amdgpu_amdkfd.c 17.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
/*
 * Copyright 2014 Advanced Micro Devices, Inc.
 *
 * 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.
 */

#include "amdgpu_amdkfd.h"
24
#include "amd_shared.h"
25

26
#include "amdgpu.h"
27
#include "amdgpu_gfx.h"
28
#include "amdgpu_dma_buf.h"
29
#include <linux/module.h>
30
#include <linux/dma-buf.h>
31
#include "amdgpu_xgmi.h"
32

33 34
static const unsigned int compute_vmid_bitmap = 0xFF00;

35 36 37 38 39
/* Total memory size in system memory and all GPU VRAM. Used to
 * estimate worst case amount of memory to reserve for page tables
 */
uint64_t amdgpu_amdkfd_total_mem_size;

40
int amdgpu_amdkfd_init(void)
41
{
42
	struct sysinfo si;
43 44
	int ret;

45 46 47 48
	si_meminfo(&si);
	amdgpu_amdkfd_total_mem_size = si.totalram - si.totalhigh;
	amdgpu_amdkfd_total_mem_size *= si.mem_unit;

49
#ifdef CONFIG_HSA_AMD
50
	ret = kgd2kfd_init();
51
	amdgpu_amdkfd_gpuvm_init_mem_limits();
52 53
#else
	ret = -ENOENT;
54
#endif
55

56
	return ret;
57 58
}

59 60
void amdgpu_amdkfd_fini(void)
{
61
	kgd2kfd_exit();
62 63 64
}

void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev)
65
{
66
	bool vf = amdgpu_sriov_vf(adev);
67

68
	adev->kfd.dev = kgd2kfd_probe((struct kgd_dev *)adev,
69
				      adev->pdev, adev->asic_type, vf);
70 71 72

	if (adev->kfd.dev)
		amdgpu_amdkfd_total_mem_size += adev->gmc.real_vram_size;
73 74
}

75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107
/**
 * amdgpu_doorbell_get_kfd_info - Report doorbell configuration required to
 *                                setup amdkfd
 *
 * @adev: amdgpu_device pointer
 * @aperture_base: output returning doorbell aperture base physical address
 * @aperture_size: output returning doorbell aperture size in bytes
 * @start_offset: output returning # of doorbell bytes reserved for amdgpu.
 *
 * amdgpu and amdkfd share the doorbell aperture. amdgpu sets it up,
 * takes doorbells required for its own rings and reports the setup to amdkfd.
 * amdgpu reserved doorbells are at the start of the doorbell aperture.
 */
static void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev,
					 phys_addr_t *aperture_base,
					 size_t *aperture_size,
					 size_t *start_offset)
{
	/*
	 * The first num_doorbells are used by amdgpu.
	 * amdkfd takes whatever's left in the aperture.
	 */
	if (adev->doorbell.size > adev->doorbell.num_doorbells * sizeof(u32)) {
		*aperture_base = adev->doorbell.base;
		*aperture_size = adev->doorbell.size;
		*start_offset = adev->doorbell.num_doorbells * sizeof(u32);
	} else {
		*aperture_base = 0;
		*aperture_size = 0;
		*start_offset = 0;
	}
}

108
void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
109
{
110
	int i;
111
	int last_valid_bit;
112 113

	if (adev->kfd.dev) {
114
		struct kgd2kfd_shared_resources gpu_resources = {
115
			.compute_vmid_bitmap = compute_vmid_bitmap,
116
			.num_pipe_per_mec = adev->gfx.mec.num_pipe_per_mec,
117 118 119
			.num_queue_per_pipe = adev->gfx.mec.num_queue_per_pipe,
			.gpuvm_size = min(adev->vm_manager.max_pfn
					  << AMDGPU_GPU_PAGE_SHIFT,
120
					  AMDGPU_GMC_HOLE_START),
121 122 123
			.drm_render_minor = adev->ddev->render->index,
			.sdma_doorbell_idx = adev->doorbell_index.sdma_engine,

124 125
		};

