gmc_v9_0.c 33.6 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 46

#include "gfxhub_v1_0.h"
#include "mmhub_v1_0.h"

47 48
#include "ivsrcid/vmc/irqsrcs_vmc_1_0.h"

49 50 51 52 53 54 55 56
/* 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
57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
/* 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,
};

74
static const struct soc15_reg_golden golden_settings_mmhub_1_0_0[] =
75
{
76 77
	SOC15_REG_GOLDEN_VALUE(MMHUB, 0, mmDAGB1_WRCLI2, 0x00000007, 0xfe5fe0fa),
	SOC15_REG_GOLDEN_VALUE(MMHUB, 0, mmMMEA1_DRAM_WR_CLI2GRP_MAP0, 0x00000030, 0x55555565)
78 79
};

80
static const struct soc15_reg_golden golden_settings_athub_1_0_0[] =
81
{
82 83
	SOC15_REG_GOLDEN_VALUE(ATHUB, 0, mmRPB_ARB_CNTL, 0x0000ff00, 0x00000800),
	SOC15_REG_GOLDEN_VALUE(ATHUB, 0, mmRPB_ARB_CNTL2, 0x00ff00ff, 0x00080008)
84 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
/* 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
201 202 203 204 205 206
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;
207
	u32 tmp, reg, bits, i, j;
A
Alex Xie 已提交
208

209 210 211 212 213 214 215 216
	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
217 218
	switch (state) {
	case AMDGPU_IRQ_STATE_DISABLE:
219 220 221 222 223 224 225 226
		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
227 228 229
		}
		break;
	case AMDGPU_IRQ_STATE_ENABLE:
230 231 232 233 234 235 236 237
		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
238 239 240 241 242 243 244 245 246 247 248 249
		}
	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)
{
250
	struct amdgpu_vmhub *hub = &adev->vmhub[entry->vmid_src];
251
	uint32_t status = 0;
A
Alex Xie 已提交
252 253 254 255 256
	u64 addr;

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

257
	if (!amdgpu_sriov_vf(adev)) {
258 259
		status = RREG32(hub->vm_l2_pro_fault_status);
		WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
260
	}
A
Alex Xie 已提交
261

262
	if (printk_ratelimit()) {
263 264 265 266
		struct amdgpu_task_info task_info = { 0 };

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

267
		dev_err(adev->dev,
268
			"[%s] VMC page fault (src_id:%u ring:%u vmid:%u pasid:%u, for process %s pid %d thread %s pid %d\n)\n",
269 270
			entry->vmid_src ? "mmhub" : "gfxhub",
			entry->src_id, entry->ring_id, entry->vmid,
271 272
			entry->pasid, task_info.process_name, task_info.tgid,
			task_info.task_name, task_info.pid);
273
		dev_err(adev->dev, "  in page starting at address 0x%016llx from %d\n",
274 275 276 277 278
			addr, entry->client_id);
		if (!amdgpu_sriov_vf(adev))
			dev_err(adev->dev,
				"VM_L2_PROTECTION_FAULT_STATUS:0x%08X\n",
				status);
279
	}
A
Alex Xie 已提交
280 281 282 283 284 285 286 287 288 289 290

	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)
{
291 292
	adev->gmc.vm_fault.num_types = 1;
	adev->gmc.vm_fault.funcs = &gmc_v9_0_irq_funcs;
A
Alex Xie 已提交
293 294
}

295
static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid)
296 297 298
{
	u32 req = 0;

299
	/* invalidate using legacy mode on vmid*/
300
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ,
301
			    PER_VMID_INVALIDATE_REQ, 1 << vmid);
302 303 304 305 306 307 308 309 310 311 312 313
	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 0);
	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;
}

