Linux Audio

Check our new training course

Linux kernel drivers training

May 6-19, 2025
Register
Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2011-2014 NVIDIA CORPORATION.  All rights reserved.
   4 */
   5
   6#include <linux/bitops.h>
   7#include <linux/debugfs.h>
   8#include <linux/err.h>
   9#include <linux/iommu.h>
  10#include <linux/kernel.h>
  11#include <linux/of.h>
  12#include <linux/of_platform.h>
  13#include <linux/pci.h>
  14#include <linux/platform_device.h>
  15#include <linux/slab.h>
  16#include <linux/spinlock.h>
  17#include <linux/dma-mapping.h>
  18
  19#include <soc/tegra/ahb.h>
  20#include <soc/tegra/mc.h>
  21
  22struct tegra_smmu_group {
  23	struct list_head list;
  24	struct tegra_smmu *smmu;
  25	const struct tegra_smmu_group_soc *soc;
  26	struct iommu_group *group;
  27	unsigned int swgroup;
  28};
  29
  30struct tegra_smmu {
  31	void __iomem *regs;
  32	struct device *dev;
  33
  34	struct tegra_mc *mc;
  35	const struct tegra_smmu_soc *soc;
  36
  37	struct list_head groups;
  38
  39	unsigned long pfn_mask;
  40	unsigned long tlb_mask;
  41
  42	unsigned long *asids;
  43	struct mutex lock;
  44
  45	struct list_head list;
  46
  47	struct dentry *debugfs;
  48
  49	struct iommu_device iommu;	/* IOMMU Core code handle */
  50};
  51
  52struct tegra_smmu_as {
  53	struct iommu_domain domain;
  54	struct tegra_smmu *smmu;
  55	unsigned int use_count;
  56	spinlock_t lock;
  57	u32 *count;
  58	struct page **pts;
  59	struct page *pd;
  60	dma_addr_t pd_dma;
  61	unsigned id;
  62	u32 attr;
  63};
  64
  65static struct tegra_smmu_as *to_smmu_as(struct iommu_domain *dom)
  66{
  67	return container_of(dom, struct tegra_smmu_as, domain);
  68}
  69
  70static inline void smmu_writel(struct tegra_smmu *smmu, u32 value,
  71			       unsigned long offset)
  72{
  73	writel(value, smmu->regs + offset);
  74}
  75
  76static inline u32 smmu_readl(struct tegra_smmu *smmu, unsigned long offset)
  77{
  78	return readl(smmu->regs + offset);
  79}
  80
  81#define SMMU_CONFIG 0x010
  82#define  SMMU_CONFIG_ENABLE (1 << 0)
  83
  84#define SMMU_TLB_CONFIG 0x14
  85#define  SMMU_TLB_CONFIG_HIT_UNDER_MISS (1 << 29)
  86#define  SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION (1 << 28)
  87#define  SMMU_TLB_CONFIG_ACTIVE_LINES(smmu) \
  88	((smmu)->soc->num_tlb_lines & (smmu)->tlb_mask)
  89
  90#define SMMU_PTC_CONFIG 0x18
  91#define  SMMU_PTC_CONFIG_ENABLE (1 << 29)
  92#define  SMMU_PTC_CONFIG_REQ_LIMIT(x) (((x) & 0x0f) << 24)
  93#define  SMMU_PTC_CONFIG_INDEX_MAP(x) ((x) & 0x3f)
  94
  95#define SMMU_PTB_ASID 0x01c
  96#define  SMMU_PTB_ASID_VALUE(x) ((x) & 0x7f)
  97
  98#define SMMU_PTB_DATA 0x020
  99#define  SMMU_PTB_DATA_VALUE(dma, attr) ((dma) >> 12 | (attr))
 100
 101#define SMMU_MK_PDE(dma, attr) ((dma) >> SMMU_PTE_SHIFT | (attr))
 102
 103#define SMMU_TLB_FLUSH 0x030
 104#define  SMMU_TLB_FLUSH_VA_MATCH_ALL     (0 << 0)
 105#define  SMMU_TLB_FLUSH_VA_MATCH_SECTION (2 << 0)
 106#define  SMMU_TLB_FLUSH_VA_MATCH_GROUP   (3 << 0)
 107#define  SMMU_TLB_FLUSH_VA_SECTION(addr) ((((addr) & 0xffc00000) >> 12) | \
 108					  SMMU_TLB_FLUSH_VA_MATCH_SECTION)
 109#define  SMMU_TLB_FLUSH_VA_GROUP(addr)   ((((addr) & 0xffffc000) >> 12) | \
 110					  SMMU_TLB_FLUSH_VA_MATCH_GROUP)
 111#define  SMMU_TLB_FLUSH_ASID_MATCH       (1 << 31)
 112
 113#define SMMU_PTC_FLUSH 0x034
 114#define  SMMU_PTC_FLUSH_TYPE_ALL (0 << 0)
 115#define  SMMU_PTC_FLUSH_TYPE_ADR (1 << 0)
 116
 117#define SMMU_PTC_FLUSH_HI 0x9b8
 118#define  SMMU_PTC_FLUSH_HI_MASK 0x3
 119
 120/* per-SWGROUP SMMU_*_ASID register */
 121#define SMMU_ASID_ENABLE (1 << 31)
 122#define SMMU_ASID_MASK 0x7f
 123#define SMMU_ASID_VALUE(x) ((x) & SMMU_ASID_MASK)
 124
 125/* page table definitions */
 126#define SMMU_NUM_PDE 1024
 127#define SMMU_NUM_PTE 1024
 128
 129#define SMMU_SIZE_PD (SMMU_NUM_PDE * 4)
 130#define SMMU_SIZE_PT (SMMU_NUM_PTE * 4)
 131
 132#define SMMU_PDE_SHIFT 22