126
		/* this is going to have a few of the MSBs set that we need to
K
Kent Russell 已提交
127 128
		 * clear
		 */
129 130 131 132 133
		bitmap_complement(gpu_resources.queue_bitmap,
				  adev->gfx.mec.queue_bitmap,
				  KGD_MAX_QUEUES);

		/* According to linux/bitmap.h we shouldn't use bitmap_clear if
K
Kent Russell 已提交
134 135
		 * nbits is not compile time constant
		 */
136
		last_valid_bit = 1 /* only first MEC can have compute queues */
137 138 139 140 141
				* adev->gfx.mec.num_pipe_per_mec
				* adev->gfx.mec.num_queue_per_pipe;
		for (i = last_valid_bit; i < KGD_MAX_QUEUES; ++i)
			clear_bit(i, gpu_resources.queue_bitmap);

142
		amdgpu_doorbell_get_kfd_info(adev,
143 144 145
				&gpu_resources.doorbell_physical_address,
				&gpu_resources.doorbell_aperture_size,
				&gpu_resources.doorbell_start_offset);
146

147 148 149 150 151 152 153
		/* Since SOC15, BIF starts to statically use the
		 * lower 12 bits of doorbell addresses for routing
		 * based on settings in registers like
		 * SDMA0_DOORBELL_RANGE etc..
		 * In order to route a doorbell to CP engine, the lower
		 * 12 bits of its address has to be outside the range
		 * set for SDMA, VCN, and IH blocks.
154
		 */
155 156 157 158 159 160
		if (adev->asic_type >= CHIP_VEGA10) {
			gpu_resources.non_cp_doorbells_start =
					adev->doorbell_index.first_non_cp;
			gpu_resources.non_cp_doorbells_end =
					adev->doorbell_index.last_non_cp;
		}
161

162
		kgd2kfd_device_init(adev->kfd.dev, adev->ddev, &gpu_resources);
163 164 165
	}
}

166
void amdgpu_amdkfd_device_fini(struct amdgpu_device *adev)
167
{
168
	if (adev->kfd.dev) {
169
		kgd2kfd_device_exit(adev->kfd.dev);
170
		adev->kfd.dev = NULL;
171 172 173
	}
}

174
void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,
175 176
		const void *ih_ring_entry)
{
177
	if (adev->kfd.dev)
178
		kgd2kfd_interrupt(adev->kfd.dev, ih_ring_entry);
179 180
}

181
void amdgpu_amdkfd_suspend(struct amdgpu_device *adev)
182
{
183
	if (adev->kfd.dev)
184
		kgd2kfd_suspend(adev->kfd.dev);
185 186
}

187
int amdgpu_amdkfd_resume(struct amdgpu_device *adev)
188 189 190
{
	int r = 0;

191
	if (adev->kfd.dev)
192
		r = kgd2kfd_resume(adev->kfd.dev);
193 194 195 196

	return r;
}

197 198 199 200
int amdgpu_amdkfd_pre_reset(struct amdgpu_device *adev)
{
	int r = 0;

201
	if (adev->kfd.dev)
202
		r = kgd2kfd_pre_reset(adev->kfd.dev);
203 204 205 206 207 208 209 210

	return r;
}

int amdgpu_amdkfd_post_reset(struct amdgpu_device *adev)
{
	int r = 0;

211
	if (adev->kfd.dev)
212
		r = kgd2kfd_post_reset(adev->kfd.dev);
213 214 215 216

	return r;
}

217 218 219 220
void amdgpu_amdkfd_gpu_reset(struct kgd_dev *kgd)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

221 222
	if (amdgpu_device_should_recover_gpu(adev))
		amdgpu_device_gpu_recover(adev, NULL);
223 224
}

225 226 227
int amdgpu_amdkfd_alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
				void **mem_obj, uint64_t *gpu_addr,
				void **cpu_ptr, bool mqd_gfx9)
228
{
229
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
230
	struct amdgpu_bo *bo = NULL;
231
	struct amdgpu_bo_param bp;
232
	int r;
233
	void *cpu_ptr_tmp = NULL;
234

