amdgpu_device.c 98.6 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 63
#include "amdgpu_xgmi.h"

64
MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
65
MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin");
66
MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin");
67
MODULE_FIRMWARE("amdgpu/picasso_gpu_info.bin");
68
MODULE_FIRMWARE("amdgpu/raven2_gpu_info.bin");
69

70 71
#define AMDGPU_RESUME_MS		2000

A
Alex Deucher 已提交
72
static const char *amdgpu_asic_name[] = {
73 74 75 76 77
	"TAHITI",
	"PITCAIRN",
	"VERDE",
	"OLAND",
	"HAINAN",
A
Alex Deucher 已提交
78 79 80 81 82 83 84
	"BONAIRE",
	"KAVERI",
	"KABINI",
	"HAWAII",
	"MULLINS",
	"TOPAZ",
	"TONGA",
85
	"FIJI",
A
Alex Deucher 已提交
86
	"CARRIZO",
S
Samuel Li 已提交
87
	"STONEY",
88 89
	"POLARIS10",
	"POLARIS11",
90
	"POLARIS12",
L
Leo Liu 已提交
91
	"VEGAM",
K
Ken Wang 已提交
92
	"VEGA10",
93
	"VEGA12",
94
	"VEGA20",
95
	"RAVEN",
A
Alex Deucher 已提交
96 97 98
	"LAST",
};

99 100
static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev);

101 102 103 104 105 106 107 108
/**
 * 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
109 110 111 112
bool amdgpu_device_is_px(struct drm_device *dev)
{
	struct amdgpu_device *adev = dev->dev_private;

113
	if (adev->flags & AMD_IS_PX)
A
Alex Deucher 已提交
114 115 116 117 118 119 120
		return true;
	return false;
}

/*
 * MMIO register access helper functions.
 */
121 122 123 124 125 126 127 128 129
/**
 * 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
130
uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg,
M
Monk Liu 已提交
131
			uint32_t acc_flags)
A
Alex Deucher 已提交
132
{
133 134
	uint32_t ret;

135
	if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev))
136 137
		return amdgpu_virt_kiq_rreg(adev, reg);

M
Monk Liu 已提交
138
	if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
139
		ret = readl(((void __iomem *)adev->rmmio) + (reg * 4));
A
Alex Deucher 已提交
140 141 142 143 144 145 146 147
	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);
	}
148 149
	trace_amdgpu_mm_rreg(adev->pdev->device, reg, ret);
	return ret;
A
Alex Deucher 已提交
150 151
}

152 153 154 155 156 157
/*
 * MMIO register read with bytes helper functions
 * @offset:bytes offset from MMIO start
 *
*/

158 159 160 161 162 163 164 165
/**
 * 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.
 */
166 167 168 169 170 171 172 173 174 175 176 177
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
 *
*/
178 179 180 181 182 183 184 185 186
/**
 * 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.
 */
187 188 189 190 191 192 193
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();
}

194 195 196 197 198 199 200 201 202 203
/**
 * 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
204
void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v,
M
Monk Liu 已提交
205
		    uint32_t acc_flags)
A
Alex Deucher 已提交
206
{
207
	trace_amdgpu_mm_wreg(adev->pdev->device, reg, v);
208

209 210 211 212
	if (adev->asic_type >= CHIP_VEGA10 && reg == 0) {
		adev->last_mm_index = v;
	}

213
	if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev))
214 215
		return amdgpu_virt_kiq_wreg(adev, reg, v);

M
Monk Liu 已提交
216
	if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
A
Alex Deucher 已提交
217 218 219 220 221 222 223 224 225
		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);
	}
226 227 228 229

	if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) {
		udelay(500);
	}
A
Alex Deucher 已提交
230 231
}

232 233 234 235 236 237 238 239
/**
 * 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
240 241 242 243 244 245 246 247 248 249
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));
	}
}

250 251 252 253 254 255 256 257 258
/**
 * 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
259 260
void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
{
261 262 263
	if (adev->asic_type >= CHIP_VEGA10 && reg == 0) {
		adev->last_mm_index = v;
	}
A
Alex Deucher 已提交
264 265 266 267 268 269 270

	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));
	}
271 272 273 274

	if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) {
		udelay(500);
	}
A
Alex Deucher 已提交
275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314
}

/**
 * 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);
	}
}

315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
/**
 * 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
353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426
/**
 * 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();
}

427 428 429 430 431 432 433 434
/**
 * 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.
 */
435
static int amdgpu_device_vram_scratch_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
436
{
437 438 439 440 441
	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
442 443
}

444 445 446 447 448 449 450
/**
 * amdgpu_device_vram_scratch_fini - Free the VRAM scratch page
 *
 * @adev: amdgpu device pointer
 *
 * Frees the VRAM scratch page.
 */
451
static void amdgpu_device_vram_scratch_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
452
{
453
	amdgpu_bo_free_kernel(&adev->vram_scratch.robj, NULL, NULL);
A
Alex Deucher 已提交
454 455 456
}

/**
457
 * amdgpu_device_program_register_sequence - program an array of registers.
A
Alex Deucher 已提交
458 459 460 461 462 463 464 465
 *
 * @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.
 */
466 467 468
void amdgpu_device_program_register_sequence(struct amdgpu_device *adev,
					     const u32 *registers,
					     const u32 array_size)
A
Alex Deucher 已提交
469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491
{
	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);
	}
}

492 493 494 495 496 497 498 499
/**
 * 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.
 */
500
void amdgpu_device_pci_config_reset(struct amdgpu_device *adev)
A
Alex Deucher 已提交
501 502 503 504 505 506 507 508
{
	pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA);
}

/*
 * GPU doorbell aperture helpers function.
 */
/**
509
 * amdgpu_device_doorbell_init - Init doorbell driver information.
A
Alex Deucher 已提交
510 511 512 513 514 515
 *
 * @adev: amdgpu_device pointer
 *
 * Init doorbell driver information (CIK)
 * Returns 0 on success, error on failure.
 */
516
static int amdgpu_device_doorbell_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
517
{
518 519
	amdgpu_asic_init_doorbell_index(adev);

520 521 522 523 524 525 526 527 528
	/* 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;
	}

529 530 531
	if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET)
		return -EINVAL;

A
Alex Deucher 已提交
532 533 534 535
	/* doorbell bar mapping */
	adev->doorbell.base = pci_resource_start(adev->pdev, 2);
	adev->doorbell.size = pci_resource_len(adev->pdev, 2);

536
	adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32),
537
					     adev->doorbell_index.max_assignment+1);
A
Alex Deucher 已提交
538 539 540
	if (adev->doorbell.num_doorbells == 0)
		return -EINVAL;

541
	/* For Vega, reserve and map two pages on doorbell BAR since SDMA
542 543 544 545
	 * paging queue doorbell use the second page. The
	 * AMDGPU_DOORBELL64_MAX_ASSIGNMENT definition assumes all the
	 * doorbells are in the first page. So with paging queue enabled,
	 * the max num_doorbells should + 1 page (0x400 in dword)
546 547
	 */
	if (adev->asic_type >= CHIP_VEGA10)
548
		adev->doorbell.num_doorbells += 0x400;
549

550 551 552 553
	adev->doorbell.ptr = ioremap(adev->doorbell.base,
				     adev->doorbell.num_doorbells *
				     sizeof(u32));
	if (adev->doorbell.ptr == NULL)
A
Alex Deucher 已提交
554 555 556 557 558 559
		return -ENOMEM;

	return 0;
}

/**
560
 * amdgpu_device_doorbell_fini - Tear down doorbell driver information.
A
Alex Deucher 已提交
561 562 563 564 565
 *
 * @adev: amdgpu_device pointer
 *
 * Tear down doorbell driver information (CIK)
 */
566
static void amdgpu_device_doorbell_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
567 568 569 570 571
{
	iounmap(adev->doorbell.ptr);
	adev->doorbell.ptr = NULL;
}

572

A
Alex Deucher 已提交
573 574

/*
575
 * amdgpu_device_wb_*()
576
 * Writeback is the method by which the GPU updates special pages in memory
A
Alex Xie 已提交
577
 * with the status of certain GPU events (fences, ring pointers,etc.).
A
Alex Deucher 已提交
578 579 580
 */

/**
581
 * amdgpu_device_wb_fini - Disable Writeback and free memory
A
Alex Deucher 已提交
582 583 584 585 586 587
 *
 * @adev: amdgpu_device pointer
 *
 * Disables Writeback and frees the Writeback memory (all asics).
 * Used at driver shutdown.
 */
588
static void amdgpu_device_wb_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
589 590
{
	if (adev->wb.wb_obj) {
591 592 593
		amdgpu_bo_free_kernel(&adev->wb.wb_obj,
				      &adev->wb.gpu_addr,
				      (void **)&adev->wb.wb);
A
Alex Deucher 已提交
594 595 596 597 598
		adev->wb.wb_obj = NULL;
	}
}

/**
599
 * amdgpu_device_wb_init- Init Writeback driver info and allocate memory
A
Alex Deucher 已提交
600 601 602
 *
 * @adev: amdgpu_device pointer
 *
603
 * Initializes writeback and allocates writeback memory (all asics).
A
Alex Deucher 已提交
604 605 606
 * Used at driver startup.
 * Returns 0 on success or an -error on failure.
 */
607
static int amdgpu_device_wb_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
608 609 610 611
{
	int r;

	if (adev->wb.wb_obj == NULL) {
612 613
		/* 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,
614 615 616
					    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
					    &adev->wb.wb_obj, &adev->wb.gpu_addr,
					    (void **)&adev->wb.wb);
A
Alex Deucher 已提交
617 618 619 620 621 622 623 624 625
		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
626
		memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t) * 8);
A
Alex Deucher 已提交
627 628 629 630 631 632
	}

	return 0;
}

/**
633
 * amdgpu_device_wb_get - Allocate a wb entry
A
Alex Deucher 已提交
634 635 636 637 638 639 640
 *
 * @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.
 */
641
int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb)
A
Alex Deucher 已提交
642 643 644
{
	unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb);

645
	if (offset < adev->wb.num_wb) {
K
Ken Wang 已提交
646
		__set_bit(offset, adev->wb.used);
M
Monk Liu 已提交
647
		*wb = offset << 3; /* convert to dw offset */
648 649 650 651 652 653
		return 0;
	} else {
		return -EINVAL;
	}
}

