gmc_v9_0.c 33.0 KB
Newer Older
A
Alex Xie 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
/*
 * Copyright 2016 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 <linux/firmware.h>
24
#include <drm/drm_cache.h>
A
Alex Xie 已提交
25 26
#include "amdgpu.h"
#include "gmc_v9_0.h"
27
#include "amdgpu_atomfirmware.h"
28
#include "amdgpu_gem.h"
A
Alex Xie 已提交
29

30 31
#include "hdp/hdp_4_0_offset.h"
#include "hdp/hdp_4_0_sh_mask.h"
32
#include "gc/gc_9_0_sh_mask.h"
33 34
#include "dce/dce_12_0_offset.h"
#include "dce/dce_12_0_sh_mask.h"
35
#include "vega10_enum.h"
36
#include "mmhub/mmhub_1_0_offset.h"
37
#include "athub/athub_1_0_offset.h"
38
#include "oss/osssys_4_0_offset.h"
A
Alex Xie 已提交
39

40
#include "soc15.h"
A
Alex Xie 已提交
41
#include "soc15_common.h"
42
#include "umc/umc_6_0_sh_mask.h"
A
Alex Xie 已提交
43 44 45

#include "gfxhub_v1_0.h"
#include "mmhub_v1_0.h"
46
#include "gfxhub_v1_1.h"
A
Alex Xie 已提交
47

48 49
#include "ivsrcid/vmc/irqsrcs_vmc_1_0.h"

50 51 52 53 54 55 56 57
/* add these here since we already include dce12 headers and these are for DCN */
#define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION                                                          0x055d
#define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION_BASE_IDX                                                 2
#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_WIDTH__SHIFT                                        0x0
#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_HEIGHT__SHIFT                                       0x10
#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_WIDTH_MASK                                          0x00003FFFL
#define HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION__PRI_VIEWPORT_HEIGHT_MASK                                         0x3FFF0000L

A
Alex Xie 已提交
58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
/* XXX Move this macro to VEGA10 header file, which is like vid.h for VI.*/
#define AMDGPU_NUM_OF_VMIDS			8

static const u32 golden_settings_vega10_hdp[] =
{
	0xf64, 0x0fffffff, 0x00000000,
	0xf65, 0x0fffffff, 0x00000000,
	0xf66, 0x0fffffff, 0x00000000,
	0xf67, 0x0fffffff, 0x00000000,
	0xf68, 0x0fffffff, 0x00000000,
	0xf6a, 0x0fffffff, 0x00000000,
	0xf6b, 0x0fffffff, 0x00000000,
	0xf6c, 0x0fffffff, 0x00000000,
	0xf6d, 0x0fffffff, 0x00000000,
	0xf6e, 0x0fffffff, 0x00000000,
};

75
static const struct soc15_reg_golden golden_settings_mmhub_1_0_0[] =
76
{
77 78
	SOC15_REG_GOLDEN_VALUE(MMHUB, 0, mmDAGB1_WRCLI2, 0x00000007, 0xfe5fe0fa),
	SOC15_REG_GOLDEN_VALUE(MMHUB, 0, mmMMEA1_DRAM_WR_CLI2GRP_MAP0, 0x00000030, 0x55555565)
79 80
};

81
static const struct soc15_reg_golden golden_settings_athub_1_0_0[] =
82
{
83 84
	SOC15_REG_GOLDEN_VALUE(ATHUB, 0, mmRPB_ARB_CNTL, 0x0000ff00, 0x00000800),
	SOC15_REG_GOLDEN_VALUE(ATHUB, 0, mmRPB_ARB_CNTL2, 0x00ff00ff, 0x00080008)
85 86
};

87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
/* Ecc related register addresses, (BASE + reg offset) */
/* Universal Memory Controller caps (may be fused). */
/* UMCCH:UmcLocalCap */
#define UMCLOCALCAPS_ADDR0	(0x00014306 + 0x00000000)
#define UMCLOCALCAPS_ADDR1	(0x00014306 + 0x00000800)
#define UMCLOCALCAPS_ADDR2	(0x00014306 + 0x00001000)
#define UMCLOCALCAPS_ADDR3	(0x00014306 + 0x00001800)
#define UMCLOCALCAPS_ADDR4	(0x00054306 + 0x00000000)
#define UMCLOCALCAPS_ADDR5	(0x00054306 + 0x00000800)
#define UMCLOCALCAPS_ADDR6	(0x00054306 + 0x00001000)
#define UMCLOCALCAPS_ADDR7	(0x00054306 + 0x00001800)
#define UMCLOCALCAPS_ADDR8	(0x00094306 + 0x00000000)
#define UMCLOCALCAPS_ADDR9	(0x00094306 + 0x00000800)
#define UMCLOCALCAPS_ADDR10	(0x00094306 + 0x00001000)
#define UMCLOCALCAPS_ADDR11	(0x00094306 + 0x00001800)
#define UMCLOCALCAPS_ADDR12	(0x000d4306 + 0x00000000)
#define UMCLOCALCAPS_ADDR13	(0x000d4306 + 0x00000800)
#define UMCLOCALCAPS_ADDR14	(0x000d4306 + 0x00001000)
#define UMCLOCALCAPS_ADDR15	(0x000d4306 + 0x00001800)

/* Universal Memory Controller Channel config. */
/* UMCCH:UMC_CONFIG */
#define UMCCH_UMC_CONFIG_ADDR0	(0x00014040 + 0x00000000)
#define UMCCH_UMC_CONFIG_ADDR1	(0x00014040 + 0x00000800)
#define UMCCH_UMC_CONFIG_ADDR2	(0x00014040 + 0x00001000)
#define UMCCH_UMC_CONFIG_ADDR3	(0x00014040 + 0x00001800)
#define UMCCH_UMC_CONFIG_ADDR4	(0x00054040 + 0x00000000)
#define UMCCH_UMC_CONFIG_ADDR5	(0x00054040 + 0x00000800)
#define UMCCH_UMC_CONFIG_ADDR6	(0x00054040 + 0x00001000)
#define UMCCH_UMC_CONFIG_ADDR7	(0x00054040 + 0x00001800)
#define UMCCH_UMC_CONFIG_ADDR8	(0x00094040 + 0x00000000)
#define UMCCH_UMC_CONFIG_ADDR9	(0x00094040 + 0x00000800)
#define UMCCH_UMC_CONFIG_ADDR10	(0x00094040 + 0x00001000)
#define UMCCH_UMC_CONFIG_ADDR11	(0x00094040 + 0x00001800)
#define UMCCH_UMC_CONFIG_ADDR12	(0x000d4040 + 0x00000000)
#define UMCCH_UMC_CONFIG_ADDR13	(0x000d4040 + 0x00000800)
#define UMCCH_UMC_CONFIG_ADDR14	(0x000d4040 + 0x00001000)
#define UMCCH_UMC_CONFIG_ADDR15	(0x000d4040 + 0x00001800)