 133#define SMMU_PTE_SHIFT 12
 134
 135#define SMMU_PAGE_MASK		(~(SMMU_SIZE_PT-1))
 136#define SMMU_OFFSET_IN_PAGE(x)	((unsigned long)(x) & ~SMMU_PAGE_MASK)
 137#define SMMU_PFN_PHYS(x)	((phys_addr_t)(x) << SMMU_PTE_SHIFT)
 138#define SMMU_PHYS_PFN(x)	((unsigned long)((x) >> SMMU_PTE_SHIFT))
 139
 140#define SMMU_PD_READABLE	(1 << 31)
 141#define SMMU_PD_WRITABLE	(1 << 30)
 142#define SMMU_PD_NONSECURE	(1 << 29)
 143
 144#define SMMU_PDE_READABLE	(1 << 31)
 145#define SMMU_PDE_WRITABLE	(1 << 30)
 146#define SMMU_PDE_NONSECURE	(1 << 29)
 147#define SMMU_PDE_NEXT		(1 << 28)
 148
 149#define SMMU_PTE_READABLE	(1 << 31)
 150#define SMMU_PTE_WRITABLE	(1 << 30)
 151#define SMMU_PTE_NONSECURE	(1 << 29)
 152
 153#define SMMU_PDE_ATTR		(SMMU_PDE_READABLE | SMMU_PDE_WRITABLE | \
 154				 SMMU_PDE_NONSECURE)
 155
 156static unsigned int iova_pd_index(unsigned long iova)
 157{
 158	return (iova >> SMMU_PDE_SHIFT) & (SMMU_NUM_PDE - 1);
 159}
 160
 161static unsigned int iova_pt_index(unsigned long iova)
 162{
 163	return (iova >> SMMU_PTE_SHIFT) & (SMMU_NUM_PTE - 1);
 164}
 165
 166static bool smmu_dma_addr_valid(struct tegra_smmu *smmu, dma_addr_t addr)
 167{
 168	addr >>= 12;
 169	return (addr & smmu->pfn_mask) == addr;
 170}
 171
 172static dma_addr_t smmu_pde_to_dma(struct tegra_smmu *smmu, u32 pde)
 173{
 174	return (dma_addr_t)(pde & smmu->pfn_mask) << 12;
 175}
 176
 177static void smmu_flush_ptc_all(struct tegra_smmu *smmu)
 178{
 179	smmu_writel(smmu, SMMU_PTC_FLUSH_TYPE_ALL, SMMU_PTC_FLUSH);
 180}
 181
 182static inline void smmu_flush_ptc(struct tegra_smmu *smmu, dma_addr_t dma,
 183				  unsigned long offset)
 184{
 185	u32 value;
 186
 187	offset &= ~(smmu->mc->soc->atom_size - 1);
 188
 189	if (smmu->mc->soc->num_address_bits > 32) {
 190#ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
 191		value = (dma >> 32) & SMMU_PTC_FLUSH_HI_MASK;
 192#else
 193		value = 0;
 194#endif
 195		smmu_writel(smmu, value, SMMU_PTC_FLUSH_HI);
 196	}
 197
 198	value = (dma + offset) | SMMU_PTC_FLUSH_TYPE_ADR;
 199	smmu_writel(smmu, value, SMMU_PTC_FLUSH);
 200}
 201
 202static inline void smmu_flush_tlb(struct tegra_smmu *smmu)
 203{
 204	smmu_writel(smmu, SMMU_TLB_FLUSH_VA_MATCH_ALL, SMMU_TLB_FLUSH);
 205}
 206
 207static inline void smmu_flush_tlb_asid(struct tegra_smmu *smmu,
 208				       unsigned long asid)
 209{
 210	u32 value;
 211
 212	if (smmu->soc->num_asids == 4)
 213		value = (asid & 0x3) << 29;
 214	else
 215		value = (asid & 0x7f) << 24;
 216
 217	value |= SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_VA_MATCH_ALL;
 218	smmu_writel(smmu, value, SMMU_TLB_FLUSH);
 219}
 220
 221static inline void smmu_flush_tlb_section(struct tegra_smmu *smmu,
 222					  unsigned long asid,
 223					  unsigned long iova)
 224{
 225	u32 value;
 226
 227	if (smmu->soc->num_asids == 4)
 228		value = (asid & 0x3) << 29;
 229	else
 230		value = (asid & 0x7f) << 24;
 231
 232	value |= SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_VA_SECTION(iova);
 233	smmu_writel(smmu, value, SMMU_TLB_FLUSH);
 234}
 235
 236static inline void smmu_flush_tlb_group(struct tegra_smmu *smmu,
 237					unsigned long asid,
 238					unsigned long iova)
 239{
 240	u32 value;
 241
 242	if (smmu->soc->num_asids == 4)
 243		value = (asid & 0x3) << 29;
 244	else
 245		value = (asid & 0x7f) << 24;
 246
 247	value |= SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_VA_GROUP(iova);
 248	smmu_writel(smmu, value, SMMU_TLB_FLUSH);
 249}
 250
 251static inline void smmu_flush(struct tegra_smmu *smmu)
 252{
 253	smmu_readl(smmu, SMMU_PTB_ASID);
 254}
 255
 256static int tegra_smmu_alloc_asid(struct tegra_smmu *smmu, unsigned int *idp)
 257{
 258	unsigned long id;
 259
 260	id = find_first_zero_bit(smmu->asids, smmu->soc->num_asids);
 261	if (id >= smmu->soc->num_asids)
 262		return -ENOSPC;
 263
 264	set_bit(id, smmu->asids);
 265	*idp = id;
 266
 267	return 0;
 268}
 269
