xref: /linux/drivers/iommu/exynos-iommu.c (revision 249ebf3f65f8530beb2cbfb91bff1d83ba88d23c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2011,2016 Samsung Electronics Co., Ltd.
4  *		http://www.samsung.com
5  */
6 
7 #ifdef CONFIG_EXYNOS_IOMMU_DEBUG
8 #define DEBUG
9 #endif
10 
11 #include <linux/clk.h>
12 #include <linux/dma-mapping.h>
13 #include <linux/err.h>
14 #include <linux/io.h>
15 #include <linux/iommu.h>
16 #include <linux/interrupt.h>
17 #include <linux/kmemleak.h>
18 #include <linux/list.h>
19 #include <linux/of.h>
20 #include <linux/of_platform.h>
21 #include <linux/platform_device.h>
22 #include <linux/pm_runtime.h>
23 #include <linux/slab.h>
24 
25 #include "iommu-pages.h"
26 
27 typedef u32 sysmmu_iova_t;
28 typedef u32 sysmmu_pte_t;
29 static struct iommu_domain exynos_identity_domain;
30 
31 /* We do not consider super section mapping (16MB) */
32 #define SECT_ORDER 20
33 #define LPAGE_ORDER 16
34 #define SPAGE_ORDER 12
35 
36 #define SECT_SIZE (1 << SECT_ORDER)
37 #define LPAGE_SIZE (1 << LPAGE_ORDER)
38 #define SPAGE_SIZE (1 << SPAGE_ORDER)
39 
40 #define SECT_MASK (~(SECT_SIZE - 1))
41 #define LPAGE_MASK (~(LPAGE_SIZE - 1))
42 #define SPAGE_MASK (~(SPAGE_SIZE - 1))
43 
44 #define lv1ent_fault(sent) ((*(sent) == ZERO_LV2LINK) || \
45 			   ((*(sent) & 3) == 0) || ((*(sent) & 3) == 3))
46 #define lv1ent_zero(sent) (*(sent) == ZERO_LV2LINK)
47 #define lv1ent_page_zero(sent) ((*(sent) & 3) == 1)
48 #define lv1ent_page(sent) ((*(sent) != ZERO_LV2LINK) && \
49 			  ((*(sent) & 3) == 1))
50 #define lv1ent_section(sent) ((*(sent) & 3) == 2)
51 
52 #define lv2ent_fault(pent) ((*(pent) & 3) == 0)
53 #define lv2ent_small(pent) ((*(pent) & 2) == 2)
54 #define lv2ent_large(pent) ((*(pent) & 3) == 1)
55 
56 /*
57  * v1.x - v3.x SYSMMU supports 32bit physical and 32bit virtual address spaces
58  * v5.0 introduced support for 36bit physical address space by shifting
59  * all page entry values by 4 bits.
60  * All SYSMMU controllers in the system support the address spaces of the same
61  * size, so PG_ENT_SHIFT can be initialized on first SYSMMU probe to proper
62  * value (0 or 4).
63  */
64 static short PG_ENT_SHIFT = -1;
65 #define SYSMMU_PG_ENT_SHIFT 0
66 #define SYSMMU_V5_PG_ENT_SHIFT 4
67 
68 static const sysmmu_pte_t *LV1_PROT;
69 static const sysmmu_pte_t SYSMMU_LV1_PROT[] = {
70 	((0 << 15) | (0 << 10)), /* no access */
71 	((1 << 15) | (1 << 10)), /* IOMMU_READ only */
72 	((0 << 15) | (1 << 10)), /* IOMMU_WRITE not supported, use read/write */
73 	((0 << 15) | (1 << 10)), /* IOMMU_READ | IOMMU_WRITE */
74 };
75 static const sysmmu_pte_t SYSMMU_V5_LV1_PROT[] = {
76 	(0 << 4), /* no access */
77 	(1 << 4), /* IOMMU_READ only */
78 	(2 << 4), /* IOMMU_WRITE only */
79 	(3 << 4), /* IOMMU_READ | IOMMU_WRITE */
80 };
81 
82 static const sysmmu_pte_t *LV2_PROT;
83 static const sysmmu_pte_t SYSMMU_LV2_PROT[] = {
84 	((0 << 9) | (0 << 4)), /* no access */
85 	((1 << 9) | (1 << 4)), /* IOMMU_READ only */
86 	((0 << 9) | (1 << 4)), /* IOMMU_WRITE not supported, use read/write */
87 	((0 << 9) | (1 << 4)), /* IOMMU_READ | IOMMU_WRITE */
88 };
89 static const sysmmu_pte_t SYSMMU_V5_LV2_PROT[] = {
90 	(0 << 2), /* no access */
91 	(1 << 2), /* IOMMU_READ only */
92 	(2 << 2), /* IOMMU_WRITE only */
93 	(3 << 2), /* IOMMU_READ | IOMMU_WRITE */
94 };
95 
96 #define SYSMMU_SUPPORTED_PROT_BITS (IOMMU_READ | IOMMU_WRITE)
97 
98 #define sect_to_phys(ent) (((phys_addr_t) ent) << PG_ENT_SHIFT)
99 #define section_phys(sent) (sect_to_phys(*(sent)) & SECT_MASK)
100 #define section_offs(iova) (iova & (SECT_SIZE - 1))
101 #define lpage_phys(pent) (sect_to_phys(*(pent)) & LPAGE_MASK)
102 #define lpage_offs(iova) (iova & (LPAGE_SIZE - 1))
103 #define spage_phys(pent) (sect_to_phys(*(pent)) & SPAGE_MASK)
104 #define spage_offs(iova) (iova & (SPAGE_SIZE - 1))
105 
106 #define NUM_LV1ENTRIES 4096
107 #define NUM_LV2ENTRIES (SECT_SIZE / SPAGE_SIZE)
108 
109 static u32 lv1ent_offset(sysmmu_iova_t iova)
110 {
111 	return iova >> SECT_ORDER;
112 }
113 
114 static u32 lv2ent_offset(sysmmu_iova_t iova)
115 {
116 	return (iova >> SPAGE_ORDER) & (NUM_LV2ENTRIES - 1);
117 }
118 
119 #define LV1TABLE_SIZE (NUM_LV1ENTRIES * sizeof(sysmmu_pte_t))
120 #define LV2TABLE_SIZE (NUM_LV2ENTRIES * sizeof(sysmmu_pte_t))
121 
122 #define SPAGES_PER_LPAGE (LPAGE_SIZE / SPAGE_SIZE)
123 #define lv2table_base(sent) (sect_to_phys(*(sent) & 0xFFFFFFC0))
124 
125 #define mk_lv1ent_sect(pa, prot) ((pa >> PG_ENT_SHIFT) | LV1_PROT[prot] | 2)
126 #define mk_lv1ent_page(pa) ((pa >> PG_ENT_SHIFT) | 1)
127 #define mk_lv2ent_lpage(pa, prot) ((pa >> PG_ENT_SHIFT) | LV2_PROT[prot] | 1)
128 #define mk_lv2ent_spage(pa, prot) ((pa >> PG_ENT_SHIFT) | LV2_PROT[prot] | 2)
129 
130 #define CTRL_ENABLE	0x5
131 #define CTRL_BLOCK	0x7
132 #define CTRL_DISABLE	0x0
133 
134 #define CFG_LRU		0x1
135 #define CFG_EAP		(1 << 2)
136 #define CFG_QOS(n)	((n & 0xF) << 7)
137 #define CFG_ACGEN	(1 << 24) /* System MMU 3.3 only */
138 #define CFG_SYSSEL	(1 << 22) /* System MMU 3.2 only */
139 #define CFG_FLPDCACHE	(1 << 20) /* System MMU 3.2+ only */
140 
141 #define CTRL_VM_ENABLE			BIT(0)
142 #define CTRL_VM_FAULT_MODE_STALL	BIT(3)
143 #define CAPA0_CAPA1_EXIST		BIT(11)
144 #define CAPA1_VCR_ENABLED		BIT(14)
145 
146 /* common registers */
147 #define REG_MMU_CTRL		0x000
148 #define REG_MMU_CFG		0x004
149 #define REG_MMU_STATUS		0x008
