xref: /openbmc/linux/drivers/iommu/ipmmu-vmsa.c (revision 24ce659d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * IOMMU API for Renesas VMSA-compatible IPMMU
4  * Author: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
5  *
6  * Copyright (C) 2014 Renesas Electronics Corporation
7  */
8 
9 #include <linux/bitmap.h>
10 #include <linux/delay.h>
11 #include <linux/dma-iommu.h>
12 #include <linux/dma-mapping.h>
13 #include <linux/err.h>
14 #include <linux/export.h>
15 #include <linux/init.h>
16 #include <linux/interrupt.h>
17 #include <linux/io.h>
18 #include <linux/io-pgtable.h>
19 #include <linux/iommu.h>
20 #include <linux/of.h>
21 #include <linux/of_device.h>
22 #include <linux/of_iommu.h>
23 #include <linux/of_platform.h>
24 #include <linux/platform_device.h>
25 #include <linux/sizes.h>
26 #include <linux/slab.h>
27 #include <linux/sys_soc.h>
28 
29 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
30 #include <asm/dma-iommu.h>
31 #include <asm/pgalloc.h>
32 #else
33 #define arm_iommu_create_mapping(...)	NULL
34 #define arm_iommu_attach_device(...)	-ENODEV
35 #define arm_iommu_release_mapping(...)	do {} while (0)
36 #define arm_iommu_detach_device(...)	do {} while (0)
37 #endif
38 
39 #define IPMMU_CTX_MAX		8U
40 #define IPMMU_CTX_INVALID	-1
41 
42 #define IPMMU_UTLB_MAX		48U
43 
44 struct ipmmu_features {
45 	bool use_ns_alias_offset;
46 	bool has_cache_leaf_nodes;
47 	unsigned int number_of_contexts;
48 	unsigned int num_utlbs;
49 	bool setup_imbuscr;
50 	bool twobit_imttbcr_sl0;
51 	bool reserved_context;
52 	bool cache_snoop;
53 	unsigned int ctx_offset_base;
54 	unsigned int ctx_offset_stride;
55 	unsigned int utlb_offset_base;
56 };
57 
58 struct ipmmu_vmsa_device {
59 	struct device *dev;
60 	void __iomem *base;
61 	struct iommu_device iommu;
62 	struct ipmmu_vmsa_device *root;
63 	const struct ipmmu_features *features;
64 	unsigned int num_ctx;
65 	spinlock_t lock;			/* Protects ctx and domains[] */
66 	DECLARE_BITMAP(ctx, IPMMU_CTX_MAX);
67 	struct ipmmu_vmsa_domain *domains[IPMMU_CTX_MAX];
68 	s8 utlb_ctx[IPMMU_UTLB_MAX];
69 
70 	struct iommu_group *group;
71 	struct dma_iommu_mapping *mapping;
72 };
73 
74 struct ipmmu_vmsa_domain {
75 	struct ipmmu_vmsa_device *mmu;
76 	struct iommu_domain io_domain;
77 
78 	struct io_pgtable_cfg cfg;
79 	struct io_pgtable_ops *iop;
80 
81 	unsigned int context_id;
82 	struct mutex mutex;			/* Protects mappings */
83 };
84 
85 static struct ipmmu_vmsa_domain *to_vmsa_domain(struct iommu_domain *dom)
86 {
87 	return container_of(dom, struct ipmmu_vmsa_domain, io_domain);
88 }
89 
90 static struct ipmmu_vmsa_device *to_ipmmu(struct device *dev)
91 {
92 	return dev_iommu_priv_get(dev);
93 }
94 
95 #define TLB_LOOP_TIMEOUT		100	/* 100us */
96 
97 /* -----------------------------------------------------------------------------
98  * Registers Definition
99  */
100 
101 #define IM_NS_ALIAS_OFFSET		0x800
102 
103 /* MMU "context" registers */
104 #define IMCTR				0x0000		/* R-Car Gen2/3 */
105 #define IMCTR_INTEN			(1 << 2)	/* R-Car Gen2/3 */
106 #define IMCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
107 #define IMCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
108 
