xref: /openbmc/linux/drivers/iommu/exynos-iommu.c (revision ac4dfccb)
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 #include <linux/dma-iommu.h>
25 
26 typedef u32 sysmmu_iova_t;
27 typedef u32 sysmmu_pte_t;
28 
29 /* We do not consider super section mapping (16MB) */
30 #define SECT_ORDER 20
31 #define LPAGE_ORDER 16
32 #define SPAGE_ORDER 12
33 
34 #define SECT_SIZE (1 << SECT_ORDER)
35 #define LPAGE_SIZE (1 << LPAGE_ORDER)
36 #define SPAGE_SIZE (1 << SPAGE_ORDER)
37 
38 #define SECT_MASK (~(SECT_SIZE - 1))
39 #define LPAGE_MASK (~(LPAGE_SIZE - 1))
40 #define SPAGE_MASK (~(SPAGE_SIZE - 1))
41 
42 #define lv1ent_fault(sent) ((*(sent) == ZERO_LV2LINK) || \
43 			   ((*(sent) & 3) == 0) || ((*(sent) & 3) == 3))
44 #define lv1ent_zero(sent) (*(sent) == ZERO_LV2LINK)
45 #define lv1ent_page_zero(sent) ((*(sent) & 3) == 1)
46 #define lv1ent_page(sent) ((*(sent) != ZERO_LV2LINK) && \
47 			  ((*(sent) & 3) == 1))
48 #define lv1ent_section(sent) ((*(sent) & 3) == 2)
49 
50 #define lv2ent_fault(pent) ((*(pent) & 3) == 0)
51 #define lv2ent_small(pent) ((*(pent) & 2) == 2)
52 #define lv2ent_large(pent) ((*(pent) & 3) == 1)
53 
54 /*
55  * v1.x - v3.x SYSMMU supports 32bit physical and 32bit virtual address spaces
56  * v5.0 introduced support for 36bit physical address space by shifting
57  * all page entry values by 4 bits.
58  * All SYSMMU controllers in the system support the address spaces of the same
59  * size, so PG_ENT_SHIFT can be initialized on first SYSMMU probe to proper
60  * value (0 or 4).
61  */
62 static short PG_ENT_SHIFT = -1;
63 #define SYSMMU_PG_ENT_SHIFT 0
64 #define SYSMMU_V5_PG_ENT_SHIFT 4
65 
66 static const sysmmu_pte_t *LV1_PROT;
67 static const sysmmu_pte_t SYSMMU_LV1_PROT[] = {
68 	((0 << 15) | (0 << 10)), /* no access */
69 	((1 << 15) | (1 << 10)), /* IOMMU_READ only */
70 	((0 << 15) | (1 << 10)), /* IOMMU_WRITE not supported, use read/write */
71 	((0 << 15) | (1 << 10)), /* IOMMU_READ | IOMMU_WRITE */
72 };
73 static const sysmmu_pte_t SYSMMU_V5_LV1_PROT[] = {
74 	(0 << 4), /* no access */
75 	(1 << 4), /* IOMMU_READ only */
76 	(2 << 4), /* IOMMU_WRITE only */
77 	(3 << 4), /* IOMMU_READ | IOMMU_WRITE */
78 };
79 
80 static const sysmmu_pte_t *LV2_PROT;
81 static const sysmmu_pte_t SYSMMU_LV2_PROT[] = {
82 	((0 << 9) | (0 << 4)), /* no access */
83 	((1 << 9) | (1 << 4)), /* IOMMU_READ only */
84 	((0 << 9) | (1 << 4)), /* IOMMU_WRITE not supported, use read/write */
85 	((0 << 9) | (1 << 4)), /* IOMMU_READ | IOMMU_WRITE */
86 };
87 static const sysmmu_pte_t SYSMMU_V5_LV2_PROT[] = {
88 	(0 << 2), /* no access */
89 	(1 << 2), /* IOMMU_READ only */
90 	(2 << 2), /* IOMMU_WRITE only */
91 	(3 << 2), /* IOMMU_READ | IOMMU_WRITE */
92 };
93 
94 #define SYSMMU_SUPPORTED_PROT_BITS (IOMMU_READ | IOMMU_WRITE)
95 
96 #define sect_to_phys(ent) (((phys_addr_t) ent) << PG_ENT_SHIFT)
97 #define section_phys(sent) (sect_to_phys(*(sent)) & SECT_MASK)
98 #define section_offs(iova) (iova & (SECT_SIZE - 1))
99 #define lpage_phys(pent) (sect_to_phys(*(pent)) & LPAGE_MASK)
100 #define lpage_offs(iova) (iova & (LPAGE_SIZE - 1))
101 #define spage_phys(pent) (sect_to_phys(*(pent)) & SPAGE_MASK)
102 #define spage_offs(iova) (iova & (SPAGE_SIZE - 1))
103 
104 #define NUM_LV1ENTRIES 4096
105 #define NUM_LV2ENTRIES (SECT_SIZE / SPAGE_SIZE)
106 
107 static u32 lv1ent_offset(sysmmu_iova_t iova)
108 {
109 	return iova >> SECT_ORDER;
110 }
111 
112 static u32 lv2ent_offset(sysmmu_iova_t iova)
113 {
114 	return (iova >> SPAGE_ORDER) & (NUM_LV2ENTRIES - 1);
115 }
116 
117 #define LV1TABLE_SIZE (NUM_LV1ENTRIES * sizeof(sysmmu_pte_t))
118 #define LV2TABLE_SIZE (NUM_LV2ENTRIES * sizeof(sysmmu_pte_t))
119 
120 #define SPAGES_PER_LPAGE (LPAGE_SIZE / SPAGE_SIZE)
121 #define lv2table_base(sent) (sect_to_phys(*(sent) & 0xFFFFFFC0))
122 
123 #define mk_lv1ent_sect(pa, prot) ((pa >> PG_ENT_SHIFT) | LV1_PROT[prot] | 2)
124 #define mk_lv1ent_page(pa) ((pa >> PG_ENT_SHIFT) | 1)
125 #define mk_lv2ent_lpage(pa, prot) ((pa >> PG_ENT_SHIFT) | LV2_PROT[prot] | 1)
126 #define mk_lv2ent_spage(pa, prot) ((pa >> PG_ENT_SHIFT) | LV2_PROT[prot] | 2)
127 
128 #define CTRL_ENABLE	0x5
129 #define CTRL_BLOCK	0x7
130 #define CTRL_DISABLE	0x0
131 
132 #define CFG_LRU		0x1
133 #define CFG_EAP		(1 << 2)
134 #define CFG_QOS(n)	((n & 0xF) << 7)
135 #define CFG_ACGEN	(1 << 24) /* System MMU 3.3 only */
136 #define CFG_SYSSEL	(1 << 22) /* System MMU 3.2 only */
137 #define CFG_FLPDCACHE	(1 << 20) /* System MMU 3.2+ only */
138 
139 /* common registers */
140 #define REG_MMU_CTRL		0x000
141 #define REG_MMU_CFG		0x004
142 #define REG_MMU_STATUS		0x008
143 #define REG_MMU_VERSION		0x034
144 
145 #define MMU_MAJ_VER(val)	((val) >> 7)
146 #define MMU_MIN_VER(val)	((val) & 0x7F)