314
static signed long  amdgpu_kiq_reg_write_reg_wait(struct amdgpu_device *adev,
315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
						  uint32_t reg0, uint32_t reg1,
						  uint32_t ref, uint32_t mask)
{
	signed long r, cnt = 0;
	unsigned long flags;
	uint32_t seq;
	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
	struct amdgpu_ring *ring = &kiq->ring;

	spin_lock_irqsave(&kiq->ring_lock, flags);

	amdgpu_ring_alloc(ring, 32);
	amdgpu_ring_emit_reg_write_reg_wait(ring, reg0, reg1,
					    ref, mask);
	amdgpu_fence_emit_polling(ring, &seq);
	amdgpu_ring_commit(ring);
	spin_unlock_irqrestore(&kiq->ring_lock, flags);

	r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);

335 336
	/* don't wait anymore for IRQ context */
	if (r < 1 && in_interrupt())
337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355
		goto failed_kiq;

	might_sleep();

	while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
		msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
		r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
	}

	if (cnt > MAX_KIQ_REG_TRY)
		goto failed_kiq;

	return 0;

failed_kiq:
	pr_err("failed to invalidate tlb with kiq\n");
	return r;
}

A
Alex Xie 已提交
356 357 358 359 360 361 362 363
/*
 * 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.
 */

/**
364
 * gmc_v9_0_flush_gpu_tlb - gart tlb flush callback
A
Alex Xie 已提交
365 366 367 368 369 370
 *
 * @adev: amdgpu_device pointer
 * @vmid: vm instance to flush
 *
 * Flush the TLB for the requested page table.
 */
371
static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
A
Alex Xie 已提交
372 373 374 375 376
					uint32_t vmid)
{
	/* Use register 17 for GART */
	const unsigned eng = 17;
	unsigned i, j;
377
	int r;
A
Alex Xie 已提交
378 379 380

	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
		struct amdgpu_vmhub *hub = &adev->vmhub[i];
381
		u32 tmp = gmc_v9_0_get_invalidate_req(vmid);
A
Alex Xie 已提交
382

383
		if (adev->gfx.kiq.ring.ready &&
384 385
		    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) &&
		    !adev->in_gpu_reset) {
386 387 388 389 390
			r = amdgpu_kiq_reg_write_reg_wait(adev, hub->vm_inv_eng0_req + eng,
				hub->vm_inv_eng0_ack + eng, tmp, 1 << vmid);
			if (!r)
				continue;
		}
391 392 393

		spin_lock(&adev->gmc.invalidate_lock);

394
		WREG32_NO_KIQ(hub->vm_inv_eng0_req + eng, tmp);
A
Alex Xie 已提交
395 396 397

		/* Busy wait for ACK.*/
		for (j = 0; j < 100; j++) {
398
			tmp = RREG32_NO_KIQ(hub->vm_inv_eng0_ack + eng);
A
Alex Xie 已提交
399 400 401 402 403
			tmp &= 1 << vmid;
			if (tmp)
				break;
			cpu_relax();
		}
404 405
		if (j < 100) {
			spin_unlock(&adev->gmc.invalidate_lock);
A
Alex Xie 已提交
406
			continue;
407
		}
A
Alex Xie 已提交
408 409 410

		/* Wait for ACK with a delay.*/
		for (j = 0; j < adev->usec_timeout; j++) {
411
			tmp = RREG32_NO_KIQ(hub->vm_inv_eng0_ack + eng);
A
Alex Xie 已提交
412 413 414 415 416
			tmp &= 1 << vmid;
			if (tmp)
				break;
			udelay(1);
		}
417 418
		if (j < adev->usec_timeout) {
			spin_unlock(&adev->gmc.invalidate_lock);
A
Alex Xie 已提交
419
			continue;
420 421
		}
		spin_unlock(&adev->gmc.invalidate_lock);
A
Alex Xie 已提交
422 423 424 425
		DRM_ERROR("Timeout waiting for VM flush ACK!\n");
	}
}

426
static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
427
					    unsigned vmid, uint64_t pd_addr)
428
{
429 430
	struct amdgpu_device *adev = ring->adev;
	struct amdgpu_vmhub *hub = &adev->vmhub[ring->funcs->vmhub];
431 432 433 434 435 436 437 438 439
	uint32_t req = gmc_v9_0_get_invalidate_req(vmid);
	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));