150 #define REG_MMU_VERSION		0x034
151 
152 #define MMU_MAJ_VER(val)	((val) >> 7)
153 #define MMU_MIN_VER(val)	((val) & 0x7F)
154 #define MMU_RAW_VER(reg)	(((reg) >> 21) & ((1 << 11) - 1)) /* 11 bits */
155 
156 #define MAKE_MMU_VER(maj, min)	((((maj) & 0xF) << 7) | ((min) & 0x7F))
157 
158 /* v1.x - v3.x registers */
159 #define REG_PAGE_FAULT_ADDR	0x024
160 #define REG_AW_FAULT_ADDR	0x028
161 #define REG_AR_FAULT_ADDR	0x02C
162 #define REG_DEFAULT_SLAVE_ADDR	0x030
163 
164 /* v5.x registers */
165 #define REG_V5_FAULT_AR_VA	0x070
166 #define REG_V5_FAULT_AW_VA	0x080
167 
168 /* v7.x registers */
169 #define REG_V7_CAPA0		0x870
170 #define REG_V7_CAPA1		0x874
171 #define REG_V7_CTRL_VM		0x8000
172 
173 #define has_sysmmu(dev)		(dev_iommu_priv_get(dev) != NULL)
174 
175 static struct device *dma_dev;
176 static struct kmem_cache *lv2table_kmem_cache;
177 static sysmmu_pte_t *zero_lv2_table;
178 #define ZERO_LV2LINK mk_lv1ent_page(virt_to_phys(zero_lv2_table))
179 
180 static sysmmu_pte_t *section_entry(sysmmu_pte_t *pgtable, sysmmu_iova_t iova)
181 {
182 	return pgtable + lv1ent_offset(iova);
183 }
184 
185 static sysmmu_pte_t *page_entry(sysmmu_pte_t *sent, sysmmu_iova_t iova)
186 {
187 	return (sysmmu_pte_t *)phys_to_virt(
188 				lv2table_base(sent)) + lv2ent_offset(iova);
189 }
190 
191 struct sysmmu_fault {
192 	sysmmu_iova_t addr;	/* IOVA address that caused fault */
193 	const char *name;	/* human readable fault name */
194 	unsigned int type;	/* fault type for report_iommu_fault() */
195 };
196 
197 struct sysmmu_v1_fault_info {
198 	unsigned short addr_reg; /* register to read IOVA fault address */
199 	const char *name;	/* human readable fault name */
200 	unsigned int type;	/* fault type for report_iommu_fault */
201 };
202 
203 static const struct sysmmu_v1_fault_info sysmmu_v1_faults[] = {
204 	{ REG_PAGE_FAULT_ADDR, "PAGE", IOMMU_FAULT_READ },
205 	{ REG_AR_FAULT_ADDR, "MULTI-HIT", IOMMU_FAULT_READ },
206 	{ REG_AW_FAULT_ADDR, "MULTI-HIT", IOMMU_FAULT_WRITE },
207 	{ REG_DEFAULT_SLAVE_ADDR, "BUS ERROR", IOMMU_FAULT_READ },
208 	{ REG_AR_FAULT_ADDR, "SECURITY PROTECTION", IOMMU_FAULT_READ },
209 	{ REG_AR_FAULT_ADDR, "ACCESS PROTECTION", IOMMU_FAULT_READ },
210 	{ REG_AW_FAULT_ADDR, "SECURITY PROTECTION", IOMMU_FAULT_WRITE },
211 	{ REG_AW_FAULT_ADDR, "ACCESS PROTECTION", IOMMU_FAULT_WRITE },
212 };
213 
214 /* SysMMU v5 has the same faults for AR (0..4 bits) and AW (16..20 bits) */
215 static const char * const sysmmu_v5_fault_names[] = {
216 	"PTW",
217 	"PAGE",
218 	"MULTI-HIT",
219 	"ACCESS PROTECTION",
220 	"SECURITY PROTECTION"
221 };
222 
223 static const char * const sysmmu_v7_fault_names[] = {
224 	"PTW",
225 	"PAGE",
226 	"ACCESS PROTECTION",
227 	"RESERVED"
228 };
229 
230 /*
231  * This structure is attached to dev->iommu->priv of the master device
232  * on device add, contains a list of SYSMMU controllers defined by device tree,
233  * which are bound to given master device. It is usually referenced by 'owner'
234  * pointer.
235 */
236 struct exynos_iommu_owner {
237 	struct list_head controllers;	/* list of sysmmu_drvdata.owner_node */
238 	struct iommu_domain *domain;	/* domain this device is attached */
239 	struct mutex rpm_lock;		/* for runtime pm of all sysmmus */
240 };
241 
242 /*
243  * This structure exynos specific generalization of struct iommu_domain.
244  * It contains list of SYSMMU controllers from all master devices, which has
245  * been attached to this domain and page tables of IO address space defined by
246  * it. It is usually referenced by 'domain' pointer.
247  */
248 struct exynos_iommu_domain {
249 	struct list_head clients; /* list of sysmmu_drvdata.domain_node */
250 	sysmmu_pte_t *pgtable;	/* lv1 page table, 16KB */
251 	short *lv2entcnt;	/* free lv2 entry counter for each section */
252 	spinlock_t lock;	/* lock for modyfying list of clients */
253 	spinlock_t pgtablelock;	/* lock for modifying page table @ pgtable */
254 	struct iommu_domain domain; /* generic domain data structure */
255 };
256 
257 struct sysmmu_drvdata;
258 
259 /*
260  * SysMMU version specific data. Contains offsets for the registers which can
261  * be found in different SysMMU variants, but have different offset values.
262  * Also contains version specific callbacks to abstract the hardware.
263  */
264 struct sysmmu_variant {
265 	u32 pt_base;		/* page table base address (physical) */
266 	u32 flush_all;		/* invalidate all TLB entries */
267 	u32 flush_entry;	/* invalidate specific TLB entry */
268 	u32 flush_range;	/* invalidate TLB entries in specified range */
269 	u32 flush_start;	/* start address of range invalidation */
270 	u32 flush_end;		/* end address of range invalidation */
271 	u32 int_status;		/* interrupt status information */
272 	u32 int_clear;		/* clear the interrupt */
273 	u32 fault_va;		/* IOVA address that caused fault */
274 	u32 fault_info;		/* fault transaction info */
275 
276 	int (*get_fault_info)(struct sysmmu_drvdata *data, unsigned int itype,
277 			      struct sysmmu_fault *fault);
278 };
279 
280 /*
281  * This structure hold all data of a single SYSMMU controller, this includes
282  * hw resources like registers and clocks, pointers and list nodes to connect
283  * it to all other structures, internal state and parameters read from device
284  * tree. It is usually referenced by 'data' pointer.