/* Universal Memory Controller Channel Ecc config. */
/* UMCCH:EccCtrl */
#define UMCCH_ECCCTRL_ADDR0	(0x00014053 + 0x00000000)
#define UMCCH_ECCCTRL_ADDR1	(0x00014053 + 0x00000800)
#define UMCCH_ECCCTRL_ADDR2	(0x00014053 + 0x00001000)
#define UMCCH_ECCCTRL_ADDR3	(0x00014053 + 0x00001800)
#define UMCCH_ECCCTRL_ADDR4	(0x00054053 + 0x00000000)
#define UMCCH_ECCCTRL_ADDR5	(0x00054053 + 0x00000800)
#define UMCCH_ECCCTRL_ADDR6	(0x00054053 + 0x00001000)
#define UMCCH_ECCCTRL_ADDR7	(0x00054053 + 0x00001800)
#define UMCCH_ECCCTRL_ADDR8	(0x00094053 + 0x00000000)
#define UMCCH_ECCCTRL_ADDR9	(0x00094053 + 0x00000800)
#define UMCCH_ECCCTRL_ADDR10	(0x00094053 + 0x00001000)
#define UMCCH_ECCCTRL_ADDR11	(0x00094053 + 0x00001800)
#define UMCCH_ECCCTRL_ADDR12	(0x000d4053 + 0x00000000)
#define UMCCH_ECCCTRL_ADDR13	(0x000d4053 + 0x00000800)
#define UMCCH_ECCCTRL_ADDR14	(0x000d4053 + 0x00001000)
#define UMCCH_ECCCTRL_ADDR15	(0x000d4053 + 0x00001800)

static const uint32_t ecc_umclocalcap_addrs[] = {
	UMCLOCALCAPS_ADDR0,
	UMCLOCALCAPS_ADDR1,
	UMCLOCALCAPS_ADDR2,
	UMCLOCALCAPS_ADDR3,
	UMCLOCALCAPS_ADDR4,
	UMCLOCALCAPS_ADDR5,
	UMCLOCALCAPS_ADDR6,
	UMCLOCALCAPS_ADDR7,
	UMCLOCALCAPS_ADDR8,
	UMCLOCALCAPS_ADDR9,
	UMCLOCALCAPS_ADDR10,
	UMCLOCALCAPS_ADDR11,
	UMCLOCALCAPS_ADDR12,
	UMCLOCALCAPS_ADDR13,
	UMCLOCALCAPS_ADDR14,
	UMCLOCALCAPS_ADDR15,
};

static const uint32_t ecc_umcch_umc_config_addrs[] = {
	UMCCH_UMC_CONFIG_ADDR0,
	UMCCH_UMC_CONFIG_ADDR1,
	UMCCH_UMC_CONFIG_ADDR2,
	UMCCH_UMC_CONFIG_ADDR3,
	UMCCH_UMC_CONFIG_ADDR4,
	UMCCH_UMC_CONFIG_ADDR5,
	UMCCH_UMC_CONFIG_ADDR6,
	UMCCH_UMC_CONFIG_ADDR7,
	UMCCH_UMC_CONFIG_ADDR8,
	UMCCH_UMC_CONFIG_ADDR9,
	UMCCH_UMC_CONFIG_ADDR10,
	UMCCH_UMC_CONFIG_ADDR11,
	UMCCH_UMC_CONFIG_ADDR12,
	UMCCH_UMC_CONFIG_ADDR13,
	UMCCH_UMC_CONFIG_ADDR14,
	UMCCH_UMC_CONFIG_ADDR15,
};

static const uint32_t ecc_umcch_eccctrl_addrs[] = {
	UMCCH_ECCCTRL_ADDR0,
	UMCCH_ECCCTRL_ADDR1,
	UMCCH_ECCCTRL_ADDR2,
	UMCCH_ECCCTRL_ADDR3,
	UMCCH_ECCCTRL_ADDR4,
	UMCCH_ECCCTRL_ADDR5,
	UMCCH_ECCCTRL_ADDR6,
	UMCCH_ECCCTRL_ADDR7,
	UMCCH_ECCCTRL_ADDR8,
	UMCCH_ECCCTRL_ADDR9,
	UMCCH_ECCCTRL_ADDR10,
	UMCCH_ECCCTRL_ADDR11,
	UMCCH_ECCCTRL_ADDR12,
	UMCCH_ECCCTRL_ADDR13,
	UMCCH_ECCCTRL_ADDR14,
	UMCCH_ECCCTRL_ADDR15,
};

A
Alex Xie 已提交
202 203 204 205 206 207
static int gmc_v9_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
					struct amdgpu_irq_src *src,
					unsigned type,
					enum amdgpu_interrupt_state state)
{
	struct amdgpu_vmhub *hub;
208
	u32 tmp, reg, bits, i, j;
A
Alex Xie 已提交
209

210 211 212 213 214 215 216 217
	bits = VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
		VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
		VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
		VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
		VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
		VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
		VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK;

A
Alex Xie 已提交
218 219
	switch (state) {
	case AMDGPU_IRQ_STATE_DISABLE:
220 221 222 223 224 225 226 227
		for (j = 0; j < AMDGPU_MAX_VMHUBS; j++) {
			hub = &adev->vmhub[j];
			for (i = 0; i < 16; i++) {
				reg = hub->vm_context0_cntl + i;
				tmp = RREG32(reg);
				tmp &= ~bits;
				WREG32(reg, tmp);
			}
A
Alex Xie 已提交
228 229 230
		}
		break;
	case AMDGPU_IRQ_STATE_ENABLE:
231 232 233 234 235 236 237 238
		for (j = 0; j < AMDGPU_MAX_VMHUBS; j++) {
			hub = &adev->vmhub[j];
			for (i = 0; i < 16; i++) {
				reg = hub->vm_context0_cntl + i;
				tmp = RREG32(reg);
				tmp |= bits;
				WREG32(reg, tmp);
			}
A
Alex Xie 已提交
239 240 241 242 243 244 245 246 247 248 249 250
		}
	default:
		break;
	}

	return 0;
}

static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
				struct amdgpu_irq_src *source,
				struct amdgpu_iv_entry *entry)
{
251
	struct amdgpu_vmhub *hub = &adev->vmhub[entry->vmid_src];
252
	uint32_t status = 0;
A
Alex Xie 已提交
253 254 255 256 257
	u64 addr;

	addr = (u64)entry->src_data[0] << 12;
	addr |= ((u64)entry->src_data[1] & 0xf) << 44;