A
Alex Deucher 已提交
654
/**
655
 * amdgpu_device_wb_free - Free a wb entry
A
Alex Deucher 已提交
656 657 658 659 660 661
 *
 * @adev: amdgpu_device pointer
 * @wb: wb index
 *
 * Free a wb slot allocated for use by the driver (all asics)
 */
662
void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb)
A
Alex Deucher 已提交
663
{
M
Monk Liu 已提交
664
	wb >>= 3;
A
Alex Deucher 已提交
665
	if (wb < adev->wb.num_wb)
M
Monk Liu 已提交
666
		__clear_bit(wb, adev->wb.used);
A
Alex Deucher 已提交
667 668
}

669 670 671 672 673 674 675 676 677 678 679
/**
 * 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)
{
680
	u64 space_needed = roundup_pow_of_two(adev->gmc.real_vram_size);
681
	u32 rbar_size = order_base_2(((space_needed >> 20) | 1)) - 1;
682 683 684
	struct pci_bus *root;
	struct resource *res;
	unsigned i;
685 686 687
	u16 cmd;
	int r;

688 689 690 691
	/* Bypass for VF */
	if (amdgpu_sriov_vf(adev))
		return 0;

692 693 694 695 696 697
	/* 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) {
698
		if (res && res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) &&
699 700 701 702 703 704 705 706
		    res->start > 0x100000000ull)
			break;
	}

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

707 708 709 710 711 712
	/* 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. */
713
	amdgpu_device_doorbell_fini(adev);
714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729
	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.
	 */
730
	r = amdgpu_device_doorbell_init(adev);
731 732 733 734 735 736 737
	if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET))
		return -ENODEV;

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

	return 0;
}
738

A
Alex Deucher 已提交
739 740 741 742
/*
 * GPU helpers function.
 */
/**
A
Alex Deucher 已提交
743
 * amdgpu_device_need_post - check if the hw need post or not
A
Alex Deucher 已提交
744 745 746
 *
 * @adev: amdgpu_device pointer
 *
747 748 749
 * 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
750
 */
A
Alex Deucher 已提交
751
bool amdgpu_device_need_post(struct amdgpu_device *adev)
A
Alex Deucher 已提交
752 753 754
{
	uint32_t reg;

755 756 757 758
	if (amdgpu_sriov_vf(adev))
		return false;

	if (amdgpu_passthrough(adev)) {
M
Monk Liu 已提交
759 760 761 762
		/* 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
763 764 765 766 767 768 769 770 771 772
		 */
		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
773 774
			if (fw_ver < 0x00160e00)
				return true;
775 776
		}
	}
777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793

	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;
794 795
}

A
Alex Deucher 已提交
796 797
/* if we get transitioned to only one device, take VGA back */
/**
798
 * amdgpu_device_vga_set_decode - enable/disable vga decode
A
Alex Deucher 已提交
799 800 801 802 803 804 805
 *
 * @cookie: amdgpu_device pointer
 * @state: enable/disable vga decode
 *
 * Enable/disable vga decode (all asics).
 * Returns VGA resource flags.
 */
806
static unsigned int amdgpu_device_vga_set_decode(void *cookie, bool state)
A
Alex Deucher 已提交
807 808 809 810 811 812 813 814 815 816
{
	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;
}

817 818 819 820 821 822 823 824 825 826
/**
 * 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.
 */
827
static void amdgpu_device_check_block_size(struct amdgpu_device *adev)
828 829 830 831
{
	/* 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 */
832 833
	if (amdgpu_vm_block_size == -1)
		return;
834

835
	if (amdgpu_vm_block_size < 9) {
836 837
		dev_warn(adev->dev, "VM page table size (%d) too small\n",
			 amdgpu_vm_block_size);
838
		amdgpu_vm_block_size = -1;
839 840 841
	}
}

842 843 844 845 846 847 848 849
/**
 * 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.
 */
850
static void amdgpu_device_check_vm_size(struct amdgpu_device *adev)
851
{
852 853 854 855
	/* no need to check the default value */
	if (amdgpu_vm_size == -1)
		return;

856 857 858
	if (amdgpu_vm_size < 1) {
		dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n",
			 amdgpu_vm_size);
859
		amdgpu_vm_size = -1;
860 861 862
	}
}

863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902
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
903
/**
904
 * amdgpu_device_check_arguments - validate module params
A
Alex Deucher 已提交
905 906 907 908 909 910
 *
 * @adev: amdgpu_device pointer
 *
 * Validates certain module parameters and updates
 * the associated values used by the driver (all asics).
 */
911
static void amdgpu_device_check_arguments(struct amdgpu_device *adev)
A
Alex Deucher 已提交
912
{
913 914 915 916
	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;
917
	} else if (!is_power_of_2(amdgpu_sched_jobs)){
918 919 920 921
		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
922

923
	if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) {
924 925 926
		/* gart size must be greater or equal to 32M */
		dev_warn(adev->dev, "gart size (%d) too small\n",
			 amdgpu_gart_size);
927
		amdgpu_gart_size = -1;
A
Alex Deucher 已提交
928 929
	}

930
	if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) {
931
		/* gtt size must be greater or equal to 32M */
932 933 934
		dev_warn(adev->dev, "gtt size (%d) too small\n",
				 amdgpu_gtt_size);
		amdgpu_gtt_size = -1;
A
Alex Deucher 已提交
935 936
	}

937 938 939 940 941 942 943
	/* 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;
	}

944 945
	amdgpu_device_check_smu_prv_buffer_size(adev);

946
	amdgpu_device_check_vm_size(adev);
A
Alex Deucher 已提交
947

948
	amdgpu_device_check_block_size(adev);
C
Christian König 已提交
949

950
	if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 ||
951
	    !is_power_of_2(amdgpu_vram_page_split))) {
C
Christian König 已提交
952 953 954 955
		dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
			 amdgpu_vram_page_split);
		amdgpu_vram_page_split = 1024;
	}
956 957 958 959 960

	if (amdgpu_lockup_timeout == 0) {
		dev_warn(adev->dev, "lockup_timeout msut be > 0, adjusting to 10000\n");
		amdgpu_lockup_timeout = 10000;
	}
961 962

	adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type);
A
Alex Deucher 已提交
963 964 965 966 967 968
}

/**
 * amdgpu_switcheroo_set_state - set switcheroo state
 *
 * @pdev: pci dev pointer
969
 * @state: vga_switcheroo state
A
Alex Deucher 已提交
970 971 972 973 974 975 976 977 978 979 980 981
 *
 * 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) {
982
		pr_info("amdgpu: switched on\n");
A
Alex Deucher 已提交
983 984 985
		/* don't suspend or resume card normally */
		dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;

986
		amdgpu_device_resume(dev, true, true);
A
Alex Deucher 已提交
987 988 989 990

		dev->switch_power_state = DRM_SWITCH_POWER_ON;
		drm_kms_helper_poll_enable(dev);
	} else {
991
		pr_info("amdgpu: switched off\n");
A
Alex Deucher 已提交
992 993
		drm_kms_helper_poll_disable(dev);
		dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
994
		amdgpu_device_suspend(dev, true, true);
A
Alex Deucher 已提交
995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025
		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,
};

1026 1027 1028
/**
 * amdgpu_device_ip_set_clockgating_state - set the CG state
 *
1029
 * @dev: amdgpu_device pointer
1030 1031 1032 1033 1034 1035 1036
 * @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.
 */
1037
int amdgpu_device_ip_set_clockgating_state(void *dev,
1038 1039
					   enum amd_ip_block_type block_type,
					   enum amd_clockgating_state state)
A
Alex Deucher 已提交
1040
{
1041
	struct amdgpu_device *adev = dev;
A
Alex Deucher 已提交
1042 1043 1044
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1045
		if (!adev->ip_blocks[i].status.valid)
1046
			continue;
1047 1048 1049 1050 1051 1052 1053 1054 1055
		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
1056 1057 1058 1059
	}
	return r;
}

1060 1061 1062
/**
 * amdgpu_device_ip_set_powergating_state - set the PG state
 *
1063
 * @dev: amdgpu_device pointer
1064 1065 1066 1067 1068 1069 1070
 * @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.
 */
1071
int amdgpu_device_ip_set_powergating_state(void *dev,
1072 1073
					   enum amd_ip_block_type block_type,
					   enum amd_powergating_state state)
A
Alex Deucher 已提交
1074
{
1075
	struct amdgpu_device *adev = dev;
A
Alex Deucher 已提交
1076 1077 1078
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1079
		if (!adev->ip_blocks[i].status.valid)
1080
			continue;
1081 1082 1083 1084 1085 1086 1087 1088 1089
		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
1090 1091 1092 1093
	}
	return r;
}

1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104
/**
 * 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.
 */
1105 1106
void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev,
					    u32 *flags)
1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117
{
	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);
	}
}

1118 1119 1120 1121 1122 1123 1124 1125 1126
/**
 * 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.
 */
1127 1128
int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev,
				   enum amd_ip_block_type block_type)
1129 1130 1131 1132
{
	int i, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1133
		if (!adev->ip_blocks[i].status.valid)
1134
			continue;
1135 1136
		if (adev->ip_blocks[i].version->type == block_type) {
			r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev);
1137 1138 1139 1140 1141 1142 1143 1144 1145
			if (r)
				return r;
			break;
		}
	}
	return 0;

}

1146 1147 1148 1149 1150 1151 1152 1153 1154
/**
 * 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.
 */
1155 1156
bool amdgpu_device_ip_is_idle(struct amdgpu_device *adev,
			      enum amd_ip_block_type block_type)
1157 1158 1159 1160
{
	int i;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1161
		if (!adev->ip_blocks[i].status.valid)
1162
			continue;
1163 1164
		if (adev->ip_blocks[i].version->type == block_type)
			return adev->ip_blocks[i].version->funcs->is_idle((void *)adev);
1165 1166 1167 1168 1169
	}
	return true;

}

1170 1171 1172 1173
/**
 * amdgpu_device_ip_get_ip_block - get a hw IP pointer
 *
 * @adev: amdgpu_device pointer
1174
 * @type: Type of hardware IP (SMU, GFX, UVD, etc.)
1175 1176 1177 1178
 *
 * Returns a pointer to the hardware IP block structure
 * if it exists for the asic, otherwise NULL.
 */
1179 1180 1181
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
1182 1183 1184 1185
{
	int i;

	for (i = 0; i < adev->num_ip_blocks; i++)
1186
		if (adev->ip_blocks[i].version->type == type)
A
Alex Deucher 已提交
1187 1188 1189 1190 1191 1192
			return &adev->ip_blocks[i];

	return NULL;
}

/**
1193
 * amdgpu_device_ip_block_version_cmp
A
Alex Deucher 已提交
1194 1195
 *
 * @adev: amdgpu_device pointer
1196
 * @type: enum amd_ip_block_type
A
Alex Deucher 已提交
1197 1198 1199 1200 1201 1202
 * @major: major version
 * @minor: minor version
 *
 * return 0 if equal or greater
 * return 1 if smaller or the ip_block doesn't exist
 */
1203 1204 1205
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
1206
{
1207
	struct amdgpu_ip_block *ip_block = amdgpu_device_ip_get_ip_block(adev, type);
A
Alex Deucher 已提交
1208