285  */
286 struct sysmmu_drvdata {
287 	struct device *sysmmu;		/* SYSMMU controller device */
288 	struct device *master;		/* master device (owner) */
289 	struct device_link *link;	/* runtime PM link to master */
290 	void __iomem *sfrbase;		/* our registers */
291 	struct clk *clk;		/* SYSMMU's clock */
292 	struct clk *aclk;		/* SYSMMU's aclk clock */
293 	struct clk *pclk;		/* SYSMMU's pclk clock */
294 	struct clk *clk_master;		/* master's device clock */
295 	spinlock_t lock;		/* lock for modyfying state */
296 	bool active;			/* current status */
297 	struct exynos_iommu_domain *domain; /* domain we belong to */
298 	struct list_head domain_node;	/* node for domain clients list */
299 	struct list_head owner_node;	/* node for owner controllers list */
300 	phys_addr_t pgtable;		/* assigned page table structure */
301 	unsigned int version;		/* our version */
302 
303 	struct iommu_device iommu;	/* IOMMU core handle */
304 	const struct sysmmu_variant *variant; /* version specific data */
305 
306 	/* v7 fields */
307 	bool has_vcr;			/* virtual machine control register */
308 };
309 
310 #define SYSMMU_REG(data, reg) ((data)->sfrbase + (data)->variant->reg)
311 
312 static int exynos_sysmmu_v1_get_fault_info(struct sysmmu_drvdata *data,
313 					   unsigned int itype,
314 					   struct sysmmu_fault *fault)
315 {
316 	const struct sysmmu_v1_fault_info *finfo;
317 
318 	if (itype >= ARRAY_SIZE(sysmmu_v1_faults))
319 		return -ENXIO;
320 
321 	finfo = &sysmmu_v1_faults[itype];
322 	fault->addr = readl(data->sfrbase + finfo->addr_reg);
323 	fault->name = finfo->name;
324 	fault->type = finfo->type;
325 
326 	return 0;
327 }
328 
329 static int exynos_sysmmu_v5_get_fault_info(struct sysmmu_drvdata *data,
330 					   unsigned int itype,
331 					   struct sysmmu_fault *fault)
332 {
333 	unsigned int addr_reg;
334 
335 	if (itype < ARRAY_SIZE(sysmmu_v5_fault_names)) {
336 		fault->type = IOMMU_FAULT_READ;
337 		addr_reg = REG_V5_FAULT_AR_VA;
338 	} else if (itype >= 16 && itype <= 20) {
339 		fault->type = IOMMU_FAULT_WRITE;
340 		addr_reg = REG_V5_FAULT_AW_VA;
341 		itype -= 16;
342 	} else {
343 		return -ENXIO;
344 	}
345 
346 	fault->name = sysmmu_v5_fault_names[itype];
347 	fault->addr = readl(data->sfrbase + addr_reg);
348 
349 	return 0;
350 }
351 
352 static int exynos_sysmmu_v7_get_fault_info(struct sysmmu_drvdata *data,
353 					   unsigned int itype,
354 					   struct sysmmu_fault *fault)
355 {
356 	u32 info = readl(SYSMMU_REG(data, fault_info));
357 
358 	fault->addr = readl(SYSMMU_REG(data, fault_va));
359 	fault->name = sysmmu_v7_fault_names[itype % 4];
360 	fault->type = (info & BIT(20)) ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
361 
362 	return 0;
363 }
364 
365 /* SysMMU v1..v3 */
366 static const struct sysmmu_variant sysmmu_v1_variant = {
367 	.flush_all	= 0x0c,
368 	.flush_entry	= 0x10,
369 	.pt_base	= 0x14,
370 	.int_status	= 0x18,
371 	.int_clear	= 0x1c,
372 
373 	.get_fault_info	= exynos_sysmmu_v1_get_fault_info,
374 };
375 
376 /* SysMMU v5 */
377 static const struct sysmmu_variant sysmmu_v5_variant = {
378 	.pt_base	= 0x0c,
379 	.flush_all	= 0x10,
380 	.flush_entry	= 0x14,
381 	.flush_range	= 0x18,
382 	.flush_start	= 0x20,
383 	.flush_end	= 0x24,
384 	.int_status	= 0x60,
385 	.int_clear	= 0x64,
386 
387 	.get_fault_info	= exynos_sysmmu_v5_get_fault_info,
388 };
389 
390 /* SysMMU v7: non-VM capable register layout */
391 static const struct sysmmu_variant sysmmu_v7_variant = {
392 	.pt_base	= 0x0c,
393 	.flush_all	= 0x10,
394 	.flush_entry	= 0x14,
395 	.flush_range	= 0x18,
396 	.flush_start	= 0x20,
397 	.flush_end	= 0x24,
398 	.int_status	= 0x60,
399 	.int_clear	= 0x64,
400 	.fault_va	= 0x70,
401 	.fault_info	= 0x78,
402 
403 	.get_fault_info	= exynos_sysmmu_v7_get_fault_info,
404 };
405 
406 /* SysMMU v7: VM capable register layout */
407 static const struct sysmmu_variant sysmmu_v7_vm_variant = {
408 	.pt_base	= 0x800c,
409 	.flush_all	= 0x8010,
410 	.flush_entry	= 0x8014,
411 	.flush_range	= 0x8018,
412 	.flush_start	= 0x8020,
413 	.flush_end	= 0x8024,
414 	.int_status	= 0x60,
415 	.int_clear	= 0x64,
416 	.fault_va	= 0x1000,
417 	.fault_info	= 0x1004,
418 
419 	.get_fault_info	= exynos_sysmmu_v7_get_fault_info,
420 };
421 
422 static struct exynos_iommu_domain *to_exynos_domain(struct iommu_domain *dom)
423 {
424 	return container_of(dom, struct exynos_iommu_domain, domain);
425 }
426 
427 static void sysmmu_unblock(struct sysmmu_drvdata *data)
428 {
429 	writel(CTRL_ENABLE, data->sfrbase + REG_MMU_CTRL);
430 }
431 
432 static bool sysmmu_block(struct sysmmu_drvdata *data)
433 {
434 	int i = 120;
435 
436 	writel(CTRL_BLOCK, data->sfrbase + REG_MMU_CTRL);
437 	while ((i > 0) && !(readl(data->sfrbase + REG_MMU_STATUS) & 1))
438 		--i;
439 
440 	if (!(readl(data->sfrbase + REG_MMU_STATUS) & 1)) {
441 		sysmmu_unblock(data);
442 		return false;
443 	}
444 
445 	return true;
446 }
447 
448 static void __sysmmu_tlb_invalidate(struct sysmmu_drvdata *data)
449 {
450 	writel(0x1, SYSMMU_REG(data, flush_all));
451 }
452 
453 static void __sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
454 				sysmmu_iova_t iova, unsigned int num_inv)
455 {
456 	unsigned int i;
457 
458 	if (MMU_MAJ_VER(data->version) < 5 || num_inv == 1) {
459 		for (i = 0; i < num_inv; i++) {
460 			writel((iova & SPAGE_MASK) | 1,
461 			       SYSMMU_REG(data, flush_entry));
462 			iova += SPAGE_SIZE;
463 		}
464 	} else {
465 		writel(iova & SPAGE_MASK, SYSMMU_REG(data, flush_start));
466 		writel((iova & SPAGE_MASK) + (num_inv - 1) * SPAGE_SIZE,
467 		       SYSMMU_REG(data, flush_end));
468 		writel(0x1, SYSMMU_REG(data, flush_range));
469 	}
470 }
471 
472 static void __sysmmu_set_ptbase(struct sysmmu_drvdata *data, phys_addr_t pgd)
473 {
474 	u32 pt_base;
475 
476 	if (MMU_MAJ_VER(data->version) < 5)
477 		pt_base = pgd;
478 	else
479 		pt_base = pgd >> SPAGE_ORDER;
480 
481 	writel(pt_base, SYSMMU_REG(data, pt_base));
482 	__sysmmu_tlb_invalidate(data);
483 }
484 
485 static void __sysmmu_enable_clocks(struct sysmmu_drvdata *data)
486 {
487 	BUG_ON(clk_prepare_enable(data->clk_master));
488 	BUG_ON(clk_prepare_enable(data->clk));
489 	BUG_ON(clk_prepare_enable(data->pclk));
490 	BUG_ON(clk_prepare_enable(data->aclk));
491 }
492 
493 static void __sysmmu_disable_clocks(struct sysmmu_drvdata *data)
494 {
495 	clk_disable_unprepare(data->aclk);
496 	clk_disable_unprepare(data->pclk);
