);
			if (r) {
				DRM_ERROR("resume of IP block <%s> failed %d\n",
					  adev->ip_blocks[i].version->funcs->name, r);
				return r;
			}
2153 2154 2155 2156 2157 2158
		}
	}

	return 0;
}

2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171
/**
 * amdgpu_device_ip_resume_phase2 - run resume for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * First resume function for hardware IPs.  The list of all the hardware
 * IPs that make up the asic is walked and the resume callbacks are run for
 * all blocks except COMMON, GMC, and IH.  resume puts the hardware into a
 * functional state after a suspend and updates the software state as
 * necessary.  This function is also used for restoring the GPU after a GPU
 * reset.
 * Returns 0 on success, negative error code on failure.
 */
2172
static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev)
A
Alex Deucher 已提交
2173 2174 2175 2176
{
	int i, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
2177
		if (!adev->ip_blocks[i].status.valid)
A
Alex Deucher 已提交
2178
			continue;
2179
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
2180 2181
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH)
2182
			continue;
2183
		r = adev->ip_blocks[i].version->funcs->resume(adev);
2184
		if (r) {
2185 2186
			DRM_ERROR("resume of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
A
Alex Deucher 已提交
2187
			return r;
2188
		}
A
Alex Deucher 已提交
2189 2190 2191 2192 2193
	}

	return 0;
}

2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205
/**
 * amdgpu_device_ip_resume - run resume for hardware IPs
 *
 * @adev: amdgpu_device pointer
 *
 * Main resume function for hardware IPs.  The hardware IPs
 * are split into two resume functions because they are
 * are also used in in recovering from a GPU reset and some additional
 * steps need to be take between them.  In this case (S3/S4) they are
 * run sequentially.
 * Returns 0 on success, negative error code on failure.
 */
2206
static int amdgpu_device_ip_resume(struct amdgpu_device *adev)
2207 2208 2209
{
	int r;

2210
	r = amdgpu_device_ip_resume_phase1(adev);
2211 2212
	if (r)
		return r;
2213
	r = amdgpu_device_ip_resume_phase2(adev);
2214 2215 2216 2217

	return r;
}

2218 2219 2220 2221 2222 2223 2224
/**
 * amdgpu_device_detect_sriov_bios - determine if the board supports SR-IOV
 *
 * @adev: amdgpu_device pointer
 *
 * Query the VBIOS data tables to determine if the board supports SR-IOV.
 */
2225
static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev)
2226
{
M
Monk Liu 已提交
2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237
	if (amdgpu_sriov_vf(adev)) {
		if (adev->is_atom_fw) {
			if (amdgpu_atomfirmware_gpu_supports_virtualization(adev))
				adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
		} else {
			if (amdgpu_atombios_has_gpu_virtualization_table(adev))
				adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
		}

		if (!(adev->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS))
			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0);
2238
	}
2239 2240
}

2241 2242 2243 2244 2245 2246 2247 2248
/**
 * amdgpu_device_asic_has_dc_support - determine if DC supports the asic
 *
 * @asic_type: AMD asic type
 *
 * Check if there is DC (new modesetting infrastructre) support for an asic.
 * returns true if DC has support, false if not.
 */
2249 2250 2251 2252 2253
bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type)
{
	switch (asic_type) {
#if defined(CONFIG_DRM_AMD_DC)
	case CHIP_BONAIRE:
2254
	case CHIP_KAVERI:
2255 2256
	case CHIP_KABINI:
	case CHIP_MULLINS:
2257 2258 2259 2260 2261 2262 2263 2264 2265
		/*
		 * We have systems in the wild with these ASICs that require
		 * LVDS and VGA support which is not supported with DC.
		 *
		 * Fallback to the non-DC driver here by default so as not to
		 * cause regressions.
		 */
		return amdgpu_dc > 0;
	case CHIP_HAWAII:
2266 2267 2268
	case CHIP_CARRIZO:
	case CHIP_STONEY:
	case CHIP_POLARIS10:
L
Leo Liu 已提交
2269
	case CHIP_POLARIS11:
2270
	case CHIP_POLARIS12:
L
Leo Liu 已提交
2271
	case CHIP_VEGAM:
2272 2273
	case CHIP_TONGA:
	case CHIP_FIJI:
2274
	case CHIP_VEGA10:
2275
	case CHIP_VEGA12:
2276
	case CHIP_VEGA20:
2277
#ifdef CONFIG_X86
2278
	case CHIP_RAVEN:
2279
#endif
2280
		return amdgpu_dc != 0;
2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295
#endif
	default:
		return false;
	}
}

/**
 * amdgpu_device_has_dc_support - check if dc is supported
 *
 * @adev: amdgpu_device_pointer
 *
 * Returns true for supported, false for not supported
 */
bool amdgpu_device_has_dc_support(struct amdgpu_device *adev)
{
X
Xiangliang Yu 已提交
2296 2297 2298
	if (amdgpu_sriov_vf(adev))
		return false;

2299 2300 2301
	return amdgpu_device_asic_has_dc_support(adev->asic_type);
}

A
Alex Deucher 已提交
2302 2303 2304 2305
/**
 * amdgpu_device_init - initialize the driver
 *
 * @adev: amdgpu_device pointer
2306
 * @ddev: drm dev pointer
A
Alex Deucher 已提交
2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320
 * @pdev: pci dev pointer
 * @flags: driver flags
 *
 * Initializes the driver info and hw (all asics).
 * Returns 0 for success or an error on failure.
 * Called at driver startup.
 */
int amdgpu_device_init(struct amdgpu_device *adev,
		       struct drm_device *ddev,
		       struct pci_dev *pdev,
		       uint32_t flags)
{
	int r, i;
	bool runtime = false;
2321
	u32 max_MBps;
A
Alex Deucher 已提交
2322 2323 2324 2325 2326 2327

	adev->shutdown = false;
	adev->dev = &pdev->dev;
	adev->ddev = ddev;
	adev->pdev = pdev;
	adev->flags = flags;
2328
	adev->asic_type = flags & AMD_ASIC_MASK;
A
Alex Deucher 已提交
2329
	adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT;
2330 2331
	if (amdgpu_emu_mode == 1)
		adev->usec_timeout *= 2;
2332
	adev->gmc.gart_size = 512 * 1024 * 1024;
A
Alex Deucher 已提交
2333 2334 2335 2336 2337
	adev->accel_working = false;
	adev->num_rings = 0;
	adev->mman.buffer_funcs = NULL;
	adev->mman.buffer_funcs_ring = NULL;
	adev->vm_manager.vm_pte_funcs = NULL;
2338
	adev->vm_manager.vm_pte_num_rings = 0;
2339
	adev->gmc.gmc_funcs = NULL;
2340
	adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS);
2341
	bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
A
Alex Deucher 已提交
2342 2343 2344 2345 2346