 270static void tegra_smmu_free_asid(struct tegra_smmu *smmu, unsigned int id)
 271{
 272	clear_bit(id, smmu->asids);
 273}
 274
 275static struct iommu_domain *tegra_smmu_domain_alloc_paging(struct device *dev)
 276{
 277	struct tegra_smmu_as *as;
 278
 279	as = kzalloc(sizeof(*as), GFP_KERNEL);
 280	if (!as)
 281		return NULL;
 282
 283	as->attr = SMMU_PD_READABLE | SMMU_PD_WRITABLE | SMMU_PD_NONSECURE;
 284
 285	as->pd = alloc_page(GFP_KERNEL | __GFP_DMA | __GFP_ZERO);
 286	if (!as->pd) {
 287		kfree(as);
 288		return NULL;
 289	}
 290
 291	as->count = kcalloc(SMMU_NUM_PDE, sizeof(u32), GFP_KERNEL);
 292	if (!as->count) {
 293		__free_page(as->pd);
 294		kfree(as);
 295		return NULL;
 296	}
 297
 298	as->pts = kcalloc(SMMU_NUM_PDE, sizeof(*as->pts), GFP_KERNEL);
 299	if (!as->pts) {
 300		kfree(as->count);
 301		__free_page(as->pd);
 302		kfree(as);
 303		return NULL;
 304	}
 305
 306	spin_lock_init(&as->lock);
 307
 308	/* setup aperture */
 309	as->domain.geometry.aperture_start = 0;
 310	as->domain.geometry.aperture_end = 0xffffffff;
 311	as->domain.geometry.force_aperture = true;
 312
 313	return &as->domain;
 314}
 315
 316static void tegra_smmu_domain_free(struct iommu_domain *domain)
 317{
 318	struct tegra_smmu_as *as = to_smmu_as(domain);
 319
 320	/* TODO: free page directory and page tables */
 321
 322	WARN_ON_ONCE(as->use_count);
 323	kfree(as->count);
 324	kfree(as->pts);
 325	kfree(as);
 326}
 327
 328static const struct tegra_smmu_swgroup *
 329tegra_smmu_find_swgroup(struct tegra_smmu *smmu, unsigned int swgroup)
 330{
 331	const struct tegra_smmu_swgroup *group = NULL;
 332	unsigned int i;
 333
 334	for (i = 0; i < smmu->soc->num_swgroups; i++) {
 335		if (smmu->soc->swgroups[i].swgroup == swgroup) {
 336			group = &smmu->soc->swgroups[i];
 337			break;
 338		}
 339	}
 340
 341	return group;
 342}
 343
 344static void tegra_smmu_enable(struct tegra_smmu *smmu, unsigned int swgroup,
 345			      unsigned int asid)
 346{
 347	const struct tegra_smmu_swgroup *group;
 348	unsigned int i;
 349	u32 value;
 350
 351	group = tegra_smmu_find_swgroup(smmu, swgroup);
 352	if (group) {
 353		value = smmu_readl(smmu, group->reg);
 354		value &= ~SMMU_ASID_MASK;
 355		value |= SMMU_ASID_VALUE(asid);
 356		value |= SMMU_ASID_ENABLE;
 357		smmu_writel(smmu, value, group->reg);
 358	} else {
 359		pr_warn("%s group from swgroup %u not found\n", __func__,
 360				swgroup);
 361		/* No point moving ahead if group was not found */
 362		return;
 363	}
 364
 365	for (i = 0; i < smmu->soc->num_clients; i++) {
 366		const struct tegra_mc_client *client = &smmu->soc->clients[i];
 367
 368		if (client->swgroup != swgroup)
 369			continue;
 370
 371		value = smmu_readl(smmu, client->regs.smmu.reg);
 372		value |= BIT(client->regs.smmu.bit);
 373		smmu_writel(smmu, value, client->regs.smmu.reg);
 374	}
 375}
 376
 377static void tegra_smmu_disable(struct tegra_smmu *smmu, unsigned int swgroup,
 378			       unsigned int asid)
 379{
 380	const struct tegra_smmu_swgroup *group;
 381	unsigned int i;
 382	u32 value;
 383
 384	group = tegra_smmu_find_swgroup(smmu, swgroup);
 385	if (group) {
 386		value = smmu_readl(smmu, group->reg);
 387		value &= ~SMMU_ASID_MASK;
 388		value |= SMMU_ASID_VALUE(asid);
 389		value &= ~SMMU_ASID_ENABLE;
 390		smmu_writel(smmu, value, group->reg);
 391	}
 392
 393	for (i = 0; i < smmu->soc->num_clients; i++) {
 394		const struct tegra_mc_client *client = &smmu->soc->clients[i];
 395
 396		if (client->swgroup != swgroup)
 397			continue;
 398
 399		value = smmu_readl(smmu, client->regs.smmu.reg);
 400		value &= ~BIT(client->regs.smmu.bit);
 401		smmu_writel(smmu, value, client->regs.smmu.reg);
 402	}
 403}
 404
 405static int tegra_smmu_as_prepare(struct tegra_smmu *smmu,
 406				 struct tegra_smmu_as *as)
 407{
 408	u32 value;
 409	int err = 0;
 410
 411	mutex_lock(&smmu->lock);
 412
 413	if (as->use_count > 0) {
 414		as->use_count++;
 415		goto unlock;
 416	}
 417