1209 1210 1211
	if (ip_block && ((ip_block->version->major > major) ||
			((ip_block->version->major == major) &&
			(ip_block->version->minor >= minor))))
A
Alex Deucher 已提交
1212 1213 1214 1215 1216
		return 0;

	return 1;
}

1217
/**
1218
 * amdgpu_device_ip_block_add
1219 1220 1221 1222 1223 1224 1225
 *
 * @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.
 */
1226 1227
int amdgpu_device_ip_block_add(struct amdgpu_device *adev,
			       const struct amdgpu_ip_block_version *ip_block_version)
1228 1229 1230 1231
{
	if (!ip_block_version)
		return -EINVAL;

1232
	DRM_INFO("add ip block number %d <%s>\n", adev->num_ip_blocks,
1233 1234
		  ip_block_version->funcs->name);

1235 1236 1237 1238 1239
	adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version;

	return 0;
}

1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
/**
 * 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.
 */
1252
static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev)
1253 1254 1255 1256 1257 1258
{
	adev->enable_virtual_display = false;

	if (amdgpu_virtual_display) {
		struct drm_device *ddev = adev->ddev;
		const char *pci_address_name = pci_name(ddev->pdev);
1259
		char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname;
1260 1261 1262

		pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL);
		pciaddstr_tmp = pciaddstr;
1263 1264
		while ((pciaddname_tmp = strsep(&pciaddstr_tmp, ";"))) {
			pciaddname = strsep(&pciaddname_tmp, ",");
1265 1266
			if (!strcmp("all", pciaddname)
			    || !strcmp(pci_address_name, pciaddname)) {
1267 1268 1269
				long num_crtc;
				int res = -1;

1270
				adev->enable_virtual_display = true;
1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284

				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;
				}
1285 1286 1287 1288
				break;
			}
		}

1289 1290 1291
		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);
1292 1293 1294 1295 1296

		kfree(pciaddstr);
	}
}

1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
/**
 * 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.
 */
1307 1308 1309 1310 1311 1312 1313
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;

1314 1315
	adev->firmware.gpu_info_fw = NULL;

1316 1317 1318 1319 1320
	switch (adev->asic_type) {
	case CHIP_TOPAZ:
	case CHIP_TONGA:
	case CHIP_FIJI:
	case CHIP_POLARIS10:
1321
	case CHIP_POLARIS11:
1322
	case CHIP_POLARIS12:
1323
	case CHIP_VEGAM:
1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
	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
1340
	case CHIP_VEGA20:
1341 1342 1343 1344 1345
	default:
		return 0;
	case CHIP_VEGA10:
		chip_name = "vega10";
		break;
1346 1347 1348
	case CHIP_VEGA12:
		chip_name = "vega12";
		break;
1349
	case CHIP_RAVEN:
1350 1351
		if (adev->rev_id >= 8)
			chip_name = "raven2";
1352 1353
		else if (adev->pdev->device == 0x15d8)
			chip_name = "picasso";
1354 1355
		else
			chip_name = "raven";
1356
		break;
1357 1358 1359
	}

	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name);
1360
	err = request_firmware(&adev->firmware.gpu_info_fw, fw_name, adev->dev);
1361 1362 1363 1364 1365 1366
	if (err) {
		dev_err(adev->dev,
			"Failed to load gpu_info firmware \"%s\"\n",
			fw_name);
		goto out;
	}
1367
	err = amdgpu_ucode_validate(adev->firmware.gpu_info_fw);
1368 1369 1370 1371 1372 1373 1374
	if (err) {
		dev_err(adev->dev,
			"Failed to validate gpu_info firmware \"%s\"\n",
			fw_name);
		goto out;
	}

1375
	hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data;
1376 1377 1378 1379 1380 1381
	amdgpu_ucode_print_gpu_info_hdr(&hdr->header);

	switch (hdr->version_major) {
	case 1:
	{
		const struct gpu_info_firmware_v1_0 *gpu_info_fw =
1382
			(const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data +
1383 1384
								le32_to_cpu(hdr->header.ucode_array_offset_bytes));

1385 1386 1387 1388
		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);
1389
		adev->gfx.config.max_texture_channel_caches =
1390 1391 1392 1393 1394
			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);
1395
		adev->gfx.config.double_offchip_lds_buf =
1396 1397
			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);
1398 1399 1400 1401 1402
		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);
1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414
		break;
	}
	default:
		dev_err(adev->dev,
			"Unsupported gpu_info table %d\n", hdr->header.ucode_version);
		err = -EINVAL;
		goto out;
	}
out:
	return err;
}

1415 1416 1417 1418 1419 1420 1421 1422 1423 1424
/**
 * 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.
 */
1425
static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1426
{
1427
	int i, r;
A
Alex Deucher 已提交
1428

1429
	amdgpu_device_enable_virtual_display(adev);
1430

A
Alex Deucher 已提交
1431
	switch (adev->asic_type) {
1432 1433
	case CHIP_TOPAZ:
	case CHIP_TONGA:
1434
	case CHIP_FIJI:
1435
	case CHIP_POLARIS10:
1436
	case CHIP_POLARIS11:
1437
	case CHIP_POLARIS12:
1438
	case CHIP_VEGAM:
1439
	case CHIP_CARRIZO:
1440 1441
	case CHIP_STONEY:
		if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY)
1442 1443 1444 1445 1446 1447 1448 1449
			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
1450 1451 1452 1453 1454 1455
#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
1456
		adev->family = AMDGPU_FAMILY_SI;
K
Ken Wang 已提交
1457 1458 1459 1460 1461
		r = si_set_ip_blocks(adev);
		if (r)
			return r;
		break;
#endif
1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477
#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
1478 1479
	case CHIP_VEGA10:
	case CHIP_VEGA12:
1480
	case CHIP_VEGA20:
1481
	case CHIP_RAVEN:
1482
		if (adev->asic_type == CHIP_RAVEN)
1483 1484 1485
			adev->family = AMDGPU_FAMILY_RV;
		else
			adev->family = AMDGPU_FAMILY_AI;
1486 1487 1488 1489 1490

		r = soc15_set_ip_blocks(adev);
		if (r)
			return r;
		break;
A
Alex Deucher 已提交
1491 1492 1493 1494 1495
	default:
		/* FIXME: not supported yet */
		return -EINVAL;
	}

1496 1497 1498 1499
	r = amdgpu_device_parse_gpu_info_fw(adev);
	if (r)
		return r;

1500 1501
	amdgpu_amdkfd_device_probe(adev);

1502 1503 1504
	if (amdgpu_sriov_vf(adev)) {
		r = amdgpu_virt_request_full_gpu(adev, true);
		if (r)
1505
			return -EAGAIN;
1506 1507
	}

1508 1509
	adev->powerplay.pp_feature = amdgpu_pp_feature_mask;

A
Alex Deucher 已提交
1510 1511
	for (i = 0; i < adev->num_ip_blocks; i++) {
		if ((amdgpu_ip_block_mask & (1 << i)) == 0) {
1512 1513
			DRM_ERROR("disabled ip block: %d <%s>\n",
				  i, adev->ip_blocks[i].version->funcs->name);
1514
			adev->ip_blocks[i].status.valid = false;
A
Alex Deucher 已提交
1515
		} else {
1516 1517
			if (adev->ip_blocks[i].version->funcs->early_init) {
				r = adev->ip_blocks[i].version->funcs->early_init((void *)adev);
1518
				if (r == -ENOENT) {
1519
					adev->ip_blocks[i].status.valid = false;
1520
				} else if (r) {
1521 1522
					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
1523
					return r;
1524
				} else {
1525
					adev->ip_blocks[i].status.valid = true;
1526
				}
1527
			} else {
1528
				adev->ip_blocks[i].status.valid = true;
A
Alex Deucher 已提交
1529 1530 1531 1532
			}
		}
	}

1533 1534 1535
	adev->cg_flags &= amdgpu_cg_mask;
	adev->pg_flags &= amdgpu_pg_mask;

A
Alex Deucher 已提交
1536 1537 1538
	return 0;
}

1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583
static int amdgpu_device_ip_hw_init_phase1(struct amdgpu_device *adev)
{
	int i, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
		if (!adev->ip_blocks[i].status.sw)
			continue;
		if (adev->ip_blocks[i].status.hw)
			continue;
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) {
			r = adev->ip_blocks[i].version->funcs->hw_init(adev);
			if (r) {
				DRM_ERROR("hw_init of IP block <%s> failed %d\n",
					  adev->ip_blocks[i].version->funcs->name, r);
				return r;
			}
			adev->ip_blocks[i].status.hw = true;
		}
	}

	return 0;
}

static int amdgpu_device_ip_hw_init_phase2(struct amdgpu_device *adev)
{
	int i, r;

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

	return 0;
}

1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
static int amdgpu_device_fw_loading(struct amdgpu_device *adev)
{
	int r = 0;
	int i;

	if (adev->asic_type >= CHIP_VEGA10) {
		for (i = 0; i < adev->num_ip_blocks; i++) {
			if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) {
				if (adev->in_gpu_reset || adev->in_suspend) {
					if (amdgpu_sriov_vf(adev) && adev->in_gpu_reset)
						break; /* sriov gpu reset, psp need to do hw_init before IH because of hw limit */
					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;
					}
				} else {
					r = adev->ip_blocks[i].version->funcs->hw_init(adev);
					if (r) {
						DRM_ERROR("hw_init of IP block <%s> failed %d\n",
						  adev->ip_blocks[i].version->funcs->name, r);
						return r;
					}
				}
				adev->ip_blocks[i].status.hw = true;
			}
		}
	}

1614
	if (adev->powerplay.pp_funcs && adev->powerplay.pp_funcs->load_firmware) {
1615 1616 1617 1618 1619 1620 1621 1622 1623 1624
		r = adev->powerplay.pp_funcs->load_firmware(adev->powerplay.pp_handle);
		if (r) {
			pr_err("firmware loading failed\n");
			return r;
		}
	}

	return 0;
}

1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
/**
 * 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.
 */
1636
static int amdgpu_device_ip_init(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1637 1638 1639 1640
{
	int i, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1641
		if (!adev->ip_blocks[i].status.valid)
A
Alex Deucher 已提交
1642
			continue;
1643
		r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev);
1644
		if (r) {
1645 1646
			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
1647
			return r;
1648
		}
1649
		adev->ip_blocks[i].status.sw = true;
1650