258
	if (!amdgpu_sriov_vf(adev)) {
259 260
		status = RREG32(hub->vm_l2_pro_fault_status);
		WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
261
	}
A
Alex Xie 已提交
262

263
	if (printk_ratelimit()) {
264 265 266 267
		struct amdgpu_task_info task_info = { 0 };

		amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);

268
		dev_err(adev->dev,
269
			"[%s] VMC page fault (src_id:%u ring:%u vmid:%u pasid:%u, for process %s pid %d thread %s pid %d)\n",
270 271
			entry->vmid_src ? "mmhub" : "gfxhub",
			entry->src_id, entry->ring_id, entry->vmid,
272 273
			entry->pasid, task_info.process_name, task_info.tgid,
			task_info.task_name, task_info.pid);
274
		dev_err(adev->dev, "  in page starting at address 0x%016llx from %d\n",
275 276 277 278 279
			addr, entry->client_id);
		if (!amdgpu_sriov_vf(adev))
			dev_err(adev->dev,
				"VM_L2_PROTECTION_FAULT_STATUS:0x%08X\n",
				status);
280
	}
A
Alex Xie 已提交
281 282 283 284 285 286 287 288 289 290 291

	return 0;
}

static const struct amdgpu_irq_src_funcs gmc_v9_0_irq_funcs = {
	.set = gmc_v9_0_vm_fault_interrupt_state,
	.process = gmc_v9_0_process_interrupt,
};

static void gmc_v9_0_set_irq_funcs(struct amdgpu_device *adev)
{
292 293
	adev->gmc.vm_fault.num_types = 1;
	adev->gmc.vm_fault.funcs = &gmc_v9_0_irq_funcs;
A
Alex Xie 已提交
294 295
}

296 297
static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid,
					uint32_t flush_type)
298 299 300 301
{
	u32 req = 0;

	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ,
302
			    PER_VMID_INVALIDATE_REQ, 1 << vmid);
303
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, flush_type);
304 305 306 307 308 309 310 311 312 313 314
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PTES, 1);
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE0, 1);
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE1, 1);
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE2, 1);
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L1_PTES, 1);
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ,
			    CLEAR_PROTECTION_FAULT_STATUS_ADDR,	0);

	return req;
}

A
Alex Xie 已提交
315 316 317 318 319 320 321 322
/*
 * GART
 * VMID 0 is the physical GPU addresses as used by the kernel.
 * VMIDs 1-15 are used for userspace clients and are handled
 * by the amdgpu vm/hsa code.
 */

/**
323
 * gmc_v9_0_flush_gpu_tlb - tlb flush with certain type
A
Alex Xie 已提交
324 325 326
 *
 * @adev: amdgpu_device pointer
 * @vmid: vm instance to flush
327
 * @flush_type: the flush type
A
Alex Xie 已提交
328
 *
329
 * Flush the TLB for the requested page table using certain type.
A
Alex Xie 已提交
330
 */
331
static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
332
				uint32_t vmid, uint32_t flush_type)
A
Alex Xie 已提交
333 334 335 336 337 338
{
	const unsigned eng = 17;
	unsigned i, j;

	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
		struct amdgpu_vmhub *hub = &adev->vmhub[i];
339
		u32 tmp = gmc_v9_0_get_invalidate_req(vmid, flush_type);
A
Alex Xie 已提交
340

341 342 343
		if (i == AMDGPU_GFXHUB && !adev->in_gpu_reset &&
		    adev->gfx.kiq.ring.sched.ready &&
		    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev))) {
344 345 346 347 348 349
			uint32_t req = hub->vm_inv_eng0_req + eng;
			uint32_t ack = hub->vm_inv_eng0_ack + eng;

			amdgpu_virt_kiq_reg_write_reg_wait(adev, req, ack, tmp,
							   1 << vmid);
			continue;
350
		}
351 352

		spin_lock(&adev->gmc.invalidate_lock);
353
		WREG32_NO_KIQ(hub->vm_inv_eng0_req + eng, tmp);
A
Alex Xie 已提交
354
		for (j = 0; j < adev->usec_timeout; j++) {
355
			tmp = RREG32_NO_KIQ(hub->vm_inv_eng0_ack + eng);
356
			if (tmp & (1 << vmid))
A
Alex Xie 已提交
357 358 359
				break;
			udelay(1);
		}
360
		spin_unlock(&adev->gmc.invalidate_lock);
361 362 363
		if (j < adev->usec_timeout)
			continue;

A
Alex Xie 已提交
364 365 366 367
		DRM_ERROR("Timeout waiting for VM flush ACK!\n");
	}
}

368
static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
369
					    unsigned vmid, uint64_t pd_addr)
370
{
371 372
	struct amdgpu_device *adev = ring->adev;
	struct amdgpu_vmhub *hub = &adev->vmhub[ring->funcs->vmhub];
373
	uint32_t req = gmc_v9_0_get_invalidate_req(vmid, 0);
374 375 376 377 378 379 380 381
	unsigned eng = ring->vm_inv_eng;

	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_lo32 + (2 * vmid),
			      lower_32_bits(pd_addr));

	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_hi32 + (2 * vmid),
			      upper_32_bits(pd_addr));

382 383 384
	amdgpu_ring_emit_reg_write_reg_wait(ring, hub->vm_inv_eng0_req + eng,
					    hub->vm_inv_eng0_ack + eng,
					    req, 1 << vmid);
385