147 #define MMU_RAW_VER(reg)	(((reg) >> 21) & ((1 << 11) - 1)) /* 11 bits */
148 
149 #define MAKE_MMU_VER(maj, min)	((((maj) & 0xF) << 7) | ((min) & 0x7F))
150 
151 /* v1.x - v3.x registers */
152 #define REG_MMU_FLUSH		0x00C
153 #define REG_MMU_FLUSH_ENTRY	0x010
154 #define REG_PT_BASE_ADDR	0x014
155 #define REG_INT_STATUS		0x018
156 #define REG_INT_CLEAR		0x01C
157 
158 #define REG_PAGE_FAULT_ADDR	0x024
159 #define REG_AW_FAULT_ADDR	0x028
160 #define REG_AR_FAULT_ADDR	0x02C
161 #define REG_DEFAULT_SLAVE_ADDR	0x030
162 
163 /* v5.x registers */
164 #define REG_V5_PT_BASE_PFN	0x00C
165 #define REG_V5_MMU_FLUSH_ALL	0x010
166 #define REG_V5_MMU_FLUSH_ENTRY	0x014
167 #define REG_V5_MMU_FLUSH_RANGE	0x018
168 #define REG_V5_MMU_FLUSH_START	0x020
169 #define REG_V5_MMU_FLUSH_END	0x024
170 #define REG_V5_INT_STATUS	0x060
171 #define REG_V5_INT_CLEAR	0x064
172 #define REG_V5_FAULT_AR_VA	0x070
173 #define REG_V5_FAULT_AW_VA	0x080
174 
175 #define has_sysmmu(dev)		(dev_iommu_priv_get(dev) != NULL)
176 
177 static struct device *dma_dev;
178 static struct kmem_cache *lv2table_kmem_cache;
179 static sysmmu_pte_t *zero_lv2_table;
180 #define ZERO_LV2LINK mk_lv1ent_page(virt_to_phys(zero_lv2_table))
181 
182 static sysmmu_pte_t *section_entry(sysmmu_pte_t *pgtable, sysmmu_iova_t iova)
183 {
184 	return pgtable + lv1ent_offset(iova);
185 }
186 
187 static sysmmu_pte_t *page_entry(sysmmu_pte_t *sent, sysmmu_iova_t iova)
188 {
189 	return (sysmmu_pte_t *)phys_to_virt(
190 				lv2table_base(sent)) + lv2ent_offset(iova);
191 }
192 
193 /*
194  * IOMMU fault information register
195  */
196 struct sysmmu_fault_info {
197 	unsigned int bit;	/* bit number in STATUS register */
198 	unsigned short addr_reg; /* register to read VA 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_fault_info sysmmu_faults[] = {
204 	{ 0, REG_PAGE_FAULT_ADDR, "PAGE", IOMMU_FAULT_READ },
205 	{ 1, REG_AR_FAULT_ADDR, "AR MULTI-HIT", IOMMU_FAULT_READ },
206 	{ 2, REG_AW_FAULT_ADDR, "AW MULTI-HIT", IOMMU_FAULT_WRITE },
207 	{ 3, REG_DEFAULT_SLAVE_ADDR, "BUS ERROR", IOMMU_FAULT_READ },
208 	{ 4, REG_AR_FAULT_ADDR, "AR SECURITY PROTECTION", IOMMU_FAULT_READ },
209 	{ 5, REG_AR_FAULT_ADDR, "AR ACCESS PROTECTION", IOMMU_FAULT_READ },
210 	{ 6, REG_AW_FAULT_ADDR, "AW SECURITY PROTECTION", IOMMU_FAULT_WRITE },
211 	{ 7, REG_AW_FAULT_ADDR, "AW ACCESS PROTECTION", IOMMU_FAULT_WRITE },
212 };
213 
214 static const struct sysmmu_fault_info sysmmu_v5_faults[] = {
215 	{ 0, REG_V5_FAULT_AR_VA, "AR PTW", IOMMU_FAULT_READ },
216 	{ 1, REG_V5_FAULT_AR_VA, "AR PAGE", IOMMU_FAULT_READ },
217 	{ 2, REG_V5_FAULT_AR_VA, "AR MULTI-HIT", IOMMU_FAULT_READ },
218 	{ 3, REG_V5_FAULT_AR_VA, "AR ACCESS PROTECTION", IOMMU_FAULT_READ },
219 	{ 4, REG_V5_FAULT_AR_VA, "AR SECURITY PROTECTION", IOMMU_FAULT_READ },
220 	{ 16, REG_V5_FAULT_AW_VA, "AW PTW", IOMMU_FAULT_WRITE },
221 	{ 17, REG_V5_FAULT_AW_VA, "AW PAGE", IOMMU_FAULT_WRITE },
222 	{ 18, REG_V5_FAULT_AW_VA, "AW MULTI-HIT", IOMMU_FAULT_WRITE },
223 	{ 19, REG_V5_FAULT_AW_VA, "AW ACCESS PROTECTION", IOMMU_FAULT_WRITE },
224 	{ 20, REG_V5_FAULT_AW_VA, "AW SECURITY PROTECTION", IOMMU_FAULT_WRITE },
225 };
226 
227 /*
228  * This structure is attached to dev->iommu->priv of the master device
229  * on device add, contains a list of SYSMMU controllers defined by device tree,
230  * which are bound to given master device. It is usually referenced by 'owner'
231  * pointer.
232 */
233 struct exynos_iommu_owner {
234 	struct list_head controllers;	/* list of sysmmu_drvdata.owner_node */
235 	struct iommu_domain *domain;	/* domain this device is attached */
236 	struct mutex rpm_lock;		/* for runtime pm of all sysmmus */
237 };
238 
239 /*
240  * This structure exynos specific generalization of struct iommu_domain.
241  * It contains list of SYSMMU controllers from all master devices, which has
242  * been attached to this domain and page tables of IO address space defined by
243  * it. It is usually referenced by 'domain' pointer.
244  */
245 struct exynos_iommu_domain {
246 	struct list_head clients; /* list of sysmmu_drvdata.domain_node */
247 	sysmmu_pte_t *pgtable;	/* lv1 page table, 16KB */
248 	short *lv2entcnt;	/* free lv2 entry counter for each section */
249 	spinlock_t lock;	/* lock for modyfying list of clients */
250 	spinlock_t pgtablelock;	/* lock for modifying page table @ pgtable */
251 	struct iommu_domain domain; /* generic domain data structure */
252 };
253 
254 /*
255  * This structure hold all data of a single SYSMMU controller, this includes
256  * hw resources like registers and clocks, pointers and list nodes to connect
257  * it to all other structures, internal state and parameters read from device
258  * tree. It is usually referenced by 'data' pointer.