497 	clk_disable_unprepare(data->clk);
498 	clk_disable_unprepare(data->clk_master);
499 }
500 
501 static bool __sysmmu_has_capa1(struct sysmmu_drvdata *data)
502 {
503 	u32 capa0 = readl(data->sfrbase + REG_V7_CAPA0);
504 
505 	return capa0 & CAPA0_CAPA1_EXIST;
506 }
507 
508 static void __sysmmu_get_vcr(struct sysmmu_drvdata *data)
509 {
510 	u32 capa1 = readl(data->sfrbase + REG_V7_CAPA1);
511 
512 	data->has_vcr = capa1 & CAPA1_VCR_ENABLED;
513 }
514 
515 static void __sysmmu_get_version(struct sysmmu_drvdata *data)
516 {
517 	u32 ver;
518 
519 	__sysmmu_enable_clocks(data);
520 
521 	ver = readl(data->sfrbase + REG_MMU_VERSION);
522 
523 	/* controllers on some SoCs don't report proper version */
524 	if (ver == 0x80000001u)
525 		data->version = MAKE_MMU_VER(1, 0);
526 	else
527 		data->version = MMU_RAW_VER(ver);
528 
529 	dev_dbg(data->sysmmu, "hardware version: %d.%d\n",
530 		MMU_MAJ_VER(data->version), MMU_MIN_VER(data->version));
531 
532 	if (MMU_MAJ_VER(data->version) < 5) {
533 		data->variant = &sysmmu_v1_variant;
534 	} else if (MMU_MAJ_VER(data->version) < 7) {
535 		data->variant = &sysmmu_v5_variant;
536 	} else {
537 		if (__sysmmu_has_capa1(data))
538 			__sysmmu_get_vcr(data);
539 		if (data->has_vcr)
540 			data->variant = &sysmmu_v7_vm_variant;
541 		else
542 			data->variant = &sysmmu_v7_variant;
543 	}
544 
545 	__sysmmu_disable_clocks(data);
546 }
547 
548 static void show_fault_information(struct sysmmu_drvdata *data,
549 				   const struct sysmmu_fault *fault)
550 {
551 	sysmmu_pte_t *ent;
552 
553 	dev_err(data->sysmmu, "%s: [%s] %s FAULT occurred at %#x\n",
554 		dev_name(data->master),
555 		fault->type == IOMMU_FAULT_READ ? "READ" : "WRITE",
556 		fault->name, fault->addr);
557 	dev_dbg(data->sysmmu, "Page table base: %pa\n", &data->pgtable);
558 	ent = section_entry(phys_to_virt(data->pgtable), fault->addr);
559 	dev_dbg(data->sysmmu, "\tLv1 entry: %#x\n", *ent);
560 	if (lv1ent_page(ent)) {
561 		ent = page_entry(ent, fault->addr);
562 		dev_dbg(data->sysmmu, "\t Lv2 entry: %#x\n", *ent);
563 	}
564 }
565 
566 static irqreturn_t exynos_sysmmu_irq(int irq, void *dev_id)
567 {
568 	struct sysmmu_drvdata *data = dev_id;
569 	unsigned int itype;
570 	struct sysmmu_fault fault;
571 	int ret = -ENOSYS;
572 
573 	WARN_ON(!data->active);
574 
575 	spin_lock(&data->lock);
576 	clk_enable(data->clk_master);
577 
578 	itype = __ffs(readl(SYSMMU_REG(data, int_status)));
579 	ret = data->variant->get_fault_info(data, itype, &fault);
580 	if (ret) {
581 		dev_err(data->sysmmu, "Unhandled interrupt bit %u\n", itype);
582 		goto out;
583 	}
584 	show_fault_information(data, &fault);
585 
586 	if (data->domain) {
587 		ret = report_iommu_fault(&data->domain->domain, data->master,
588 					 fault.addr, fault.type);
589 	}
590 	if (ret)
591 		panic("Unrecoverable System MMU Fault!");
592 
593 out:
594 	writel(1 << itype, SYSMMU_REG(data, int_clear));
595 
596 	/* SysMMU is in blocked state when interrupt occurred */
597 	sysmmu_unblock(data);
598 	clk_disable(data->clk_master);
599 	spin_unlock(&data->lock);
600 
601 	return IRQ_HANDLED;
602 }
603 
604 static void __sysmmu_disable(struct sysmmu_drvdata *data)
605 {
606 	unsigned long flags;
607 
608 	clk_enable(data->clk_master);
609 
610 	spin_lock_irqsave(&data->lock, flags);
611 	writel(CTRL_DISABLE, data->sfrbase + REG_MMU_CTRL);
612 	writel(0, data->sfrbase + REG_MMU_CFG);
613 	data->active = false;
614 	spin_unlock_irqrestore(&data->lock, flags);
615 
616 	__sysmmu_disable_clocks(data);
617 }
618 
619 static void __sysmmu_init_config(struct sysmmu_drvdata *data)
620 {
621 	unsigned int cfg;
622 
623 	if (data->version <= MAKE_MMU_VER(3, 1))
624 		cfg = CFG_LRU | CFG_QOS(15);
625 	else if (data->version <= MAKE_MMU_VER(3, 2))
626 		cfg = CFG_LRU | CFG_QOS(15) | CFG_FLPDCACHE | CFG_SYSSEL;
627 	else
628 		cfg = CFG_QOS(15) | CFG_FLPDCACHE | CFG_ACGEN;
629 
630 	cfg |= CFG_EAP; /* enable access protection bits check */
631 
632 	writel(cfg, data->sfrbase + REG_MMU_CFG);
633 }
634 
635 static void __sysmmu_enable_vid(struct sysmmu_drvdata *data)
636 {
637 	u32 ctrl;
638 
639 	if (MMU_MAJ_VER(data->version) < 7 || !data->has_vcr)
640 		return;
641 
642 	ctrl = readl(data->sfrbase + REG_V7_CTRL_VM);
643 	ctrl |= CTRL_VM_ENABLE | CTRL_VM_FAULT_MODE_STALL;
644 	writel(ctrl, data->sfrbase + REG_V7_CTRL_VM);
645 }
646 
647 static void __sysmmu_enable(struct sysmmu_drvdata *data)
648 {
649 	unsigned long flags;
650 
651 	__sysmmu_enable_clocks(data);
652 
653 	spin_lock_irqsave(&data->lock, flags);
654 	writel(CTRL_BLOCK, data->sfrbase + REG_MMU_CTRL);
655 	__sysmmu_init_config(data);
656 	__sysmmu_set_ptbase(data, data->pgtable);
657 	__sysmmu_enable_vid(data);
658 	writel(CTRL_ENABLE, data->sfrbase + REG_MMU_CTRL);
659 	data->active = true;
660 	spin_unlock_irqrestore(&data->lock, flags);
661 
662 	/*
663 	 * SYSMMU driver keeps master's clock enabled only for the short
664 	 * time, while accessing the registers. For performing address
665 	 * translation during DMA transaction it relies on the client
666 	 * driver to enable it.
667 	 */
668 	clk_disable(data->clk_master);
669 }
670 
671 static void sysmmu_tlb_invalidate_flpdcache(struct sysmmu_drvdata *data,
672 					    sysmmu_iova_t iova)
673 {
674 	unsigned long flags;
675 
676 	spin_lock_irqsave(&data->lock, flags);
677 	if (data->active && data->version >= MAKE_MMU_VER(3, 3)) {
678 		clk_enable(data->clk_master);
679 		if (sysmmu_block(data)) {
680 			if (data->version >= MAKE_MMU_VER(5, 0))
681 				__sysmmu_tlb_invalidate(data);
682 			else
683 				__sysmmu_tlb_invalidate_entry(data, iova, 1);
684 			sysmmu_unblock(data);
685 		}
686 		clk_disable(data->clk_master);
687 	}
688 	spin_unlock_irqrestore(&data->lock, flags);
689 }
690 
691 static void sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
692 					sysmmu_iova_t iova, size_t size)
693 {
694 	unsigned long flags;
695 
696 	spin_lock_irqsave(&data->lock, flags);
697 	if (data->active) {
698 		unsigned int num_inv = 1;
699 
700 		clk_enable(data->clk_master);
701 
702 		/*
703 		 * L2TLB invalidation required
704 		 * 4KB page: 1 invalidation
705 		 * 64KB page: 16 invalidations
706 		 * 1MB page: 64 invalidations
707 		 * because it is set-associative TLB
708 		 * with 8-way and 64 sets.
709 		 * 1MB page can be cached in one of all sets.
710 		 * 64KB page can be one of 16 consecutive sets.