	adev->smc_rreg = &amdgpu_invalid_rreg;
	adev->smc_wreg = &amdgpu_invalid_wreg;
	adev->pcie_rreg = &amdgpu_invalid_rreg;
	adev->pcie_wreg = &amdgpu_invalid_wreg;
2347 2348
	adev->pciep_rreg = &amdgpu_invalid_rreg;
	adev->pciep_wreg = &amdgpu_invalid_wreg;
A
Alex Deucher 已提交
2349 2350 2351 2352
	adev->uvd_ctx_rreg = &amdgpu_invalid_rreg;
	adev->uvd_ctx_wreg = &amdgpu_invalid_wreg;
	adev->didt_rreg = &amdgpu_invalid_rreg;
	adev->didt_wreg = &amdgpu_invalid_wreg;
2353 2354
	adev->gc_cac_rreg = &amdgpu_invalid_rreg;
	adev->gc_cac_wreg = &amdgpu_invalid_wreg;
A
Alex Deucher 已提交
2355 2356 2357
	adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg;
	adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg;

2358 2359 2360
	DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n",
		 amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device,
		 pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision);
A
Alex Deucher 已提交
2361 2362 2363 2364

	/* mutex initialization are all done here so we
	 * can recall function without having locking issues */
	atomic_set(&adev->irq.ih.lock, 0);
2365
	mutex_init(&adev->firmware.mutex);
A
Alex Deucher 已提交
2366 2367 2368
	mutex_init(&adev->pm.mutex);
	mutex_init(&adev->gfx.gpu_clock_mutex);
	mutex_init(&adev->srbm_mutex);
2369
	mutex_init(&adev->gfx.pipe_reserve_mutex);
A
Alex Deucher 已提交
2370 2371
	mutex_init(&adev->grbm_idx_mutex);
	mutex_init(&adev->mn_lock);
A
Alex Deucher 已提交
2372
	mutex_init(&adev->virt.vf_errors.lock);
A
Alex Deucher 已提交
2373
	hash_init(adev->mn_hash);
2374
	mutex_init(&adev->lock_reset);
A
Alex Deucher 已提交
2375

2376
	amdgpu_device_check_arguments(adev);
A
Alex Deucher 已提交
2377 2378 2379 2380 2381 2382

	spin_lock_init(&adev->mmio_idx_lock);
	spin_lock_init(&adev->smc_idx_lock);
	spin_lock_init(&adev->pcie_idx_lock);
	spin_lock_init(&adev->uvd_ctx_idx_lock);
	spin_lock_init(&adev->didt_idx_lock);
2383
	spin_lock_init(&adev->gc_cac_idx_lock);
2384
	spin_lock_init(&adev->se_cac_idx_lock);
A
Alex Deucher 已提交
2385
	spin_lock_init(&adev->audio_endpt_idx_lock);
2386
	spin_lock_init(&adev->mm_stats.lock);
A
Alex Deucher 已提交
2387

2388 2389 2390
	INIT_LIST_HEAD(&adev->shadow_list);
	mutex_init(&adev->shadow_list_lock);

2391 2392 2393
	INIT_LIST_HEAD(&adev->ring_lru_list);
	spin_lock_init(&adev->ring_lru_list_lock);

2394 2395
	INIT_DELAYED_WORK(&adev->late_init_work,
			  amdgpu_device_ip_late_init_func_handler);
2396

2397 2398
	adev->pm.ac_power = power_supply_is_system_supplied() > 0 ? true : false;

2399 2400
	/* Registers mapping */
	/* TODO: block userspace mapping of io register */
2401 2402 2403 2404 2405 2406 2407
	if (adev->asic_type >= CHIP_BONAIRE) {
		adev->rmmio_base = pci_resource_start(adev->pdev, 5);
		adev->rmmio_size = pci_resource_len(adev->pdev, 5);
	} else {
		adev->rmmio_base = pci_resource_start(adev->pdev, 2);
		adev->rmmio_size = pci_resource_len(adev->pdev, 2);
	}
A
Alex Deucher 已提交
2408 2409 2410 2411 2412 2413 2414 2415

	adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size);
	if (adev->rmmio == NULL) {
		return -ENOMEM;
	}
	DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base);
	DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size);

2416
	/* doorbell bar mapping */
2417
	amdgpu_device_doorbell_init(adev);
A
Alex Deucher 已提交
2418 2419 2420 2421 2422 2423 2424 2425 2426 2427

	/* io port mapping */
	for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
		if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) {
			adev->rio_mem_size = pci_resource_len(adev->pdev, i);
			adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size);
			break;
		}
	}
	if (adev->rio_mem == NULL)
2428
		DRM_INFO("PCI I/O BAR is not found.\n");
A
Alex Deucher 已提交
2429

2430 2431
	amdgpu_device_get_pcie_info(adev);

A
Alex Deucher 已提交
2432
	/* early init functions */
2433
	r = amdgpu_device_ip_early_init(adev);
A
Alex Deucher 已提交
2434 2435 2436 2437 2438 2439
	if (r)
		return r;

	/* if we have > 1 VGA cards, then disable the amdgpu VGA resources */
	/* this will fail for cards that aren't VGA class devices, just
	 * ignore it */
2440
	vga_client_register(adev->pdev, adev, NULL, amdgpu_device_vga_set_decode);
A
Alex Deucher 已提交
2441

2442
	if (amdgpu_device_is_px(ddev))
A
Alex Deucher 已提交
2443
		runtime = true;
2444 2445 2446
	if (!pci_is_thunderbolt_attached(adev->pdev))
		vga_switcheroo_register_client(adev->pdev,
					       &amdgpu_switcheroo_ops, runtime);
A
Alex Deucher 已提交
2447 2448 2449
	if (runtime)
		vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain);

2450 2451 2452
	if (amdgpu_emu_mode == 1) {
		/* post the asic on emulation mode */
		emu_soc_asic_init(adev);
2453
		goto fence_driver_init;
2454
	}
2455