 418	as->pd_dma = dma_map_page(smmu->dev, as->pd, 0, SMMU_SIZE_PD,
 419				  DMA_TO_DEVICE);
 420	if (dma_mapping_error(smmu->dev, as->pd_dma)) {
 421		err = -ENOMEM;
 422		goto unlock;
 423	}
 424
 425	/* We can't handle 64-bit DMA addresses */
 426	if (!smmu_dma_addr_valid(smmu, as->pd_dma)) {
 427		err = -ENOMEM;
 428		goto err_unmap;
 429	}
 430
 431	err = tegra_smmu_alloc_asid(smmu, &as->id);
 432	if (err < 0)
 433		goto err_unmap;
 434
 435	smmu_flush_ptc(smmu, as->pd_dma, 0);
 436	smmu_flush_tlb_asid(smmu, as->id);
 437
 438	smmu_writel(smmu, as->id & 0x7f, SMMU_PTB_ASID);
 439	value = SMMU_PTB_DATA_VALUE(as->pd_dma, as->attr);
 440	smmu_writel(smmu, value, SMMU_PTB_DATA);
 441	smmu_flush(smmu);
 442
 443	as->smmu = smmu;
 444	as->use_count++;
 445
 446	mutex_unlock(&smmu->lock);
 447
 448	return 0;
 449
 450err_unmap:
 451	dma_unmap_page(smmu->dev, as->pd_dma, SMMU_SIZE_PD, DMA_TO_DEVICE);
 452unlock:
 453	mutex_unlock(&smmu->lock);
 454
 455	return err;
 456}
 457
 458static void tegra_smmu_as_unprepare(struct tegra_smmu *smmu,
 459				    struct tegra_smmu_as *as)
 460{
 461	mutex_lock(&smmu->lock);
 462
 463	if (--as->use_count > 0) {
 464		mutex_unlock(&smmu->lock);
 465		return;
 466	}
 467
 468	tegra_smmu_free_asid(smmu, as->id);
 469
 470	dma_unmap_page(smmu->dev, as->pd_dma, SMMU_SIZE_PD, DMA_TO_DEVICE);
 471
 472	as->smmu = NULL;
 473
 474	mutex_unlock(&smmu->lock);
 475}
 476
 477static int tegra_smmu_attach_dev(struct iommu_domain *domain,
 478				 struct device *dev)
 479{
 480	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
 481	struct tegra_smmu *smmu = dev_iommu_priv_get(dev);
 482	struct tegra_smmu_as *as = to_smmu_as(domain);
 483	unsigned int index;
 484	int err;
 485
 486	if (!fwspec)
 487		return -ENOENT;
 488
 489	for (index = 0; index < fwspec->num_ids; index++) {
 490		err = tegra_smmu_as_prepare(smmu, as);
 491		if (err)
 492			goto disable;
 493
 494		tegra_smmu_enable(smmu, fwspec->ids[index], as->id);
 495	}
 496
 497	if (index == 0)
 498		return -ENODEV;
 499
 500	return 0;
 501
 502disable:
 503	while (index--) {
 504		tegra_smmu_disable(smmu, fwspec->ids[index], as->id);
 505		tegra_smmu_as_unprepare(smmu, as);
 506	}
 507
 508	return err;
 509}
 510
 511static int tegra_smmu_identity_attach(struct iommu_domain *identity_domain,
 512				      struct device *dev)
 513{
 514	struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
 515	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
 516	struct tegra_smmu_as *as;
 517	struct tegra_smmu *smmu;
 518	unsigned int index;
 519
 520	if (!fwspec)
 521		return -ENODEV;
 522
 523	if (domain == identity_domain || !domain)
 524		return 0;
 525
 526	as = to_smmu_as(domain);
 527	smmu = as->smmu;
 528	for (index = 0; index < fwspec->num_ids; index++) {
 529		tegra_smmu_disable(smmu, fwspec->ids[index], as->id);
 530		tegra_smmu_as_unprepare(smmu, as);
 531	}
 532	return 0;
 533}
 534
 535static struct iommu_domain_ops tegra_smmu_identity_ops = {
 536	.attach_dev = tegra_smmu_identity_attach,
 537};
 538
 539static struct iommu_domain tegra_smmu_identity_domain = {
 540	.type = IOMMU_DOMAIN_IDENTITY,
 541	.ops = &tegra_smmu_identity_ops,
 542};
 543
 544static void tegra_smmu_set_pde(struct tegra_smmu_as *as, unsigned long iova,
 545			       u32 value)
 546{
 547	unsigned int pd_index = iova_pd_index(iova);
 548	struct tegra_smmu *smmu = as->smmu;
 549	u32 *pd = page_address(as->pd);
 550	unsigned long offset = pd_index * sizeof(*pd);
 551
 552	/* Set the page directory entry first */
 553	pd[pd_index] = value;
 554
 555	/* The flush the page directory entry from caches */
 556	dma_sync_single_range_for_device(smmu->dev, as->pd_dma, offset,
 557					 sizeof(*pd), DMA_TO_DEVICE);
 558
 559	/* And flush the iommu */
 560	smmu_flush_ptc(smmu, as->pd_dma, offset);
 561	smmu_flush_tlb_section(smmu, as->id, iova);
 562	smmu_flush(smmu);
 563}
 564
 565static u32 *tegra_smmu_pte_offset(struct page *pt_page, unsigned long iova)