711 		 */
712 		if (MMU_MAJ_VER(data->version) == 2)
713 			num_inv = min_t(unsigned int, size / SPAGE_SIZE, 64);
714 
715 		if (sysmmu_block(data)) {
716 			__sysmmu_tlb_invalidate_entry(data, iova, num_inv);
717 			sysmmu_unblock(data);
718 		}
719 		clk_disable(data->clk_master);
720 	}
721 	spin_unlock_irqrestore(&data->lock, flags);
722 }
723 
724 static const struct iommu_ops exynos_iommu_ops;
725 
726 static int exynos_sysmmu_probe(struct platform_device *pdev)
727 {
728 	int irq, ret;
729 	struct device *dev = &pdev->dev;
730 	struct sysmmu_drvdata *data;
731 	struct resource *res;
732 
733 	data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL);
734 	if (!data)
735 		return -ENOMEM;
736 
737 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
738 	data->sfrbase = devm_ioremap_resource(dev, res);
739 	if (IS_ERR(data->sfrbase))
740 		return PTR_ERR(data->sfrbase);
741 
742 	irq = platform_get_irq(pdev, 0);
743 	if (irq <= 0)
744 		return irq;
745 
746 	ret = devm_request_irq(dev, irq, exynos_sysmmu_irq, 0,
747 				dev_name(dev), data);
748 	if (ret) {
749 		dev_err(dev, "Unabled to register handler of irq %d\n", irq);
750 		return ret;
751 	}
752 
753 	data->clk = devm_clk_get_optional(dev, "sysmmu");
754 	if (IS_ERR(data->clk))
755 		return PTR_ERR(data->clk);
756 
757 	data->aclk = devm_clk_get_optional(dev, "aclk");
758 	if (IS_ERR(data->aclk))
759 		return PTR_ERR(data->aclk);
760 
761 	data->pclk = devm_clk_get_optional(dev, "pclk");
762 	if (IS_ERR(data->pclk))
763 		return PTR_ERR(data->pclk);
764 
765 	if (!data->clk && (!data->aclk || !data->pclk)) {
766 		dev_err(dev, "Failed to get device clock(s)!\n");
767 		return -ENOSYS;
768 	}
769 
770 	data->clk_master = devm_clk_get_optional(dev, "master");
771 	if (IS_ERR(data->clk_master))
772 		return PTR_ERR(data->clk_master);
773 
774 	data->sysmmu = dev;
775 	spin_lock_init(&data->lock);
776 
777 	__sysmmu_get_version(data);
778 
779 	ret = iommu_device_sysfs_add(&data->iommu, &pdev->dev, NULL,
780 				     dev_name(data->sysmmu));
781 	if (ret)
782 		return ret;
783 
784 	platform_set_drvdata(pdev, data);
785 
786 	if (PG_ENT_SHIFT < 0) {
787 		if (MMU_MAJ_VER(data->version) < 5) {
788 			PG_ENT_SHIFT = SYSMMU_PG_ENT_SHIFT;
789 			LV1_PROT = SYSMMU_LV1_PROT;
790 			LV2_PROT = SYSMMU_LV2_PROT;
791 		} else {
792 			PG_ENT_SHIFT = SYSMMU_V5_PG_ENT_SHIFT;
793 			LV1_PROT = SYSMMU_V5_LV1_PROT;
794 			LV2_PROT = SYSMMU_V5_LV2_PROT;
795 		}
796 	}
797 
798 	if (MMU_MAJ_VER(data->version) >= 5) {
799 		ret = dma_set_mask(dev, DMA_BIT_MASK(36));
800 		if (ret) {
801 			dev_err(dev, "Unable to set DMA mask: %d\n", ret);
802 			goto err_dma_set_mask;
803 		}
804 	}
805 
806 	/*
807 	 * use the first registered sysmmu device for performing
808 	 * dma mapping operations on iommu page tables (cpu cache flush)
809 	 */
810 	if (!dma_dev)
811 		dma_dev = &pdev->dev;
812 
813 	pm_runtime_enable(dev);
814 
815 	ret = iommu_device_register(&data->iommu, &exynos_iommu_ops, dev);
816 	if (ret)
817 		goto err_dma_set_mask;
818 
819 	return 0;
820 
821 err_dma_set_mask:
822 	iommu_device_sysfs_remove(&data->iommu);
823 	return ret;
824 }
825 
826 static int __maybe_unused exynos_sysmmu_suspend(struct device *dev)
827 {
828 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
829 	struct device *master = data->master;
830 
831 	if (master) {
832 		struct exynos_iommu_owner *owner = dev_iommu_priv_get(master);
833 
834 		mutex_lock(&owner->rpm_lock);
835 		if (&data->domain->domain != &exynos_identity_domain) {
836 			dev_dbg(data->sysmmu, "saving state\n");
837 			__sysmmu_disable(data);
838 		}
839 		mutex_unlock(&owner->rpm_lock);
840 	}
841 	return 0;
842 }
843 
844 static int __maybe_unused exynos_sysmmu_resume(struct device *dev)
845 {
846 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
847 	struct device *master = data->master;
848 
849 	if (master) {
850 		struct exynos_iommu_owner *owner = dev_iommu_priv_get(master);
851 
852 		mutex_lock(&owner->rpm_lock);
853 		if (&data->domain->domain != &exynos_identity_domain) {
854 			dev_dbg(data->sysmmu, "restoring state\n");
855 			__sysmmu_enable(data);
856 		}
857 		mutex_unlock(&owner->rpm_lock);
858 	}
859 	return 0;
860 }
861 
862 static const struct dev_pm_ops sysmmu_pm_ops = {
863 	SET_RUNTIME_PM_OPS(exynos_sysmmu_suspend, exynos_sysmmu_resume, NULL)
864 	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
865 				pm_runtime_force_resume)
866 };
867 
868 static const struct of_device_id sysmmu_of_match[] = {
869 	{ .compatible	= "samsung,exynos-sysmmu", },
870 	{ },
871 };
872 
873 static struct platform_driver exynos_sysmmu_driver __refdata = {
874 	.probe	= exynos_sysmmu_probe,
875 	.driver	= {
876 		.name		= "exynos-sysmmu",
877 		.of_match_table	= sysmmu_of_match,
878 		.pm		= &sysmmu_pm_ops,
879 		.suppress_bind_attrs = true,
880 	}
881 };
882 
883 static inline void exynos_iommu_set_pte(sysmmu_pte_t *ent, sysmmu_pte_t val)
884 {
885 	dma_sync_single_for_cpu(dma_dev, virt_to_phys(ent), sizeof(*ent),
886 				DMA_TO_DEVICE);
887 	*ent = cpu_to_le32(val);
888 	dma_sync_single_for_device(dma_dev, virt_to_phys(ent), sizeof(*ent),
889 				   DMA_TO_DEVICE);
890 }
891 
892 static struct iommu_domain *exynos_iommu_domain_alloc_paging(struct device *dev)
893 {
894 	struct exynos_iommu_domain *domain;
895 	dma_addr_t handle;
896 	int i;
897 
898 	/* Check if correct PTE offsets are initialized */
899 	BUG_ON(PG_ENT_SHIFT < 0 || !dma_dev);
900 
901 	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
902 	if (!domain)
903 		return NULL;
904 
905 	domain->pgtable = iommu_alloc_pages(GFP_KERNEL, 2);
906 	if (!domain->pgtable)
907 		goto err_pgtable;
908 
909 	domain->lv2entcnt = iommu_alloc_pages(GFP_KERNEL, 1);
910 	if (!domain->lv2entcnt)
911 		goto err_counter;
912 
913 	/* Workaround for System MMU v3.3 to prevent caching 1MiB mapping */
914 	for (i = 0; i < NUM_LV1ENTRIES; i++)
915 		domain->pgtable[i] = ZERO_LV2LINK;
916 
917 	handle = dma_map_single(dma_dev, domain->pgtable, LV1TABLE_SIZE,
918 				DMA_TO_DEVICE);
919 	/* For mapping page table entries we rely on dma == phys */
920 	BUG_ON(handle != virt_to_phys(domain->pgtable));
921 	if (dma_mapping_error(dma_dev, handle))
922 		goto err_lv2ent;
923 
924 	spin_lock_init(&domain->lock);
925 	spin_lock_init(&domain->pgtablelock);
926 	INIT_LIST_HEAD(&domain->clients);