235 236 237 238 239 240 241
	memset(&bp, 0, sizeof(bp));
	bp.size = size;
	bp.byte_align = PAGE_SIZE;
	bp.domain = AMDGPU_GEM_DOMAIN_GTT;
	bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
	bp.type = ttm_bo_type_kernel;
	bp.resv = NULL;
242 243 244 245

	if (mqd_gfx9)
		bp.flags |= AMDGPU_GEM_CREATE_MQD_GFX9;

246
	r = amdgpu_bo_create(adev, &bp, &bo);
247
	if (r) {
248
		dev_err(adev->dev,
249 250 251 252 253
			"failed to allocate BO for amdkfd (%d)\n", r);
		return r;
	}

	/* map the buffer */
254
	r = amdgpu_bo_reserve(bo, true);
255
	if (r) {
256
		dev_err(adev->dev, "(%d) failed to reserve bo for amdkfd\n", r);
257 258 259
		goto allocate_mem_reserve_bo_failed;
	}

260
	r = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT);
261
	if (r) {
262
		dev_err(adev->dev, "(%d) failed to pin bo for amdkfd\n", r);
263 264 265
		goto allocate_mem_pin_bo_failed;
	}

266 267 268 269 270 271
	r = amdgpu_ttm_alloc_gart(&bo->tbo);
	if (r) {
		dev_err(adev->dev, "%p bind failed\n", bo);
		goto allocate_mem_kmap_bo_failed;
	}

272
	r = amdgpu_bo_kmap(bo, &cpu_ptr_tmp);
273
	if (r) {
274
		dev_err(adev->dev,
275 276 277 278
			"(%d) failed to map bo to kernel for amdkfd\n", r);
		goto allocate_mem_kmap_bo_failed;
	}

279
	*mem_obj = bo;
280
	*gpu_addr = amdgpu_bo_gpu_offset(bo);
281 282 283
	*cpu_ptr = cpu_ptr_tmp;

	amdgpu_bo_unreserve(bo);
284 285 286 287

	return 0;

allocate_mem_kmap_bo_failed:
288
	amdgpu_bo_unpin(bo);
289
allocate_mem_pin_bo_failed:
290
	amdgpu_bo_unreserve(bo);
291
allocate_mem_reserve_bo_failed:
292
	amdgpu_bo_unref(&bo);
293 294 295 296

	return r;
}

297
void amdgpu_amdkfd_free_gtt_mem(struct kgd_dev *kgd, void *mem_obj)
298
{
299 300 301 302 303 304 305
	struct amdgpu_bo *bo = (struct amdgpu_bo *) mem_obj;

	amdgpu_bo_reserve(bo, true);
	amdgpu_bo_kunmap(bo);
	amdgpu_bo_unpin(bo);
	amdgpu_bo_unreserve(bo);
	amdgpu_bo_unref(&(bo));
306 307
}

308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341
int amdgpu_amdkfd_alloc_gws(struct kgd_dev *kgd, size_t size,
				void **mem_obj)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
	struct amdgpu_bo *bo = NULL;
	struct amdgpu_bo_param bp;
	int r;

	memset(&bp, 0, sizeof(bp));
	bp.size = size;
	bp.byte_align = 1;
	bp.domain = AMDGPU_GEM_DOMAIN_GWS;
	bp.flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
	bp.type = ttm_bo_type_device;
	bp.resv = NULL;

	r = amdgpu_bo_create(adev, &bp, &bo);
	if (r) {
		dev_err(adev->dev,
			"failed to allocate gws BO for amdkfd (%d)\n", r);
		return r;
	}