 566{
 567	u32 *pt = page_address(pt_page);
 568
 569	return pt + iova_pt_index(iova);
 570}
 571
 572static u32 *tegra_smmu_pte_lookup(struct tegra_smmu_as *as, unsigned long iova,
 573				  dma_addr_t *dmap)
 574{
 575	unsigned int pd_index = iova_pd_index(iova);
 576	struct tegra_smmu *smmu = as->smmu;
 577	struct page *pt_page;
 578	u32 *pd;
 579
 580	pt_page = as->pts[pd_index];
 581	if (!pt_page)
 582		return NULL;
 583
 584	pd = page_address(as->pd);
 585	*dmap = smmu_pde_to_dma(smmu, pd[pd_index]);
 586
 587	return tegra_smmu_pte_offset(pt_page, iova);
 588}
 589
 590static u32 *as_get_pte(struct tegra_smmu_as *as, dma_addr_t iova,
 591		       dma_addr_t *dmap, struct page *page)
 592{
 593	unsigned int pde = iova_pd_index(iova);
 594	struct tegra_smmu *smmu = as->smmu;
 595
 596	if (!as->pts[pde]) {
 597		dma_addr_t dma;
 598
 599		dma = dma_map_page(smmu->dev, page, 0, SMMU_SIZE_PT,
 600				   DMA_TO_DEVICE);
 601		if (dma_mapping_error(smmu->dev, dma)) {
 602			__free_page(page);
 603			return NULL;
 604		}
 605
 606		if (!smmu_dma_addr_valid(smmu, dma)) {
 607			dma_unmap_page(smmu->dev, dma, SMMU_SIZE_PT,
 608				       DMA_TO_DEVICE);
 609			__free_page(page);
 610			return NULL;
 611		}
 612
 613		as->pts[pde] = page;
 614
 615		tegra_smmu_set_pde(as, iova, SMMU_MK_PDE(dma, SMMU_PDE_ATTR |
 616							      SMMU_PDE_NEXT));
 617
 618		*dmap = dma;
 619	} else {
 620		u32 *pd = page_address(as->pd);
 621
 622		*dmap = smmu_pde_to_dma(smmu, pd[pde]);
 623	}
 624
 625	return tegra_smmu_pte_offset(as->pts[pde], iova);
 626}
 627
 628static void tegra_smmu_pte_get_use(struct tegra_smmu_as *as, unsigned long iova)
 629{
 630	unsigned int pd_index = iova_pd_index(iova);
 631
 632	as->count[pd_index]++;
 633}
 634
 635static void tegra_smmu_pte_put_use(struct tegra_smmu_as *as, unsigned long iova)
 636{
 637	unsigned int pde = iova_pd_index(iova);
 638	struct page *page = as->pts[pde];
 639
 640	/*
 641	 * When no entries in this page table are used anymore, return the
 642	 * memory page to the system.
 643	 */
 644	if (--as->count[pde] == 0) {
 645		struct tegra_smmu *smmu = as->smmu;
 646		u32 *pd = page_address(as->pd);
 647		dma_addr_t pte_dma = smmu_pde_to_dma(smmu, pd[pde]);
 648
 649		tegra_smmu_set_pde(as, iova, 0);
 650
 651		dma_unmap_page(smmu->dev, pte_dma, SMMU_SIZE_PT, DMA_TO_DEVICE);
 652		__free_page(page);
 653		as->pts[pde] = NULL;
 654	}
 655}
 656
 657static void tegra_smmu_set_pte(struct tegra_smmu_as *as, unsigned long iova,
 658			       u32 *pte, dma_addr_t pte_dma, u32 val)
 659{
 660	struct tegra_smmu *smmu = as->smmu;
 661	unsigned long offset = SMMU_OFFSET_IN_PAGE(pte);
 662
 663	*pte = val;
 664
 665	dma_sync_single_range_for_device(smmu->dev, pte_dma, offset,
 666					 4, DMA_TO_DEVICE);
 667	smmu_flush_ptc(smmu, pte_dma, offset);
 668	smmu_flush_tlb_group(smmu, as->id, iova);
 669	smmu_flush(smmu);
 670}
 671
 672static struct page *as_get_pde_page(struct tegra_smmu_as *as,
 673				    unsigned long iova, gfp_t gfp,
 674				    unsigned long *flags)
 675{
 676	unsigned int pde = iova_pd_index(iova);
 677	struct page *page = as->pts[pde];
 678
 679	/* at first check whether allocation needs to be done at all */
 680	if (page)
 681		return page;
 682
 683	/*
 684	 * In order to prevent exhaustion of the atomic memory pool, we
 685	 * allocate page in a sleeping context if GFP flags permit. Hence
 686	 * spinlock needs to be unlocked and re-locked after allocation.
 687	 */
 688	if (gfpflags_allow_blocking(gfp))
 689		spin_unlock_irqrestore(&as->lock, *flags);
 690
 691	page = alloc_page(gfp | __GFP_DMA | __GFP_ZERO);
 692
 693	if (gfpflags_allow_blocking(gfp))
 694		spin_lock_irqsave(&as->lock, *flags);
 695
 696	/*
 697	 * In a case of blocking allocation, a concurrent mapping may win
 698	 * the PDE allocation. In this case the allocated page isn't needed
 699	 * if allocation succeeded and the allocation failure isn't fatal.