440 441 442
	amdgpu_ring_emit_reg_write_reg_wait(ring, hub->vm_inv_eng0_req + eng,
					    hub->vm_inv_eng0_ack + eng,
					    req, 1 << vmid);
443

444 445 446
	return pd_addr;
}

447 448 449 450 451 452 453 454 455 456 457 458 459 460
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
461
/**
462
 * gmc_v9_0_set_pte_pde - update the page tables using MMIO
A
Alex Xie 已提交
463 464 465 466 467 468 469 470 471
 *
 * @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.
 */
472 473 474
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
475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559
{
	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;
}

560 561
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
562
{
563
	if (!(*flags & AMDGPU_PDE_PTE) && !(*flags & AMDGPU_PTE_SYSTEM))
564
		*addr = adev->vm_manager.vram_base_offset + *addr -
565
			adev->gmc.vram_start;
566
	BUG_ON(*addr & 0xFFFF00000000003FULL);
567

568
	if (!adev->gmc.translate_further)
569 570 571 572 573 574 575 576 577 578 579 580 581
		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
582 583
}

584 585
static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
	.flush_gpu_tlb = gmc_v9_0_flush_gpu_tlb,
586
	.emit_flush_gpu_tlb = gmc_v9_0_emit_flush_gpu_tlb,
587
	.emit_pasid_mapping = gmc_v9_0_emit_pasid_mapping,
588
	.set_pte_pde = gmc_v9_0_set_pte_pde,
589 590
	.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
591 592
};

593
static void gmc_v9_0_set_gmc_funcs(struct amdgpu_device *adev)
A
Alex Xie 已提交
594
{
595 596
	if (adev->gmc.gmc_funcs == NULL)
		adev->gmc.gmc_funcs = &gmc_v9_0_gmc_funcs;
A
Alex Xie 已提交
597 598 599 600 601 602
}

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

603
	gmc_v9_0_set_gmc_funcs(adev);
A
Alex Xie 已提交
604 605
	gmc_v9_0_set_irq_funcs(adev);

606 607 608
	adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
	adev->gmc.shared_aperture_end =
		adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
609
	adev->gmc.private_aperture_start = 0x1000000000000000ULL;
610 611
	adev->gmc.private_aperture_end =
		adev->gmc.private_aperture_start + (4ULL << 30) - 1;
612

A
Alex Xie 已提交
613 614 615
	return 0;
}

616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681
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) {
682
			DRM_DEBUG("ecc: WrEccEn is not set\n");
683 684 685
			++lost_sheep;
		}
		if (!fv2) {
686
			DRM_DEBUG("ecc: RdEccEn is not set\n");
687 688 689 690 691 692 693 694
			++lost_sheep;
		}
	}

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

A
Alex Xie 已提交
695 696 697
static int gmc_v9_0_late_init(void *handle)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
698 699 700 701 702 703 704 705 706 707
	/*
	 * 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 };
708
	unsigned i;
709
	int r;
710

711 712 713 714 715
	/*
	 * TODO - Uncomment once GART corruption issue is fixed.
	 */
	/* amdgpu_bo_late_init(adev); */

716 717 718 719 720
	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]++;
721 722 723
		dev_info(adev->dev, "ring %u(%s) uses VM inv eng %u on hub %u\n",
			 ring->idx, ring->name, ring->vm_inv_eng,
			 ring->funcs->vmhub);
724 725
	}

726
	/* Engine 16 is used for KFD and 17 for GART flushes */
727
	for(i = 0; i < AMDGPU_MAX_VMHUBS; ++i)
728
		BUG_ON(vm_inv_eng[i] > 16);
729