386 387 388
	return pd_addr;
}

389 390 391 392 393 394 395 396 397 398 399 400 401 402
static void gmc_v9_0_emit_pasid_mapping(struct amdgpu_ring *ring, unsigned vmid,
					unsigned pasid)
{
	struct amdgpu_device *adev = ring->adev;
	uint32_t reg;

	if (ring->funcs->vmhub == AMDGPU_GFXHUB)
		reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT) + vmid;
	else
		reg = SOC15_REG_OFFSET(OSSSYS, 0, mmIH_VMID_0_LUT_MM) + vmid;

	amdgpu_ring_emit_wreg(ring, reg, pasid);
}

A
Alex Xie 已提交
403
/**
404
 * gmc_v9_0_set_pte_pde - update the page tables using MMIO
A
Alex Xie 已提交
405 406 407 408 409 410 411 412 413
 *
 * @adev: amdgpu_device pointer
 * @cpu_pt_addr: cpu address of the page table
 * @gpu_page_idx: entry in the page table to update
 * @addr: dst addr to write into pte/pde
 * @flags: access flags
 *
 * Update the page tables using the CPU.
 */
414 415 416
static int gmc_v9_0_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
				uint32_t gpu_page_idx, uint64_t addr,
				uint64_t flags)
A
Alex Xie 已提交
417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
{
	void __iomem *ptr = (void *)cpu_pt_addr;
	uint64_t value;

	/*
	 * PTE format on VEGA 10:
	 * 63:59 reserved
	 * 58:57 mtype
	 * 56 F
	 * 55 L
	 * 54 P
	 * 53 SW
	 * 52 T
	 * 50:48 reserved
	 * 47:12 4k physical page base address
	 * 11:7 fragment
	 * 6 write
	 * 5 read
	 * 4 exe
	 * 3 Z
	 * 2 snooped
	 * 1 system
	 * 0 valid
	 *
	 * PDE format on VEGA 10:
	 * 63:59 block fragment size
	 * 58:55 reserved
	 * 54 P
	 * 53:48 reserved
	 * 47:6 physical base address of PD or PTE
	 * 5:3 reserved
	 * 2 C
	 * 1 system
	 * 0 valid
	 */

	/*
	 * The following is for PTE only. GART does not have PDEs.
	*/
	value = addr & 0x0000FFFFFFFFF000ULL;
	value |= flags;
	writeq(value, ptr + (gpu_page_idx * 8));
	return 0;
}

static uint64_t gmc_v9_0_get_vm_pte_flags(struct amdgpu_device *adev,
						uint32_t flags)

{
	uint64_t pte_flag = 0;

	if (flags & AMDGPU_VM_PAGE_EXECUTABLE)
		pte_flag |= AMDGPU_PTE_EXECUTABLE;
	if (flags & AMDGPU_VM_PAGE_READABLE)
		pte_flag |= AMDGPU_PTE_READABLE;
	if (flags & AMDGPU_VM_PAGE_WRITEABLE)
		pte_flag |= AMDGPU_PTE_WRITEABLE;

	switch (flags & AMDGPU_VM_MTYPE_MASK) {
	case AMDGPU_VM_MTYPE_DEFAULT:
		pte_flag |= AMDGPU_PTE_MTYPE(MTYPE_NC);
		break;
	case AMDGPU_VM_MTYPE_NC:
		pte_flag |= AMDGPU_PTE_MTYPE(MTYPE_NC);
		break;
	case AMDGPU_VM_MTYPE_WC:
		pte_flag |= AMDGPU_PTE_MTYPE(MTYPE_WC);
		break;
	case AMDGPU_VM_MTYPE_CC:
		pte_flag |= AMDGPU_PTE_MTYPE(MTYPE_CC);
		break;
	case AMDGPU_VM_MTYPE_UC:
		pte_flag |= AMDGPU_PTE_MTYPE(MTYPE_UC);
		break;
	default:
		pte_flag |= AMDGPU_PTE_MTYPE(MTYPE_NC);
		break;
	}

	if (flags & AMDGPU_VM_PAGE_PRT)
		pte_flag |= AMDGPU_PTE_PRT;

	return pte_flag;
}

502 503
static void gmc_v9_0_get_vm_pde(struct amdgpu_device *adev, int level,
				uint64_t *addr, uint64_t *flags)
A
Alex Xie 已提交
504
{
505
	if (!(*flags & AMDGPU_PDE_PTE) && !(*flags & AMDGPU_PTE_SYSTEM))
506
		*addr = adev->vm_manager.vram_base_offset + *addr -
507
			adev->gmc.vram_start;
508
	BUG_ON(*addr & 0xFFFF00000000003FULL);
509

510
	if (!adev->gmc.translate_further)
511 512 513 514 515 516 517 518 519 520 521 522 523
		return;

	if (level == AMDGPU_VM_PDB1) {
		/* Set the block fragment size */
		if (!(*flags & AMDGPU_PDE_PTE))
			*flags |= AMDGPU_PDE_BFS(0x9);

	} else if (level == AMDGPU_VM_PDB0) {
		if (*flags & AMDGPU_PDE_PTE)
			*flags &= ~AMDGPU_PDE_PTE;
		else
			*flags |= AMDGPU_PTE_TF;
	}
A
Alex Xie 已提交
524 525
}

526 527
static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
	.flush_gpu_tlb = gmc_v9_0_flush_gpu_tlb,
528
	.emit_flush_gpu_tlb = gmc_v9_0_emit_flush_gpu_tlb,
529
	.emit_pasid_mapping = gmc_v9_0_emit_pasid_mapping,
530
	.set_pte_pde = gmc_v9_0_set_pte_pde,
531 532
	.get_vm_pte_flags = gmc_v9_0_get_vm_pte_flags,
	.get_vm_pde = gmc_v9_0_get_vm_pde
A
Alex Xie 已提交
533 534
};

535
static void gmc_v9_0_set_gmc_funcs(struct amdgpu_device *adev)
A
Alex Xie 已提交
536
{
537
	adev->gmc.gmc_funcs = &gmc_v9_0_gmc_funcs;
A
Alex Xie 已提交
538 539 540 541 542 543
}

static int gmc_v9_0_early_init(void *handle)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

544
	gmc_v9_0_set_gmc_funcs(adev);
A
Alex Xie 已提交
545 546
	gmc_v9_0_set_irq_funcs(adev);

547 548 549
	adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
	adev->gmc.shared_aperture_end =
		adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
550
	adev->gmc.private_aperture_start = 0x1000000000000000ULL;
551 552
	adev->gmc.private_aperture_end =
		adev->gmc.private_aperture_start + (4ULL << 30) - 1;
553