109 #define IMTTBCR				0x0008		/* R-Car Gen2/3 */
110 #define IMTTBCR_EAE			(1 << 31)	/* R-Car Gen2/3 */
111 #define IMTTBCR_SH0_INNER_SHAREABLE	(3 << 12)	/* R-Car Gen2 only */
112 #define IMTTBCR_ORGN0_WB_WA		(1 << 10)	/* R-Car Gen2 only */
113 #define IMTTBCR_IRGN0_WB_WA		(1 << 8)	/* R-Car Gen2 only */
114 #define IMTTBCR_SL0_TWOBIT_LVL_1	(2 << 6)	/* R-Car Gen3 only */
115 #define IMTTBCR_SL0_LVL_1		(1 << 4)	/* R-Car Gen2 only */
116 
117 #define IMBUSCR				0x000c		/* R-Car Gen2 only */
118 #define IMBUSCR_DVM			(1 << 2)	/* R-Car Gen2 only */
119 #define IMBUSCR_BUSSEL_MASK		(3 << 0)	/* R-Car Gen2 only */
120 
121 #define IMTTLBR0			0x0010		/* R-Car Gen2/3 */
122 #define IMTTUBR0			0x0014		/* R-Car Gen2/3 */
123 
124 #define IMSTR				0x0020		/* R-Car Gen2/3 */
125 #define IMSTR_MHIT			(1 << 4)	/* R-Car Gen2/3 */
126 #define IMSTR_ABORT			(1 << 2)	/* R-Car Gen2/3 */
127 #define IMSTR_PF			(1 << 1)	/* R-Car Gen2/3 */
128 #define IMSTR_TF			(1 << 0)	/* R-Car Gen2/3 */
129 
130 #define IMMAIR0				0x0028		/* R-Car Gen2/3 */
131 
132 #define IMELAR				0x0030		/* R-Car Gen2/3, IMEAR on R-Car Gen2 */
133 #define IMEUAR				0x0034		/* R-Car Gen3 only */
134 
135 /* uTLB registers */
136 #define IMUCTR(n)			((n) < 32 ? IMUCTR0(n) : IMUCTR32(n))
137 #define IMUCTR0(n)			(0x0300 + ((n) * 16))		/* R-Car Gen2/3 */
138 #define IMUCTR32(n)			(0x0600 + (((n) - 32) * 16))	/* R-Car Gen3 only */
139 #define IMUCTR_TTSEL_MMU(n)		((n) << 4)	/* R-Car Gen2/3 */
140 #define IMUCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
141 #define IMUCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
142 
143 #define IMUASID(n)			((n) < 32 ? IMUASID0(n) : IMUASID32(n))
144 #define IMUASID0(n)			(0x0308 + ((n) * 16))		/* R-Car Gen2/3 */
145 #define IMUASID32(n)			(0x0608 + (((n) - 32) * 16))	/* R-Car Gen3 only */
146 
147 /* -----------------------------------------------------------------------------
148  * Root device handling
149  */
150 
151 static struct platform_driver ipmmu_driver;
152 
153 static bool ipmmu_is_root(struct ipmmu_vmsa_device *mmu)
154 {
155 	return mmu->root == mmu;
156 }
157 
158 static int __ipmmu_check_device(struct device *dev, void *data)
159 {
160 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
161 	struct ipmmu_vmsa_device **rootp = data;
162 
163 	if (ipmmu_is_root(mmu))
164 		*rootp = mmu;
165 
166 	return 0;
167 }
168 
169 static struct ipmmu_vmsa_device *ipmmu_find_root(void)
170 {
171 	struct ipmmu_vmsa_device *root = NULL;
172 
173 	return driver_for_each_device(&ipmmu_driver.driver, NULL, &root,
174 				      __ipmmu_check_device) == 0 ? root : NULL;
175 }
176 
177 /* -----------------------------------------------------------------------------
178  * Read/Write Access
179  */
180 
181 static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
182 {
183 	return ioread32(mmu->base + offset);
184 }
185 
186 static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
187 			u32 data)
188 {
189 	iowrite32(data, mmu->base + offset);
190 }
191 
192 static unsigned int ipmmu_ctx_reg(struct ipmmu_vmsa_device *mmu,
193 				  unsigned int context_id, unsigned int reg)
194 {
195 	return mmu->features->ctx_offset_base +
196 	       context_id * mmu->features->ctx_offset_stride + reg;