A
Alex Deucher 已提交
2456
	/* Read BIOS */
2457 2458 2459 2460
	if (!amdgpu_get_bios(adev)) {
		r = -EINVAL;
		goto failed;
	}
2461

A
Alex Deucher 已提交
2462
	r = amdgpu_atombios_init(adev);
2463 2464
	if (r) {
		dev_err(adev->dev, "amdgpu_atombios_init failed\n");
A
Alex Deucher 已提交
2465
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0);
2466
		goto failed;
2467
	}
A
Alex Deucher 已提交
2468

2469 2470
	/* detect if we are with an SRIOV vbios */
	amdgpu_device_detect_sriov_bios(adev);
2471

A
Alex Deucher 已提交
2472
	/* Post card if necessary */
A
Alex Deucher 已提交
2473
	if (amdgpu_device_need_post(adev)) {
A
Alex Deucher 已提交
2474
		if (!adev->bios) {
2475
			dev_err(adev->dev, "no vBIOS found\n");
2476 2477
			r = -EINVAL;
			goto failed;
A
Alex Deucher 已提交
2478
		}
2479
		DRM_INFO("GPU posting now...\n");
2480 2481 2482 2483 2484
		r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
		if (r) {
			dev_err(adev->dev, "gpu post error!\n");
			goto failed;
		}
A
Alex Deucher 已提交
2485 2486
	}

2487 2488 2489 2490 2491
	if (adev->is_atom_fw) {
		/* Initialize clocks */
		r = amdgpu_atomfirmware_get_clock_info(adev);
		if (r) {
			dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n");
A
Alex Deucher 已提交
2492
			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
2493 2494 2495
			goto failed;
		}
	} else {
2496 2497 2498 2499
		/* Initialize clocks */
		r = amdgpu_atombios_get_clock_info(adev);
		if (r) {
			dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n");
A
Alex Deucher 已提交
2500
			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
2501
			goto failed;
2502 2503
		}
		/* init i2c buses */
2504 2505
		if (!amdgpu_device_has_dc_support(adev))
			amdgpu_atombios_i2c_init(adev);
2506
	}
A
Alex Deucher 已提交
2507

2508
fence_driver_init:
A
Alex Deucher 已提交
2509 2510
	/* Fence driver */
	r = amdgpu_fence_driver_init(adev);
2511 2512
	if (r) {
		dev_err(adev->dev, "amdgpu_fence_driver_init failed\n");
A
Alex Deucher 已提交
2513
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0);
2514
		goto failed;
2515
	}
A
Alex Deucher 已提交
2516 2517 2518 2519

	/* init the mode config */
	drm_mode_config_init(adev->ddev);

2520
	r = amdgpu_device_ip_init(adev);
A
Alex Deucher 已提交
2521
	if (r) {
2522 2523 2524 2525 2526 2527
		/* failed in exclusive mode due to timeout */
		if (amdgpu_sriov_vf(adev) &&
		    !amdgpu_sriov_runtime(adev) &&
		    amdgpu_virt_mmio_blocked(adev) &&
		    !amdgpu_virt_wait_reset(adev)) {
			dev_err(adev->dev, "VF exclusive mode timeout\n");
2528 2529 2530
			/* Don't send request since VF is inactive. */
			adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME;
			adev->virt.ops = NULL;
2531 2532 2533
			r = -EAGAIN;
			goto failed;
		}
2534
		dev_err(adev->dev, "amdgpu_device_ip_init failed\n");
A
Alex Deucher 已提交
2535
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
2536
		goto failed;
A
Alex Deucher 已提交
2537 2538 2539 2540
	}

	adev->accel_working = true;

2541 2542
	amdgpu_vm_check_compute_bug(adev);

2543 2544 2545 2546 2547 2548 2549 2550
	/* Initialize the buffer migration limit. */
	if (amdgpu_moverate >= 0)
		max_MBps = amdgpu_moverate;
	else
		max_MBps = 8; /* Allow 8 MB/s. */
	/* Get a log2 for easy divisions. */
	adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps));

A
Alex Deucher 已提交
2551 2552 2553
	r = amdgpu_ib_pool_init(adev);
	if (r) {
		dev_err(adev->dev, "IB initialization failed (%d).\n", r);
A
Alex Deucher 已提交
2554
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r);
2555
		goto failed;
A
Alex Deucher 已提交
2556 2557
	}

2558 2559 2560
	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_init_data_exchange(adev);

2561 2562
	amdgpu_fbdev_init(adev);

2563 2564 2565 2566
	r = amdgpu_pm_sysfs_init(adev);
	if (r)
		DRM_ERROR("registering pm debugfs failed (%d).\n", r);

2567
	r = amdgpu_debugfs_gem_init(adev);
M
Monk Liu 已提交
2568
	if (r)
A
Alex Deucher 已提交
2569 2570 2571
		DRM_ERROR("registering gem debugfs failed (%d).\n", r);

	r = amdgpu_debugfs_regs_init(adev);
M
Monk Liu 已提交
2572
	if (r)
A
Alex Deucher 已提交
2573 2574
		DRM_ERROR("registering register debugfs failed (%d).\n", r);

2575
	r = amdgpu_debugfs_firmware_init(adev);
M
Monk Liu 已提交
2576
	if (r)
2577 2578
		DRM_ERROR("registering firmware debugfs failed (%d).\n", r);

2579
	r = amdgpu_debugfs_init(adev);
2580
	if (r)
2581
		DRM_ERROR("Creating debugfs files failed (%d).\n", r);
2582

A
Alex Deucher 已提交
2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598
	if ((amdgpu_testing & 1)) {
		if (adev->accel_working)
			amdgpu_test_moves(adev);
		else
			DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n");
	}
	if (amdgpu_benchmarking) {
		if (adev->accel_working)
			amdgpu_benchmark(adev, amdgpu_benchmarking);
		else
			DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n");
	}

	/* enable clockgating, etc. after ib tests, etc. since some blocks require
	 * explicit gating rather than handling it automatically.
	 */
2599
	r = amdgpu_device_ip_late_init(adev);
2600
	if (r) {
2601
		dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n");
A
Alex Deucher 已提交
2602
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
2603
		goto failed;
2604
	}
A
Alex Deucher 已提交
2605 2606

	return 0;
2607 2608

failed:
2609
	amdgpu_vf_error_trans_all(adev);
2610 2611
	if (runtime)
		vga_switcheroo_fini_domain_pm_ops(adev->dev);
2612