A
Alex Deucher 已提交
1651
		/* need to do gmc hw init early so we can allocate gpu mem */
1652
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
1653
			r = amdgpu_device_vram_scratch_init(adev);
1654 1655
			if (r) {
				DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r);
A
Alex Deucher 已提交
1656
				return r;
1657
			}
1658
			r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev);
1659 1660
			if (r) {
				DRM_ERROR("hw_init %d failed %d\n", i, r);
A
Alex Deucher 已提交
1661
				return r;
1662
			}
1663
			r = amdgpu_device_wb_init(adev);
1664
			if (r) {
1665
				DRM_ERROR("amdgpu_device_wb_init failed %d\n", r);
A
Alex Deucher 已提交
1666
				return r;
1667
			}
1668
			adev->ip_blocks[i].status.hw = true;
M
Monk Liu 已提交
1669 1670 1671

			/* right after GMC hw init, we create CSA */
			if (amdgpu_sriov_vf(adev)) {
R
Rex Zhu 已提交
1672 1673 1674
				r = amdgpu_allocate_static_csa(adev, &adev->virt.csa_obj,
								AMDGPU_GEM_DOMAIN_VRAM,
								AMDGPU_CSA_SIZE);
M
Monk Liu 已提交
1675 1676 1677 1678 1679
				if (r) {
					DRM_ERROR("allocate CSA failed %d\n", r);
					return r;
				}
			}
A
Alex Deucher 已提交
1680 1681 1682
		}
	}

1683 1684 1685
	r = amdgpu_ucode_create_bo(adev); /* create ucode bo when sw_init complete*/
	if (r)
		return r;
1686 1687 1688 1689 1690

	r = amdgpu_device_ip_hw_init_phase1(adev);
	if (r)
		return r;

1691 1692 1693 1694
	r = amdgpu_device_fw_loading(adev);
	if (r)
		return r;

1695 1696 1697
	r = amdgpu_device_ip_hw_init_phase2(adev);
	if (r)
		return r;
A
Alex Deucher 已提交
1698

1699 1700
	if (adev->gmc.xgmi.num_physical_nodes > 1)
		amdgpu_xgmi_add_device(adev);
1701
	amdgpu_amdkfd_device_init(adev);
1702 1703 1704 1705

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

A
Alex Deucher 已提交
1706 1707 1708
	return 0;
}

1709 1710 1711 1712 1713 1714 1715 1716 1717
/**
 * 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.
 */
1718
static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev)
1719 1720 1721 1722
{
	memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM);
}

1723 1724 1725 1726 1727 1728 1729 1730 1731 1732
/**
 * 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.
 */
1733
static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev)
1734 1735 1736 1737 1738
{
	return !!memcmp(adev->gart.ptr, adev->reset_magic,
			AMDGPU_RESET_MAGIC_NUM);
}

1739
/**
1740
 * amdgpu_device_set_cg_state - set clockgating for amdgpu device
1741 1742 1743 1744
 *
 * @adev: amdgpu_device pointer
 *
 * The list of all the hardware IPs that make up the asic is walked and the
1745 1746 1747
 * set_clockgating_state callbacks are run.
 * Late initialization pass enabling clockgating for hardware IPs.
 * Fini or suspend, pass disabling clockgating for hardware IPs.
1748 1749
 * Returns 0 on success, negative error code on failure.
 */
1750

1751 1752
static int amdgpu_device_set_cg_state(struct amdgpu_device *adev,
						enum amd_clockgating_state state)
A
Alex Deucher 已提交
1753
{
1754
	int i, j, r;
A
Alex Deucher 已提交
1755

1756 1757 1758
	if (amdgpu_emu_mode == 1)
		return 0;

1759 1760
	for (j = 0; j < adev->num_ip_blocks; j++) {
		i = state == AMD_CG_STATE_GATE ? j : adev->num_ip_blocks - j - 1;
1761
		if (!adev->ip_blocks[i].status.late_initialized)
A
Alex Deucher 已提交
1762
			continue;
1763
		/* skip CG for VCE/UVD, it's handled specially */
1764
		if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
1765
		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
1766
		    adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
1767
		    adev->ip_blocks[i].version->funcs->set_clockgating_state) {
1768
			/* enable clockgating to save power */
1769
			r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
1770
										     state);
1771 1772
			if (r) {
				DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n",
1773
					  adev->ip_blocks[i].version->funcs->name, r);
1774 1775
				return r;
			}
1776
		}
A
Alex Deucher 已提交
1777
	}
1778

1779 1780 1781
	return 0;
}

1782
static int amdgpu_device_set_pg_state(struct amdgpu_device *adev, enum amd_powergating_state state)
1783
{
1784
	int i, j, r;
1785

1786 1787 1788
	if (amdgpu_emu_mode == 1)
		return 0;

1789 1790
	for (j = 0; j < adev->num_ip_blocks; j++) {
		i = state == AMD_PG_STATE_GATE ? j : adev->num_ip_blocks - j - 1;
1791
		if (!adev->ip_blocks[i].status.late_initialized)
1792 1793 1794 1795 1796 1797 1798 1799
			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,
1800
											state);
1801 1802 1803 1804 1805 1806 1807
			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;
			}
		}
	}
1808 1809 1810
	return 0;
}

1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822
/**
 * 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.
 */
1823
static int amdgpu_device_ip_late_init(struct amdgpu_device *adev)
1824 1825 1826 1827
{
	int i = 0, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
1828
		if (!adev->ip_blocks[i].status.hw)
1829 1830 1831 1832 1833 1834 1835 1836 1837
			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;
			}
		}
1838
		adev->ip_blocks[i].status.late_initialized = true;
1839 1840
	}

1841 1842
	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE);
	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE);
1843

1844 1845
	queue_delayed_work(system_wq, &adev->late_init_work,
			   msecs_to_jiffies(AMDGPU_RESUME_MS));
A
Alex Deucher 已提交
1846

1847
	amdgpu_device_fill_reset_magic(adev);
A
Alex Deucher 已提交
1848 1849 1850 1851

	return 0;
}

1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862
/**
 * 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.
 */
1863
static int amdgpu_device_ip_fini(struct amdgpu_device *adev)
A
Alex Deucher 已提交
1864 1865 1866
{
	int i, r;

1867
	amdgpu_amdkfd_device_fini(adev);
1868 1869

	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
1870 1871
	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);

1872 1873
	/* need to disable SMC first */
	for (i = 0; i < adev->num_ip_blocks; i++) {
1874
		if (!adev->ip_blocks[i].status.hw)
1875
			continue;
1876
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) {
1877
			r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
1878 1879 1880
			/* XXX handle errors */
			if (r) {
				DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
1881
					  adev->ip_blocks[i].version->funcs->name, r);
1882
			}
1883
			adev->ip_blocks[i].status.hw = false;
1884 1885 1886 1887
			break;
		}
	}

A
Alex Deucher 已提交
1888
	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
1889
		if (!adev->ip_blocks[i].status.hw)
A
Alex Deucher 已提交
1890
			continue;
1891

1892
		r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
A
Alex Deucher 已提交
1893
		/* XXX handle errors */
1894
		if (r) {
1895 1896
			DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
1897
		}
1898

1899
		adev->ip_blocks[i].status.hw = false;
A
Alex Deucher 已提交
1900 1901
	}

1902

A
Alex Deucher 已提交
1903
	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
1904
		if (!adev->ip_blocks[i].status.sw)
A
Alex Deucher 已提交
1905
			continue;
1906 1907

		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
1908
			amdgpu_ucode_free_bo(adev);
R
Rex Zhu 已提交
1909
			amdgpu_free_static_csa(&adev->virt.csa_obj);
1910 1911 1912 1913
			amdgpu_device_wb_fini(adev);
			amdgpu_device_vram_scratch_fini(adev);
		}

1914
		r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev);
A
Alex Deucher 已提交
1915
		/* XXX handle errors */
1916
		if (r) {
1917 1918
			DRM_DEBUG("sw_fini of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
1919
		}
1920 1921
		adev->ip_blocks[i].status.sw = false;
		adev->ip_blocks[i].status.valid = false;
A
Alex Deucher 已提交
1922 1923
	}

M
Monk Liu 已提交
1924
	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
1925
		if (!adev->ip_blocks[i].status.late_initialized)
1926
			continue;
1927 1928 1929
		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
1930 1931
	}

1932
	if (amdgpu_sriov_vf(adev))
1933 1934
		if (amdgpu_virt_release_full_gpu(adev, false))
			DRM_ERROR("failed to release exclusive mode on fini\n");
M
Monk Liu 已提交
1935

A
Alex Deucher 已提交
1936 1937 1938
	return 0;
}

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
static int amdgpu_device_enable_mgpu_fan_boost(void)
{
	struct amdgpu_gpu_instance *gpu_ins;
	struct amdgpu_device *adev;
	int i, ret = 0;

	mutex_lock(&mgpu_info.mutex);

	/*
	 * MGPU fan boost feature should be enabled
	 * only when there are two or more dGPUs in
	 * the system
	 */
	if (mgpu_info.num_dgpu < 2)
		goto out;

	for (i = 0; i < mgpu_info.num_dgpu; i++) {
		gpu_ins = &(mgpu_info.gpu_ins[i]);
		adev = gpu_ins->adev;
		if (!(adev->flags & AMD_IS_APU) &&
		    !gpu_ins->mgpu_fan_enabled &&
		    adev->powerplay.pp_funcs &&
		    adev->powerplay.pp_funcs->enable_mgpu_fan_boost) {
			ret = amdgpu_dpm_enable_mgpu_fan_boost(adev);
			if (ret)
				break;

			gpu_ins->mgpu_fan_enabled = 1;
		}
	}

out:
	mutex_unlock(&mgpu_info.mutex);

	return ret;
}

1976
/**
1977
 * amdgpu_device_ip_late_init_func_handler - work handler for ib test
1978
 *
1979
 * @work: work_struct.
1980
 */
1981
static void amdgpu_device_ip_late_init_func_handler(struct work_struct *work)
1982 1983 1984
{
	struct amdgpu_device *adev =
		container_of(work, struct amdgpu_device, late_init_work.work);
1985 1986 1987 1988 1989
	int r;

	r = amdgpu_ib_ring_tests(adev);
	if (r)
		DRM_ERROR("ib ring test failed (%d).\n", r);
1990 1991 1992 1993

	r = amdgpu_device_enable_mgpu_fan_boost();
	if (r)
		DRM_ERROR("enable mgpu fan boost failed (%d).\n", r);
1994 1995
}