	*mem_obj = bo;
	return 0;
}

void amdgpu_amdkfd_free_gws(struct kgd_dev *kgd, void *mem_obj)
{
	struct amdgpu_bo *bo = (struct amdgpu_bo *)mem_obj;

	amdgpu_bo_unref(&bo);
}

342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
uint32_t amdgpu_amdkfd_get_fw_version(struct kgd_dev *kgd,
				      enum kgd_engine_type type)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

	switch (type) {
	case KGD_ENGINE_PFP:
		return adev->gfx.pfp_fw_version;

	case KGD_ENGINE_ME:
		return adev->gfx.me_fw_version;

	case KGD_ENGINE_CE:
		return adev->gfx.ce_fw_version;

	case KGD_ENGINE_MEC1:
		return adev->gfx.mec_fw_version;

	case KGD_ENGINE_MEC2:
		return adev->gfx.mec2_fw_version;

	case KGD_ENGINE_RLC:
		return adev->gfx.rlc_fw_version;

	case KGD_ENGINE_SDMA1:
		return adev->sdma.instance[0].fw_version;

	case KGD_ENGINE_SDMA2:
		return adev->sdma.instance[1].fw_version;

	default:
		return 0;
	}

	return 0;
}

379 380
void amdgpu_amdkfd_get_local_mem_info(struct kgd_dev *kgd,
				      struct kfd_local_mem_info *mem_info)
381 382 383 384
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
	uint64_t address_mask = adev->dev->dma_mask ? ~*adev->dev->dma_mask :
					     ~((1ULL << 32) - 1);
385
	resource_size_t aper_limit = adev->gmc.aper_base + adev->gmc.aper_size;
386 387

	memset(mem_info, 0, sizeof(*mem_info));
388 389 390 391
	if (!(adev->gmc.aper_base & address_mask || aper_limit & address_mask)) {
		mem_info->local_mem_size_public = adev->gmc.visible_vram_size;
		mem_info->local_mem_size_private = adev->gmc.real_vram_size -
				adev->gmc.visible_vram_size;
392 393
	} else {
		mem_info->local_mem_size_public = 0;
394
		mem_info->local_mem_size_private = adev->gmc.real_vram_size;
395
	}
396
	mem_info->vram_width = adev->gmc.vram_width;
397

398
	pr_debug("Address base: %pap limit %pap public 0x%llx private 0x%llx\n",
399
			&adev->gmc.aper_base, &aper_limit,
400 401 402 403 404
			mem_info->local_mem_size_public,
			mem_info->local_mem_size_private);

	if (amdgpu_sriov_vf(adev))
		mem_info->mem_clk_max = adev->clock.default_mclk / 100;
405 406 407 408 409 410
	else if (adev->powerplay.pp_funcs) {
		if (amdgpu_emu_mode == 1)
			mem_info->mem_clk_max = 0;
		else
			mem_info->mem_clk_max = amdgpu_dpm_get_mclk(adev, false) / 100;
	} else
411
		mem_info->mem_clk_max = 100;
412 413
}

414
uint64_t amdgpu_amdkfd_get_gpu_clock_counter(struct kgd_dev *kgd)
415
{
416
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
417

418 419
	if (adev->gfx.funcs->get_gpu_clock_counter)
		return adev->gfx.funcs->get_gpu_clock_counter(adev);
420 421 422
	return 0;
}

423
uint32_t amdgpu_amdkfd_get_max_engine_clock_in_mhz(struct kgd_dev *kgd)
424
{
425
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
426

427 428 429
	/* the sclk is in quantas of 10kHz */
	if (amdgpu_sriov_vf(adev))
		return adev->clock.default_sclk / 100;
430 431 432 433
	else if (adev->powerplay.pp_funcs)
		return amdgpu_dpm_get_sclk(adev, false) / 100;
	else
		return 100;
434
}
435