A
Alex Xie 已提交
554 555 556
	return 0;
}

557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622
static int gmc_v9_0_ecc_available(struct amdgpu_device *adev)
{
	uint32_t reg_val;
	uint32_t reg_addr;
	uint32_t field_val;
	size_t i;
	uint32_t fv2;
	size_t lost_sheep;

	DRM_DEBUG("ecc: gmc_v9_0_ecc_available()\n");

	lost_sheep = 0;
	for (i = 0; i < ARRAY_SIZE(ecc_umclocalcap_addrs); ++i) {
		reg_addr = ecc_umclocalcap_addrs[i];
		DRM_DEBUG("ecc: "
			  "UMCCH_UmcLocalCap[%zu]: reg_addr: 0x%08x\n",
			  i, reg_addr);
		reg_val = RREG32(reg_addr);
		field_val = REG_GET_FIELD(reg_val, UMCCH0_0_UmcLocalCap,
					  EccDis);
		DRM_DEBUG("ecc: "
			  "reg_val: 0x%08x, "
			  "EccDis: 0x%08x, ",
			  reg_val, field_val);
		if (field_val) {
			DRM_ERROR("ecc: UmcLocalCap:EccDis is set.\n");
			++lost_sheep;
		}
	}

	for (i = 0; i < ARRAY_SIZE(ecc_umcch_umc_config_addrs); ++i) {
		reg_addr = ecc_umcch_umc_config_addrs[i];
		DRM_DEBUG("ecc: "
			  "UMCCH0_0_UMC_CONFIG[%zu]: reg_addr: 0x%08x",
			  i, reg_addr);
		reg_val = RREG32(reg_addr);
		field_val = REG_GET_FIELD(reg_val, UMCCH0_0_UMC_CONFIG,
					  DramReady);
		DRM_DEBUG("ecc: "
			  "reg_val: 0x%08x, "
			  "DramReady: 0x%08x\n",
			  reg_val, field_val);

		if (!field_val) {
			DRM_ERROR("ecc: UMC_CONFIG:DramReady is not set.\n");
			++lost_sheep;
		}
	}

	for (i = 0; i < ARRAY_SIZE(ecc_umcch_eccctrl_addrs); ++i) {
		reg_addr = ecc_umcch_eccctrl_addrs[i];
		DRM_DEBUG("ecc: "
			  "UMCCH_EccCtrl[%zu]: reg_addr: 0x%08x, ",
			  i, reg_addr);
		reg_val = RREG32(reg_addr);
		field_val = REG_GET_FIELD(reg_val, UMCCH0_0_EccCtrl,
					  WrEccEn);
		fv2 = REG_GET_FIELD(reg_val, UMCCH0_0_EccCtrl,
				    RdEccEn);
		DRM_DEBUG("ecc: "
			  "reg_val: 0x%08x, "
			  "WrEccEn: 0x%08x, "
			  "RdEccEn: 0x%08x\n",
			  reg_val, field_val, fv2);

		if (!field_val) {
623
			DRM_DEBUG("ecc: WrEccEn is not set\n");
624 625 626
			++lost_sheep;
		}
		if (!fv2) {
627
			DRM_DEBUG("ecc: RdEccEn is not set\n");
628 629 630 631 632 633 634 635
			++lost_sheep;
		}
	}

	DRM_DEBUG("ecc: lost_sheep: %zu\n", lost_sheep);
	return lost_sheep == 0;
}

636 637 638 639 640 641 642 643 644 645 646 647 648
static bool gmc_v9_0_keep_stolen_memory(struct amdgpu_device *adev)
{

	/*
	 * TODO:
	 * Currently there is a bug where some memory client outside
	 * of the driver writes to first 8M of VRAM on S3 resume,
	 * this overrides GART which by default gets placed in first 8M and
	 * causes VM_FAULTS once GTT is accessed.
	 * Keep the stolen memory reservation until the while this is not solved.
	 * Also check code in gmc_v9_0_get_vbios_fb_size and gmc_v9_0_late_init
	 */
	switch (adev->asic_type) {
649 650
	case CHIP_VEGA10:
		return true;
651 652 653 654
	case CHIP_RAVEN:
	case CHIP_VEGA12:
	case CHIP_VEGA20:
	default:
655
		return false;
656 657 658
	}
}

A
Alex Xie 已提交
659 660 661
static int gmc_v9_0_late_init(void *handle)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
662 663 664 665 666 667 668 669 670 671
	/*
	 * The latest engine allocation on gfx9 is:
	 * Engine 0, 1: idle
	 * Engine 2, 3: firmware
	 * Engine 4~13: amdgpu ring, subject to change when ring number changes
	 * Engine 14~15: idle
	 * Engine 16: kfd tlb invalidation
	 * Engine 17: Gart flushes
	 */
	unsigned vm_inv_eng[AMDGPU_MAX_VMHUBS] = { 4, 4 };
672
	unsigned i;
673
	int r;
674

675 676
	if (!gmc_v9_0_keep_stolen_memory(adev))
		amdgpu_bo_late_init(adev);
677

678 679 680 681 682
	for(i = 0; i < adev->num_rings; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];
		unsigned vmhub = ring->funcs->vmhub;

		ring->vm_inv_eng = vm_inv_eng[vmhub]++;
683 684
		dev_info(adev->dev, "ring %s uses VM inv eng %u on hub %u\n",
			 ring->name, ring->vm_inv_eng, ring->funcs->vmhub);
685 686
	}

687
	/* Engine 16 is used for KFD and 17 for GART flushes */
688
	for(i = 0; i < AMDGPU_MAX_VMHUBS; ++i)
689
		BUG_ON(vm_inv_eng[i] > 16);
690

691
	if (adev->asic_type == CHIP_VEGA10 && !amdgpu_sriov_vf(adev)) {
692 693 694 695 696
		r = gmc_v9_0_ecc_available(adev);
		if (r == 1) {
			DRM_INFO("ECC is active.\n");
		} else if (r == 0) {
			DRM_INFO("ECC is not present.\n");
697
			adev->df_funcs->enable_ecc_force_par_wr_rmw(adev, false);
698 699 700 701
		} else {
			DRM_ERROR("gmc_v9_0_ecc_available() failed. r: %d\n", r);
			return r;
		}
702 703
	}