730
	if (adev->asic_type == CHIP_VEGA10 && !amdgpu_sriov_vf(adev)) {
731 732 733 734 735
		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");
736
			adev->df_funcs->enable_ecc_force_par_wr_rmw(adev, false);
737 738 739 740
		} else {
			DRM_ERROR("gmc_v9_0_ecc_available() failed. r: %d\n", r);
			return r;
		}
741 742
	}

743
	return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
A
Alex Xie 已提交
744 745 746
}

static void gmc_v9_0_vram_gtt_location(struct amdgpu_device *adev,
747
					struct amdgpu_gmc *mc)
A
Alex Xie 已提交
748
{
749 750 751
	u64 base = 0;
	if (!amdgpu_sriov_vf(adev))
		base = mmhub_v1_0_get_fb_location(adev);
752
	amdgpu_device_vram_location(adev, &adev->gmc, base);
753
	amdgpu_device_gart_location(adev, mc);
754
	/* base offset of vram pages */
755
	adev->vm_manager.vram_base_offset = gfxhub_v1_0_get_mc_fb_offset(adev);
A
Alex Xie 已提交
756 757 758 759 760 761 762 763 764 765 766 767 768 769
}

/**
 * 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;
770
	int r;
A
Alex Xie 已提交
771

772 773
	if (amdgpu_emu_mode != 1)
		adev->gmc.vram_width = amdgpu_atomfirmware_get_vram_width(adev);
774
	if (!adev->gmc.vram_width) {
775
		/* hbm memory channel size */
776 777 778 779
		if (adev->flags & AMD_IS_APU)
			chansize = 64;
		else
			chansize = 128;
780

781
		numchan = adev->df_funcs->get_hbm_channel_number(adev);
782
		adev->gmc.vram_width = numchan * chansize;
A
Alex Xie 已提交
783 784 785
	}

	/* size in MB on si */
786
	adev->gmc.mc_vram_size =
787
		adev->nbio_funcs->get_memsize(adev) * 1024ULL * 1024ULL;
788
	adev->gmc.real_vram_size = adev->gmc.mc_vram_size;
789 790 791 792 793 794

	if (!(adev->flags & AMD_IS_APU)) {
		r = amdgpu_device_resize_fb_bar(adev);
		if (r)
			return r;
	}
795 796
	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
797

798 799 800 801 802 803
#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
804
	/* In case the PCI BAR is larger than the actual amount of vram */
805 806 807
	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
808

809 810 811 812
	/* set the gart size */
	if (amdgpu_gart_size == -1) {
		switch (adev->asic_type) {
		case CHIP_VEGA10:  /* all engines support GPUVM */
813
		case CHIP_VEGA12:  /* all engines support GPUVM */
F
Feifei Xu 已提交
814
		case CHIP_VEGA20:
815
		default:
816
			adev->gmc.gart_size = 512ULL << 20;
817 818
			break;
		case CHIP_RAVEN:   /* DCE SG support */
819
			adev->gmc.gart_size = 1024ULL << 20;
820 821 822
			break;
		}
	} else {
823
		adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
824 825
	}

826
	gmc_v9_0_vram_gtt_location(adev, &adev->gmc);
A
Alex Xie 已提交
827 828 829 830 831 832 833 834

	return 0;
}

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

835
	if (adev->gart.bo) {
A
Alex Xie 已提交
836 837 838 839 840 841 842 843 844 845 846 847 848
		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);
}

849 850 851 852 853 854 855
static unsigned gmc_v9_0_get_vbios_fb_size(struct amdgpu_device *adev)
{
#if 0
	u32 d1vga_control = RREG32_SOC15(DCE, 0, mmD1VGA_CONTROL);
#endif
	unsigned size;

856 857 858 859 860 861 862
	/*
	 * TODO Remove once GART corruption is resolved
	 * Check related code in gmc_v9_0_sw_fini
	 * */
	size = 9 * 1024 * 1024;

#if 0
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
	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:
		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;
890 891

#endif
892 893 894
	return size;
}

A
Alex Xie 已提交
895 896 897 898 899 900
static int gmc_v9_0_sw_init(void *handle)
{
	int r;
	int dma_bits;
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