 700	 */
 701	if (as->pts[pde]) {
 702		if (page)
 703			__free_page(page);
 704
 705		page = as->pts[pde];
 706	}
 707
 708	return page;
 709}
 710
 711static int
 712__tegra_smmu_map(struct iommu_domain *domain, unsigned long iova,
 713		 phys_addr_t paddr, size_t size, int prot, gfp_t gfp,
 714		 unsigned long *flags)
 715{
 716	struct tegra_smmu_as *as = to_smmu_as(domain);
 717	dma_addr_t pte_dma;
 718	struct page *page;
 719	u32 pte_attrs;
 720	u32 *pte;
 721
 722	page = as_get_pde_page(as, iova, gfp, flags);
 723	if (!page)
 724		return -ENOMEM;
 725
 726	pte = as_get_pte(as, iova, &pte_dma, page);
 727	if (!pte)
 728		return -ENOMEM;
 729
 730	/* If we aren't overwriting a pre-existing entry, increment use */
 731	if (*pte == 0)
 732		tegra_smmu_pte_get_use(as, iova);
 733
 734	pte_attrs = SMMU_PTE_NONSECURE;
 735
 736	if (prot & IOMMU_READ)
 737		pte_attrs |= SMMU_PTE_READABLE;
 738
 739	if (prot & IOMMU_WRITE)
 740		pte_attrs |= SMMU_PTE_WRITABLE;
 741
 742	tegra_smmu_set_pte(as, iova, pte, pte_dma,
 743			   SMMU_PHYS_PFN(paddr) | pte_attrs);
 744
 745	return 0;
 746}
 747
 748static size_t
 749__tegra_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
 750		   size_t size, struct iommu_iotlb_gather *gather)
 751{
 752	struct tegra_smmu_as *as = to_smmu_as(domain);
 753	dma_addr_t pte_dma;
 754	u32 *pte;
 755
 756	pte = tegra_smmu_pte_lookup(as, iova, &pte_dma);
 757	if (!pte || !*pte)
 758		return 0;
 759
 760	tegra_smmu_set_pte(as, iova, pte, pte_dma, 0);
 761	tegra_smmu_pte_put_use(as, iova);
 762
 763	return size;
 764}
 765
 766static int tegra_smmu_map(struct iommu_domain *domain, unsigned long iova,
 767			  phys_addr_t paddr, size_t size, size_t count,
 768			  int prot, gfp_t gfp, size_t *mapped)
 769{
 770	struct tegra_smmu_as *as = to_smmu_as(domain);
 771	unsigned long flags;
 772	int ret;
 773
 774	spin_lock_irqsave(&as->lock, flags);
 775	ret = __tegra_smmu_map(domain, iova, paddr, size, prot, gfp, &flags);
 776	spin_unlock_irqrestore(&as->lock, flags);
 777
 778	if (!ret)
 779		*mapped = size;
 780
 781	return ret;
 782}
 783
 784static size_t tegra_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
 785			       size_t size, size_t count, struct iommu_iotlb_gather *gather)
 786{
 787	struct tegra_smmu_as *as = to_smmu_as(domain);
 788	unsigned long flags;
 789
 790	spin_lock_irqsave(&as->lock, flags);
 791	size = __tegra_smmu_unmap(domain, iova, size, gather);
 792	spin_unlock_irqrestore(&as->lock, flags);
 793
 794	return size;
 795}
 796
 797static phys_addr_t tegra_smmu_iova_to_phys(struct iommu_domain *domain,
 798					   dma_addr_t iova)
 799{
 800	struct tegra_smmu_as *as = to_smmu_as(domain);
 801	unsigned long pfn;
 802	dma_addr_t pte_dma;
 803	u32 *pte;
 804
 805	pte = tegra_smmu_pte_lookup(as, iova, &pte_dma);
 806	if (!pte || !*pte)
 807		return 0;
 808
 809	pfn = *pte & as->smmu->pfn_mask;
 810
 811	return SMMU_PFN_PHYS(pfn) + SMMU_OFFSET_IN_PAGE(iova);
 812}
 813
 814static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
 815{
 816	struct platform_device *pdev;
 817	struct tegra_mc *mc;
 818
 819	pdev = of_find_device_by_node(np);
 820	if (!pdev)
 821		return NULL;
 822
 823	mc = platform_get_drvdata(pdev);
 824	if (!mc) {
 825		put_device(&pdev->dev);
 826		return NULL;
 827	}
 828
 829	return mc->smmu;
 830}
 831
 832static int tegra_smmu_configure(struct tegra_smmu *smmu, struct device *dev,
 833				const struct of_phandle_args *args)
 834{
 835	const struct iommu_ops *ops = smmu->iommu.ops;
 836	int err;
 837
 838	err = iommu_fwspec_init(dev, &dev->of_node->fwnode, ops);
 839	if (err < 0) {
 840		dev_err(dev, "failed to initialize fwspec: %d\n", err);
 841		return err;
 842	}
 843
 844	err = ops->of_xlate(dev, args);
 845	if (err < 0) {
 846		dev_err(dev, "failed to parse SW group ID: %d\n", err);
 847		iommu_fwspec_free(dev);
 848		return err;
 849	}
 850
 851	return 0;
 852}
 853
 854static struct iommu_device *tegra_smmu_probe_device(struct device *dev)
 855{
 856	struct device_node *np = dev->of_node;
 857	struct tegra_smmu *smmu = NULL;
 858	struct of_phandle_args args;
 859	unsigned int index = 0;
 860	int err;
 861
 862	while (of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
 863					  &args) == 0) {
 864		smmu = tegra_smmu_find(args.np);
 865		if (smmu) {
 866			err = tegra_smmu_configure(smmu, dev, &args);
 867
 868			if (err < 0) {
 869				of_node_put(args.np);
 870				return ERR_PTR(err);
 871			}
 872		}
 873
 874		of_node_put(args.np);
 875		index++;
 876	}
 877
 878	smmu = dev_iommu_priv_get(dev);
 879	if (!smmu)