704
	return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
A
Alex Xie 已提交
705 706 707
}

static void gmc_v9_0_vram_gtt_location(struct amdgpu_device *adev,
708
					struct amdgpu_gmc *mc)
A
Alex Xie 已提交
709
{
710 711 712
	u64 base = 0;
	if (!amdgpu_sriov_vf(adev))
		base = mmhub_v1_0_get_fb_location(adev);
713 714
	/* add the xgmi offset of the physical node */
	base += adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
715 716
	amdgpu_gmc_vram_location(adev, &adev->gmc, base);
	amdgpu_gmc_gart_location(adev, mc);
717 718
	if (!amdgpu_sriov_vf(adev))
		amdgpu_gmc_agp_location(adev, mc);
719
	/* base offset of vram pages */
720
	adev->vm_manager.vram_base_offset = gfxhub_v1_0_get_mc_fb_offset(adev);
721 722 723 724

	/* XXX: add the xgmi offset of the physical node? */
	adev->vm_manager.vram_base_offset +=
		adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
A
Alex Xie 已提交
725 726 727 728 729 730 731 732 733 734 735 736 737 738
}

/**
 * gmc_v9_0_mc_init - initialize the memory controller driver params
 *
 * @adev: amdgpu_device pointer
 *
 * Look up the amount of vram, vram width, and decide how to place
 * vram and gart within the GPU's physical address space.
 * Returns 0 for success.
 */
static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
{
	int chansize, numchan;
739
	int r;
A
Alex Xie 已提交
740

741 742
	if (amdgpu_emu_mode != 1)
		adev->gmc.vram_width = amdgpu_atomfirmware_get_vram_width(adev);
743
	if (!adev->gmc.vram_width) {
744
		/* hbm memory channel size */
745 746 747 748
		if (adev->flags & AMD_IS_APU)
			chansize = 64;
		else
			chansize = 128;
749

750
		numchan = adev->df_funcs->get_hbm_channel_number(adev);
751
		adev->gmc.vram_width = numchan * chansize;
A
Alex Xie 已提交
752 753 754
	}

	/* size in MB on si */
755
	adev->gmc.mc_vram_size =
756
		adev->nbio_funcs->get_memsize(adev) * 1024ULL * 1024ULL;
757
	adev->gmc.real_vram_size = adev->gmc.mc_vram_size;
758 759 760 761 762 763

	if (!(adev->flags & AMD_IS_APU)) {
		r = amdgpu_device_resize_fb_bar(adev);
		if (r)
			return r;
	}
764 765
	adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
	adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
A
Alex Xie 已提交
766

767 768 769 770 771 772
#ifdef CONFIG_X86_64
	if (adev->flags & AMD_IS_APU) {
		adev->gmc.aper_base = gfxhub_v1_0_get_mc_fb_offset(adev);
		adev->gmc.aper_size = adev->gmc.real_vram_size;
	}
#endif
A
Alex Xie 已提交
773
	/* In case the PCI BAR is larger than the actual amount of vram */
774 775 776
	adev->gmc.visible_vram_size = adev->gmc.aper_size;
	if (adev->gmc.visible_vram_size > adev->gmc.real_vram_size)
		adev->gmc.visible_vram_size = adev->gmc.real_vram_size;
A
Alex Xie 已提交
777

778 779 780 781
	/* set the gart size */
	if (amdgpu_gart_size == -1) {
		switch (adev->asic_type) {
		case CHIP_VEGA10:  /* all engines support GPUVM */
782
		case CHIP_VEGA12:  /* all engines support GPUVM */
F
Feifei Xu 已提交
783
		case CHIP_VEGA20:
784
		default:
785
			adev->gmc.gart_size = 512ULL << 20;
786 787
			break;
		case CHIP_RAVEN:   /* DCE SG support */
788
			adev->gmc.gart_size = 1024ULL << 20;
789 790 791
			break;
		}
	} else {
792
		adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
793 794
	}

795
	gmc_v9_0_vram_gtt_location(adev, &adev->gmc);
A
Alex Xie 已提交
796 797 798 799 800 801 802 803

	return 0;
}

static int gmc_v9_0_gart_init(struct amdgpu_device *adev)
{
	int r;

804
	if (adev->gart.bo) {
A
Alex Xie 已提交
805 806 807 808 809 810 811 812 813 814 815 816 817
		WARN(1, "VEGA10 PCIE GART already initialized\n");
		return 0;
	}
	/* Initialize common gart structure */
	r = amdgpu_gart_init(adev);
	if (r)
		return r;
	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
	adev->gart.gart_pte_flags = AMDGPU_PTE_MTYPE(MTYPE_UC) |
				 AMDGPU_PTE_EXECUTABLE;
	return amdgpu_gart_table_vram_alloc(adev);
}

818 819 820 821 822
static unsigned gmc_v9_0_get_vbios_fb_size(struct amdgpu_device *adev)
{
	u32 d1vga_control = RREG32_SOC15(DCE, 0, mmD1VGA_CONTROL);
	unsigned size;

823 824 825 826
	/*
	 * TODO Remove once GART corruption is resolved
	 * Check related code in gmc_v9_0_sw_fini
	 * */
827 828
	if (gmc_v9_0_keep_stolen_memory(adev))
		return 9 * 1024 * 1024;
829

830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845
	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
		size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
	} else {
		u32 viewport;

		switch (adev->asic_type) {
		case CHIP_RAVEN:
			viewport = RREG32_SOC15(DCE, 0, mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION);
			size = (REG_GET_FIELD(viewport,
					      HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION, PRI_VIEWPORT_HEIGHT) *
				REG_GET_FIELD(viewport,
					      HUBP0_DCSURF_PRI_VIEWPORT_DIMENSION, PRI_VIEWPORT_WIDTH) *
				4);
			break;
		case CHIP_VEGA10:
		case CHIP_VEGA12:
846
		case CHIP_VEGA20:
847 848 849 850 851 852 853 854 855 856 857
		default:
			viewport = RREG32_SOC15(DCE, 0, mmSCL0_VIEWPORT_SIZE);
			size = (REG_GET_FIELD(viewport, SCL0_VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
				REG_GET_FIELD(viewport, SCL0_VIEWPORT_SIZE, VIEWPORT_WIDTH) *
				4);
			break;
		}
	}
	/* return 0 if the pre-OS buffer uses up most of vram */
	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
		return 0;
858