259  */
260 struct sysmmu_drvdata {
261 	struct device *sysmmu;		/* SYSMMU controller device */
262 	struct device *master;		/* master device (owner) */
263 	struct device_link *link;	/* runtime PM link to master */
264 	void __iomem *sfrbase;		/* our registers */
265 	struct clk *clk;		/* SYSMMU's clock */
266 	struct clk *aclk;		/* SYSMMU's aclk clock */
267 	struct clk *pclk;		/* SYSMMU's pclk clock */
268 	struct clk *clk_master;		/* master's device clock */
269 	spinlock_t lock;		/* lock for modyfying state */
270 	bool active;			/* current status */
271 	struct exynos_iommu_domain *domain; /* domain we belong to */
272 	struct list_head domain_node;	/* node for domain clients list */
273 	struct list_head owner_node;	/* node for owner controllers list */
274 	phys_addr_t pgtable;		/* assigned page table structure */
275 	unsigned int version;		/* our version */
276 
277 	struct iommu_device iommu;	/* IOMMU core handle */
278 };
279 
280 static struct exynos_iommu_domain *to_exynos_domain(struct iommu_domain *dom)
281 {
282 	return container_of(dom, struct exynos_iommu_domain, domain);
283 }
284 
285 static void sysmmu_unblock(struct sysmmu_drvdata *data)
286 {
287 	writel(CTRL_ENABLE, data->sfrbase + REG_MMU_CTRL);
288 }
289 
290 static bool sysmmu_block(struct sysmmu_drvdata *data)
291 {
292 	int i = 120;
293 
294 	writel(CTRL_BLOCK, data->sfrbase + REG_MMU_CTRL);
295 	while ((i > 0) && !(readl(data->sfrbase + REG_MMU_STATUS) & 1))
296 		--i;
297 
298 	if (!(readl(data->sfrbase + REG_MMU_STATUS) & 1)) {
299 		sysmmu_unblock(data);
300 		return false;
301 	}
302 
303 	return true;
304 }
305 
306 static void __sysmmu_tlb_invalidate(struct sysmmu_drvdata *data)
307 {
308 	if (MMU_MAJ_VER(data->version) < 5)
309 		writel(0x1, data->sfrbase + REG_MMU_FLUSH);
310 	else
311 		writel(0x1, data->sfrbase + REG_V5_MMU_FLUSH_ALL);
312 }
313 
314 static void __sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
315 				sysmmu_iova_t iova, unsigned int num_inv)
316 {
317 	unsigned int i;
318 
319 	if (MMU_MAJ_VER(data->version) < 5) {
320 		for (i = 0; i < num_inv; i++) {
321 			writel((iova & SPAGE_MASK) | 1,
322 				     data->sfrbase + REG_MMU_FLUSH_ENTRY);
323 			iova += SPAGE_SIZE;
324 		}
325 	} else {
326 		if (num_inv == 1) {
327 			writel((iova & SPAGE_MASK) | 1,
328 				     data->sfrbase + REG_V5_MMU_FLUSH_ENTRY);
329 		} else {
330 			writel((iova & SPAGE_MASK),
331 				     data->sfrbase + REG_V5_MMU_FLUSH_START);
332 			writel((iova & SPAGE_MASK) + (num_inv - 1) * SPAGE_SIZE,
333 				     data->sfrbase + REG_V5_MMU_FLUSH_END);
334 			writel(1, data->sfrbase + REG_V5_MMU_FLUSH_RANGE);
335 		}
336 	}
337 }
338 
339 static void __sysmmu_set_ptbase(struct sysmmu_drvdata *data, phys_addr_t pgd)
340 {
341 	if (MMU_MAJ_VER(data->version) < 5)
342 		writel(pgd, data->sfrbase + REG_PT_BASE_ADDR);
343 	else
344 		writel(pgd >> PAGE_SHIFT,
345 			     data->sfrbase + REG_V5_PT_BASE_PFN);
346 
347 	__sysmmu_tlb_invalidate(data);
348 }
349 
350 static void __sysmmu_enable_clocks(struct sysmmu_drvdata *data)
351 {
352 	BUG_ON(clk_prepare_enable(data->clk_master));
353 	BUG_ON(clk_prepare_enable(data->clk));
354 	BUG_ON(clk_prepare_enable(data->pclk));
355 	BUG_ON(clk_prepare_enable(data->aclk));
356 }
357 
358 static void __sysmmu_disable_clocks(struct sysmmu_drvdata *data)
359 {
360 	clk_disable_unprepare(data->aclk);
361 	clk_disable_unprepare(data->pclk);
362 	clk_disable_unprepare(data->clk);
363 	clk_disable_unprepare(data->clk_master);
364 }
365 
366 static void __sysmmu_get_version(struct sysmmu_drvdata *data)
367 {
368 	u32 ver;
369 
370 	__sysmmu_enable_clocks(data);
371 
372 	ver = readl(data->sfrbase + REG_MMU_VERSION);
373 
374 	/* controllers on some SoCs don't report proper version */
375 	if (ver == 0x80000001u)
376 		data->version = MAKE_MMU_VER(1, 0);
377 	else
378 		data->version = MMU_RAW_VER(ver);
379 
380 	dev_dbg(data->sysmmu, "hardware version: %d.%d\n",
381 		MMU_MAJ_VER(data->version), MMU_MIN_VER(data->version));
382 
383 	__sysmmu_disable_clocks(data);
384 }
385 
386 static void show_fault_information(struct sysmmu_drvdata *data,
387 				   const struct sysmmu_fault_info *finfo,
388 				   sysmmu_iova_t fault_addr)
389 {
390 	sysmmu_pte_t *ent;
391 
392 	dev_err(data->sysmmu, "%s: %s FAULT occurred at %#x\n",
393 		dev_name(data->master), finfo->name, fault_addr);
394 	dev_dbg(data->sysmmu, "Page table base: %pa\n", &data->pgtable);
395 	ent = section_entry(phys_to_virt(data->pgtable), fault_addr);
396 	dev_dbg(data->sysmmu, "\tLv1 entry: %#x\n", *ent);
397 	if (lv1ent_page(ent)) {
398 		ent = page_entry(ent, fault_addr);
399 		dev_dbg(data->sysmmu, "\t Lv2 entry: %#x\n", *ent);
400 	}
401 }
402 
403 static irqreturn_t exynos_sysmmu_irq(int irq, void *dev_id)
404 {
405 	/* SYSMMU is in blocked state when interrupt occurred. */
406 	struct sysmmu_drvdata *data = dev_id;
407 	const struct sysmmu_fault_info *finfo;
408 	unsigned int i, n, itype;
409 	sysmmu_iova_t fault_addr;
410 	unsigned short reg_status, reg_clear;
411 	int ret = -ENOSYS;
412 
413 	WARN_ON(!data->active);
414 
415 	if (MMU_MAJ_VER(data->version) < 5) {
416 		reg_status = REG_INT_STATUS;
417 		reg_clear = REG_INT_CLEAR;
418 		finfo = sysmmu_faults;
419 		n = ARRAY_SIZE(sysmmu_faults);
420 	} else {
421 		reg_status = REG_V5_INT_STATUS;
422 		reg_clear = REG_V5_INT_CLEAR;
423 		finfo = sysmmu_v5_faults;
424 		n = ARRAY_SIZE(sysmmu_v5_faults);
425 	}
426 
427 	spin_lock(&data->lock);
428 
429 	clk_enable(data->clk_master);
430 
431 	itype = __ffs(readl(data->sfrbase + reg_status));
432 	for (i = 0; i < n; i++, finfo++)
433 		if (finfo->bit == itype)
434 			break;
435 	/* unknown/unsupported fault */
436 	BUG_ON(i == n);
437 
438 	/* print debug message */
439 	fault_addr = readl(data->sfrbase + finfo->addr_reg);
440 	show_fault_information(data, finfo, fault_addr);
441 
442 	if (data->domain)
443 		ret = report_iommu_fault(&data->domain->domain,
444 					data->master, fault_addr, finfo->type);
445 	/* fault is not recovered by fault handler */
446 	BUG_ON(ret != 0);
447 
448 	writel(1 << itype, data->sfrbase + reg_clear);
449 
450 	sysmmu_unblock(data);
451 
452 	clk_disable(data->clk_master);
453 
454 	spin_unlock(&data->lock);
455 
456 	return IRQ_HANDLED;
457 }
458 
459 static void __sysmmu_disable(struct sysmmu_drvdata *data)
460 {
461 	unsigned long flags;
462 
463 	clk_enable(data->clk_master);
464 
465 	spin_lock_irqsave(&data->lock, flags);
466 	writel(CTRL_DISABLE, data->sfrbase + REG_MMU_CTRL);
467 	writel(0, data->sfrbase + REG_MMU_CFG);
468 	data->active = false;
469 	spin_unlock_irqrestore(&data->lock, flags);
470 
471 	__sysmmu_disable_clocks(data);
472 }
473 
474 static void __sysmmu_init_config(struct sysmmu_drvdata *data)
475 {
476 	unsigned int cfg;
477 
478 	if (data->version <= MAKE_MMU_VER(3, 1))
479 		cfg = CFG_LRU | CFG_QOS(15);
480 	else if (data->version <= MAKE_MMU_VER(3, 2))
481 		cfg = CFG_LRU | CFG_QOS(15) | CFG_FLPDCACHE | CFG_SYSSEL;
482 	else
483 		cfg = CFG_QOS(15) | CFG_FLPDCACHE | CFG_ACGEN;
484 
485 	cfg |= CFG_EAP; /* enable access protection bits check */
486 
487 	writel(cfg, data->sfrbase + REG_MMU_CFG);
488 }
489 
490 static void __sysmmu_enable(struct sysmmu_drvdata *data)
491 {
492 	unsigned long flags;
493 
494 	__sysmmu_enable_clocks(data);
495 
496 	spin_lock_irqsave(&data->lock, flags);
497 	writel(CTRL_BLOCK, data->sfrbase + REG_MMU_CTRL);
498 	__sysmmu_init_config(data);
499 	__sysmmu_set_ptbase(data, data->pgtable);
500 	writel(CTRL_ENABLE, data->sfrbase + REG_MMU_CTRL);
501 	data->active = true;
502 	spin_unlock_irqrestore(&data->lock, flags);
503 
504 	/*
505 	 * SYSMMU driver keeps master's clock enabled only for the short
506 	 * time, while accessing the registers. For performing address
507 	 * translation during DMA transaction it relies on the client
508 	 * driver to enable it.