901
	gfxhub_v1_0_init(adev);
902
	mmhub_v1_0_init(adev);
903

904
	spin_lock_init(&adev->gmc.invalidate_lock);
A
Alex Xie 已提交
905

906
	adev->gmc.vram_type = amdgpu_atomfirmware_get_vram_type(adev);
907 908
	switch (adev->asic_type) {
	case CHIP_RAVEN:
909
		if (adev->rev_id == 0x0 || adev->rev_id == 0x1) {
910
			amdgpu_vm_adjust_size(adev, 256 * 1024, 9, 3, 48);
911 912 913
		} else {
			/* vm_size is 128TB + 512GB for legacy 3-level page support */
			amdgpu_vm_adjust_size(adev, 128 * 1024 + 512, 9, 2, 48);
914
			adev->gmc.translate_further =
915 916
				adev->vm_manager.num_level > 1;
		}
917 918
		break;
	case CHIP_VEGA10:
919
	case CHIP_VEGA12:
F
Feifei Xu 已提交
920
	case CHIP_VEGA20:
921 922 923 924 925
		/*
		 * To fulfill 4-level page support,
		 * vm size is 256TB (48bit), maximum size of Vega10,
		 * block size 512 (9bit)
		 */
926
		amdgpu_vm_adjust_size(adev, 256 * 1024, 9, 3, 48);
927 928 929
		break;
	default:
		break;
A
Alex Xie 已提交
930 931 932
	}

	/* This interrupt is VMC page fault.*/
933
	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VMC, VMC_1_0__SRCID__VM_FAULT,
934
				&adev->gmc.vm_fault);
935
	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_UTCL2, UTCL2_1_0__SRCID__FAULT,
936
				&adev->gmc.vm_fault);
A
Alex Xie 已提交
937 938 939 940 941 942 943 944

	if (r)
		return r;

	/* Set the internal MC address mask
	 * This is the max address of the GPU's
	 * internal address space.
	 */
945
	adev->gmc.mc_mask = 0xffffffffffffULL; /* 48 bit MC */
A
Alex Xie 已提交
946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964

	/* 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");
	}
965
	adev->need_swiotlb = drm_get_max_iomem() > ((u64)1 << dma_bits);
A
Alex Xie 已提交
966 967 968 969 970

	r = gmc_v9_0_mc_init(adev);
	if (r)
		return r;

971 972
	adev->gmc.stolen_size = gmc_v9_0_get_vbios_fb_size(adev);

A
Alex Xie 已提交
973 974 975 976 977 978 979 980 981
	/* Memory manager */
	r = amdgpu_bo_init(adev);
	if (r)
		return r;

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

982 983 984 985 986 987 988 989 990 991 992 993
	/*
	 * 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
994 995 996 997 998 999
}

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

1000
	amdgpu_gem_force_release(adev);
1001
	amdgpu_vm_manager_fini(adev);
1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013

	/*
	* 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
	*/
	amdgpu_bo_free_kernel(&adev->stolen_vga_memory, NULL, NULL);

1014
	amdgpu_gart_table_vram_free(adev);
A
Alex Xie 已提交
1015
	amdgpu_bo_fini(adev);
1016
	amdgpu_gart_fini(adev);
A
Alex Xie 已提交
1017 1018 1019 1020 1021 1022

	return 0;
}

static void gmc_v9_0_init_golden_registers(struct amdgpu_device *adev)
{
1023

A
Alex Xie 已提交
1024 1025
	switch (adev->asic_type) {
	case CHIP_VEGA10:
F
Feifei Xu 已提交
1026
	case CHIP_VEGA20:
1027
		soc15_program_register_sequence(adev,
1028
						golden_settings_mmhub_1_0_0,
1029
						ARRAY_SIZE(golden_settings_mmhub_1_0_0));
1030
		soc15_program_register_sequence(adev,
1031
						golden_settings_athub_1_0_0,
1032
						ARRAY_SIZE(golden_settings_athub_1_0_0));
A
Alex Xie 已提交
1033
		break;
1034 1035
	case CHIP_VEGA12:
		break;
1036
	case CHIP_RAVEN:
1037
		soc15_program_register_sequence(adev,
1038
						golden_settings_athub_1_0_0,
1039
						ARRAY_SIZE(golden_settings_athub_1_0_0));
1040
		break;
A
Alex Xie 已提交
1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056
	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;