2613
	return r;
A
Alex Deucher 已提交
2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629
}

/**
 * amdgpu_device_fini - tear down the driver
 *
 * @adev: amdgpu_device pointer
 *
 * Tear down the driver info (all asics).
 * Called at driver shutdown.
 */
void amdgpu_device_fini(struct amdgpu_device *adev)
{
	int r;

	DRM_INFO("amdgpu: finishing device.\n");
	adev->shutdown = true;
2630 2631
	/* disable all interrupts */
	amdgpu_irq_disable_all(adev);
2632 2633 2634 2635 2636 2637
	if (adev->mode_info.mode_config_initialized){
		if (!amdgpu_device_has_dc_support(adev))
			drm_crtc_force_disable_all(adev->ddev);
		else
			drm_atomic_helper_shutdown(adev->ddev);
	}
A
Alex Deucher 已提交
2638 2639
	amdgpu_ib_pool_fini(adev);
	amdgpu_fence_driver_fini(adev);
2640
	amdgpu_pm_sysfs_fini(adev);
A
Alex Deucher 已提交
2641
	amdgpu_fbdev_fini(adev);
2642
	r = amdgpu_device_ip_fini(adev);
2643 2644 2645 2646
	if (adev->firmware.gpu_info_fw) {
		release_firmware(adev->firmware.gpu_info_fw);
		adev->firmware.gpu_info_fw = NULL;
	}
A
Alex Deucher 已提交
2647
	adev->accel_working = false;
2648
	cancel_delayed_work_sync(&adev->late_init_work);
A
Alex Deucher 已提交
2649
	/* free i2c buses */
2650 2651
	if (!amdgpu_device_has_dc_support(adev))
		amdgpu_i2c_fini(adev);
2652 2653 2654 2655

	if (amdgpu_emu_mode != 1)
		amdgpu_atombios_fini(adev);

A
Alex Deucher 已提交
2656 2657
	kfree(adev->bios);
	adev->bios = NULL;
2658 2659
	if (!pci_is_thunderbolt_attached(adev->pdev))
		vga_switcheroo_unregister_client(adev->pdev);
2660 2661
	if (adev->flags & AMD_IS_PX)
		vga_switcheroo_fini_domain_pm_ops(adev->dev);
A
Alex Deucher 已提交
2662 2663 2664 2665 2666 2667
	vga_client_register(adev->pdev, NULL, NULL, NULL);
	if (adev->rio_mem)
		pci_iounmap(adev->pdev, adev->rio_mem);
	adev->rio_mem = NULL;
	iounmap(adev->rmmio);
	adev->rmmio = NULL;
2668
	amdgpu_device_doorbell_fini(adev);
A
Alex Deucher 已提交
2669 2670 2671 2672 2673 2674 2675 2676
	amdgpu_debugfs_regs_cleanup(adev);
}


/*
 * Suspend & resume.
 */
/**
2677
 * amdgpu_device_suspend - initiate device suspend
A
Alex Deucher 已提交
2678
 *
2679 2680 2681
 * @dev: drm dev pointer
 * @suspend: suspend state
 * @fbcon : notify the fbdev of suspend
A
Alex Deucher 已提交
2682 2683 2684 2685 2686
 *
 * Puts the hw in the suspend state (all asics).
 * Returns 0 for success or an error on failure.
 * Called at driver suspend.
 */
2687
int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon)
A
Alex Deucher 已提交
2688 2689 2690 2691
{
	struct amdgpu_device *adev;
	struct drm_crtc *crtc;
	struct drm_connector *connector;
2692
	int r;
A
Alex Deucher 已提交
2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704

	if (dev == NULL || dev->dev_private == NULL) {
		return -ENODEV;
	}

	adev = dev->dev_private;

	if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
		return 0;

	drm_kms_helper_poll_disable(dev);

2705 2706 2707 2708 2709 2710 2711
	if (!amdgpu_device_has_dc_support(adev)) {
		/* turn off display hw */
		drm_modeset_lock_all(dev);
		list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
			drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
		}
		drm_modeset_unlock_all(dev);
2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724
			/* unpin the front buffers and cursors */
		list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
			struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
			struct drm_framebuffer *fb = crtc->primary->fb;
			struct amdgpu_bo *robj;

			if (amdgpu_crtc->cursor_bo) {
				struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
				r = amdgpu_bo_reserve(aobj, true);
				if (r == 0) {
					amdgpu_bo_unpin(aobj);
					amdgpu_bo_unreserve(aobj);
				}
2725 2726
			}

2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737
			if (fb == NULL || fb->obj[0] == NULL) {
				continue;
			}
			robj = gem_to_amdgpu_bo(fb->obj[0]);
			/* don't unpin kernel fb objects */
			if (!amdgpu_fbdev_robj_is_fb(adev, robj)) {
				r = amdgpu_bo_reserve(robj, true);
				if (r == 0) {
					amdgpu_bo_unpin(robj);
					amdgpu_bo_unreserve(robj);
				}
A
Alex Deucher 已提交
2738 2739 2740
			}
		}
	}
2741 2742 2743 2744 2745

	amdgpu_amdkfd_suspend(adev);

	r = amdgpu_device_ip_suspend_phase1(adev);

A
Alex Deucher 已提交
2746 2747 2748
	/* evict vram memory */
	amdgpu_bo_evict_vram(adev);

2749
	amdgpu_fence_driver_suspend(adev);
A
Alex Deucher 已提交
2750

2751
	r = amdgpu_device_ip_suspend_phase2(adev);
A
Alex Deucher 已提交
2752

2753 2754 2755 2756
	/* evict remaining vram memory
	 * This second call to evict vram is to evict the gart page table
	 * using the CPU.
	 */
A
Alex Deucher 已提交
2757 2758 2759 2760 2761 2762 2763
	amdgpu_bo_evict_vram(adev);

	pci_save_state(dev->pdev);
	if (suspend) {
		/* Shut down the device */
		pci_disable_device(dev->pdev);
		pci_set_power_state(dev->pdev, PCI_D3hot);
J
jimqu 已提交
2764 2765 2766 2767
	} else {
		r = amdgpu_asic_reset(adev);
		if (r)
			DRM_ERROR("amdgpu asic reset failed\n");
A
Alex Deucher 已提交
2768 2769
	}

2770
	if (fbcon)
A
Alex Deucher 已提交
2771
		amdgpu_fbdev_set_suspend(adev, 1);
2772