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work)
{
	struct amdgpu_device *adev =
		container_of(work, struct amdgpu_device, gfx.gfx_off_delay_work.work);

	mutex_lock(&adev->gfx.gfx_off_mutex);
	if (!adev->gfx.gfx_off_state && !adev->gfx.gfx_off_req_count) {
		if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true))
			adev->gfx.gfx_off_state = true;
	}
	mutex_unlock(&adev->gfx.gfx_off_mutex);
}

2009
/**
2010
 * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1)
2011 2012 2013 2014 2015 2016 2017 2018 2019
 *
 * @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.
 */
2020 2021 2022 2023
static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev)
{
	int i, r;

2024
	amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
2025
	amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
2026

2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056
	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) {
			/* 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);
			}
		}
	}

	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
2057 2058 2059 2060
{
	int i, r;

	for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
2061
		if (!adev->ip_blocks[i].status.valid)
A
Alex Deucher 已提交
2062
			continue;
2063 2064 2065
		/* displays are handled in phase1 */
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE)
			continue;
A
Alex Deucher 已提交
2066
		/* XXX handle errors */
2067
		r = adev->ip_blocks[i].version->funcs->suspend(adev);
A
Alex Deucher 已提交
2068
		/* XXX handle errors */
2069
		if (r) {
2070 2071
			DRM_ERROR("suspend of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
2072
		}
A
Alex Deucher 已提交
2073 2074 2075 2076 2077
	}

	return 0;
}

2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092
/**
 * 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;

2093 2094 2095
	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_request_full_gpu(adev, false);

2096 2097 2098 2099 2100
	r = amdgpu_device_ip_suspend_phase1(adev);
	if (r)
		return r;
	r = amdgpu_device_ip_suspend_phase2(adev);

2101 2102 2103
	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_release_full_gpu(adev, false);

2104 2105 2106
	return r;
}

2107
static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev)
2108 2109 2110
{
	int i, r;

2111 2112 2113
	static enum amd_ip_block_type ip_order[] = {
		AMD_IP_BLOCK_TYPE_GMC,
		AMD_IP_BLOCK_TYPE_COMMON,
2114
		AMD_IP_BLOCK_TYPE_PSP,
2115 2116
		AMD_IP_BLOCK_TYPE_IH,
	};
2117

2118 2119 2120
	for (i = 0; i < ARRAY_SIZE(ip_order); i++) {
		int j;
		struct amdgpu_ip_block *block;
2121

2122 2123 2124 2125 2126 2127 2128 2129
		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);
2130
			DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded");
2131 2132
			if (r)
				return r;
2133 2134 2135 2136 2137 2138
		}
	}

	return 0;
}

2139
static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev)
2140 2141 2142
{
	int i, r;

2143 2144 2145 2146 2147
	static enum amd_ip_block_type ip_order[] = {
		AMD_IP_BLOCK_TYPE_SMC,
		AMD_IP_BLOCK_TYPE_DCE,
		AMD_IP_BLOCK_TYPE_GFX,
		AMD_IP_BLOCK_TYPE_SDMA,
2148 2149
		AMD_IP_BLOCK_TYPE_UVD,
		AMD_IP_BLOCK_TYPE_VCE
2150
	};
2151

2152 2153 2154
	for (i = 0; i < ARRAY_SIZE(ip_order); i++) {
		int j;
		struct amdgpu_ip_block *block;
2155

2156 2157 2158 2159 2160 2161 2162 2163
		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);
2164
			DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded");
2165 2166
			if (r)
				return r;
2167 2168 2169 2170 2171 2172
		}
	}

	return 0;
}

2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184
/**
 * 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.
 */
2185
static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev)
A
Alex Deucher 已提交
2186 2187 2188
{
	int i, r;

2189 2190 2191 2192
	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 ||
2193 2194
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) {
2195 2196 2197 2198 2199 2200
			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;
			}
2201 2202 2203 2204 2205 2206
		}
	}

	return 0;
}

2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219
/**
 * 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.
 */
2220
static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev)
A
Alex Deucher 已提交
2221 2222 2223 2224
{
	int i, r;

	for (i = 0; i < adev->num_ip_blocks; i++) {
2225
		if (!adev->ip_blocks[i].status.valid)
A
Alex Deucher 已提交
2226
			continue;
2227
		if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
2228
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
2229 2230
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH ||
		    adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP)
2231
			continue;
2232
		r = adev->ip_blocks[i].version->funcs->resume(adev);
2233
		if (r) {
2234 2235
			DRM_ERROR("resume of IP block <%s> failed %d\n",
				  adev->ip_blocks[i].version->funcs->name, r);
A
Alex Deucher 已提交
2236
			return r;
2237
		}
A
Alex Deucher 已提交
2238 2239 2240 2241 2242
	}

	return 0;
}

2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254
/**
 * 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.
 */
2255
static int amdgpu_device_ip_resume(struct amdgpu_device *adev)
2256 2257 2258
{
	int r;

2259
	r = amdgpu_device_ip_resume_phase1(adev);
2260 2261
	if (r)
		return r;
2262 2263 2264 2265 2266

	r = amdgpu_device_fw_loading(adev);
	if (r)
		return r;

2267
	r = amdgpu_device_ip_resume_phase2(adev);
2268 2269 2270 2271

	return r;
}

2272 2273 2274 2275 2276 2277 2278
/**
 * 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.
 */
2279
static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev)
2280
{
M
Monk Liu 已提交
2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291
	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);
2292
	}
2293 2294
}

2295 2296 2297 2298 2299 2300 2301 2302
/**
 * 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.
 */
2303 2304 2305 2306 2307
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:
2308
	case CHIP_KAVERI:
2309 2310
	case CHIP_KABINI:
	case CHIP_MULLINS:
2311 2312 2313 2314 2315 2316 2317 2318 2319
		/*
		 * 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:
2320 2321 2322
	case CHIP_CARRIZO:
	case CHIP_STONEY:
	case CHIP_POLARIS10:
L
Leo Liu 已提交
2323
	case CHIP_POLARIS11:
2324
	case CHIP_POLARIS12:
L
Leo Liu 已提交
2325
	case CHIP_VEGAM:
2326 2327
	case CHIP_TONGA:
	case CHIP_FIJI:
2328
	case CHIP_VEGA10:
2329
	case CHIP_VEGA12:
2330
	case CHIP_VEGA20:
2331
#if defined(CONFIG_DRM_AMD_DC_DCN1_0)
2332
	case CHIP_RAVEN:
2333
#endif
2334
		return amdgpu_dc != 0;
2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349
#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
2350 2351 2352
	if (amdgpu_sriov_vf(adev))
		return false;

2353 2354 2355
	return amdgpu_device_asic_has_dc_support(adev->asic_type);
}

A
Alex Deucher 已提交
2356 2357 2358 2359
/**
 * amdgpu_device_init - initialize the driver
 *
 * @adev: amdgpu_device pointer
2360
 * @ddev: drm dev pointer
A
Alex Deucher 已提交
2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374
 * @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;
2375
	u32 max_MBps;
A
Alex Deucher 已提交
2376 2377 2378 2379 2380 2381

	adev->shutdown = false;
	adev->dev = &pdev->dev;
	adev->ddev = ddev;
	adev->pdev = pdev;
	adev->flags = flags;
2382
	adev->asic_type = flags & AMD_ASIC_MASK;
A
Alex Deucher 已提交
2383
	adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT;
2384 2385
	if (amdgpu_emu_mode == 1)
		adev->usec_timeout *= 2;
2386
	adev->gmc.gart_size = 512 * 1024 * 1024;
A
Alex Deucher 已提交
2387 2388 2389 2390 2391
	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;
2392
	adev->vm_manager.vm_pte_num_rqs = 0;
2393
	adev->gmc.gmc_funcs = NULL;
2394
	adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS);
2395
	bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
A
Alex Deucher 已提交
2396 2397 2398 2399 2400

	adev->smc_rreg = &amdgpu_invalid_rreg;
	adev->smc_wreg = &amdgpu_invalid_wreg;
	adev->pcie_rreg = &amdgpu_invalid_rreg;
	adev->pcie_wreg = &amdgpu_invalid_wreg;
2401 2402
	adev->pciep_rreg = &amdgpu_invalid_rreg;
	adev->pciep_wreg = &amdgpu_invalid_wreg;
A
Alex Deucher 已提交
2403 2404 2405 2406
	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;
2407 2408
	adev->gc_cac_rreg = &amdgpu_invalid_rreg;
	adev->gc_cac_wreg = &amdgpu_invalid_wreg;
A
Alex Deucher 已提交
2409 2410 2411
	adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg;
	adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg;

2412 2413 2414
	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
2415 2416 2417 2418

	/* mutex initialization are all done here so we
	 * can recall function without having locking issues */
	atomic_set(&adev->irq.ih.lock, 0);
2419
	mutex_init(&adev->firmware.mutex);
A
Alex Deucher 已提交
2420 2421 2422
	mutex_init(&adev->pm.mutex);
	mutex_init(&adev->gfx.gpu_clock_mutex);
	mutex_init(&adev->srbm_mutex);
2423
	mutex_init(&adev->gfx.pipe_reserve_mutex);
2424
	mutex_init(&adev->gfx.gfx_off_mutex);
A
Alex Deucher 已提交
2425 2426
	mutex_init(&adev->grbm_idx_mutex);
	mutex_init(&adev->mn_lock);
A
Alex Deucher 已提交
2427
	mutex_init(&adev->virt.vf_errors.lock);
A
Alex Deucher 已提交
2428
	hash_init(adev->mn_hash);
2429
	mutex_init(&adev->lock_reset);
A
Alex Deucher 已提交
2430

2431
	amdgpu_device_check_arguments(adev);
A
Alex Deucher 已提交
2432 2433 2434 2435 2436 2437

	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);
2438
	spin_lock_init(&adev->gc_cac_idx_lock);
2439
	spin_lock_init(&adev->se_cac_idx_lock);
A
Alex Deucher 已提交
2440
	spin_lock_init(&adev->audio_endpt_idx_lock);
2441
	spin_lock_init(&adev->mm_stats.lock);
A
Alex Deucher 已提交
2442

2443 2444 2445
	INIT_LIST_HEAD(&adev->shadow_list);
	mutex_init(&adev->shadow_list_lock);

2446 2447 2448
	INIT_LIST_HEAD(&adev->ring_lru_list);
	spin_lock_init(&adev->ring_lru_list_lock);

2449 2450
	INIT_DELAYED_WORK(&adev->late_init_work,
			  amdgpu_device_ip_late_init_func_handler);
2451 2452
	INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
			  amdgpu_device_delay_enable_gfx_off);
2453

2454
	adev->gfx.gfx_off_req_count = 1;
2455 2456
	adev->pm.ac_power = power_supply_is_system_supplied() > 0 ? true : false;