509 	 */
510 	clk_disable(data->clk_master);
511 }
512 
513 static void sysmmu_tlb_invalidate_flpdcache(struct sysmmu_drvdata *data,
514 					    sysmmu_iova_t iova)
515 {
516 	unsigned long flags;
517 
518 	spin_lock_irqsave(&data->lock, flags);
519 	if (data->active && data->version >= MAKE_MMU_VER(3, 3)) {
520 		clk_enable(data->clk_master);
521 		if (sysmmu_block(data)) {
522 			if (data->version >= MAKE_MMU_VER(5, 0))
523 				__sysmmu_tlb_invalidate(data);
524 			else
525 				__sysmmu_tlb_invalidate_entry(data, iova, 1);
526 			sysmmu_unblock(data);
527 		}
528 		clk_disable(data->clk_master);
529 	}
530 	spin_unlock_irqrestore(&data->lock, flags);
531 }
532 
533 static void sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
534 					sysmmu_iova_t iova, size_t size)
535 {
536 	unsigned long flags;
537 
538 	spin_lock_irqsave(&data->lock, flags);
539 	if (data->active) {
540 		unsigned int num_inv = 1;
541 
542 		clk_enable(data->clk_master);
543 
544 		/*
545 		 * L2TLB invalidation required
546 		 * 4KB page: 1 invalidation
547 		 * 64KB page: 16 invalidations
548 		 * 1MB page: 64 invalidations
549 		 * because it is set-associative TLB
550 		 * with 8-way and 64 sets.
551 		 * 1MB page can be cached in one of all sets.
552 		 * 64KB page can be one of 16 consecutive sets.
553 		 */
554 		if (MMU_MAJ_VER(data->version) == 2)
555 			num_inv = min_t(unsigned int, size / PAGE_SIZE, 64);
556 
557 		if (sysmmu_block(data)) {
558 			__sysmmu_tlb_invalidate_entry(data, iova, num_inv);
559 			sysmmu_unblock(data);
560 		}
561 		clk_disable(data->clk_master);
562 	}
563 	spin_unlock_irqrestore(&data->lock, flags);
564 }
565 
566 static const struct iommu_ops exynos_iommu_ops;
567 
568 static int exynos_sysmmu_probe(struct platform_device *pdev)
569 {
570 	int irq, ret;
571 	struct device *dev = &pdev->dev;
572 	struct sysmmu_drvdata *data;
573 	struct resource *res;
574 
575 	data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL);
576 	if (!data)
577 		return -ENOMEM;
578 
579 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
580 	data->sfrbase = devm_ioremap_resource(dev, res);
581 	if (IS_ERR(data->sfrbase))
582 		return PTR_ERR(data->sfrbase);
583 
584 	irq = platform_get_irq(pdev, 0);
585 	if (irq <= 0)
586 		return irq;
587 
588 	ret = devm_request_irq(dev, irq, exynos_sysmmu_irq, 0,
589 				dev_name(dev), data);
590 	if (ret) {
591 		dev_err(dev, "Unabled to register handler of irq %d\n", irq);
592 		return ret;
593 	}
594 
595 	data->clk = devm_clk_get(dev, "sysmmu");
596 	if (PTR_ERR(data->clk) == -ENOENT)
597 		data->clk = NULL;
598 	else if (IS_ERR(data->clk))
599 		return PTR_ERR(data->clk);
600 
601 	data->aclk = devm_clk_get(dev, "aclk");
602 	if (PTR_ERR(data->aclk) == -ENOENT)
603 		data->aclk = NULL;
604 	else if (IS_ERR(data->aclk))
605 		return PTR_ERR(data->aclk);
606 
607 	data->pclk = devm_clk_get(dev, "pclk");
608 	if (PTR_ERR(data->pclk) == -ENOENT)
609 		data->pclk = NULL;
610 	else if (IS_ERR(data->pclk))
611 		return PTR_ERR(data->pclk);
612 
613 	if (!data->clk && (!data->aclk || !data->pclk)) {
614 		dev_err(dev, "Failed to get device clock(s)!\n");
615 		return -ENOSYS;
616 	}
617 
618 	data->clk_master = devm_clk_get(dev, "master");
619 	if (PTR_ERR(data->clk_master) == -ENOENT)
620 		data->clk_master = NULL;
621 	else if (IS_ERR(data->clk_master))
622 		return PTR_ERR(data->clk_master);
623 
624 	data->sysmmu = dev;
625 	spin_lock_init(&data->lock);
626 
627 	ret = iommu_device_sysfs_add(&data->iommu, &pdev->dev, NULL,
628 				     dev_name(data->sysmmu));
629 	if (ret)
630 		return ret;
631 
632 	ret = iommu_device_register(&data->iommu, &exynos_iommu_ops, dev);
633 	if (ret)
634 		return ret;
635 
636 	platform_set_drvdata(pdev, data);
637 
638 	__sysmmu_get_version(data);
639 	if (PG_ENT_SHIFT < 0) {
640 		if (MMU_MAJ_VER(data->version) < 5) {
641 			PG_ENT_SHIFT = SYSMMU_PG_ENT_SHIFT;
642 			LV1_PROT = SYSMMU_LV1_PROT;
643 			LV2_PROT = SYSMMU_LV2_PROT;
644 		} else {
645 			PG_ENT_SHIFT = SYSMMU_V5_PG_ENT_SHIFT;
646 			LV1_PROT = SYSMMU_V5_LV1_PROT;
647 			LV2_PROT = SYSMMU_V5_LV2_PROT;
648 		}
649 	}
650 
651 	/*
652 	 * use the first registered sysmmu device for performing
653 	 * dma mapping operations on iommu page tables (cpu cache flush)
654 	 */
655 	if (!dma_dev)
656 		dma_dev = &pdev->dev;
657 
658 	pm_runtime_enable(dev);
659 
660 	return 0;
661 }
662 
663 static int __maybe_unused exynos_sysmmu_suspend(struct device *dev)
664 {
665 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
666 	struct device *master = data->master;
667 
668 	if (master) {
669 		struct exynos_iommu_owner *owner = dev_iommu_priv_get(master);
670 
671 		mutex_lock(&owner->rpm_lock);
672 		if (data->domain) {
673 			dev_dbg(data->sysmmu, "saving state\n");
674 			__sysmmu_disable(data);
675 		}
676 		mutex_unlock(&owner->rpm_lock);
677 	}
678 	return 0;
679 }
680 
681 static int __maybe_unused exynos_sysmmu_resume(struct device *dev)
682 {
683 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
684 	struct device *master = data->master;