A
Alex Deucher 已提交
2773 2774 2775 2776
	return 0;
}

/**
2777
 * amdgpu_device_resume - initiate device resume
A
Alex Deucher 已提交
2778
 *
2779 2780 2781
 * @dev: drm dev pointer
 * @resume: resume state
 * @fbcon : notify the fbdev of resume
A
Alex Deucher 已提交
2782 2783 2784 2785 2786
 *
 * Bring the hw back to operating state (all asics).
 * Returns 0 for success or an error on failure.
 * Called at driver resume.
 */
2787
int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon)
A
Alex Deucher 已提交
2788 2789 2790
{
	struct drm_connector *connector;
	struct amdgpu_device *adev = dev->dev_private;
2791
	struct drm_crtc *crtc;
2792
	int r = 0;
A
Alex Deucher 已提交
2793 2794 2795 2796 2797 2798 2799

	if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
		return 0;

	if (resume) {
		pci_set_power_state(dev->pdev, PCI_D0);
		pci_restore_state(dev->pdev);
J
jimqu 已提交
2800
		r = pci_enable_device(dev->pdev);
2801
		if (r)
2802
			return r;
A
Alex Deucher 已提交
2803 2804 2805
	}

	/* post card */
A
Alex Deucher 已提交
2806
	if (amdgpu_device_need_post(adev)) {
J
jimqu 已提交
2807 2808 2809 2810
		r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
		if (r)
			DRM_ERROR("amdgpu asic init failed\n");
	}
A
Alex Deucher 已提交
2811

2812
	r = amdgpu_device_ip_resume(adev);
2813
	if (r) {
2814
		DRM_ERROR("amdgpu_device_ip_resume failed (%d).\n", r);
2815
		return r;
2816
	}
2817 2818
	amdgpu_fence_driver_resume(adev);

A
Alex Deucher 已提交
2819

2820
	r = amdgpu_device_ip_late_init(adev);
2821
	if (r)
2822
		return r;
A
Alex Deucher 已提交
2823

2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838
	if (!amdgpu_device_has_dc_support(adev)) {
		/* pin cursors */
		list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
			struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);

			if (amdgpu_crtc->cursor_bo) {
				struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
				r = amdgpu_bo_reserve(aobj, true);
				if (r == 0) {
					r = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM);
					if (r != 0)
						DRM_ERROR("Failed to pin cursor BO (%d)\n", r);
					amdgpu_crtc->cursor_addr = amdgpu_bo_gpu_offset(aobj);
					amdgpu_bo_unreserve(aobj);
				}
2839 2840 2841
			}
		}
	}
2842 2843 2844
	r = amdgpu_amdkfd_resume(adev);
	if (r)
		return r;
2845

A
Alex Deucher 已提交
2846 2847
	/* blat the mode back in */
	if (fbcon) {
2848 2849 2850 2851 2852 2853 2854 2855 2856 2857
		if (!amdgpu_device_has_dc_support(adev)) {
			/* pre DCE11 */
			drm_helper_resume_force_mode(dev);

			/* turn on display hw */
			drm_modeset_lock_all(dev);
			list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
				drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON);
			}
			drm_modeset_unlock_all(dev);
A
Alex Deucher 已提交
2858
		}
2859
		amdgpu_fbdev_set_suspend(adev, 0);
A
Alex Deucher 已提交
2860 2861 2862
	}

	drm_kms_helper_poll_enable(dev);
2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875

	/*
	 * Most of the connector probing functions try to acquire runtime pm
	 * refs to ensure that the GPU is powered on when connector polling is
	 * performed. Since we're calling this from a runtime PM callback,
	 * trying to acquire rpm refs will cause us to deadlock.
	 *
	 * Since we're guaranteed to be holding the rpm lock, it's safe to
	 * temporarily disable the rpm helpers so this doesn't deadlock us.
	 */
#ifdef CONFIG_PM
	dev->dev->power.disable_depth++;
#endif
2876 2877 2878 2879
	if (!amdgpu_device_has_dc_support(adev))
		drm_helper_hpd_irq_event(dev);
	else
		drm_kms_helper_hotplug_event(dev);
2880 2881 2882
#ifdef CONFIG_PM
	dev->dev->power.disable_depth--;
#endif
2883
	return 0;
A
Alex Deucher 已提交
2884 2885
}

2886 2887 2888 2889 2890 2891 2892 2893 2894 2895
/**
 * amdgpu_device_ip_check_soft_reset - did soft reset succeed
 *
 * @adev: amdgpu_device pointer
 *
 * The list of all the hardware IPs that make up the asic is walked and
 * the check_soft_reset callbacks are run.  check_soft_reset determines
 * if the asic is still hung or not.
 * Returns true if any of the IPs are still in a hung state, false if not.
 */
2896
static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev)
2897 2898 2899 2900
{
	int i;
	bool asic_hang = false;

2901 2902 2903
	if (amdgpu_sriov_vf(adev))
		return true;

2904 2905 2906
	if (amdgpu_asic_need_full_reset(adev))
		return true;

2907
	for (i = 0; i < adev->num_ip_blocks; i++) {
2908
		if (!adev->ip_blocks[i].status.valid)
2909
			continue;
2910 2911 2912 2913 2914
		if (adev->ip_blocks[i].version->funcs->check_soft_reset)
			adev->ip_blocks[i].status.hang =
				adev->ip_blocks[i].version->funcs->check_soft_reset(adev);
		if (adev->ip_blocks[i].status.hang) {
			DRM_INFO("IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name);
2915 2916 2917 2918 2919 2920
			asic_hang = true;
		}
	}
	return asic_hang;
}

2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931
/**
 * amdgpu_device_ip_pre_soft_reset - prepare for soft reset
 *
 * @adev: amdgpu_device pointer
 *
 * The list of all the hardware IPs that make up the asic is walked and the
 * pre_soft_reset callbacks are run if the block is hung.  pre_soft_reset
 * handles any IP specific hardware or software state changes that are
 * necessary for a soft reset to succeed.
 * Returns 0 on success, negative error code on failure.
 */
2932
static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev)
2933 2934 2935 2936
{
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
2937
		if (!adev->ip_blocks[i].status.valid)
2938
			continue;
2939 2940 2941
		if (adev->ip_blocks[i].status.hang &&
		    adev->ip_blocks[i].version->funcs->pre_soft_reset) {
			r = adev->ip_blocks[i].version->funcs->pre_soft_reset(adev);
2942 2943 2944 2945 2946 2947 2948 2949
			if (r)
				return r;
		}
	}

	return 0;
}

2950 2951 2952 2953 2954 2955 2956 2957 2958
/**
 * amdgpu_device_ip_need_full_reset - check if a full asic reset is needed
 *
 * @adev: amdgpu_device pointer
 *
 * Some hardware IPs cannot be soft reset.  If they are hung, a full gpu
 * reset is necessary to recover.
 * Returns true if a full asic reset is required, false if not.
 */
2959
static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev)
2960
{
2961 2962
	int i;