2457 2458
	/* Registers mapping */
	/* TODO: block userspace mapping of io register */
2459 2460 2461 2462 2463 2464 2465
	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
2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482

	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);

	/* 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)
2483
		DRM_INFO("PCI I/O BAR is not found.\n");
A
Alex Deucher 已提交
2484

2485 2486
	amdgpu_device_get_pcie_info(adev);

A
Alex Deucher 已提交
2487
	/* early init functions */
2488
	r = amdgpu_device_ip_early_init(adev);
A
Alex Deucher 已提交
2489 2490 2491
	if (r)
		return r;

2492 2493 2494
	/* doorbell bar mapping and doorbell index init*/
	amdgpu_device_doorbell_init(adev);

A
Alex Deucher 已提交
2495 2496 2497
	/* 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 */
2498
	vga_client_register(adev->pdev, adev, NULL, amdgpu_device_vga_set_decode);
A
Alex Deucher 已提交
2499

2500
	if (amdgpu_device_is_px(ddev))
A
Alex Deucher 已提交
2501
		runtime = true;
2502 2503 2504
	if (!pci_is_thunderbolt_attached(adev->pdev))
		vga_switcheroo_register_client(adev->pdev,
					       &amdgpu_switcheroo_ops, runtime);
A
Alex Deucher 已提交
2505 2506 2507
	if (runtime)
		vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain);

2508 2509 2510
	if (amdgpu_emu_mode == 1) {
		/* post the asic on emulation mode */
		emu_soc_asic_init(adev);
2511
		goto fence_driver_init;
2512
	}
2513

A
Alex Deucher 已提交
2514
	/* Read BIOS */
2515 2516 2517 2518
	if (!amdgpu_get_bios(adev)) {
		r = -EINVAL;
		goto failed;
	}
2519

A
Alex Deucher 已提交
2520
	r = amdgpu_atombios_init(adev);
2521 2522
	if (r) {
		dev_err(adev->dev, "amdgpu_atombios_init failed\n");
A
Alex Deucher 已提交
2523
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0);
2524
		goto failed;
2525
	}
A
Alex Deucher 已提交
2526

2527 2528
	/* detect if we are with an SRIOV vbios */
	amdgpu_device_detect_sriov_bios(adev);
2529

A
Alex Deucher 已提交
2530
	/* Post card if necessary */
A
Alex Deucher 已提交
2531
	if (amdgpu_device_need_post(adev)) {
A
Alex Deucher 已提交
2532
		if (!adev->bios) {
2533
			dev_err(adev->dev, "no vBIOS found\n");
2534 2535
			r = -EINVAL;
			goto failed;
A
Alex Deucher 已提交
2536
		}
2537
		DRM_INFO("GPU posting now...\n");
2538 2539 2540 2541 2542
		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
2543 2544
	}

2545 2546 2547 2548 2549
	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
2550
			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
2551 2552 2553
			goto failed;
		}
	} else {
2554 2555 2556 2557
		/* 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
2558
			amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
2559
			goto failed;
2560 2561
		}
		/* init i2c buses */
2562 2563
		if (!amdgpu_device_has_dc_support(adev))
			amdgpu_atombios_i2c_init(adev);
2564
	}
A
Alex Deucher 已提交
2565

2566
fence_driver_init:
A
Alex Deucher 已提交
2567 2568
	/* Fence driver */
	r = amdgpu_fence_driver_init(adev);
2569 2570
	if (r) {
		dev_err(adev->dev, "amdgpu_fence_driver_init failed\n");
A
Alex Deucher 已提交
2571
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0);
2572
		goto failed;
2573
	}
A
Alex Deucher 已提交
2574 2575 2576 2577

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

2578
	r = amdgpu_device_ip_init(adev);
A
Alex Deucher 已提交
2579
	if (r) {
2580 2581 2582 2583 2584 2585
		/* 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");
2586 2587 2588
			/* Don't send request since VF is inactive. */
			adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME;
			adev->virt.ops = NULL;
2589 2590 2591
			r = -EAGAIN;
			goto failed;
		}
2592
		dev_err(adev->dev, "amdgpu_device_ip_init failed\n");
A
Alex Deucher 已提交
2593
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
2594
		goto failed;
A
Alex Deucher 已提交
2595 2596 2597 2598
	}

	adev->accel_working = true;

2599 2600
	amdgpu_vm_check_compute_bug(adev);

2601 2602 2603 2604 2605 2606 2607 2608
	/* 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
2609 2610 2611
	r = amdgpu_ib_pool_init(adev);
	if (r) {
		dev_err(adev->dev, "IB initialization failed (%d).\n", r);
A
Alex Deucher 已提交
2612
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r);
2613
		goto failed;
A
Alex Deucher 已提交
2614 2615
	}

2616 2617 2618
	if (amdgpu_sriov_vf(adev))
		amdgpu_virt_init_data_exchange(adev);

2619 2620
	amdgpu_fbdev_init(adev);

2621 2622 2623 2624
	r = amdgpu_pm_sysfs_init(adev);
	if (r)
		DRM_ERROR("registering pm debugfs failed (%d).\n", r);

2625
	r = amdgpu_debugfs_gem_init(adev);
M
Monk Liu 已提交
2626
	if (r)
A
Alex Deucher 已提交
2627 2628 2629
		DRM_ERROR("registering gem debugfs failed (%d).\n", r);

	r = amdgpu_debugfs_regs_init(adev);
M
Monk Liu 已提交
2630
	if (r)
A
Alex Deucher 已提交
2631 2632
		DRM_ERROR("registering register debugfs failed (%d).\n", r);

2633
	r = amdgpu_debugfs_firmware_init(adev);
M
Monk Liu 已提交
2634
	if (r)
2635 2636
		DRM_ERROR("registering firmware debugfs failed (%d).\n", r);

2637
	r = amdgpu_debugfs_init(adev);
2638
	if (r)
2639
		DRM_ERROR("Creating debugfs files failed (%d).\n", r);
2640

A
Alex Deucher 已提交
2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656
	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.
	 */
2657
	r = amdgpu_device_ip_late_init(adev);
2658
	if (r) {
2659
		dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n");
A
Alex Deucher 已提交
2660
		amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
2661
		goto failed;
2662
	}
A
Alex Deucher 已提交
2663 2664

	return 0;
2665 2666

failed:
2667
	amdgpu_vf_error_trans_all(adev);
2668 2669
	if (runtime)
		vga_switcheroo_fini_domain_pm_ops(adev->dev);
2670

2671
	return r;
A
Alex Deucher 已提交
2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687
}

/**
 * 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;
2688 2689
	/* disable all interrupts */
	amdgpu_irq_disable_all(adev);
2690 2691 2692 2693 2694 2695
	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
2696 2697
	amdgpu_ib_pool_fini(adev);
	amdgpu_fence_driver_fini(adev);
2698
	amdgpu_pm_sysfs_fini(adev);
A
Alex Deucher 已提交
2699
	amdgpu_fbdev_fini(adev);
2700
	r = amdgpu_device_ip_fini(adev);
2701 2702 2703 2704
	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
2705
	adev->accel_working = false;
2706
	cancel_delayed_work_sync(&adev->late_init_work);
A
Alex Deucher 已提交
2707
	/* free i2c buses */
2708 2709
	if (!amdgpu_device_has_dc_support(adev))
		amdgpu_i2c_fini(adev);
2710 2711 2712 2713

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

A
Alex Deucher 已提交
2714 2715
	kfree(adev->bios);
	adev->bios = NULL;
2716 2717
	if (!pci_is_thunderbolt_attached(adev->pdev))
		vga_switcheroo_unregister_client(adev->pdev);
2718 2719
	if (adev->flags & AMD_IS_PX)
		vga_switcheroo_fini_domain_pm_ops(adev->dev);
A
Alex Deucher 已提交
2720 2721 2722 2723 2724 2725
	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;
2726
	amdgpu_device_doorbell_fini(adev);
A
Alex Deucher 已提交
2727 2728 2729 2730 2731 2732 2733 2734
	amdgpu_debugfs_regs_cleanup(adev);
}


/*
 * Suspend & resume.
 */
/**
2735
 * amdgpu_device_suspend - initiate device suspend
A
Alex Deucher 已提交
2736
 *
2737 2738 2739
 * @dev: drm dev pointer
 * @suspend: suspend state
 * @fbcon : notify the fbdev of suspend
A
Alex Deucher 已提交
2740 2741 2742 2743 2744
 *
 * Puts the hw in the suspend state (all asics).
 * Returns 0 for success or an error on failure.
 * Called at driver suspend.
 */
2745
int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon)
A
Alex Deucher 已提交
2746 2747 2748 2749
{
	struct amdgpu_device *adev;
	struct drm_crtc *crtc;
	struct drm_connector *connector;
2750
	int r;
A
Alex Deucher 已提交
2751 2752 2753 2754 2755 2756 2757 2758 2759 2760

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

	adev = dev->dev_private;

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

2761
	adev->in_suspend = true;
A
Alex Deucher 已提交
2762 2763
	drm_kms_helper_poll_disable(dev);

2764 2765 2766
	if (fbcon)
		amdgpu_fbdev_set_suspend(adev, 1);

2767 2768
	cancel_delayed_work_sync(&adev->late_init_work);

2769 2770 2771 2772 2773 2774 2775
	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);
2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788
			/* 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);
				}
2789 2790
			}

2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801
			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
2802 2803 2804
			}
		}
	}
2805 2806 2807 2808 2809

	amdgpu_amdkfd_suspend(adev);

	r = amdgpu_device_ip_suspend_phase1(adev);

A
Alex Deucher 已提交
2810 2811 2812
	/* evict vram memory */
	amdgpu_bo_evict_vram(adev);

2813
	amdgpu_fence_driver_suspend(adev);
A
Alex Deucher 已提交
2814

2815
	r = amdgpu_device_ip_suspend_phase2(adev);
A
Alex Deucher 已提交
2816

2817 2818 2819 2820
	/* 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
2821 2822 2823 2824 2825 2826 2827
	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
2828 2829 2830 2831
	} else {
		r = amdgpu_asic_reset(adev);
		if (r)
			DRM_ERROR("amdgpu asic reset failed\n");
A
Alex Deucher 已提交
2832 2833 2834 2835 2836 2837
	}

	return 0;
}

/**
2838
 * amdgpu_device_resume - initiate device resume
A
Alex Deucher 已提交
2839
 *
2840 2841 2842
 * @dev: drm dev pointer
 * @resume: resume state
 * @fbcon : notify the fbdev of resume
A
Alex Deucher 已提交
2843 2844 2845 2846 2847
 *
 * Bring the hw back to operating state (all asics).
 * Returns 0 for success or an error on failure.
 * Called at driver resume.
 */
2848
int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon)
A
Alex Deucher 已提交
2849 2850 2851
{
	struct drm_connector *connector;
	struct amdgpu_device *adev = dev->dev_private;
2852
	struct drm_crtc *crtc;
2853
	int r = 0;
A
Alex Deucher 已提交
2854 2855 2856 2857 2858 2859 2860