685 
686 	if (master) {
687 		struct exynos_iommu_owner *owner = dev_iommu_priv_get(master);
688 
689 		mutex_lock(&owner->rpm_lock);
690 		if (data->domain) {
691 			dev_dbg(data->sysmmu, "restoring state\n");
692 			__sysmmu_enable(data);
693 		}
694 		mutex_unlock(&owner->rpm_lock);
695 	}
696 	return 0;
697 }
698 
699 static const struct dev_pm_ops sysmmu_pm_ops = {
700 	SET_RUNTIME_PM_OPS(exynos_sysmmu_suspend, exynos_sysmmu_resume, NULL)
701 	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
702 				pm_runtime_force_resume)
703 };
704 
705 static const struct of_device_id sysmmu_of_match[] = {
706 	{ .compatible	= "samsung,exynos-sysmmu", },
707 	{ },
708 };
709 
710 static struct platform_driver exynos_sysmmu_driver __refdata = {
711 	.probe	= exynos_sysmmu_probe,
712 	.driver	= {
713 		.name		= "exynos-sysmmu",
714 		.of_match_table	= sysmmu_of_match,
715 		.pm		= &sysmmu_pm_ops,
716 		.suppress_bind_attrs = true,
717 	}
718 };
719 
720 static inline void exynos_iommu_set_pte(sysmmu_pte_t *ent, sysmmu_pte_t val)
721 {
722 	dma_sync_single_for_cpu(dma_dev, virt_to_phys(ent), sizeof(*ent),
723 				DMA_TO_DEVICE);
724 	*ent = cpu_to_le32(val);
725 	dma_sync_single_for_device(dma_dev, virt_to_phys(ent), sizeof(*ent),
726 				   DMA_TO_DEVICE);
727 }
728 
729 static struct iommu_domain *exynos_iommu_domain_alloc(unsigned type)
730 {
731 	struct exynos_iommu_domain *domain;
732 	dma_addr_t handle;
733 	int i;
734 
735 	/* Check if correct PTE offsets are initialized */
736 	BUG_ON(PG_ENT_SHIFT < 0 || !dma_dev);
737 
738 	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
739 	if (!domain)
740 		return NULL;
741 
742 	if (type == IOMMU_DOMAIN_DMA) {
743 		if (iommu_get_dma_cookie(&domain->domain) != 0)
744 			goto err_pgtable;
745 	} else if (type != IOMMU_DOMAIN_UNMANAGED) {
746 		goto err_pgtable;
747 	}
748 
749 	domain->pgtable = (sysmmu_pte_t *)__get_free_pages(GFP_KERNEL, 2);
750 	if (!domain->pgtable)
751 		goto err_dma_cookie;
752 
753 	domain->lv2entcnt = (short *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
754 	if (!domain->lv2entcnt)
755 		goto err_counter;
756 
757 	/* Workaround for System MMU v3.3 to prevent caching 1MiB mapping */
758 	for (i = 0; i < NUM_LV1ENTRIES; i++)
759 		domain->pgtable[i] = ZERO_LV2LINK;
760 
761 	handle = dma_map_single(dma_dev, domain->pgtable, LV1TABLE_SIZE,
762 				DMA_TO_DEVICE);
763 	/* For mapping page table entries we rely on dma == phys */
764 	BUG_ON(handle != virt_to_phys(domain->pgtable));
765 	if (dma_mapping_error(dma_dev, handle))
766 		goto err_lv2ent;
767 
768 	spin_lock_init(&domain->lock);
769 	spin_lock_init(&domain->pgtablelock);
770 	INIT_LIST_HEAD(&domain->clients);
771 
772 	domain->domain.geometry.aperture_start = 0;
773 	domain->domain.geometry.aperture_end   = ~0UL;
774 	domain->domain.geometry.force_aperture = true;
775 
776 	return &domain->domain;
777 
778 err_lv2ent:
779 	free_pages((unsigned long)domain->lv2entcnt, 1);
780 err_counter:
781 	free_pages((unsigned long)domain->pgtable, 2);
782 err_dma_cookie:
783 	if (type == IOMMU_DOMAIN_DMA)
784 		iommu_put_dma_cookie(&domain->domain);
785 err_pgtable:
786 	kfree(domain);
787 	return NULL;
788 }
789 
790 static void exynos_iommu_domain_free(struct iommu_domain *iommu_domain)
791 {
792 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
793 	struct sysmmu_drvdata *data, *next;
794 	unsigned long flags;
795 	int i;
796 
797 	WARN_ON(!list_empty(&domain->clients));
798 
799 	spin_lock_irqsave(&domain->lock, flags);
800 
801 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
802 		spin_lock(&data->lock);
803 		__sysmmu_disable(data);
804 		data->pgtable = 0;
805 		data->domain = NULL;
806 		list_del_init(&data->domain_node);
807 		spin_unlock(&data->lock);
808 	}
809 
810 	spin_unlock_irqrestore(&domain->lock, flags);
811 
812 	if (iommu_domain->type == IOMMU_DOMAIN_DMA)
813 		iommu_put_dma_cookie(iommu_domain);
814 
815 	dma_unmap_single(dma_dev, virt_to_phys(domain->pgtable), LV1TABLE_SIZE,
816 			 DMA_TO_DEVICE);
817 
818 	for (i = 0; i < NUM_LV1ENTRIES; i++)
819 		if (lv1ent_page(domain->pgtable + i)) {
820 			phys_addr_t base = lv2table_base(domain->pgtable + i);
821 
822 			dma_unmap_single(dma_dev, base, LV2TABLE_SIZE,
823 					 DMA_TO_DEVICE);
824 			kmem_cache_free(lv2table_kmem_cache,
825 					phys_to_virt(base));
826 		}
827 
828 	free_pages((unsigned long)domain->pgtable, 2);
829 	free_pages((unsigned long)domain->lv2entcnt, 1);
830 	kfree(domain);
831 }
832 
833 static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
834 				    struct device *dev)
835 {
836 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
837 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
838 	phys_addr_t pagetable = virt_to_phys(domain->pgtable);
839 	struct sysmmu_drvdata *data, *next;
840 	unsigned long flags;
841 
842 	if (!has_sysmmu(dev) || owner->domain != iommu_domain)
843 		return;
844 
845 	mutex_lock(&owner->rpm_lock);
846 
847 	list_for_each_entry(data, &owner->controllers, owner_node) {