436
void amdgpu_amdkfd_get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info)
437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
	struct amdgpu_cu_info acu_info = adev->gfx.cu_info;

	memset(cu_info, 0, sizeof(*cu_info));
	if (sizeof(cu_info->cu_bitmap) != sizeof(acu_info.bitmap))
		return;

	cu_info->cu_active_number = acu_info.number;
	cu_info->cu_ao_mask = acu_info.ao_cu_mask;
	memcpy(&cu_info->cu_bitmap[0], &acu_info.bitmap[0],
	       sizeof(acu_info.bitmap));
	cu_info->num_shader_engines = adev->gfx.config.max_shader_engines;
	cu_info->num_shader_arrays_per_engine = adev->gfx.config.max_sh_per_se;
	cu_info->num_cu_per_sh = adev->gfx.config.max_cu_per_sh;
	cu_info->simd_per_cu = acu_info.simd_per_cu;
	cu_info->max_waves_per_simd = acu_info.max_waves_per_simd;
	cu_info->wave_front_size = acu_info.wave_front_size;
	cu_info->max_scratch_slots_per_cu = acu_info.max_scratch_slots_per_cu;
	cu_info->lds_size = acu_info.lds_size;
}
458

459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514
int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int dma_buf_fd,
				  struct kgd_dev **dma_buf_kgd,
				  uint64_t *bo_size, void *metadata_buffer,
				  size_t buffer_size, uint32_t *metadata_size,
				  uint32_t *flags)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
	struct dma_buf *dma_buf;
	struct drm_gem_object *obj;
	struct amdgpu_bo *bo;
	uint64_t metadata_flags;
	int r = -EINVAL;

	dma_buf = dma_buf_get(dma_buf_fd);
	if (IS_ERR(dma_buf))
		return PTR_ERR(dma_buf);

	if (dma_buf->ops != &amdgpu_dmabuf_ops)
		/* Can't handle non-graphics buffers */
		goto out_put;

	obj = dma_buf->priv;
	if (obj->dev->driver != adev->ddev->driver)
		/* Can't handle buffers from different drivers */
		goto out_put;

	adev = obj->dev->dev_private;
	bo = gem_to_amdgpu_bo(obj);
	if (!(bo->preferred_domains & (AMDGPU_GEM_DOMAIN_VRAM |
				    AMDGPU_GEM_DOMAIN_GTT)))
		/* Only VRAM and GTT BOs are supported */
		goto out_put;

	r = 0;
	if (dma_buf_kgd)
		*dma_buf_kgd = (struct kgd_dev *)adev;
	if (bo_size)
		*bo_size = amdgpu_bo_size(bo);
	if (metadata_size)
		*metadata_size = bo->metadata_size;
	if (metadata_buffer)
		r = amdgpu_bo_get_metadata(bo, metadata_buffer, buffer_size,
					   metadata_size, &metadata_flags);
	if (flags) {
		*flags = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) ?
			ALLOC_MEM_FLAGS_VRAM : ALLOC_MEM_FLAGS_GTT;

		if (bo->flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)
			*flags |= ALLOC_MEM_FLAGS_PUBLIC;
	}

out_put:
	dma_buf_put(dma_buf);
	return r;
}

515 516 517 518 519 520
uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

	return amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
}
521

522 523 524 525 526 527
uint64_t amdgpu_amdkfd_get_hive_id(struct kgd_dev *kgd)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

	return adev->gmc.xgmi.hive_id;
}
528 529 530 531 532 533 534 535 536 537 538 539 540 541
uint8_t amdgpu_amdkfd_get_xgmi_hops_count(struct kgd_dev *dst, struct kgd_dev *src)
{
	struct amdgpu_device *peer_adev = (struct amdgpu_device *)src;
	struct amdgpu_device *adev = (struct amdgpu_device *)dst;
	int ret = amdgpu_xgmi_get_hops_count(adev, peer_adev);

	if (ret < 0) {
		DRM_ERROR("amdgpu: failed to get  xgmi hops count between node %d and %d. ret = %d\n",
			adev->gmc.xgmi.physical_node_id,
			peer_adev->gmc.xgmi.physical_node_id, ret);
		ret = 0;
	}
	return  (uint8_t)ret;
}
542