859 860 861
	return size;
}

A
Alex Xie 已提交
862 863 864 865 866 867
static int gmc_v9_0_sw_init(void *handle)
{
	int r;
	int dma_bits;
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

868
	gfxhub_v1_0_init(adev);
869
	mmhub_v1_0_init(adev);
870

871
	spin_lock_init(&adev->gmc.invalidate_lock);
A
Alex Xie 已提交
872

873
	adev->gmc.vram_type = amdgpu_atomfirmware_get_vram_type(adev);
874 875
	switch (adev->asic_type) {
	case CHIP_RAVEN:
876
		if (adev->rev_id == 0x0 || adev->rev_id == 0x1) {
877
			amdgpu_vm_adjust_size(adev, 256 * 1024, 9, 3, 48);
878 879 880
		} else {
			/* vm_size is 128TB + 512GB for legacy 3-level page support */
			amdgpu_vm_adjust_size(adev, 128 * 1024 + 512, 9, 2, 48);
881
			adev->gmc.translate_further =
882 883
				adev->vm_manager.num_level > 1;
		}
884 885
		break;
	case CHIP_VEGA10:
886
	case CHIP_VEGA12:
F
Feifei Xu 已提交
887
	case CHIP_VEGA20:
888 889 890 891 892
		/*
		 * To fulfill 4-level page support,
		 * vm size is 256TB (48bit), maximum size of Vega10,
		 * block size 512 (9bit)
		 */
893
		amdgpu_vm_adjust_size(adev, 256 * 1024, 9, 3, 48);
894 895 896
		break;
	default:
		break;
A
Alex Xie 已提交
897 898 899
	}

	/* This interrupt is VMC page fault.*/
900
	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VMC, VMC_1_0__SRCID__VM_FAULT,
901
				&adev->gmc.vm_fault);
902
	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_UTCL2, UTCL2_1_0__SRCID__FAULT,
903
				&adev->gmc.vm_fault);
A
Alex Xie 已提交
904 905 906 907 908 909 910 911

	if (r)
		return r;

	/* Set the internal MC address mask
	 * This is the max address of the GPU's
	 * internal address space.
	 */
912
	adev->gmc.mc_mask = 0xffffffffffffULL; /* 48 bit MC */
A
Alex Xie 已提交
913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931

	/* set DMA mask + need_dma32 flags.
	 * PCIE - can handle 44-bits.
	 * IGP - can handle 44-bits
	 * PCI - dma32 for legacy pci gart, 44 bits on vega10
	 */
	adev->need_dma32 = false;
	dma_bits = adev->need_dma32 ? 32 : 44;
	r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
	if (r) {
		adev->need_dma32 = true;
		dma_bits = 32;
		printk(KERN_WARNING "amdgpu: No suitable DMA available.\n");
	}
	r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
	if (r) {
		pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
		printk(KERN_WARNING "amdgpu: No coherent DMA available.\n");
	}
932
	adev->need_swiotlb = drm_get_max_iomem() > ((u64)1 << dma_bits);
A
Alex Xie 已提交
933

934 935 936 937 938 939
	if (adev->asic_type == CHIP_VEGA20) {
		r = gfxhub_v1_1_get_xgmi_info(adev);
		if (r)
			return r;
	}

A
Alex Xie 已提交
940 941 942 943
	r = gmc_v9_0_mc_init(adev);
	if (r)
		return r;

944 945
	adev->gmc.stolen_size = gmc_v9_0_get_vbios_fb_size(adev);

A
Alex Xie 已提交
946 947 948 949 950 951 952 953 954
	/* Memory manager */
	r = amdgpu_bo_init(adev);
	if (r)
		return r;

	r = gmc_v9_0_gart_init(adev);
	if (r)
		return r;

955 956 957 958 959 960 961 962 963 964 965 966
	/*
	 * number of VMs
	 * VMID 0 is reserved for System
	 * amdgpu graphics/compute will use VMIDs 1-7
	 * amdkfd will use VMIDs 8-15
	 */
	adev->vm_manager.id_mgr[AMDGPU_GFXHUB].num_ids = AMDGPU_NUM_OF_VMIDS;
	adev->vm_manager.id_mgr[AMDGPU_MMHUB].num_ids = AMDGPU_NUM_OF_VMIDS;

	amdgpu_vm_manager_init(adev);

	return 0;
A
Alex Xie 已提交
967 968 969 970 971 972
}

static int gmc_v9_0_sw_fini(void *handle)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

973
	amdgpu_gem_force_release(adev);
974
	amdgpu_vm_manager_fini(adev);
975

976 977
	if (gmc_v9_0_keep_stolen_memory(adev))
		amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL);
978

979
	amdgpu_gart_table_vram_free(adev);
A
Alex Xie 已提交
980
	amdgpu_bo_fini(adev);
981
	amdgpu_gart_fini(adev);
A
Alex Xie 已提交
982 983 984 985 986 987

	return 0;
}

static void gmc_v9_0_init_golden_registers(struct amdgpu_device *adev)
{
988

A
Alex Xie 已提交
989 990
	switch (adev->asic_type) {
	case CHIP_VEGA10:
F
Feifei Xu 已提交
991
	case CHIP_VEGA20:
992
		soc15_program_register_sequence(adev,
993
						golden_settings_mmhub_1_0_0,
994
						ARRAY_SIZE(golden_settings_mmhub_1_0_0));
995
		soc15_program_register_sequence(adev,
996
						golden_settings_athub_1_0_0,
997
						ARRAY_SIZE(golden_settings_athub_1_0_0));
A
Alex Xie 已提交
998
		break;
999 1000
	case CHIP_VEGA12:
		break;
1001
	case CHIP_RAVEN:
1002
		soc15_program_register_sequence(adev,
1003
						golden_settings_athub_1_0_0,
1004
						ARRAY_SIZE(golden_settings_athub_1_0_0));
1005
		break;
A
Alex Xie 已提交
1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021
	default:
		break;
	}
}

/**
 * gmc_v9_0_gart_enable - gart enable
 *
 * @adev: amdgpu_device pointer
 */
static int gmc_v9_0_gart_enable(struct amdgpu_device *adev)
{
	int r;
	bool value;
	u32 tmp;

1022 1023 1024
	amdgpu_device_program_register_sequence(adev,
						golden_settings_vega10_hdp,
						ARRAY_SIZE(golden_settings_vega10_hdp));
A
Alex Xie 已提交
1025