 880		return ERR_PTR(-ENODEV);
 881
 882	return &smmu->iommu;
 883}
 884
 885static const struct tegra_smmu_group_soc *
 886tegra_smmu_find_group(struct tegra_smmu *smmu, unsigned int swgroup)
 887{
 888	unsigned int i, j;
 889
 890	for (i = 0; i < smmu->soc->num_groups; i++)
 891		for (j = 0; j < smmu->soc->groups[i].num_swgroups; j++)
 892			if (smmu->soc->groups[i].swgroups[j] == swgroup)
 893				return &smmu->soc->groups[i];
 894
 895	return NULL;
 896}
 897
 898static void tegra_smmu_group_release(void *iommu_data)
 899{
 900	struct tegra_smmu_group *group = iommu_data;
 901	struct tegra_smmu *smmu = group->smmu;
 902
 903	mutex_lock(&smmu->lock);
 904	list_del(&group->list);
 905	mutex_unlock(&smmu->lock);
 906}
 907
 908static struct iommu_group *tegra_smmu_device_group(struct device *dev)
 909{
 910	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
 911	struct tegra_smmu *smmu = dev_iommu_priv_get(dev);
 912	const struct tegra_smmu_group_soc *soc;
 913	unsigned int swgroup = fwspec->ids[0];
 914	struct tegra_smmu_group *group;
 915	struct iommu_group *grp;
 916
 917	/* Find group_soc associating with swgroup */
 918	soc = tegra_smmu_find_group(smmu, swgroup);
 919
 920	mutex_lock(&smmu->lock);
 921
 922	/* Find existing iommu_group associating with swgroup or group_soc */
 923	list_for_each_entry(group, &smmu->groups, list)
 924		if ((group->swgroup == swgroup) || (soc && group->soc == soc)) {
 925			grp = iommu_group_ref_get(group->group);
 926			mutex_unlock(&smmu->lock);
 927			return grp;
 928		}
 929
 930	group = devm_kzalloc(smmu->dev, sizeof(*group), GFP_KERNEL);
 931	if (!group) {
 932		mutex_unlock(&smmu->lock);
 933		return NULL;
 934	}
 935
 936	INIT_LIST_HEAD(&group->list);
 937	group->swgroup = swgroup;
 938	group->smmu = smmu;
 939	group->soc = soc;
 940
 941	if (dev_is_pci(dev))
 942		group->group = pci_device_group(dev);
 943	else
 944		group->group = generic_device_group(dev);
 945
 946	if (IS_ERR(group->group)) {
 947		devm_kfree(smmu->dev, group);
 948		mutex_unlock(&smmu->lock);
 949		return NULL;
 950	}
 951
 952	iommu_group_set_iommudata(group->group, group, tegra_smmu_group_release);
 953	if (soc)
 954		iommu_group_set_name(group->group, soc->name);
 955	list_add_tail(&group->list, &smmu->groups);
 956	mutex_unlock(&smmu->lock);
 957
 958	return group->group;
 959}
 960
 961static int tegra_smmu_of_xlate(struct device *dev,
 962			       const struct of_phandle_args *args)
 963{
 964	struct platform_device *iommu_pdev = of_find_device_by_node(args->np);
 965	struct tegra_mc *mc = platform_get_drvdata(iommu_pdev);
 966	u32 id = args->args[0];
 967
 968	/*
 969	 * Note: we are here releasing the reference of &iommu_pdev->dev, which
 970	 * is mc->dev. Although some functions in tegra_smmu_ops may keep using
 971	 * its private data beyond this point, it's still safe to do so because
 972	 * the SMMU parent device is the same as the MC, so the reference count
 973	 * isn't strictly necessary.
 974	 */
 975	put_device(&iommu_pdev->dev);
 976
 977	dev_iommu_priv_set(dev, mc->smmu);
 978
 979	return iommu_fwspec_add_ids(dev, &id, 1);
 980}
 981
 982static int tegra_smmu_def_domain_type(struct device *dev)
 983{
 984	/*
 985	 * FIXME: For now we want to run all translation in IDENTITY mode, due
 986	 * to some device quirks. Better would be to just quirk the troubled
 987	 * devices.
 988	 */
 989	return IOMMU_DOMAIN_IDENTITY;
 990}
 991
 992static const struct iommu_ops tegra_smmu_ops = {
 993	.identity_domain = &tegra_smmu_identity_domain,
 994	.def_domain_type = &tegra_smmu_def_domain_type,
 995	.domain_alloc_paging = tegra_smmu_domain_alloc_paging,
 996	.probe_device = tegra_smmu_probe_device,
 997	.device_group = tegra_smmu_device_group,
 998	.of_xlate = tegra_smmu_of_xlate,
 999	.pgsize_bitmap = SZ_4K,
1000	.default_domain_ops = &(const struct iommu_domain_ops) {
1001		.attach_dev	= tegra_smmu_attach_dev,
1002		.map_pages	= tegra_smmu_map,
1003		.unmap_pages	= tegra_smmu_unmap,
1004		.iova_to_phys	= tegra_smmu_iova_to_phys,
1005		.free		= tegra_smmu_domain_free,
1006	}
1007};
1008
1009static void tegra_smmu_ahb_enable(void)
1010{
1011	static const struct of_device_id ahb_match[] = {
1012		{ .compatible = "nvidia,tegra30-ahb", },
1013		{ }
1014	};
1015	struct device_node *ahb;
1016
1017	ahb = of_find_matching_node(NULL, ahb_match);
1018	if (ahb) {
1019		tegra_ahb_enable_smmu(ahb);
1020		of_node_put(ahb);
1021	}
1022}
1023
1024static int tegra_smmu_swgroups_show(struct seq_file *s, void *data)
1025{
1026	struct tegra_smmu *smmu = s->private;
1027	unsigned int i;
1028	u32 value;
1029