927 
928 	domain->domain.geometry.aperture_start = 0;
929 	domain->domain.geometry.aperture_end   = ~0UL;
930 	domain->domain.geometry.force_aperture = true;
931 
932 	return &domain->domain;
933 
934 err_lv2ent:
935 	iommu_free_pages(domain->lv2entcnt, 1);
936 err_counter:
937 	iommu_free_pages(domain->pgtable, 2);
938 err_pgtable:
939 	kfree(domain);
940 	return NULL;
941 }
942 
943 static void exynos_iommu_domain_free(struct iommu_domain *iommu_domain)
944 {
945 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
946 	struct sysmmu_drvdata *data, *next;
947 	unsigned long flags;
948 	int i;
949 
950 	WARN_ON(!list_empty(&domain->clients));
951 
952 	spin_lock_irqsave(&domain->lock, flags);
953 
954 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
955 		spin_lock(&data->lock);
956 		__sysmmu_disable(data);
957 		data->pgtable = 0;
958 		data->domain = NULL;
959 		list_del_init(&data->domain_node);
960 		spin_unlock(&data->lock);
961 	}
962 
963 	spin_unlock_irqrestore(&domain->lock, flags);
964 
965 	dma_unmap_single(dma_dev, virt_to_phys(domain->pgtable), LV1TABLE_SIZE,
966 			 DMA_TO_DEVICE);
967 
968 	for (i = 0; i < NUM_LV1ENTRIES; i++)
969 		if (lv1ent_page(domain->pgtable + i)) {
970 			phys_addr_t base = lv2table_base(domain->pgtable + i);
971 
972 			dma_unmap_single(dma_dev, base, LV2TABLE_SIZE,
973 					 DMA_TO_DEVICE);
974 			kmem_cache_free(lv2table_kmem_cache,
975 					phys_to_virt(base));
976 		}
977 
978 	iommu_free_pages(domain->pgtable, 2);
979 	iommu_free_pages(domain->lv2entcnt, 1);
980 	kfree(domain);
981 }
982 
983 static int exynos_iommu_identity_attach(struct iommu_domain *identity_domain,
984 					struct device *dev)
985 {
986 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
987 	struct exynos_iommu_domain *domain;
988 	phys_addr_t pagetable;
989 	struct sysmmu_drvdata *data, *next;
990 	unsigned long flags;
991 
992 	if (owner->domain == identity_domain)
993 		return 0;
994 
995 	domain = to_exynos_domain(owner->domain);
996 	pagetable = virt_to_phys(domain->pgtable);
997 
998 	mutex_lock(&owner->rpm_lock);
999 
1000 	list_for_each_entry(data, &owner->controllers, owner_node) {
1001 		pm_runtime_get_noresume(data->sysmmu);
1002 		if (pm_runtime_active(data->sysmmu))
1003 			__sysmmu_disable(data);
1004 		pm_runtime_put(data->sysmmu);
1005 	}
1006 
1007 	spin_lock_irqsave(&domain->lock, flags);
1008 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
1009 		spin_lock(&data->lock);
1010 		data->pgtable = 0;
1011 		data->domain = NULL;
1012 		list_del_init(&data->domain_node);
1013 		spin_unlock(&data->lock);
1014 	}
1015 	owner->domain = identity_domain;
1016 	spin_unlock_irqrestore(&domain->lock, flags);
1017 
1018 	mutex_unlock(&owner->rpm_lock);
1019 
1020 	dev_dbg(dev, "%s: Restored IOMMU to IDENTITY from pgtable %pa\n",
1021 		__func__, &pagetable);
1022 	return 0;
1023 }
1024 
1025 static struct iommu_domain_ops exynos_identity_ops = {
1026 	.attach_dev = exynos_iommu_identity_attach,
1027 };
1028 
1029 static struct iommu_domain exynos_identity_domain = {
1030 	.type = IOMMU_DOMAIN_IDENTITY,
1031 	.ops = &exynos_identity_ops,
1032 };
1033 
1034 static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
1035 				   struct device *dev)
1036 {
1037 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
1038 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
1039 	struct sysmmu_drvdata *data;
1040 	phys_addr_t pagetable = virt_to_phys(domain->pgtable);
1041 	unsigned long flags;
1042 	int err;
1043 
1044 	err = exynos_iommu_identity_attach(&exynos_identity_domain, dev);
1045 	if (err)
1046 		return err;
1047 
1048 	mutex_lock(&owner->rpm_lock);
1049 
1050 	spin_lock_irqsave(&domain->lock, flags);
1051 	list_for_each_entry(data, &owner->controllers, owner_node) {
1052 		spin_lock(&data->lock);
1053 		data->pgtable = pagetable;
1054 		data->domain = domain;
1055 		list_add_tail(&data->domain_node, &domain->clients);
1056 		spin_unlock(&data->lock);
1057 	}
1058 	owner->domain = iommu_domain;
1059 	spin_unlock_irqrestore(&domain->lock, flags);
1060 
1061 	list_for_each_entry(data, &owner->controllers, owner_node) {
1062 		pm_runtime_get_noresume(data->sysmmu);
1063 		if (pm_runtime_active(data->sysmmu))
1064 			__sysmmu_enable(data);
1065 		pm_runtime_put(data->sysmmu);
1066 	}
1067 
1068 	mutex_unlock(&owner->rpm_lock);
1069 
1070 	dev_dbg(dev, "%s: Attached IOMMU with pgtable %pa\n", __func__,
1071 		&pagetable);
1072 
1073 	return 0;
1074 }
1075 
1076 static sysmmu_pte_t *alloc_lv2entry(struct exynos_iommu_domain *domain,
1077 		sysmmu_pte_t *sent, sysmmu_iova_t iova, short *pgcounter)
1078 {
1079 	if (lv1ent_section(sent)) {
1080 		WARN(1, "Trying mapping on %#08x mapped with 1MiB page", iova);
1081 		return ERR_PTR(-EADDRINUSE);
1082 	}
1083 
1084 	if (lv1ent_fault(sent)) {
1085 		dma_addr_t handle;
1086 		sysmmu_pte_t *pent;
1087 		bool need_flush_flpd_cache = lv1ent_zero(sent);
1088 
1089 		pent = kmem_cache_zalloc(lv2table_kmem_cache, GFP_ATOMIC);
1090 		BUG_ON((uintptr_t)pent & (LV2TABLE_SIZE - 1));
1091 		if (!pent)
1092 			return ERR_PTR(-ENOMEM);
1093 
1094 		exynos_iommu_set_pte(sent, mk_lv1ent_page(virt_to_phys(pent)));
1095 		kmemleak_ignore(pent);
1096 		*pgcounter = NUM_LV2ENTRIES;
1097 		handle = dma_map_single(dma_dev, pent, LV2TABLE_SIZE,
1098 					DMA_TO_DEVICE);
1099 		if (dma_mapping_error(dma_dev, handle)) {
1100 			kmem_cache_free(lv2table_kmem_cache, pent);
1101 			return ERR_PTR(-EADDRINUSE);
1102 		}
1103 
1104 		/*
1105 		 * If pre-fetched SLPD is a faulty SLPD in zero_l2_table,
1106 		 * FLPD cache may cache the address of zero_l2_table. This
1107 		 * function replaces the zero_l2_table with new L2 page table
1108 		 * to write valid mappings.
1109 		 * Accessing the valid area may cause page fault since FLPD
1110 		 * cache may still cache zero_l2_table for the valid area
1111 		 * instead of new L2 page table that has the mapping
1112 		 * information of the valid area.
1113 		 * Thus any replacement of zero_l2_table with other valid L2
1114 		 * page table must involve FLPD cache invalidation for System
1115 		 * MMU v3.3.
1116 		 * FLPD cache invalidation is performed with TLB invalidation
1117 		 * by VPN without blocking. It is safe to invalidate TLB without
1118 		 * blocking because the target address of TLB invalidation is
1119 		 * not currently mapped.