197 }
198 
199 static u32 ipmmu_ctx_read(struct ipmmu_vmsa_device *mmu,
200 			  unsigned int context_id, unsigned int reg)
201 {
202 	return ipmmu_read(mmu, ipmmu_ctx_reg(mmu, context_id, reg));
203 }
204 
205 static void ipmmu_ctx_write(struct ipmmu_vmsa_device *mmu,
206 			    unsigned int context_id, unsigned int reg, u32 data)
207 {
208 	ipmmu_write(mmu, ipmmu_ctx_reg(mmu, context_id, reg), data);
209 }
210 
211 static u32 ipmmu_ctx_read_root(struct ipmmu_vmsa_domain *domain,
212 			       unsigned int reg)
213 {
214 	return ipmmu_ctx_read(domain->mmu->root, domain->context_id, reg);
215 }
216 
217 static void ipmmu_ctx_write_root(struct ipmmu_vmsa_domain *domain,
218 				 unsigned int reg, u32 data)
219 {
220 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
221 }
222 
223 static void ipmmu_ctx_write_all(struct ipmmu_vmsa_domain *domain,
224 				unsigned int reg, u32 data)
225 {
226 	if (domain->mmu != domain->mmu->root)
227 		ipmmu_ctx_write(domain->mmu, domain->context_id, reg, data);
228 
229 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
230 }
231 
232 static u32 ipmmu_utlb_reg(struct ipmmu_vmsa_device *mmu, unsigned int reg)
233 {
234 	return mmu->features->utlb_offset_base + reg;
235 }
236 
237 static void ipmmu_imuasid_write(struct ipmmu_vmsa_device *mmu,
238 				unsigned int utlb, u32 data)
239 {
240 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUASID(utlb)), data);
241 }
242 
243 static void ipmmu_imuctr_write(struct ipmmu_vmsa_device *mmu,
244 			       unsigned int utlb, u32 data)
245 {
246 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUCTR(utlb)), data);
247 }
248 
249 /* -----------------------------------------------------------------------------
250  * TLB and microTLB Management
251  */
252 
253 /* Wait for any pending TLB invalidations to complete */
254 static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
255 {
256 	unsigned int count = 0;
257 
258 	while (ipmmu_ctx_read_root(domain, IMCTR) & IMCTR_FLUSH) {
259 		cpu_relax();
260 		if (++count == TLB_LOOP_TIMEOUT) {
261 			dev_err_ratelimited(domain->mmu->dev,
262 			"TLB sync timed out -- MMU may be deadlocked\n");
263 			return;
264 		}
265 		udelay(1);
266 	}
267 }
268 
269 static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
270 {
271 	u32 reg;
272 
273 	reg = ipmmu_ctx_read_root(domain, IMCTR);
274 	reg |= IMCTR_FLUSH;
275 	ipmmu_ctx_write_all(domain, IMCTR, reg);
276 
277 	ipmmu_tlb_sync(domain);
278 }
279 
280 /*
281  * Enable MMU translation for the microTLB.
282  */
283 static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
284 			      unsigned int utlb)
285 {
286 	struct ipmmu_vmsa_device *mmu = domain->mmu;
287 
288 	/*
289 	 * TODO: Reference-count the microTLB as several bus masters can be
290 	 * connected to the same microTLB.
291 	 */
292 
293 	/* TODO: What should we set the ASID to ? */
294 	ipmmu_imuasid_write(mmu, utlb, 0);
295 	/* TODO: Do we need to flush the microTLB ? */
296 	ipmmu_imuctr_write(mmu, utlb, IMUCTR_TTSEL_MMU(domain->context_id) |
297 				      IMUCTR_FLUSH | IMUCTR_MMUEN);
298 	mmu->utlb_ctx[utlb] = domain->context_id;
299 }
300 
301 /*
302  * Disable MMU translation for the microTLB.