848 		pm_runtime_get_noresume(data->sysmmu);
849 		if (pm_runtime_active(data->sysmmu))
850 			__sysmmu_disable(data);
851 		pm_runtime_put(data->sysmmu);
852 	}
853 
854 	spin_lock_irqsave(&domain->lock, flags);
855 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
856 		spin_lock(&data->lock);
857 		data->pgtable = 0;
858 		data->domain = NULL;
859 		list_del_init(&data->domain_node);
860 		spin_unlock(&data->lock);
861 	}
862 	owner->domain = NULL;
863 	spin_unlock_irqrestore(&domain->lock, flags);
864 
865 	mutex_unlock(&owner->rpm_lock);
866 
867 	dev_dbg(dev, "%s: Detached IOMMU with pgtable %pa\n", __func__,
868 		&pagetable);
869 }
870 
871 static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
872 				   struct device *dev)
873 {
874 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
875 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
876 	struct sysmmu_drvdata *data;
877 	phys_addr_t pagetable = virt_to_phys(domain->pgtable);
878 	unsigned long flags;
879 
880 	if (!has_sysmmu(dev))
881 		return -ENODEV;
882 
883 	if (owner->domain)
884 		exynos_iommu_detach_device(owner->domain, dev);
885 
886 	mutex_lock(&owner->rpm_lock);
887 
888 	spin_lock_irqsave(&domain->lock, flags);
889 	list_for_each_entry(data, &owner->controllers, owner_node) {
890 		spin_lock(&data->lock);
891 		data->pgtable = pagetable;
892 		data->domain = domain;
893 		list_add_tail(&data->domain_node, &domain->clients);
894 		spin_unlock(&data->lock);
895 	}
896 	owner->domain = iommu_domain;
897 	spin_unlock_irqrestore(&domain->lock, flags);
898 
899 	list_for_each_entry(data, &owner->controllers, owner_node) {
900 		pm_runtime_get_noresume(data->sysmmu);
901 		if (pm_runtime_active(data->sysmmu))
902 			__sysmmu_enable(data);
903 		pm_runtime_put(data->sysmmu);
904 	}
905 
906 	mutex_unlock(&owner->rpm_lock);
907 
908 	dev_dbg(dev, "%s: Attached IOMMU with pgtable %pa\n", __func__,
909 		&pagetable);
910 
911 	return 0;
912 }
913 
914 static sysmmu_pte_t *alloc_lv2entry(struct exynos_iommu_domain *domain,
915 		sysmmu_pte_t *sent, sysmmu_iova_t iova, short *pgcounter)
916 {
917 	if (lv1ent_section(sent)) {
918 		WARN(1, "Trying mapping on %#08x mapped with 1MiB page", iova);
919 		return ERR_PTR(-EADDRINUSE);
920 	}
921 
922 	if (lv1ent_fault(sent)) {
923 		dma_addr_t handle;
924 		sysmmu_pte_t *pent;
925 		bool need_flush_flpd_cache = lv1ent_zero(sent);
926 
927 		pent = kmem_cache_zalloc(lv2table_kmem_cache, GFP_ATOMIC);
928 		BUG_ON((uintptr_t)pent & (LV2TABLE_SIZE - 1));
929 		if (!pent)
930 			return ERR_PTR(-ENOMEM);
931 
932 		exynos_iommu_set_pte(sent, mk_lv1ent_page(virt_to_phys(pent)));
933 		kmemleak_ignore(pent);
934 		*pgcounter = NUM_LV2ENTRIES;
935 		handle = dma_map_single(dma_dev, pent, LV2TABLE_SIZE,
936 					DMA_TO_DEVICE);
937 		if (dma_mapping_error(dma_dev, handle)) {
938 			kmem_cache_free(lv2table_kmem_cache, pent);
939 			return ERR_PTR(-EADDRINUSE);
940 		}
941 
942 		/*
943 		 * If pre-fetched SLPD is a faulty SLPD in zero_l2_table,
944 		 * FLPD cache may cache the address of zero_l2_table. This
945 		 * function replaces the zero_l2_table with new L2 page table
946 		 * to write valid mappings.
947 		 * Accessing the valid area may cause page fault since FLPD
948 		 * cache may still cache zero_l2_table for the valid area
949 		 * instead of new L2 page table that has the mapping
950 		 * information of the valid area.
951 		 * Thus any replacement of zero_l2_table with other valid L2
952 		 * page table must involve FLPD cache invalidation for System
953 		 * MMU v3.3.
954 		 * FLPD cache invalidation is performed with TLB invalidation
955 		 * by VPN without blocking. It is safe to invalidate TLB without
956 		 * blocking because the target address of TLB invalidation is
957 		 * not currently mapped.
958 		 */
959 		if (need_flush_flpd_cache) {
960 			struct sysmmu_drvdata *data;
961 
962 			spin_lock(&domain->lock);
963 			list_for_each_entry(data, &domain->clients, domain_node)
964 				sysmmu_tlb_invalidate_flpdcache(data, iova);
965 			spin_unlock(&domain->lock);
966 		}
967 	}
968 
969 	return page_entry(sent, iova);
970 }
971 
972 static int lv1set_section(struct exynos_iommu_domain *domain,
973 			  sysmmu_pte_t *sent, sysmmu_iova_t iova,
974 			  phys_addr_t paddr, int prot, short *pgcnt)
975 {
976 	if (lv1ent_section(sent)) {
977 		WARN(1, "Trying mapping on 1MiB@%#08x that is mapped",
978 			iova);
979 		return -EADDRINUSE;
980 	}
981 
982 	if (lv1ent_page(sent)) {
983 		if (*pgcnt != NUM_LV2ENTRIES) {
984 			WARN(1, "Trying mapping on 1MiB@%#08x that is mapped",
985 				iova);
986 			return -EADDRINUSE;
987 		}
988 
989 		kmem_cache_free(lv2table_kmem_cache, page_entry(sent, 0));
990 		*pgcnt = 0;
991 	}
992 
993 	exynos_iommu_set_pte(sent, mk_lv1ent_sect(paddr, prot));
994 
995 	spin_lock(&domain->lock);
996 	if (lv1ent_page_zero(sent)) {
997 		struct sysmmu_drvdata *data;
998 		/*
999 		 * Flushing FLPD cache in System MMU v3.3 that may cache a FLPD
1000 		 * entry by speculative prefetch of SLPD which has no mapping.