2963 2964 2965
	if (amdgpu_asic_need_full_reset(adev))
		return true;

2966
	for (i = 0; i < adev->num_ip_blocks; i++) {
2967
		if (!adev->ip_blocks[i].status.valid)
2968
			continue;
2969 2970 2971
		if ((adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) ||
		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) ||
		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_ACP) ||
2972 2973
		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) ||
		     adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) {
2974
			if (adev->ip_blocks[i].status.hang) {
2975 2976 2977 2978
				DRM_INFO("Some block need full reset!\n");
				return true;
			}
		}
2979 2980 2981 2982
	}
	return false;
}

2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993
/**
 * amdgpu_device_ip_soft_reset - do a soft reset
 *
 * @adev: amdgpu_device pointer
 *
 * The list of all the hardware IPs that make up the asic is walked and the
 * soft_reset callbacks are run if the block is hung.  soft_reset handles any
 * IP specific hardware or software state changes that are necessary to soft
 * reset the IP.
 * Returns 0 on success, negative error code on failure.
 */
2994
static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev)
2995 2996 2997 2998
{
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
2999
		if (!adev->ip_blocks[i].status.valid)
3000
			continue;
3001 3002 3003
		if (adev->ip_blocks[i].status.hang &&
		    adev->ip_blocks[i].version->funcs->soft_reset) {
			r = adev->ip_blocks[i].version->funcs->soft_reset(adev);
3004 3005 3006 3007 3008 3009 3010 3011
			if (r)
				return r;
		}
	}

	return 0;
}

3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022
/**
 * amdgpu_device_ip_post_soft_reset - clean up from soft reset
 *
 * @adev: amdgpu_device pointer
 *
 * The list of all the hardware IPs that make up the asic is walked and the
 * post_soft_reset callbacks are run if the asic was hung.  post_soft_reset
 * handles any IP specific hardware or software state changes that are
 * necessary after the IP has been soft reset.
 * Returns 0 on success, negative error code on failure.
 */
3023
static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev)
3024 3025 3026 3027
{
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
3028
		if (!adev->ip_blocks[i].status.valid)
3029
			continue;
3030 3031 3032
		if (adev->ip_blocks[i].status.hang &&
		    adev->ip_blocks[i].version->funcs->post_soft_reset)
			r = adev->ip_blocks[i].version->funcs->post_soft_reset(adev);
3033 3034 3035 3036 3037 3038 3039
		if (r)
			return r;
	}

	return 0;
}

3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052
/**
 * amdgpu_device_recover_vram_from_shadow - restore shadowed VRAM buffers
 *
 * @adev: amdgpu_device pointer
 * @ring: amdgpu_ring for the engine handling the buffer operations
 * @bo: amdgpu_bo buffer whose shadow is being restored
 * @fence: dma_fence associated with the operation
 *
 * Restores the VRAM buffer contents from the shadow in GTT.  Used to
 * restore things like GPUVM page tables after a GPU reset where
 * the contents of VRAM might be lost.
 * Returns 0 on success, negative error code on failure.
 */
3053 3054 3055 3056
static int amdgpu_device_recover_vram_from_shadow(struct amdgpu_device *adev,
						  struct amdgpu_ring *ring,
						  struct amdgpu_bo *bo,
						  struct dma_fence **fence)
3057 3058 3059 3060
{
	uint32_t domain;
	int r;

R
Roger.He 已提交
3061 3062 3063
	if (!bo->shadow)
		return 0;

3064
	r = amdgpu_bo_reserve(bo, true);
R
Roger.He 已提交
3065 3066 3067 3068 3069
	if (r)
		return r;
	domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
	/* if bo has been evicted, then no need to recover */
	if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
3070 3071 3072 3073 3074 3075
		r = amdgpu_bo_validate(bo->shadow);
		if (r) {
			DRM_ERROR("bo validate failed!\n");
			goto err;
		}

R
Roger.He 已提交
3076
		r = amdgpu_bo_restore_from_shadow(adev, ring, bo,
3077
						 NULL, fence, true);
R
Roger.He 已提交
3078 3079 3080 3081 3082
		if (r) {
			DRM_ERROR("recover page table failed!\n");
			goto err;
		}
	}
3083
err:
R
Roger.He 已提交
3084 3085
	amdgpu_bo_unreserve(bo);
	return r;
3086 3087
}

3088 3089 3090 3091 3092 3093 3094 3095 3096 3097
/**
 * amdgpu_device_handle_vram_lost - Handle the loss of VRAM contents
 *
 * @adev: amdgpu_device pointer
 *
 * Restores the contents of VRAM buffers from the shadows in GTT.  Used to
 * restore things like GPUVM page tables after a GPU reset where
 * the contents of VRAM might be lost.
 * Returns 0 on success, 1 on failure.
 */
3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150
static int amdgpu_device_handle_vram_lost(struct amdgpu_device *adev)
{
	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
	struct amdgpu_bo *bo, *tmp;
	struct dma_fence *fence = NULL, *next = NULL;
	long r = 1;
	int i = 0;
	long tmo;

	if (amdgpu_sriov_runtime(adev))
		tmo = msecs_to_jiffies(amdgpu_lockup_timeout);
	else
		tmo = msecs_to_jiffies(100);

	DRM_INFO("recover vram bo from shadow start\n");
	mutex_lock(&adev->shadow_list_lock);
	list_for_each_entry_safe(bo, tmp, &adev->shadow_list, shadow_list) {
		next = NULL;
		amdgpu_device_recover_vram_from_shadow(adev, ring, bo, &next);
		if (fence) {
			r = dma_fence_wait_timeout(fence, false, tmo);
			if (r == 0)
				pr_err("wait fence %p[%d] timeout\n", fence, i);
			else if (r < 0)
				pr_err("wait fence %p[%d] interrupted\n", fence, i);
			if (r < 1) {
				dma_fence_put(fence);
				fence = next;
				break;
			}
			i++;
		}

		dma_fence_put(fence);
		fence = next;
	}
	mutex_unlock(&adev->shadow_list_lock);

	if (fence) {
		r = dma_fence_wait_timeout(fence, false, tmo);
		if (r == 0)
			pr_err("wait fence %p[%d] timeout\n", fence, i);
		else if (r < 0)
			pr_err("wait fence %p[%d] interrupted\n", fence, i);