	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
2861
		r = pci_enable_device(dev->pdev);
2862
		if (r)
2863
			return r;
A
Alex Deucher 已提交
2864 2865 2866
	}

	/* post card */
A
Alex Deucher 已提交
2867
	if (amdgpu_device_need_post(adev)) {
J
jimqu 已提交
2868 2869 2870 2871
		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
2872

2873
	r = amdgpu_device_ip_resume(adev);
2874
	if (r) {
2875
		DRM_ERROR("amdgpu_device_ip_resume failed (%d).\n", r);
2876
		return r;
2877
	}
2878 2879
	amdgpu_fence_driver_resume(adev);

A
Alex Deucher 已提交
2880

2881
	r = amdgpu_device_ip_late_init(adev);
2882
	if (r)
2883
		return r;
A
Alex Deucher 已提交
2884

2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899
	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);
				}
2900 2901 2902
			}
		}
	}
2903 2904 2905
	r = amdgpu_amdkfd_resume(adev);
	if (r)
		return r;
2906

2907 2908 2909
	/* Make sure IB tests flushed */
	flush_delayed_work(&adev->late_init_work);

A
Alex Deucher 已提交
2910 2911
	/* blat the mode back in */
	if (fbcon) {
2912 2913 2914 2915 2916 2917 2918 2919 2920 2921
		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
2922
		}
2923
		amdgpu_fbdev_set_suspend(adev, 0);
A
Alex Deucher 已提交
2924 2925 2926
	}

	drm_kms_helper_poll_enable(dev);
2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939

	/*
	 * 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
2940 2941 2942 2943
	if (!amdgpu_device_has_dc_support(adev))
		drm_helper_hpd_irq_event(dev);
	else
		drm_kms_helper_hotplug_event(dev);
2944 2945 2946
#ifdef CONFIG_PM
	dev->dev->power.disable_depth--;
#endif
2947 2948
	adev->in_suspend = false;

2949
	return 0;
A
Alex Deucher 已提交
2950 2951
}

2952 2953 2954 2955 2956 2957 2958 2959 2960 2961
/**
 * 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.
 */
2962
static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev)
2963 2964 2965 2966
{
	int i;
	bool asic_hang = false;

2967 2968 2969
	if (amdgpu_sriov_vf(adev))
		return true;

2970 2971 2972
	if (amdgpu_asic_need_full_reset(adev))
		return true;

2973
	for (i = 0; i < adev->num_ip_blocks; i++) {
2974
		if (!adev->ip_blocks[i].status.valid)
2975
			continue;
2976 2977 2978 2979 2980
		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);
2981 2982 2983 2984 2985 2986
			asic_hang = true;
		}
	}
	return asic_hang;
}

2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997
/**
 * 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.
 */
2998
static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev)
2999 3000 3001 3002
{
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
3003
		if (!adev->ip_blocks[i].status.valid)
3004
			continue;
3005 3006 3007
		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);
3008 3009 3010 3011 3012 3013 3014 3015
			if (r)
				return r;
		}
	}

	return 0;
}

3016 3017 3018 3019 3020 3021 3022 3023 3024
/**
 * 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.
 */
3025
static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev)
3026
{
3027 3028
	int i;

3029 3030 3031
	if (amdgpu_asic_need_full_reset(adev))
		return true;

3032
	for (i = 0; i < adev->num_ip_blocks; i++) {
3033
		if (!adev->ip_blocks[i].status.valid)
3034
			continue;
3035 3036 3037
		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) ||
3038 3039
		    (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) ||
		     adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) {
3040
			if (adev->ip_blocks[i].status.hang) {
3041 3042 3043 3044
				DRM_INFO("Some block need full reset!\n");
				return true;
			}
		}
3045 3046 3047 3048
	}
	return false;
}

3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059
/**
 * 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.
 */
3060
static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev)
3061 3062 3063 3064
{
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
3065
		if (!adev->ip_blocks[i].status.valid)
3066
			continue;
3067 3068 3069
		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);
3070 3071 3072 3073 3074 3075 3076 3077
			if (r)
				return r;
		}
	}

	return 0;
}

3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088
/**
 * 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.
 */
3089
static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev)
3090 3091 3092 3093
{
	int i, r = 0;

	for (i = 0; i < adev->num_ip_blocks; i++) {
3094
		if (!adev->ip_blocks[i].status.valid)
3095
			continue;
3096 3097 3098
		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);
3099 3100 3101 3102 3103 3104 3105
		if (r)
			return r;
	}

	return 0;
}

3106
/**
3107
 * amdgpu_device_recover_vram - Recover some VRAM contents
3108 3109 3110 3111 3112 3113
 *
 * @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.
3114 3115 3116
 *
 * Returns:
 * 0 on success, negative error code on failure.
3117
 */
3118
static int amdgpu_device_recover_vram(struct amdgpu_device *adev)
3119 3120
{
	struct dma_fence *fence = NULL, *next = NULL;
3121 3122
	struct amdgpu_bo *shadow;
	long r = 1, tmo;
3123 3124

	if (amdgpu_sriov_runtime(adev))
3125
		tmo = msecs_to_jiffies(8000);
3126 3127 3128 3129 3130
	else
		tmo = msecs_to_jiffies(100);

	DRM_INFO("recover vram bo from shadow start\n");
	mutex_lock(&adev->shadow_list_lock);
3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141
	list_for_each_entry(shadow, &adev->shadow_list, shadow_list) {

		/* No need to recover an evicted BO */
		if (shadow->tbo.mem.mem_type != TTM_PL_TT ||
		    shadow->parent->tbo.mem.mem_type != TTM_PL_VRAM)
			continue;

		r = amdgpu_bo_restore_shadow(shadow, &next);
		if (r)
			break;

3142 3143
		if (fence) {
			r = dma_fence_wait_timeout(fence, false, tmo);
3144 3145 3146
			dma_fence_put(fence);
			fence = next;
			if (r <= 0)
3147
				break;
3148 3149
		} else {
			fence = next;
3150 3151 3152 3153
		}
	}
	mutex_unlock(&adev->shadow_list_lock);

3154 3155
	if (fence)
		tmo = dma_fence_wait_timeout(fence, false, tmo);
3156 3157
	dma_fence_put(fence);

3158
	if (r <= 0 || tmo <= 0) {
3159
		DRM_ERROR("recover vram bo from shadow failed\n");
3160 3161
		return -EIO;
	}
3162

3163 3164
	DRM_INFO("recover vram bo from shadow done\n");
	return 0;
3165 3166
}

3167

3168
/**
3169
 * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf
3170 3171
 *
 * @adev: amdgpu device pointer
3172
 * @from_hypervisor: request from hypervisor
3173 3174
 *
 * do VF FLR and reinitialize Asic
3175
 * return 0 means succeeded otherwise failed
3176 3177 3178
 */
static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
				     bool from_hypervisor)
3179 3180 3181 3182 3183 3184 3185 3186 3187
{
	int r;

	if (from_hypervisor)
		r = amdgpu_virt_request_full_gpu(adev, true);
	else
		r = amdgpu_virt_reset_gpu(adev);
	if (r)
		return r;
3188 3189

	/* Resume IP prior to SMC */
3190
	r = amdgpu_device_ip_reinit_early_sriov(adev);
3191 3192
	if (r)
		goto error;
3193 3194

	/* we need recover gart prior to run SMC/CP/SDMA resume */
3195
	amdgpu_gtt_mgr_recover(&adev->mman.bdev.man[TTM_PL_TT]);
3196

3197 3198 3199 3200
	r = amdgpu_device_fw_loading(adev);
	if (r)
		return r;

3201
	/* now we are okay to resume SMC/CP/SDMA */
3202
	r = amdgpu_device_ip_reinit_late_sriov(adev);
3203 3204
	if (r)
		goto error;
3205 3206

	amdgpu_irq_gpu_reset_resume_helper(adev);
3207
	r = amdgpu_ib_ring_tests(adev);
3208

3209 3210
error:
	amdgpu_virt_release_full_gpu(adev, true);
3211 3212
	if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) {
		atomic_inc(&adev->vram_lost_counter);
3213
		r = amdgpu_device_recover_vram(adev);
3214 3215 3216 3217 3218
	}

	return r;
}

3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233
/**
 * amdgpu_device_should_recover_gpu - check if we should try GPU recovery
 *
 * @adev: amdgpu device pointer
 *
 * Check amdgpu_gpu_recovery and SRIOV status to see if we should try to recover
 * a hung GPU.
 */
bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev)
{
	if (!amdgpu_device_ip_check_soft_reset(adev)) {
		DRM_INFO("Timeout, but no hardware hang detected.\n");
		return false;
	}

3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255
	if (amdgpu_gpu_recovery == 0)
		goto disabled;

	if (amdgpu_sriov_vf(adev))
		return true;

	if (amdgpu_gpu_recovery == -1) {
		switch (adev->asic_type) {
		case CHIP_TOPAZ:
		case CHIP_TONGA:
		case CHIP_FIJI:
		case CHIP_POLARIS10:
		case CHIP_POLARIS11:
		case CHIP_POLARIS12:
		case CHIP_VEGAM:
		case CHIP_VEGA20:
		case CHIP_VEGA10:
		case CHIP_VEGA12:
			break;
		default:
			goto disabled;
		}
3256 3257 3258
	}

	return true;
3259 3260 3261 3262

disabled:
		DRM_INFO("GPU recovery disabled.\n");
		return false;
3263 3264
}

3265

3266 3267 3268 3269 3270 3271
static int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,
					struct amdgpu_job *job,
					bool *need_full_reset_arg)
{
	int i, r = 0;
	bool need_full_reset  = *need_full_reset_arg;
3272 3273

	/* block all schedulers and reset given job's ring */
3274 3275 3276
	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];

C
Chunming Zhou 已提交
3277
		if (!ring || !ring->sched.thread)
3278
			continue;
3279

3280 3281
		kthread_park(ring->sched.thread);

3282
		if (job && job->base.sched != &ring->sched)
3283 3284
			continue;

3285
		drm_sched_hw_job_reset(&ring->sched, job ? &job->base : NULL);
3286

M
Monk Liu 已提交
3287 3288
		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
		amdgpu_fence_driver_force_completion(ring);
3289
	}
A
Alex Deucher 已提交
3290