1001 		 */
1002 		list_for_each_entry(data, &domain->clients, domain_node)
1003 			sysmmu_tlb_invalidate_flpdcache(data, iova);
1004 	}
1005 	spin_unlock(&domain->lock);
1006 
1007 	return 0;
1008 }
1009 
1010 static int lv2set_page(sysmmu_pte_t *pent, phys_addr_t paddr, size_t size,
1011 		       int prot, short *pgcnt)
1012 {
1013 	if (size == SPAGE_SIZE) {
1014 		if (WARN_ON(!lv2ent_fault(pent)))
1015 			return -EADDRINUSE;
1016 
1017 		exynos_iommu_set_pte(pent, mk_lv2ent_spage(paddr, prot));
1018 		*pgcnt -= 1;
1019 	} else { /* size == LPAGE_SIZE */
1020 		int i;
1021 		dma_addr_t pent_base = virt_to_phys(pent);
1022 
1023 		dma_sync_single_for_cpu(dma_dev, pent_base,
1024 					sizeof(*pent) * SPAGES_PER_LPAGE,
1025 					DMA_TO_DEVICE);
1026 		for (i = 0; i < SPAGES_PER_LPAGE; i++, pent++) {
1027 			if (WARN_ON(!lv2ent_fault(pent))) {
1028 				if (i > 0)
1029 					memset(pent - i, 0, sizeof(*pent) * i);
1030 				return -EADDRINUSE;
1031 			}
1032 
1033 			*pent = mk_lv2ent_lpage(paddr, prot);
1034 		}
1035 		dma_sync_single_for_device(dma_dev, pent_base,
1036 					   sizeof(*pent) * SPAGES_PER_LPAGE,
1037 					   DMA_TO_DEVICE);
1038 		*pgcnt -= SPAGES_PER_LPAGE;
1039 	}
1040 
1041 	return 0;
1042 }
1043 
1044 /*
1045  * *CAUTION* to the I/O virtual memory managers that support exynos-iommu:
1046  *
1047  * System MMU v3.x has advanced logic to improve address translation
1048  * performance with caching more page table entries by a page table walk.
1049  * However, the logic has a bug that while caching faulty page table entries,
1050  * System MMU reports page fault if the cached fault entry is hit even though
1051  * the fault entry is updated to a valid entry after the entry is cached.
1052  * To prevent caching faulty page table entries which may be updated to valid
1053  * entries later, the virtual memory manager should care about the workaround
1054  * for the problem. The following describes the workaround.
1055  *
1056  * Any two consecutive I/O virtual address regions must have a hole of 128KiB
1057  * at maximum to prevent misbehavior of System MMU 3.x (workaround for h/w bug).
1058  *
1059  * Precisely, any start address of I/O virtual region must be aligned with
1060  * the following sizes for System MMU v3.1 and v3.2.
1061  * System MMU v3.1: 128KiB
1062  * System MMU v3.2: 256KiB
1063  *
1064  * Because System MMU v3.3 caches page table entries more aggressively, it needs
1065  * more workarounds.
1066  * - Any two consecutive I/O virtual regions must have a hole of size larger
1067  *   than or equal to 128KiB.
1068  * - Start address of an I/O virtual region must be aligned by 128KiB.
1069  */
1070 static int exynos_iommu_map(struct iommu_domain *iommu_domain,
1071 			    unsigned long l_iova, phys_addr_t paddr, size_t size,
1072 			    int prot, gfp_t gfp)
1073 {
1074 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
1075 	sysmmu_pte_t *entry;
1076 	sysmmu_iova_t iova = (sysmmu_iova_t)l_iova;
1077 	unsigned long flags;
1078 	int ret = -ENOMEM;
1079 
1080 	BUG_ON(domain->pgtable == NULL);
1081 	prot &= SYSMMU_SUPPORTED_PROT_BITS;
1082 
1083 	spin_lock_irqsave(&domain->pgtablelock, flags);
1084 
1085 	entry = section_entry(domain->pgtable, iova);
1086 
1087 	if (size == SECT_SIZE) {
1088 		ret = lv1set_section(domain, entry, iova, paddr, prot,
1089 				     &domain->lv2entcnt[lv1ent_offset(iova)]);
1090 	} else {
1091 		sysmmu_pte_t *pent;
1092 
1093 		pent = alloc_lv2entry(domain, entry, iova,
1094 				      &domain->lv2entcnt[lv1ent_offset(iova)]);
1095 
1096 		if (IS_ERR(pent))
1097 			ret = PTR_ERR(pent);
1098 		else
1099 			ret = lv2set_page(pent, paddr, size, prot,
1100 				       &domain->lv2entcnt[lv1ent_offset(iova)]);
1101 	}
1102 
1103 	if (ret)
1104 		pr_err("%s: Failed(%d) to map %#zx bytes @ %#x\n",
1105 			__func__, ret, size, iova);
1106 
1107 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1108 
1109 	return ret;
1110 }
1111 
1112 static void exynos_iommu_tlb_invalidate_entry(struct exynos_iommu_domain *domain,
1113 					      sysmmu_iova_t iova, size_t size)
1114 {
1115 	struct sysmmu_drvdata *data;
1116 	unsigned long flags;
1117 
1118 	spin_lock_irqsave(&domain->lock, flags);
1119 
1120 	list_for_each_entry(data, &domain->clients, domain_node)
1121 		sysmmu_tlb_invalidate_entry(data, iova, size);
1122 
1123 	spin_unlock_irqrestore(&domain->lock, flags);
1124 }
1125 
1126 static size_t exynos_iommu_unmap(struct iommu_domain *iommu_domain,
1127 				 unsigned long l_iova, size_t size,
1128 				 struct iommu_iotlb_gather *gather)
1129 {
1130 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
1131 	sysmmu_iova_t iova = (sysmmu_iova_t)l_iova;
1132 	sysmmu_pte_t *ent;
1133 	size_t err_pgsize;
1134 	unsigned long flags;
1135 
1136 	BUG_ON(domain->pgtable == NULL);
1137 
1138 	spin_lock_irqsave(&domain->pgtablelock, flags);
1139 
1140 	ent = section_entry(domain->pgtable, iova);
1141 
1142 	if (lv1ent_section(ent)) {
1143 		if (WARN_ON(size < SECT_SIZE)) {
1144 			err_pgsize = SECT_SIZE;
1145 			goto err;
1146 		}
1147 
1148 		/* workaround for h/w bug in System MMU v3.3 */
1149 		exynos_iommu_set_pte(ent, ZERO_LV2LINK);
1150 		size = SECT_SIZE;
1151 		goto done;
1152 	}
1153 
1154 	if (unlikely(lv1ent_fault(ent))) {
1155 		if (size > SECT_SIZE)
1156 			size = SECT_SIZE;
1157 		goto done;
1158 	}
1159 
1160 	/* lv1ent_page(sent) == true here */
1161 
1162 	ent = page_entry(ent, iova);
1163 
1164 	if (unlikely(lv2ent_fault(ent))) {
1165 		size = SPAGE_SIZE;
1166 		goto done;
1167 	}
1168 
1169 	if (lv2ent_small(ent)) {
1170 		exynos_iommu_set_pte(ent, 0);
1171 		size = SPAGE_SIZE;
1172 		domain->lv2entcnt[lv1ent_offset(iova)] += 1;
1173 		goto done;
1174 	}
1175 
1176 	/* lv1ent_large(ent) == true here */
1177 	if (WARN_ON(size < LPAGE_SIZE)) {
1178 		err_pgsize = LPAGE_SIZE;
1179 		goto err;
1180 	}
1181 
1182 	dma_sync_single_for_cpu(dma_dev, virt_to_phys(ent),
1183 				sizeof(*ent) * SPAGES_PER_LPAGE,
1184 				DMA_TO_DEVICE);
1185 	memset(ent, 0, sizeof(*ent) * SPAGES_PER_LPAGE);
1186 	dma_sync_single_for_device(dma_dev, virt_to_phys(ent),
1187 				   sizeof(*ent) * SPAGES_PER_LPAGE,
1188 				   DMA_TO_DEVICE);
1189 	size = LPAGE_SIZE;
1190 	domain->lv2entcnt[lv1ent_offset(iova)] += SPAGES_PER_LPAGE;
1191 done:
1192 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1193 
1194 	exynos_iommu_tlb_invalidate_entry(domain, iova, size);
1195 
1196 	return size;
1197 err:
1198 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1199 
1200 	pr_err("%s: Failed: size(%#zx) @ %#x is smaller than page size %#zx\n",
1201 		__func__, size, iova, err_pgsize);
1202 
1203 	return 0;
1204 }
1205 
1206 static phys_addr_t exynos_iommu_iova_to_phys(struct iommu_domain *iommu_domain,
1207 					  dma_addr_t iova)
1208 {
1209 	struct exynos_iommu_domain *domain = to_exynos_domain(iommu_domain);
1210 	sysmmu_pte_t *entry;
1211 	unsigned long flags;
1212 	phys_addr_t phys = 0;
1213 
1214 	spin_lock_irqsave(&domain->pgtablelock, flags);
1215 
1216 	entry = section_entry(domain->pgtable, iova);
1217 
1218 	if (lv1ent_section(entry)) {
1219 		phys = section_phys(entry) + section_offs(iova);
1220 	} else if (lv1ent_page(entry)) {
1221 		entry = page_entry(entry, iova);
1222 
1223 		if (lv2ent_large(entry))
1224 			phys = lpage_phys(entry) + lpage_offs(iova);
1225 		else if (lv2ent_small(entry))
1226 			phys = spage_phys(entry) + spage_offs(iova);
1227 	}
1228 
1229 	spin_unlock_irqrestore(&domain->pgtablelock, flags);
1230 
1231 	return phys;
1232 }
1233 
1234 static struct iommu_device *exynos_iommu_probe_device(struct device *dev)
1235 {
1236 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
1237 	struct sysmmu_drvdata *data;
1238 
1239 	if (!has_sysmmu(dev))
1240 		return ERR_PTR(-ENODEV);
1241 
1242 	list_for_each_entry(data, &owner->controllers, owner_node) {
1243 		/*
1244 		 * SYSMMU will be runtime activated via device link
1245 		 * (dependency) to its master device, so there are no
1246 		 * direct calls to pm_runtime_get/put in this driver.