1120 		 */
1121 		if (need_flush_flpd_cache) {
1122 			struct sysmmu_drvdata *data;
1123 
1124 			spin_lock(&domain->lock);
1125 			list_for_each_entry(data, &domain->clients, domain_node)
1126 				sysmmu_tlb_invalidate_flpdcache(data, iova);
1127 			spin_unlock(&domain->lock);
1128 		}
1129 	}
1130 
1131 	return page_entry(sent, iova);
1132 }
1133 
1134 static int lv1set_section(struct exynos_iommu_domain *domain,
1135 			  sysmmu_pte_t *sent, sysmmu_iova_t iova,
1136 			  phys_addr_t paddr, int prot, short *pgcnt)
1137 {
1138 	if (lv1ent_section(sent)) {
1139 		WARN(1, "Trying mapping on 1MiB@%#08x that is mapped",
1140 			iova);
1141 		return -EADDRINUSE;
1142 	}
1143 
1144 	if (lv1ent_page(sent)) {
1145 		if (*pgcnt != NUM_LV2ENTRIES) {
1146 			WARN(1, "Trying mapping on 1MiB@%#08x that is mapped",
1147 				iova);
1148 			return -EADDRINUSE;
1149 		}
1150 
1151 		kmem_cache_free(lv2table_kmem_cache, page_entry(sent, 0));
1152 		*pgcnt = 0;
1153 	}
1154 
1155 	exynos_iommu_set_pte(sent, mk_lv1ent_sect(paddr, prot));
1156 
1157 	spin_lock(&domain->lock);
1158 	if (lv1ent_page_zero(sent)) {
1159 		struct sysmmu_drvdata *data;
1160 		/*
1161 		 * Flushing FLPD cache in System MMU v3.3 that may cache a FLPD
1162 		 * entry by speculative prefetch of SLPD which has no mapping.
1163 		 */
1164 		list_for_each_entry(data, &domain->clients, domain_node)
1165 			sysmmu_tlb_invalidate_flpdcache(data, iova);
1166 	}
1167 	spin_unlock(&domain->lock);
1168 
1169 	return 0;
1170 }
1171 
1172 static int lv2set_page(sysmmu_pte_t *pent, phys_addr_t paddr, size_t size,
1173 		       int prot, short *pgcnt)
1174 {
1175 	if (size == SPAGE_SIZE) {
1176 		if (WARN_ON(!lv2ent_fault(pent)))
1177 			return -EADDRINUSE;
1178 
1179 		exynos_iommu_set_pte(pent, mk_lv2ent_spage(paddr, prot));
1180 		*pgcnt -= 1;
1181 	} else { /* size == LPAGE_SIZE */
1182 		int i;
1183 		dma_addr_t pent_base = virt_to_phys(pent);
1184 
1185 		dma_sync_single_for_cpu(dma_dev, pent_base,
1186 					sizeof(*pent) * SPAGES_PER_LPAGE,
1187 					DMA_TO_DEVICE);
1188 		for (i = 0; i < SPAGES_PER_LPAGE; i++, pent++) {
1189 			if (WARN_ON(!lv2ent_fault(pent))) {
1190 				if (i > 0)
1191 					memset(pent - i, 0, sizeof(*pent) * i);
1192 				return -EADDRINUSE;
1193 			}
1194 
1195 			*pent = mk_lv2ent_lpage(paddr, prot);
1196 		}
1197 		dma_sync_single_for_device(dma_dev, pent_base,
1198 					   sizeof(*pent) * SPAGES_PER_LPAGE,
1199 					   DMA_TO_DEVICE);
1200 		*pgcnt -= SPAGES_PER_LPAGE;
1201 	}
1202 
1203 	return 0;
1204 }
1205 
1206 /*
1207  * *CAUTION* to the I/O virtual memory managers that support exynos-iommu:
1208  *
1209  * System MMU v3.x has advanced logic to improve address translation
1210  * performance with caching more page table entries by a page table walk.
1211  * However, the logic has a bug that while caching faulty page table entries,
1212  * System MMU reports page fault if the cached fault entry is hit even though
1213  * the fault entry is updated to a valid entry after the entry is cached.
1214  * To prevent caching faulty page table entries which may be updated to valid
1215  * entries later, the virtual memory manager should care about the workaround
1216  * for the problem. The following describes the workaround.
1217  *
1218  * Any two consecutive I/O virtual address regions must have a hole of 128KiB
1219  * at maximum to prevent misbehavior of System MMU 3.x (workaround for h/w bug).
1220  *
1221  * Precisely, any start address of I/O virtual region must be aligned with
1222  * the following sizes for System MMU v3.1 and v3.2.
1223  * System MMU v3.1: 128KiB
1224  * System MMU v3.2: 256KiB
1225  *
1226  * Because System MMU v3.3 caches page table entries more aggressively, it needs
1227  * more workarounds.
1228  * - Any two consecutive I/O virtual regions must have a hole of size larger
1229  *   than or equal to 128KiB.
1230  * - Start address of an I/O virtual region must be aligned by 128KiB.
1231  */
1232 static int exynos_iommu_map(struct iommu_domain *iommu_domain,
1233 			    unsigned long l_iova, phys_addr_t paddr, size_t size,
1234 			    size_t count, int prot, gfp_t gfp, size_t *mapped)
1235 {
1236 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
1237 	sysmmu_pte_t *entry;
1238 	sysmmu_iova_t iova = (sysmmu_iova_t)l_iova;
1239 	unsigned long flags;
1240 	int ret = -ENOMEM;
1241 
1242 	BUG_ON(domain->pgtable == NULL);
1243 	prot &= SYSMMU_SUPPORTED_PROT_BITS;
1244 
1245 	spin_lock_irqsave(&domain->pgtablelock, flags);
1246 
1247 	entry = section_entry(domain->pgtable, iova);
1248 
1249 	if (size == SECT_SIZE) {
1250 		ret = lv1set_section(domain, entry, iova, paddr, prot,
1251 				     &domain->lv2entcnt[lv1ent_offset(iova)]);
1252 	} else {
1253 		sysmmu_pte_t *pent;
1254 
1255 		pent = alloc_lv2entry(domain, entry, iova,
1256 				      &domain->lv2entcnt[lv1ent_offset(iova)]);
1257 
1258 		if (IS_ERR(pent))
1259 			ret = PTR_ERR(pent);
1260 		else
1261 			ret = lv2set_page(pent, paddr, size, prot,
1262 				       &domain->lv2entcnt[lv1ent_offset(iova)]);
1263 	}
1264 
1265 	if (ret)
1266 		pr_err("%s: Failed(%d) to map %#zx bytes @ %#x\n",
1267 			__func__, ret, size, iova);
1268 	else
1269 		*mapped = size;
1270 
1271 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1272 
1273 	return ret;
1274 }
1275 
1276 static void exynos_iommu_tlb_invalidate_entry(struct exynos_iommu_domain *domain,
1277 					      sysmmu_iova_t iova, size_t size)
1278 {
1279 	struct sysmmu_drvdata *data;
1280 	unsigned long flags;
1281 
1282 	spin_lock_irqsave(&domain->lock, flags);
1283 
1284 	list_for_each_entry(data, &domain->clients, domain_node)
1285 		sysmmu_tlb_invalidate_entry(data, iova, size);
1286 
1287 	spin_unlock_irqrestore(&domain->lock, flags);
1288 }
1289 
1290 static size_t exynos_iommu_unmap(struct iommu_domain *iommu_domain,
1291 				 unsigned long l_iova, size_t size, size_t count,
1292 				 struct iommu_iotlb_gather *gather)
1293 {
1294 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
1295 	sysmmu_iova_t iova = (sysmmu_iova_t)l_iova;
1296 	sysmmu_pte_t *ent;
1297 	size_t err_pgsize;
1298 	unsigned long flags;
1299 
1300 	BUG_ON(domain->pgtable == NULL);
1301 
1302 	spin_lock_irqsave(&domain->pgtablelock, flags);
1303 
1304 	ent = section_entry(domain->pgtable, iova);
1305 
1306 	if (lv1ent_section(ent)) {
1307 		if (WARN_ON(size < SECT_SIZE)) {
1308 			err_pgsize = SECT_SIZE;
1309 			goto err;
1310 		}
1311 
1312 		/* workaround for h/w bug in System MMU v3.3 */
1313 		exynos_iommu_set_pte(ent, ZERO_LV2LINK);