303  */
304 static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
305 			       unsigned int utlb)
306 {
307 	struct ipmmu_vmsa_device *mmu = domain->mmu;
308 
309 	ipmmu_imuctr_write(mmu, utlb, 0);
310 	mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
311 }
312 
313 static void ipmmu_tlb_flush_all(void *cookie)
314 {
315 	struct ipmmu_vmsa_domain *domain = cookie;
316 
317 	ipmmu_tlb_invalidate(domain);
318 }
319 
320 static void ipmmu_tlb_flush(unsigned long iova, size_t size,
321 				size_t granule, void *cookie)
322 {
323 	ipmmu_tlb_flush_all(cookie);
324 }
325 
326 static const struct iommu_flush_ops ipmmu_flush_ops = {
327 	.tlb_flush_all = ipmmu_tlb_flush_all,
328 	.tlb_flush_walk = ipmmu_tlb_flush,
329 	.tlb_flush_leaf = ipmmu_tlb_flush,
330 };
331 
332 /* -----------------------------------------------------------------------------
333  * Domain/Context Management
334  */
335 
336 static int ipmmu_domain_allocate_context(struct ipmmu_vmsa_device *mmu,
337 					 struct ipmmu_vmsa_domain *domain)
338 {
339 	unsigned long flags;
340 	int ret;
341 
342 	spin_lock_irqsave(&mmu->lock, flags);
343 
344 	ret = find_first_zero_bit(mmu->ctx, mmu->num_ctx);
345 	if (ret != mmu->num_ctx) {
346 		mmu->domains[ret] = domain;
347 		set_bit(ret, mmu->ctx);
348 	} else
349 		ret = -EBUSY;
350 
351 	spin_unlock_irqrestore(&mmu->lock, flags);
352 
353 	return ret;
354 }
355 
356 static void ipmmu_domain_free_context(struct ipmmu_vmsa_device *mmu,
357 				      unsigned int context_id)
358 {
359 	unsigned long flags;
360 
361 	spin_lock_irqsave(&mmu->lock, flags);
362 
363 	clear_bit(context_id, mmu->ctx);
364 	mmu->domains[context_id] = NULL;
365 
366 	spin_unlock_irqrestore(&mmu->lock, flags);
367 }
368 
369 static void ipmmu_domain_setup_context(struct ipmmu_vmsa_domain *domain)
370 {
371 	u64 ttbr;
372 	u32 tmp;
373 
374 	/* TTBR0 */
375 	ttbr = domain->cfg.arm_lpae_s1_cfg.ttbr;
376 	ipmmu_ctx_write_root(domain, IMTTLBR0, ttbr);
377 	ipmmu_ctx_write_root(domain, IMTTUBR0, ttbr >> 32);
378 
379 	/*
380 	 * TTBCR
381 	 * We use long descriptors and allocate the whole 32-bit VA space to
382 	 * TTBR0.
383 	 */
384 	if (domain->mmu->features->twobit_imttbcr_sl0)
385 		tmp = IMTTBCR_SL0_TWOBIT_LVL_1;
386 	else
387 		tmp = IMTTBCR_SL0_LVL_1;
388 
389 	if (domain->mmu->features->cache_snoop)
390 		tmp |= IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
391 		       IMTTBCR_IRGN0_WB_WA;
392 
393 	ipmmu_ctx_write_root(domain, IMTTBCR, IMTTBCR_EAE | tmp);
394 
395 	/* MAIR0 */
396 	ipmmu_ctx_write_root(domain, IMMAIR0,
397 			     domain->cfg.arm_lpae_s1_cfg.mair);
398 
399 	/* IMBUSCR */
400 	if (domain->mmu->features->setup_imbuscr)
401 		ipmmu_ctx_write_root(domain, IMBUSCR,
402 				     ipmmu_ctx_read_root(domain, IMBUSCR) &
403 				     ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
404 
405 	/*
406 	 * IMSTR
407 	 * Clear all interrupt flags.
408 	 */
409 	ipmmu_ctx_write_root(domain, IMSTR, ipmmu_ctx_read_root(domain, IMSTR));
410 
411 	/*
412 	 * IMCTR
413 	 * Enable the MMU and interrupt generation. The long-descriptor
414 	 * translation table format doesn't use TEX remapping. Don't enable AF
415 	 * software management as we have no use for it. Flush the TLB as
416 	 * required when modifying the context registers.
417 	 */
418 	ipmmu_ctx_write_all(domain, IMCTR,
419 			    IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
420 }
421 
422 static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
423 {
424 	int ret;
425 
426 	/*
427 	 * Allocate the page table operations.
428 	 *
429 	 * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory
430 	 * access, Long-descriptor format" that the NStable bit being set in a
431 	 * table descriptor will result in the NStable and NS bits of all child
432 	 * entries being ignored and considered as being set. The IPMMU seems
433 	 * not to comply with this, as it generates a secure access page fault
434 	 * if any of the NStable and NS bits isn't set when running in
435 	 * non-secure mode.
436 	 */
437 	domain->cfg.quirks = IO_PGTABLE_QUIRK_ARM_NS;
438 	domain->cfg.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K;
439 	domain->cfg.ias = 32;
440 	domain->cfg.oas = 40;
441 	domain->cfg.tlb = &ipmmu_flush_ops;
442 	domain->io_domain.geometry.aperture_end = DMA_BIT_MASK(32);
443 	domain->io_domain.geometry.force_aperture = true;
444 	/*
445 	 * TODO: Add support for coherent walk through CCI with DVM and remove
446 	 * cache handling. For now, delegate it to the io-pgtable code.