543 544 545 546 547 548 549
uint64_t amdgpu_amdkfd_get_mmio_remap_phys_addr(struct kgd_dev *kgd)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

	return adev->rmmio_remap.bus_addr;
}

550 551 552 553 554 555 556
uint32_t amdgpu_amdkfd_get_num_gws(struct kgd_dev *kgd)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

	return adev->gds.gws_size;
}

557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611
int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
				uint32_t vmid, uint64_t gpu_addr,
				uint32_t *ib_cmd, uint32_t ib_len)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
	struct amdgpu_job *job;
	struct amdgpu_ib *ib;
	struct amdgpu_ring *ring;
	struct dma_fence *f = NULL;
	int ret;

	switch (engine) {
	case KGD_ENGINE_MEC1:
		ring = &adev->gfx.compute_ring[0];
		break;
	case KGD_ENGINE_SDMA1:
		ring = &adev->sdma.instance[0].ring;
		break;
	case KGD_ENGINE_SDMA2:
		ring = &adev->sdma.instance[1].ring;
		break;
	default:
		pr_err("Invalid engine in IB submission: %d\n", engine);
		ret = -EINVAL;
		goto err;
	}

	ret = amdgpu_job_alloc(adev, 1, &job, NULL);
	if (ret)
		goto err;

	ib = &job->ibs[0];
	memset(ib, 0, sizeof(struct amdgpu_ib));

	ib->gpu_addr = gpu_addr;
	ib->ptr = ib_cmd;
	ib->length_dw = ib_len;
	/* This works for NO_HWS. TODO: need to handle without knowing VMID */
	job->vmid = vmid;

	ret = amdgpu_ib_schedule(ring, 1, ib, job, &f);
	if (ret) {
		DRM_ERROR("amdgpu: failed to schedule IB.\n");
		goto err_ib_sched;
	}

	ret = dma_fence_wait(f, false);

err_ib_sched:
	dma_fence_put(f);
	amdgpu_job_free(job);
err:
	return ret;
}

612 613 614 615
void amdgpu_amdkfd_set_compute_idle(struct kgd_dev *kgd, bool idle)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

616 617 618 619 620 621
	if (is_support_sw_smu(adev))
		smu_switch_power_profile(&adev->smu,
					 PP_SMC_POWER_PROFILE_COMPUTE,
					 !idle);
	else if (adev->powerplay.pp_funcs &&
		 adev->powerplay.pp_funcs->switch_power_profile)
622 623 624
		amdgpu_dpm_switch_power_profile(adev,
						PP_SMC_POWER_PROFILE_COMPUTE,
						!idle);
625 626
}

627 628
bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
{
629
	if (adev->kfd.dev) {
630 631 632 633 634 635
		if ((1 << vmid) & compute_vmid_bitmap)
			return true;
	}

	return false;
}
636

637 638 639 640 641 642 643
bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;

	return adev->have_atomics_support;
}

644
#ifndef CONFIG_HSA_AMD
645 646 647 648 649
bool amdkfd_fence_check_mm(struct dma_fence *f, struct mm_struct *mm)
{
	return false;
}

650
void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo)
651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668
{
}

void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
					struct amdgpu_vm *vm)
{
}

struct amdgpu_amdkfd_fence *to_amdgpu_amdkfd_fence(struct dma_fence *f)
{
	return NULL;
}

int amdgpu_amdkfd_evict_userptr(struct kgd_mem *mem, struct mm_struct *mm)
{
	return 0;
}

669
struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, struct pci_dev *pdev,
670
			      unsigned int asic_type, bool vf)
671 672 673 674 675
{
	return NULL;
}

bool kgd2kfd_device_init(struct kfd_dev *kfd,
676
			 struct drm_device *ddev,
677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
			 const struct kgd2kfd_shared_resources *gpu_resources)
{
	return false;
}

void kgd2kfd_device_exit(struct kfd_dev *kfd)
{
}

void kgd2kfd_exit(void)
{
}

void kgd2kfd_suspend(struct kfd_dev *kfd)
{
}

int kgd2kfd_resume(struct kfd_dev *kfd)
{
	return 0;
}

int kgd2kfd_pre_reset(struct kfd_dev *kfd)
{
	return 0;
}

int kgd2kfd_post_reset(struct kfd_dev *kfd)
{
	return 0;
}

void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
{
}
712 713 714 715

void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd)
{
}
716
#endif