	}
	dma_fence_put(fence);

	if (r > 0)
		DRM_INFO("recover vram bo from shadow done\n");
	else
		DRM_ERROR("recover vram bo from shadow failed\n");

3151
	return (r > 0) ? 0 : 1;
3152 3153
}

3154
/**
3155
 * amdgpu_device_reset - reset ASIC/GPU for bare-metal or passthrough
3156 3157 3158
 *
 * @adev: amdgpu device pointer
 *
3159
 * attempt to do soft-reset or full-reset and reinitialize Asic
3160
 * return 0 means succeeded otherwise failed
3161
 */
3162
static int amdgpu_device_reset(struct amdgpu_device *adev)
3163
{
3164 3165
	bool need_full_reset, vram_lost = 0;
	int r;
3166

3167
	need_full_reset = amdgpu_device_ip_need_full_reset(adev);
3168

3169
	if (!need_full_reset) {
3170 3171 3172 3173
		amdgpu_device_ip_pre_soft_reset(adev);
		r = amdgpu_device_ip_soft_reset(adev);
		amdgpu_device_ip_post_soft_reset(adev);
		if (r || amdgpu_device_ip_check_soft_reset(adev)) {
3174 3175 3176 3177
			DRM_INFO("soft reset failed, will fallback to full reset!\n");
			need_full_reset = true;
		}
	}
3178

3179
	if (need_full_reset) {
3180
		r = amdgpu_device_ip_suspend(adev);
3181

3182 3183 3184 3185
retry:
		r = amdgpu_asic_reset(adev);
		/* post card */
		amdgpu_atom_asic_init(adev->mode_info.atom_context);
3186

3187 3188
		if (!r) {
			dev_info(adev->dev, "GPU reset succeeded, trying to resume\n");
3189
			r = amdgpu_device_ip_resume_phase1(adev);
3190 3191
			if (r)
				goto out;
3192

3193
			vram_lost = amdgpu_device_check_vram_lost(adev);
3194 3195 3196 3197 3198
			if (vram_lost) {
				DRM_ERROR("VRAM is lost!\n");
				atomic_inc(&adev->vram_lost_counter);
			}

3199 3200
			r = amdgpu_gtt_mgr_recover(
				&adev->mman.bdev.man[TTM_PL_TT]);
3201 3202 3203
			if (r)
				goto out;

3204
			r = amdgpu_device_ip_resume_phase2(adev);
3205 3206 3207 3208
			if (r)
				goto out;

			if (vram_lost)
3209
				amdgpu_device_fill_reset_magic(adev);
3210
		}
3211
	}
3212

3213 3214 3215 3216 3217 3218
out:
	if (!r) {
		amdgpu_irq_gpu_reset_resume_helper(adev);
		r = amdgpu_ib_ring_tests(adev);
		if (r) {
			dev_err(adev->dev, "ib ring test failed (%d).\n", r);
3219
			r = amdgpu_device_ip_suspend(adev);
3220 3221 3222 3223
			need_full_reset = true;
			goto retry;
		}
	}
3224

3225 3226
	if (!r && ((need_full_reset && !(adev->flags & AMD_IS_APU)) || vram_lost))
		r = amdgpu_device_handle_vram_lost(adev);
3227

3228 3229
	return r;
}
3230

3231
/**
3232
 * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf
3233 3234
 *
 * @adev: amdgpu device pointer
3235
 * @from_hypervisor: request from hypervisor
3236 3237
 *
 * do VF FLR and reinitialize Asic
3238
 * return 0 means succeeded otherwise failed
3239 3240 3241
 */
static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
				     bool from_hypervisor)
3242 3243 3244 3245 3246 3247 3248 3249 3250
{
	int r;

	if (from_hypervisor)
		r = amdgpu_virt_request_full_gpu(adev, true);
	else
		r = amdgpu_virt_reset_gpu(adev);
	if (r)
		return r;
3251 3252

	/* Resume IP prior to SMC */
3253
	r = amdgpu_device_ip_reinit_early_sriov(adev);
3254 3255
	if (r)
		goto error;
3256 3257

	/* we need recover gart prior to run SMC/CP/SDMA resume */
3258
	amdgpu_gtt_mgr_recover(&adev->mman.bdev.man[TTM_PL_TT]);
3259 3260

	/* now we are okay to resume SMC/CP/SDMA */
3261
	r = amdgpu_device_ip_reinit_late_sriov(adev);
3262 3263
	if (r)
		goto error;
3264 3265

	amdgpu_irq_gpu_reset_resume_helper(adev);
3266
	r = amdgpu_ib_ring_tests(adev);
3267

3268 3269
error:
	amdgpu_virt_release_full_gpu(adev, true);
3270 3271 3272
	if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) {
		atomic_inc(&adev->vram_lost_counter);
		r = amdgpu_device_handle_vram_lost(adev);
3273 3274 3275 3276 3277
	}

	return r;
}

A
Alex Deucher 已提交
3278
/**
3279
 * amdgpu_device_gpu_recover - reset the asic and recover scheduler
A
Alex Deucher 已提交
3280 3281
 *
 * @adev: amdgpu device pointer
3282
 * @job: which job trigger hang
3283
 * @force: forces reset regardless of amdgpu_gpu_recovery
A
Alex Deucher 已提交
3284
 *
3285
 * Attempt to reset the GPU if it has hung (all asics).
A
Alex Deucher 已提交
3286 3287
 * Returns 0 for success or an error on failure.
 */
3288 3289
int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
			      struct amdgpu_job *job, bool force)
A
Alex Deucher 已提交
3290
{
3291
	int i, r, resched;
3292

3293
	if (!force && !amdgpu_device_ip_check_soft_reset(adev)) {
3294 3295 3296
		DRM_INFO("No hardware hang detected. Did some blocks stall?\n");
		return 0;
	}
A
Alex Deucher 已提交
3297