1057 1058 1059
	amdgpu_device_program_register_sequence(adev,
						golden_settings_vega10_hdp,
						ARRAY_SIZE(golden_settings_vega10_hdp));
A
Alex Xie 已提交
1060

1061
	if (adev->gart.bo == NULL) {
A
Alex Xie 已提交
1062 1063 1064
		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
		return -EINVAL;
	}
1065 1066 1067
	r = amdgpu_gart_table_vram_pin(adev);
	if (r)
		return r;
A
Alex Xie 已提交
1068

1069 1070 1071
	switch (adev->asic_type) {
	case CHIP_RAVEN:
		mmhub_v1_0_initialize_power_gating(adev);
1072
		mmhub_v1_0_update_power_gating(adev, true);
1073 1074 1075 1076 1077
		break;
	default:
		break;
	}

A
Alex Xie 已提交
1078 1079 1080 1081 1082 1083 1084 1085
	r = gfxhub_v1_0_gart_enable(adev);
	if (r)
		return r;

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

1086
	WREG32_FIELD15(HDP, 0, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 1);
A
Alex Xie 已提交
1087

1088 1089
	tmp = RREG32_SOC15(HDP, 0, mmHDP_HOST_PATH_CNTL);
	WREG32_SOC15(HDP, 0, mmHDP_HOST_PATH_CNTL, tmp);
A
Alex Xie 已提交
1090

1091
	/* After HDP is initialized, flush HDP.*/
1092
	adev->nbio_funcs->hdp_flush(adev, NULL);
1093

A
Alex Xie 已提交
1094 1095 1096 1097 1098 1099 1100
	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);
1101
	gmc_v9_0_flush_gpu_tlb(adev, 0);
A
Alex Xie 已提交
1102 1103

	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
1104
		 (unsigned)(adev->gmc.gart_size >> 20),
1105
		 (unsigned long long)amdgpu_bo_gpu_offset(adev->gart.bo));
A
Alex Xie 已提交
1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117
	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);

1118 1119
	if (adev->mode_info.num_crtc) {
		/* Lockout access through VGA aperture*/
1120
		WREG32_FIELD15(DCE, 0, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
1121 1122

		/* disable VGA render */
1123
		WREG32_FIELD15(DCE, 0, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
1124 1125
	}

A
Alex Xie 已提交
1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
	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);
1142
	amdgpu_gart_table_vram_unpin(adev);
A
Alex Xie 已提交
1143 1144 1145 1146 1147 1148
}

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

1149 1150 1151 1152 1153 1154
	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;
	}

1155
	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
A
Alex Xie 已提交
1156 1157 1158 1159 1160 1161 1162 1163 1164
	gmc_v9_0_gart_disable(adev);

	return 0;
}

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

1165
	return gmc_v9_0_hw_fini(adev);
A
Alex Xie 已提交
1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176
}

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;

1177
	amdgpu_vmid_reset_all(adev);
A
Alex Xie 已提交
1178

1179
	return 0;
A
Alex Xie 已提交
1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202
}

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)
{
1203 1204 1205
	struct amdgpu_device *adev = (struct amdgpu_device *)handle;

	return mmhub_v1_0_set_clockgating(adev, state);
A
Alex Xie 已提交
1206 1207
}

1208 1209 1210 1211 1212 1213 1214
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
1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235
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,
1236
	.get_clockgating_state = gmc_v9_0_get_clockgating_state,
A
Alex Xie 已提交
1237 1238 1239 1240 1241 1242 1243 1244 1245 1246
};

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,
};