1030	seq_printf(s, "swgroup    enabled  ASID\n");
1031	seq_printf(s, "------------------------\n");
1032
1033	for (i = 0; i < smmu->soc->num_swgroups; i++) {
1034		const struct tegra_smmu_swgroup *group = &smmu->soc->swgroups[i];
1035		const char *status;
1036		unsigned int asid;
1037
1038		value = smmu_readl(smmu, group->reg);
1039
1040		if (value & SMMU_ASID_ENABLE)
1041			status = "yes";
1042		else
1043			status = "no";
1044
1045		asid = value & SMMU_ASID_MASK;
1046
1047		seq_printf(s, "%-9s  %-7s  %#04x\n", group->name, status,
1048			   asid);
1049	}
1050
1051	return 0;
1052}
1053
1054DEFINE_SHOW_ATTRIBUTE(tegra_smmu_swgroups);
1055
1056static int tegra_smmu_clients_show(struct seq_file *s, void *data)
1057{
1058	struct tegra_smmu *smmu = s->private;
1059	unsigned int i;
1060	u32 value;
1061
1062	seq_printf(s, "client       enabled\n");
1063	seq_printf(s, "--------------------\n");
1064
1065	for (i = 0; i < smmu->soc->num_clients; i++) {
1066		const struct tegra_mc_client *client = &smmu->soc->clients[i];
1067		const char *status;
1068
1069		value = smmu_readl(smmu, client->regs.smmu.reg);
1070
1071		if (value & BIT(client->regs.smmu.bit))
1072			status = "yes";
1073		else
1074			status = "no";
1075
1076		seq_printf(s, "%-12s %s\n", client->name, status);
1077	}
1078
1079	return 0;
1080}
1081
1082DEFINE_SHOW_ATTRIBUTE(tegra_smmu_clients);
1083
1084static void tegra_smmu_debugfs_init(struct tegra_smmu *smmu)
1085{
1086	smmu->debugfs = debugfs_create_dir("smmu", NULL);
1087
1088	debugfs_create_file("swgroups", S_IRUGO, smmu->debugfs, smmu,
1089			    &tegra_smmu_swgroups_fops);
1090	debugfs_create_file("clients", S_IRUGO, smmu->debugfs, smmu,
1091			    &tegra_smmu_clients_fops);
1092}
1093
1094static void tegra_smmu_debugfs_exit(struct tegra_smmu *smmu)
1095{
1096	debugfs_remove_recursive(smmu->debugfs);
1097}
1098
1099struct tegra_smmu *tegra_smmu_probe(struct device *dev,
1100				    const struct tegra_smmu_soc *soc,
1101				    struct tegra_mc *mc)
1102{
1103	struct tegra_smmu *smmu;
1104	u32 value;
1105	int err;
1106
1107	smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1108	if (!smmu)
1109		return ERR_PTR(-ENOMEM);
1110
1111	/*
1112	 * This is a bit of a hack. Ideally we'd want to simply return this
1113	 * value. However iommu_device_register() will attempt to add
1114	 * all devices to the IOMMU before we get that far. In order
1115	 * not to rely on global variables to track the IOMMU instance, we
1116	 * set it here so that it can be looked up from the .probe_device()
1117	 * callback via the IOMMU device's .drvdata field.
1118	 */
1119	mc->smmu = smmu;
1120
1121	smmu->asids = devm_bitmap_zalloc(dev, soc->num_asids, GFP_KERNEL);
1122	if (!smmu->asids)
1123		return ERR_PTR(-ENOMEM);
1124
1125	INIT_LIST_HEAD(&smmu->groups);
1126	mutex_init(&smmu->lock);
1127
1128	smmu->regs = mc->regs;
1129	smmu->soc = soc;
1130	smmu->dev = dev;
1131	smmu->mc = mc;
1132
1133	smmu->pfn_mask =
1134		BIT_MASK(mc->soc->num_address_bits - SMMU_PTE_SHIFT) - 1;
1135	dev_dbg(dev, "address bits: %u, PFN mask: %#lx\n",
1136		mc->soc->num_address_bits, smmu->pfn_mask);
1137	smmu->tlb_mask = (1 << fls(smmu->soc->num_tlb_lines)) - 1;
1138	dev_dbg(dev, "TLB lines: %u, mask: %#lx\n", smmu->soc->num_tlb_lines,
1139		smmu->tlb_mask);
1140
1141	value = SMMU_PTC_CONFIG_ENABLE | SMMU_PTC_CONFIG_INDEX_MAP(0x3f);
1142
1143	if (soc->supports_request_limit)
1144		value |= SMMU_PTC_CONFIG_REQ_LIMIT(8);
1145
1146	smmu_writel(smmu, value, SMMU_PTC_CONFIG);
1147
1148	value = SMMU_TLB_CONFIG_HIT_UNDER_MISS |
1149		SMMU_TLB_CONFIG_ACTIVE_LINES(smmu);
1150
1151	if (soc->supports_round_robin_arbitration)
1152		value |= SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION;
1153
1154	smmu_writel(smmu, value, SMMU_TLB_CONFIG);
1155
1156	smmu_flush_ptc_all(smmu);
1157	smmu_flush_tlb(smmu);
1158	smmu_writel(smmu, SMMU_CONFIG_ENABLE, SMMU_CONFIG);
1159	smmu_flush(smmu);
1160
1161	tegra_smmu_ahb_enable();
1162
1163	err = iommu_device_sysfs_add(&smmu->iommu, dev, NULL, dev_name(dev));
1164	if (err)
1165		return ERR_PTR(err);
1166
1167	err = iommu_device_register(&smmu->iommu, &tegra_smmu_ops, dev);
1168	if (err) {
1169		iommu_device_sysfs_remove(&smmu->iommu);
1170		return ERR_PTR(err);
1171	}
1172
1173	if (IS_ENABLED(CONFIG_DEBUG_FS))
1174		tegra_smmu_debugfs_init(smmu);
1175
1176	return smmu;
1177}
1178
1179void tegra_smmu_remove(struct tegra_smmu *smmu)
1180{
1181	iommu_device_unregister(&smmu->iommu);
1182	iommu_device_sysfs_remove(&smmu->iommu);
1183
1184	if (IS_ENABLED(CONFIG_DEBUG_FS))
1185		tegra_smmu_debugfs_exit(smmu);
1186}