1026
	if (adev->gart.bo == NULL) {
A
Alex Xie 已提交
1027 1028 1029
		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
		return -EINVAL;
	}
1030 1031 1032
	r = amdgpu_gart_table_vram_pin(adev);
	if (r)
		return r;
A
Alex Xie 已提交
1033

1034 1035
	switch (adev->asic_type) {
	case CHIP_RAVEN:
1036
		mmhub_v1_0_update_power_gating(adev, true);
1037 1038 1039 1040 1041
		break;
	default:
		break;
	}

A
Alex Xie 已提交
1042 1043 1044 1045 1046 1047 1048 1049
	r = gfxhub_v1_0_gart_enable(adev);
	if (r)
		return r;

	r = mmhub_v1_0_gart_enable(adev);
	if (r)
		return r;

1050
	WREG32_FIELD15(HDP, 0, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 1);
A
Alex Xie 已提交
1051

1052 1053
	tmp = RREG32_SOC15(HDP, 0, mmHDP_HOST_PATH_CNTL);
	WREG32_SOC15(HDP, 0, mmHDP_HOST_PATH_CNTL, tmp);
A
Alex Xie 已提交
1054

1055
	/* After HDP is initialized, flush HDP.*/
1056
	adev->nbio_funcs->hdp_flush(adev, NULL);
1057

A
Alex Xie 已提交
1058 1059 1060 1061 1062 1063 1064
	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
		value = false;
	else
		value = true;

	gfxhub_v1_0_set_fault_enable_default(adev, value);
	mmhub_v1_0_set_fault_enable_default(adev, value);
1065
	gmc_v9_0_flush_gpu_tlb(adev, 0, 0);
A
Alex Xie 已提交
1066 1067

	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
1068
		 (unsigned)(adev->gmc.gart_size >> 20),
1069
		 (unsigned long long)amdgpu_bo_gpu_offset(adev->gart.bo));
A
Alex Xie 已提交
1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081
	adev->gart.ready = true;
	return 0;
}

static int gmc_v9_0_hw_init(void *handle)
{
	int r;
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

	/* The sequence of these two function calls matters.*/
	gmc_v9_0_init_golden_registers(adev);

1082 1083
	if (adev->mode_info.num_crtc) {
		/* Lockout access through VGA aperture*/
1084
		WREG32_FIELD15(DCE, 0, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
1085 1086

		/* disable VGA render */
1087
		WREG32_FIELD15(DCE, 0, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
1088 1089
	}

A
Alex Xie 已提交
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105
	r = gmc_v9_0_gart_enable(adev);

	return r;
}

/**
 * gmc_v9_0_gart_disable - gart disable
 *
 * @adev: amdgpu_device pointer
 *
 * This disables all VM page table.
 */
static void gmc_v9_0_gart_disable(struct amdgpu_device *adev)
{
	gfxhub_v1_0_gart_disable(adev);
	mmhub_v1_0_gart_disable(adev);
1106
	amdgpu_gart_table_vram_unpin(adev);
A
Alex Xie 已提交
1107 1108 1109 1110 1111 1112
}

static int gmc_v9_0_hw_fini(void *handle)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

1113 1114 1115 1116 1117 1118
	if (amdgpu_sriov_vf(adev)) {
		/* full access mode, so don't touch any GMC register */
		DRM_DEBUG("For SRIOV client, shouldn't do anything.\n");
		return 0;
	}

1119
	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
A
Alex Xie 已提交
1120 1121 1122 1123 1124 1125 1126 1127 1128
	gmc_v9_0_gart_disable(adev);

	return 0;
}

static int gmc_v9_0_suspend(void *handle)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

1129
	return gmc_v9_0_hw_fini(adev);
A
Alex Xie 已提交
1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140
}

static int gmc_v9_0_resume(void *handle)
{
	int r;
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

	r = gmc_v9_0_hw_init(adev);
	if (r)
		return r;

1141
	amdgpu_vmid_reset_all(adev);
A
Alex Xie 已提交
1142

1143
	return 0;
A
Alex Xie 已提交
1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
}

static bool gmc_v9_0_is_idle(void *handle)
{
	/* MC is always ready in GMC v9.*/
	return true;
}

static int gmc_v9_0_wait_for_idle(void *handle)
{
	/* There is no need to wait for MC idle in GMC v9.*/
	return 0;
}

static int gmc_v9_0_soft_reset(void *handle)
{
	/* XXX for emulation.*/
	return 0;
}

static int gmc_v9_0_set_clockgating_state(void *handle,
					enum amd_clockgating_state state)
{
1167 1168 1169
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

	return mmhub_v1_0_set_clockgating(adev, state);
A
Alex Xie 已提交
1170 1171
}

1172 1173 1174 1175 1176 1177 1178
static void gmc_v9_0_get_clockgating_state(void *handle, u32 *flags)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

	mmhub_v1_0_get_clockgating(adev, flags);
}

A
Alex Xie 已提交
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
static int gmc_v9_0_set_powergating_state(void *handle,
					enum amd_powergating_state state)
{
	return 0;
}

const struct amd_ip_funcs gmc_v9_0_ip_funcs = {
	.name = "gmc_v9_0",
	.early_init = gmc_v9_0_early_init,
	.late_init = gmc_v9_0_late_init,
	.sw_init = gmc_v9_0_sw_init,
	.sw_fini = gmc_v9_0_sw_fini,
	.hw_init = gmc_v9_0_hw_init,
	.hw_fini = gmc_v9_0_hw_fini,
	.suspend = gmc_v9_0_suspend,
	.resume = gmc_v9_0_resume,
	.is_idle = gmc_v9_0_is_idle,
	.wait_for_idle = gmc_v9_0_wait_for_idle,
	.soft_reset = gmc_v9_0_soft_reset,
	.set_clockgating_state = gmc_v9_0_set_clockgating_state,
	.set_powergating_state = gmc_v9_0_set_powergating_state,
1200
	.get_clockgating_state = gmc_v9_0_get_clockgating_state,
A
Alex Xie 已提交
1201 1202 1203 1204 1205 1206 1207 1208 1209 1210
};

const struct amdgpu_ip_block_version gmc_v9_0_ip_block =
{
	.type = AMD_IP_BLOCK_TYPE_GMC,
	.major = 9,
	.minor = 0,
	.rev = 0,
	.funcs = &gmc_v9_0_ip_funcs,
};