1314 		size = SECT_SIZE;
1315 		goto done;
1316 	}
1317 
1318 	if (unlikely(lv1ent_fault(ent))) {
1319 		if (size > SECT_SIZE)
1320 			size = SECT_SIZE;
1321 		goto done;
1322 	}
1323 
1324 	/* lv1ent_page(sent) == true here */
1325 
1326 	ent = page_entry(ent, iova);
1327 
1328 	if (unlikely(lv2ent_fault(ent))) {
1329 		size = SPAGE_SIZE;
1330 		goto done;
1331 	}
1332 
1333 	if (lv2ent_small(ent)) {
1334 		exynos_iommu_set_pte(ent, 0);
1335 		size = SPAGE_SIZE;
1336 		domain->lv2entcnt[lv1ent_offset(iova)] += 1;
1337 		goto done;
1338 	}
1339 
1340 	/* lv1ent_large(ent) == true here */
1341 	if (WARN_ON(size < LPAGE_SIZE)) {
1342 		err_pgsize = LPAGE_SIZE;
1343 		goto err;
1344 	}
1345 
1346 	dma_sync_single_for_cpu(dma_dev, virt_to_phys(ent),
1347 				sizeof(*ent) * SPAGES_PER_LPAGE,
1348 				DMA_TO_DEVICE);
1349 	memset(ent, 0, sizeof(*ent) * SPAGES_PER_LPAGE);
1350 	dma_sync_single_for_device(dma_dev, virt_to_phys(ent),
1351 				   sizeof(*ent) * SPAGES_PER_LPAGE,
1352 				   DMA_TO_DEVICE);
1353 	size = LPAGE_SIZE;
1354 	domain->lv2entcnt[lv1ent_offset(iova)] += SPAGES_PER_LPAGE;
1355 done:
1356 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1357 
1358 	exynos_iommu_tlb_invalidate_entry(domain, iova, size);
1359 
1360 	return size;
1361 err:
1362 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1363 
1364 	pr_err("%s: Failed: size(%#zx) @ %#x is smaller than page size %#zx\n",
1365 		__func__, size, iova, err_pgsize);
1366 
1367 	return 0;
1368 }
1369 
1370 static phys_addr_t exynos_iommu_iova_to_phys(struct iommu_domain *iommu_domain,
1371 					  dma_addr_t iova)
1372 {
1373 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
1374 	sysmmu_pte_t *entry;
1375 	unsigned long flags;
1376 	phys_addr_t phys = 0;
1377 
1378 	spin_lock_irqsave(&domain->pgtablelock, flags);
1379 
1380 	entry = section_entry(domain->pgtable, iova);
1381 
1382 	if (lv1ent_section(entry)) {
1383 		phys = section_phys(entry) + section_offs(iova);
1384 	} else if (lv1ent_page(entry)) {
1385 		entry = page_entry(entry, iova);
1386 
1387 		if (lv2ent_large(entry))
1388 			phys = lpage_phys(entry) + lpage_offs(iova);
1389 		else if (lv2ent_small(entry))
1390 			phys = spage_phys(entry) + spage_offs(iova);
1391 	}
1392 
1393 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1394 
1395 	return phys;
1396 }
1397 
1398 static struct iommu_device *exynos_iommu_probe_device(struct device *dev)
1399 {
1400 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
1401 	struct sysmmu_drvdata *data;
1402 
1403 	if (!has_sysmmu(dev))
1404 		return ERR_PTR(-ENODEV);
1405 
1406 	list_for_each_entry(data, &owner->controllers, owner_node) {
1407 		/*
1408 		 * SYSMMU will be runtime activated via device link
1409 		 * (dependency) to its master device, so there are no
1410 		 * direct calls to pm_runtime_get/put in this driver.
1411 		 */
1412 		data->link = device_link_add(dev, data->sysmmu,
1413 					     DL_FLAG_STATELESS |
1414 					     DL_FLAG_PM_RUNTIME);
1415 	}
1416 
1417 	/* There is always at least one entry, see exynos_iommu_of_xlate() */
1418 	data = list_first_entry(&owner->controllers,
1419 				struct sysmmu_drvdata, owner_node);
1420 
1421 	return &data->iommu;
1422 }
1423 
1424 static void exynos_iommu_release_device(struct device *dev)
1425 {
1426 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
1427 	struct sysmmu_drvdata *data;
1428 
1429 	WARN_ON(exynos_iommu_identity_attach(&exynos_identity_domain, dev));
1430 
1431 	list_for_each_entry(data, &owner->controllers, owner_node)
1432 		device_link_del(data->link);
1433 }
1434 
1435 static int exynos_iommu_of_xlate(struct device *dev,
1436 				 const struct of_phandle_args *spec)
1437 {
1438 	struct platform_device *sysmmu = of_find_device_by_node(spec->np);
1439 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
1440 	struct sysmmu_drvdata *data, *entry;
1441 
1442 	if (!sysmmu)
1443 		return -ENODEV;
1444 
1445 	data = platform_get_drvdata(sysmmu);
1446 	if (!data) {
1447 		put_device(&sysmmu->dev);
1448 		return -ENODEV;
1449 	}
1450 
1451 	if (!owner) {
1452 		owner = kzalloc(sizeof(*owner), GFP_KERNEL);
1453 		if (!owner) {
1454 			put_device(&sysmmu->dev);
1455 			return -ENOMEM;
1456 		}
1457 
1458 		INIT_LIST_HEAD(&owner->controllers);
1459 		mutex_init(&owner->rpm_lock);
1460 		owner->domain = &exynos_identity_domain;
1461 		dev_iommu_priv_set(dev, owner);
1462 	}
1463 
1464 	list_for_each_entry(entry, &owner->controllers, owner_node)
1465 		if (entry == data)
1466 			return 0;
1467 
1468 	list_add_tail(&data->owner_node, &owner->controllers);
1469 	data->master = dev;
1470 
1471 	return 0;
1472 }
1473 
1474 static const struct iommu_ops exynos_iommu_ops = {
1475 	.identity_domain = &exynos_identity_domain,
1476 	.domain_alloc_paging = exynos_iommu_domain_alloc_paging,
1477 	.device_group = generic_device_group,
1478 	.probe_device = exynos_iommu_probe_device,
1479 	.release_device = exynos_iommu_release_device,
1480 	.pgsize_bitmap = SECT_SIZE | LPAGE_SIZE | SPAGE_SIZE,
1481 	.of_xlate = exynos_iommu_of_xlate,
1482 	.default_domain_ops = &(const struct iommu_domain_ops) {
1483 		.attach_dev	= exynos_iommu_attach_device,
1484 		.map_pages	= exynos_iommu_map,
1485 		.unmap_pages	= exynos_iommu_unmap,
1486 		.iova_to_phys	= exynos_iommu_iova_to_phys,
1487 		.free		= exynos_iommu_domain_free,
1488 	}
1489 };
1490 
1491 static int __init exynos_iommu_init(void)
1492 {
1493 	struct device_node *np;
1494 	int ret;
1495 
1496 	np = of_find_matching_node(NULL, sysmmu_of_match);
1497 	if (!np)
1498 		return 0;
1499 
1500 	of_node_put(np);
1501 
1502 	lv2table_kmem_cache = kmem_cache_create("exynos-iommu-lv2table",
1503 				LV2TABLE_SIZE, LV2TABLE_SIZE, 0, NULL);
1504 	if (!lv2table_kmem_cache) {
1505 		pr_err("%s: Failed to create kmem cache\n", __func__);
1506 		return -ENOMEM;
1507 	}
1508 
1509 	zero_lv2_table = kmem_cache_zalloc(lv2table_kmem_cache, GFP_KERNEL);
1510 	if (zero_lv2_table == NULL) {
1511 		pr_err("%s: Failed to allocate zero level2 page table\n",
1512 			__func__);
1513 		ret = -ENOMEM;
1514 		goto err_zero_lv2;
1515 	}
1516 
1517 	ret = platform_driver_register(&exynos_sysmmu_driver);
1518 	if (ret) {
1519 		pr_err("%s: Failed to register driver\n", __func__);
1520 		goto err_reg_driver;
1521 	}
1522 
1523 	return 0;
1524 err_reg_driver:
1525 	kmem_cache_free(lv2table_kmem_cache, zero_lv2_table);
1526 err_zero_lv2:
1527 	kmem_cache_destroy(lv2table_kmem_cache);
1528 	return ret;
1529 }
1530 core_initcall(exynos_iommu_init);
1531