1247 		 */
1248 		data->link = device_link_add(dev, data->sysmmu,
1249 					     DL_FLAG_STATELESS |
1250 					     DL_FLAG_PM_RUNTIME);
1251 	}
1252 
1253 	/* There is always at least one entry, see exynos_iommu_of_xlate() */
1254 	data = list_first_entry(&owner->controllers,
1255 				struct sysmmu_drvdata, owner_node);
1256 
1257 	return &data->iommu;
1258 }
1259 
1260 static void exynos_iommu_release_device(struct device *dev)
1261 {
1262 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
1263 	struct sysmmu_drvdata *data;
1264 
1265 	if (!has_sysmmu(dev))
1266 		return;
1267 
1268 	if (owner->domain) {
1269 		struct iommu_group *group = iommu_group_get(dev);
1270 
1271 		if (group) {
1272 			WARN_ON(owner->domain !=
1273 				iommu_group_default_domain(group));
1274 			exynos_iommu_detach_device(owner->domain, dev);
1275 			iommu_group_put(group);
1276 		}
1277 	}
1278 
1279 	list_for_each_entry(data, &owner->controllers, owner_node)
1280 		device_link_del(data->link);
1281 }
1282 
1283 static int exynos_iommu_of_xlate(struct device *dev,
1284 				 struct of_phandle_args *spec)
1285 {
1286 	struct platform_device *sysmmu = of_find_device_by_node(spec->np);
1287 	struct exynos_iommu_owner *owner = dev_iommu_priv_get(dev);
1288 	struct sysmmu_drvdata *data, *entry;
1289 
1290 	if (!sysmmu)
1291 		return -ENODEV;
1292 
1293 	data = platform_get_drvdata(sysmmu);
1294 	if (!data) {
1295 		put_device(&sysmmu->dev);
1296 		return -ENODEV;
1297 	}
1298 
1299 	if (!owner) {
1300 		owner = kzalloc(sizeof(*owner), GFP_KERNEL);
1301 		if (!owner) {
1302 			put_device(&sysmmu->dev);
1303 			return -ENOMEM;
1304 		}
1305 
1306 		INIT_LIST_HEAD(&owner->controllers);
1307 		mutex_init(&owner->rpm_lock);
1308 		dev_iommu_priv_set(dev, owner);
1309 	}
1310 
1311 	list_for_each_entry(entry, &owner->controllers, owner_node)
1312 		if (entry == data)
1313 			return 0;
1314 
1315 	list_add_tail(&data->owner_node, &owner->controllers);
1316 	data->master = dev;
1317 
1318 	return 0;
1319 }
1320 
1321 static const struct iommu_ops exynos_iommu_ops = {
1322 	.domain_alloc = exynos_iommu_domain_alloc,
1323 	.domain_free = exynos_iommu_domain_free,
1324 	.attach_dev = exynos_iommu_attach_device,
1325 	.detach_dev = exynos_iommu_detach_device,
1326 	.map = exynos_iommu_map,
1327 	.unmap = exynos_iommu_unmap,
1328 	.iova_to_phys = exynos_iommu_iova_to_phys,
1329 	.device_group = generic_device_group,
1330 	.probe_device = exynos_iommu_probe_device,
1331 	.release_device = exynos_iommu_release_device,
1332 	.pgsize_bitmap = SECT_SIZE | LPAGE_SIZE | SPAGE_SIZE,
1333 	.of_xlate = exynos_iommu_of_xlate,
1334 };
1335 
1336 static int __init exynos_iommu_init(void)
1337 {
1338 	struct device_node *np;
1339 	int ret;
1340 
1341 	np = of_find_matching_node(NULL, sysmmu_of_match);
1342 	if (!np)
1343 		return 0;
1344 
1345 	of_node_put(np);
1346 
1347 	lv2table_kmem_cache = kmem_cache_create("exynos-iommu-lv2table",
1348 				LV2TABLE_SIZE, LV2TABLE_SIZE, 0, NULL);
1349 	if (!lv2table_kmem_cache) {
1350 		pr_err("%s: Failed to create kmem cache\n", __func__);
1351 		return -ENOMEM;
1352 	}
1353 
1354 	ret = platform_driver_register(&exynos_sysmmu_driver);
1355 	if (ret) {
1356 		pr_err("%s: Failed to register driver\n", __func__);
1357 		goto err_reg_driver;
1358 	}
1359 
1360 	zero_lv2_table = kmem_cache_zalloc(lv2table_kmem_cache, GFP_KERNEL);
1361 	if (zero_lv2_table == NULL) {
1362 		pr_err("%s: Failed to allocate zero level2 page table\n",
1363 			__func__);
1364 		ret = -ENOMEM;
1365 		goto err_zero_lv2;
1366 	}
1367 
1368 	ret = bus_set_iommu(&platform_bus_type, &exynos_iommu_ops);
1369 	if (ret) {
1370 		pr_err("%s: Failed to register exynos-iommu driver.\n",
1371 								__func__);
1372 		goto err_set_iommu;
1373 	}
1374 
1375 	return 0;
1376 err_set_iommu:
1377 	kmem_cache_free(lv2table_kmem_cache, zero_lv2_table);
1378 err_zero_lv2:
1379 	platform_driver_unregister(&exynos_sysmmu_driver);
1380 err_reg_driver:
1381 	kmem_cache_destroy(lv2table_kmem_cache);
1382 	return ret;
1383 }
1384 core_initcall(exynos_iommu_init);
1385