3298 3299 3300 3301 3302 3303
	if (!force && (amdgpu_gpu_recovery == 0 ||
			(amdgpu_gpu_recovery == -1  && !amdgpu_sriov_vf(adev)))) {
		DRM_INFO("GPU recovery disabled.\n");
		return 0;
	}

3304 3305
	dev_info(adev->dev, "GPU reset begin!\n");

3306
	mutex_lock(&adev->lock_reset);
3307
	atomic_inc(&adev->gpu_reset_counter);
3308
	adev->in_gpu_reset = 1;
A
Alex Deucher 已提交
3309

3310 3311
	/* block TTM */
	resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev);
3312 3313

	/* block all schedulers and reset given job's ring */
3314 3315 3316
	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];

C
Chunming Zhou 已提交
3317
		if (!ring || !ring->sched.thread)
3318
			continue;
3319

3320 3321
		kthread_park(ring->sched.thread);

C
Christian König 已提交
3322
		if (job && job->base.sched == &ring->sched)
3323 3324
			continue;

3325
		drm_sched_hw_job_reset(&ring->sched, &job->base);
3326

M
Monk Liu 已提交
3327 3328
		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
		amdgpu_fence_driver_force_completion(ring);
3329
	}
A
Alex Deucher 已提交
3330

3331
	if (amdgpu_sriov_vf(adev))
3332
		r = amdgpu_device_reset_sriov(adev, job ? false : true);
3333
	else
3334
		r = amdgpu_device_reset(adev);
3335

3336 3337
	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];
3338

3339 3340
		if (!ring || !ring->sched.thread)
			continue;
3341

3342 3343 3344 3345
		/* only need recovery sched of the given job's ring
		 * or all rings (in the case @job is NULL)
		 * after above amdgpu_reset accomplished
		 */
C
Christian König 已提交
3346
		if ((!job || job->base.sched == &ring->sched) && !r)
3347
			drm_sched_job_recovery(&ring->sched);
3348

3349
		kthread_unpark(ring->sched.thread);
A
Alex Deucher 已提交
3350 3351
	}

3352
	if (!amdgpu_device_has_dc_support(adev)) {
3353
		drm_helper_resume_force_mode(adev->ddev);
3354
	}
A
Alex Deucher 已提交
3355 3356

	ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched);
3357

3358
	if (r) {
A
Alex Deucher 已提交
3359
		/* bad news, how to tell it to userspace ? */
3360 3361 3362
		dev_info(adev->dev, "GPU reset(%d) failed\n", atomic_read(&adev->gpu_reset_counter));
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r);
	} else {
3363
		dev_info(adev->dev, "GPU reset(%d) succeeded!\n",atomic_read(&adev->gpu_reset_counter));
3364
	}
A
Alex Deucher 已提交
3365

3366
	amdgpu_vf_error_trans_all(adev);
3367 3368
	adev->in_gpu_reset = 0;
	mutex_unlock(&adev->lock_reset);
A
Alex Deucher 已提交
3369 3370 3371
	return r;
}

3372 3373 3374 3375 3376 3377 3378 3379 3380
/**
 * amdgpu_device_get_pcie_info - fence pcie info about the PCIE slot
 *
 * @adev: amdgpu_device pointer
 *
 * Fetchs and stores in the driver the PCIE capabilities (gen speed
 * and lanes) of the slot the device is in. Handles APUs and
 * virtualized environments where PCIE config space may not be available.
 */
3381
static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev)
3382
{
3383 3384 3385
	struct pci_dev *pdev;
	enum pci_bus_speed speed_cap;
	enum pcie_link_width link_width;
3386

3387 3388
	if (amdgpu_pcie_gen_cap)
		adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap;
3389

3390 3391
	if (amdgpu_pcie_lane_cap)
		adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap;
3392

3393 3394 3395 3396 3397 3398
	/* covers APUs as well */
	if (pci_is_root_bus(adev->pdev->bus)) {
		if (adev->pm.pcie_gen_mask == 0)
			adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK;
		if (adev->pm.pcie_mlw_mask == 0)
			adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK;
3399
		return;
3400
	}
3401

3402
	if (adev->pm.pcie_gen_mask == 0) {
3403 3404 3405 3406 3407
		/* asic caps */
		pdev = adev->pdev;
		speed_cap = pcie_get_speed_cap(pdev);
		if (speed_cap == PCI_SPEED_UNKNOWN) {
			adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3408 3409 3410
						  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
						  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3);
		} else {
3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447
			if (speed_cap == PCIE_SPEED_16_0GT)
				adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 |
							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4);
			else if (speed_cap == PCIE_SPEED_8_0GT)
				adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3);
			else if (speed_cap == PCIE_SPEED_5_0GT)
				adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
							  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2);
			else
				adev->pm.pcie_gen_mask |= CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1;
		}
		/* platform caps */
		pdev = adev->ddev->pdev->bus->self;
		speed_cap = pcie_get_speed_cap(pdev);
		if (speed_cap == PCI_SPEED_UNKNOWN) {
			adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
						   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2);
		} else {
			if (speed_cap == PCIE_SPEED_16_0GT)
				adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 |
							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 |
							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4);
			else if (speed_cap == PCIE_SPEED_8_0GT)
				adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 |
							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3);
			else if (speed_cap == PCIE_SPEED_5_0GT)
				adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
							   CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2);
			else
				adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1;

3448 3449 3450
		}
	}
	if (adev->pm.pcie_mlw_mask == 0) {
3451 3452 3453 3454 3455 3456 3457
		pdev = adev->ddev->pdev->bus->self;
		link_width = pcie_get_width_cap(pdev);
		if (link_width == PCIE_LNK_WIDTH_UNKNOWN) {
			adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_PCIE_MLW_MASK;
		} else {
			switch (link_width) {
			case PCIE_LNK_X32:
3458 3459 3460 3461 3462 3463 3464 3465
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3466
			case PCIE_LNK_X16:
3467 3468 3469 3470 3471 3472 3473
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3474
			case PCIE_LNK_X12:
3475 3476 3477 3478 3479 3480
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3481
			case PCIE_LNK_X8:
3482 3483 3484 3485 3486
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3487
			case PCIE_LNK_X4:
3488 3489 3490 3491
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3492
			case PCIE_LNK_X2:
3493 3494 3495
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3496
			case PCIE_LNK_X1:
3497 3498 3499 3500 3501
				adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1;
				break;
			default:
				break;
			}
3502 3503 3504
		}
	}
}
A
Alex Deucher 已提交
3505