3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407


	if (!amdgpu_sriov_vf(adev)) {

		if (!need_full_reset)
			need_full_reset = amdgpu_device_ip_need_full_reset(adev);

		if (!need_full_reset) {
			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)) {
				DRM_INFO("soft reset failed, will fallback to full reset!\n");
				need_full_reset = true;
			}
		}

		if (need_full_reset)
			r = amdgpu_device_ip_suspend(adev);

		*need_full_reset_arg = need_full_reset;
	}

	return r;
}

static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive,
			       struct list_head *device_list_handle,
			       bool *need_full_reset_arg)
{
	struct amdgpu_device *tmp_adev = NULL;
	bool need_full_reset = *need_full_reset_arg, vram_lost = false;
	int r = 0;

	/*
	 * ASIC reset has to be done on all HGMI hive nodes ASAP
	 * to allow proper links negotiation in FW (within 1 sec)
	 */
	if (need_full_reset) {
		list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
			r = amdgpu_asic_reset(tmp_adev);
			if (r)
				DRM_WARN("ASIC reset failed with err r, %d for drm dev, %s",
					 r, tmp_adev->ddev->unique);
		}
	}


	list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
		if (need_full_reset) {
			/* post card */
			if (amdgpu_atom_asic_init(tmp_adev->mode_info.atom_context))
				DRM_WARN("asic atom init failed!");

			if (!r) {
				dev_info(tmp_adev->dev, "GPU reset succeeded, trying to resume\n");
				r = amdgpu_device_ip_resume_phase1(tmp_adev);
				if (r)
					goto out;

				vram_lost = amdgpu_device_check_vram_lost(tmp_adev);
				if (vram_lost) {
					DRM_ERROR("VRAM is lost!\n");
					atomic_inc(&tmp_adev->vram_lost_counter);
				}

				r = amdgpu_gtt_mgr_recover(
					&tmp_adev->mman.bdev.man[TTM_PL_TT]);
				if (r)
					goto out;

				r = amdgpu_device_fw_loading(tmp_adev);
				if (r)
					return r;

				r = amdgpu_device_ip_resume_phase2(tmp_adev);
				if (r)
					goto out;

				if (vram_lost)
					amdgpu_device_fill_reset_magic(tmp_adev);

				/* Update PSP FW topology after reset */
				if (hive && tmp_adev->gmc.xgmi.num_physical_nodes > 1)
					r = amdgpu_xgmi_update_topology(hive, tmp_adev);
			}
		}


out:
		if (!r) {
			amdgpu_irq_gpu_reset_resume_helper(tmp_adev);
			r = amdgpu_ib_ring_tests(tmp_adev);
			if (r) {
				dev_err(tmp_adev->dev, "ib ring test failed (%d).\n", r);
				r = amdgpu_device_ip_suspend(tmp_adev);
				need_full_reset = true;
				r = -EAGAIN;
				goto end;
			}
		}

		if (!r)
			r = amdgpu_device_recover_vram(tmp_adev);
		else
			tmp_adev->asic_reset_res = r;
	}

end:
	*need_full_reset_arg = need_full_reset;
	return r;
}

static void amdgpu_device_post_asic_reset(struct amdgpu_device *adev,
					  struct amdgpu_job *job)
{
	int i;
3408

3409 3410
	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];
3411

3412 3413
		if (!ring || !ring->sched.thread)
			continue;
3414

3415 3416 3417 3418
		/* only need recovery sched of the given job's ring
		 * or all rings (in the case @job is NULL)
		 * after above amdgpu_reset accomplished
		 */
3419
		if ((!job || job->base.sched == &ring->sched) && !adev->asic_reset_res)
3420
			drm_sched_job_recovery(&ring->sched);
3421

3422
		kthread_unpark(ring->sched.thread);
A
Alex Deucher 已提交
3423 3424
	}

3425
	if (!amdgpu_device_has_dc_support(adev)) {
3426
		drm_helper_resume_force_mode(adev->ddev);
3427
	}
A
Alex Deucher 已提交
3428

3429 3430
	adev->asic_reset_res = 0;
}
3431

3432 3433 3434 3435 3436 3437 3438 3439
static void amdgpu_device_lock_adev(struct amdgpu_device *adev)
{
	mutex_lock(&adev->lock_reset);
	atomic_inc(&adev->gpu_reset_counter);
	adev->in_gpu_reset = 1;
	/* Block kfd */
	amdgpu_amdkfd_pre_reset(adev);
}
A
Alex Deucher 已提交
3440

3441 3442
static void amdgpu_device_unlock_adev(struct amdgpu_device *adev)
{
3443 3444
	/*unlock kfd */
	amdgpu_amdkfd_post_reset(adev);
3445
	amdgpu_vf_error_trans_all(adev);
3446 3447
	adev->in_gpu_reset = 0;
	mutex_unlock(&adev->lock_reset);
3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566
}


/**
 * amdgpu_device_gpu_recover - reset the asic and recover scheduler
 *
 * @adev: amdgpu device pointer
 * @job: which job trigger hang
 *
 * Attempt to reset the GPU if it has hung (all asics).
 * Attempt to do soft-reset or full-reset and reinitialize Asic
 * Returns 0 for success or an error on failure.
 */

int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
			      struct amdgpu_job *job)
{
	int r;
	struct amdgpu_hive_info *hive = NULL;
	bool need_full_reset = false;
	struct amdgpu_device *tmp_adev = NULL;
	struct list_head device_list, *device_list_handle =  NULL;

	INIT_LIST_HEAD(&device_list);

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

	/*
	 * In case of XGMI hive disallow concurrent resets to be triggered
	 * by different nodes. No point also since the one node already executing
	 * reset will also reset all the other nodes in the hive.
	 */
	hive = amdgpu_get_xgmi_hive(adev);
	if (hive && adev->gmc.xgmi.num_physical_nodes > 1 &&
	    !mutex_trylock(&hive->hive_lock))
		return 0;

	/* Start with adev pre asic reset first for soft reset check.*/
	amdgpu_device_lock_adev(adev);
	r = amdgpu_device_pre_asic_reset(adev,
					 job,
					 &need_full_reset);
	if (r) {
		/*TODO Should we stop ?*/
		DRM_ERROR("GPU pre asic reset failed with err, %d for drm dev, %s ",
			  r, adev->ddev->unique);
		adev->asic_reset_res = r;
	}

	/* Build list of devices to reset */
	if  (need_full_reset && adev->gmc.xgmi.num_physical_nodes > 1) {
		if (!hive) {
			amdgpu_device_unlock_adev(adev);
			return -ENODEV;
		}

		/*
		 * In case we are in XGMI hive mode device reset is done for all the
		 * nodes in the hive to retrain all XGMI links and hence the reset
		 * sequence is executed in loop on all nodes.
		 */
		device_list_handle = &hive->device_list;
	} else {
		list_add_tail(&adev->gmc.xgmi.head, &device_list);
		device_list_handle = &device_list;
	}

retry:	/* Rest of adevs pre asic reset from XGMI hive. */
	list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {

		if (tmp_adev == adev)
			continue;

		dev_info(tmp_adev->dev, "GPU reset begin for drm dev %s!\n", adev->ddev->unique);

		amdgpu_device_lock_adev(tmp_adev);
		r = amdgpu_device_pre_asic_reset(tmp_adev,
						 NULL,
						 &need_full_reset);
		/*TODO Should we stop ?*/
		if (r) {
			DRM_ERROR("GPU pre asic reset failed with err, %d for drm dev, %s ",
				  r, tmp_adev->ddev->unique);
			tmp_adev->asic_reset_res = r;
		}
	}

	/* Actual ASIC resets if needed.*/
	/* TODO Implement XGMI hive reset logic for SRIOV */
	if (amdgpu_sriov_vf(adev)) {
		r = amdgpu_device_reset_sriov(adev, job ? false : true);
		if (r)
			adev->asic_reset_res = r;
	} else {
		r  = amdgpu_do_asic_reset(hive, device_list_handle, &need_full_reset);
		if (r && r == -EAGAIN)
			goto retry;
	}

	/* Post ASIC reset for all devs .*/
	list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
		amdgpu_device_post_asic_reset(tmp_adev, tmp_adev == adev ? job : NULL);

		if (r) {
			/* bad news, how to tell it to userspace ? */
			dev_info(tmp_adev->dev, "GPU reset(%d) failed\n", atomic_read(&adev->gpu_reset_counter));
			amdgpu_vf_error_put(tmp_adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r);
		} else {
			dev_info(tmp_adev->dev, "GPU reset(%d) succeeded!\n", atomic_read(&adev->gpu_reset_counter));
		}

		amdgpu_device_unlock_adev(tmp_adev);
	}

	if (hive && adev->gmc.xgmi.num_physical_nodes > 1)
		mutex_unlock(&hive->hive_lock);

	if (r)
		dev_info(adev->dev, "GPU reset end with ret = %d\n", r);
A
Alex Deucher 已提交
3567 3568 3569
	return r;
}

3570 3571 3572 3573 3574 3575 3576 3577 3578
/**
 * 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.
 */
3579
static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev)
3580
{
3581 3582 3583
	struct pci_dev *pdev;
	enum pci_bus_speed speed_cap;
	enum pcie_link_width link_width;
3584

3585 3586
	if (amdgpu_pcie_gen_cap)
		adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap;
3587

3588 3589
	if (amdgpu_pcie_lane_cap)
		adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap;
3590

3591 3592 3593 3594 3595 3596
	/* 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;
3597
		return;
3598
	}
3599

3600
	if (adev->pm.pcie_gen_mask == 0) {
3601 3602 3603 3604 3605
		/* 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 |
3606 3607 3608
						  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
						  CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3);
		} else {
3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645
			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;

3646 3647 3648
		}
	}
	if (adev->pm.pcie_mlw_mask == 0) {
3649 3650 3651 3652 3653 3654 3655
		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:
3656 3657 3658 3659 3660 3661 3662 3663
				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;
3664
			case PCIE_LNK_X16:
3665 3666 3667 3668 3669 3670 3671
				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;
3672
			case PCIE_LNK_X12:
3673 3674 3675 3676 3677 3678
				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;
3679
			case PCIE_LNK_X8:
3680 3681 3682 3683 3684
				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;
3685
			case PCIE_LNK_X4:
3686 3687 3688 3689
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3690
			case PCIE_LNK_X2:
3691 3692 3693
				adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
							  CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
				break;
3694
			case PCIE_LNK_X1:
3695 3696 3697 3698 3699
				adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1;
				break;
			default:
				break;
			}
3700 3701 3702
		}
	}
}
A
Alex Deucher 已提交
3703