447 	 */
448 	domain->cfg.coherent_walk = false;
449 	domain->cfg.iommu_dev = domain->mmu->root->dev;
450 
451 	/*
452 	 * Find an unused context.
453 	 */
454 	ret = ipmmu_domain_allocate_context(domain->mmu->root, domain);
455 	if (ret < 0)
456 		return ret;
457 
458 	domain->context_id = ret;
459 
460 	domain->iop = alloc_io_pgtable_ops(ARM_32_LPAE_S1, &domain->cfg,
461 					   domain);
462 	if (!domain->iop) {
463 		ipmmu_domain_free_context(domain->mmu->root,
464 					  domain->context_id);
465 		return -EINVAL;
466 	}
467 
468 	ipmmu_domain_setup_context(domain);
469 	return 0;
470 }
471 
472 static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
473 {
474 	if (!domain->mmu)
475 		return;
476 
477 	/*
478 	 * Disable the context. Flush the TLB as required when modifying the
479 	 * context registers.
480 	 *
481 	 * TODO: Is TLB flush really needed ?
482 	 */
483 	ipmmu_ctx_write_all(domain, IMCTR, IMCTR_FLUSH);
484 	ipmmu_tlb_sync(domain);
485 	ipmmu_domain_free_context(domain->mmu->root, domain->context_id);
486 }
487 
488 /* -----------------------------------------------------------------------------
489  * Fault Handling
490  */
491 
492 static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
493 {
494 	const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
495 	struct ipmmu_vmsa_device *mmu = domain->mmu;
496 	unsigned long iova;
497 	u32 status;
498 
499 	status = ipmmu_ctx_read_root(domain, IMSTR);
500 	if (!(status & err_mask))
501 		return IRQ_NONE;
502 
503 	iova = ipmmu_ctx_read_root(domain, IMELAR);
504 	if (IS_ENABLED(CONFIG_64BIT))
505 		iova |= (u64)ipmmu_ctx_read_root(domain, IMEUAR) << 32;
506 
507 	/*
508 	 * Clear the error status flags. Unlike traditional interrupt flag
509 	 * registers that must be cleared by writing 1, this status register
510 	 * seems to require 0. The error address register must be read before,
511 	 * otherwise its value will be 0.
512 	 */
513 	ipmmu_ctx_write_root(domain, IMSTR, 0);
514 
515 	/* Log fatal errors. */
516 	if (status & IMSTR_MHIT)
517 		dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%lx\n",
518 				    iova);
519 	if (status & IMSTR_ABORT)
520 		dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%lx\n",
521 				    iova);
522 
523 	if (!(status & (IMSTR_PF | IMSTR_TF)))
524 		return IRQ_NONE;
525 
526 	/*
527 	 * Try to handle page faults and translation faults.
528 	 *
529 	 * TODO: We need to look up the faulty device based on the I/O VA. Use
530 	 * the IOMMU device for now.
531 	 */
532 	if (!report_iommu_fault(&domain->io_domain, mmu->dev, iova, 0))
533 		return IRQ_HANDLED;
534 
535 	dev_err_ratelimited(mmu->dev,
536 			    "Unhandled fault: status 0x%08x iova 0x%lx\n",
537 			    status, iova);
538 
539 	return IRQ_HANDLED;
540 }
541 
542 static irqreturn_t ipmmu_irq(int irq, void *dev)
543 {
544 	struct ipmmu_vmsa_device *mmu = dev;
545 	irqreturn_t status = IRQ_NONE;
546 	unsigned int i;
547 	unsigned long flags;
548 
549 	spin_lock_irqsave(&mmu->lock, flags);
550 
551 	/*
552 	 * Check interrupts for all active contexts.
553 	 */
554 	for (i = 0; i < mmu->num_ctx; i++) {
555 		if (!mmu->domains[i])
556 			continue;
557 		if (ipmmu_domain_irq(mmu->domains[i]) == IRQ_HANDLED)
558 			status = IRQ_HANDLED;
559 	}
560 
561 	spin_unlock_irqrestore(&mmu->lock, flags);
562 
563 	return status;
564 }
565 
566 /* -----------------------------------------------------------------------------
567  * IOMMU Operations
568  */
569 
570 static struct iommu_domain *__ipmmu_domain_alloc(unsigned type)
571 {
572 	struct ipmmu_vmsa_domain *domain;
573 
574 	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
575 	if (!domain)
576 		return NULL;
577 
578 	mutex_init(&domain->mutex);
579 
580 	return &domain->io_domain;
581 }
582 
583 static struct iommu_domain *ipmmu_domain_alloc(unsigned type)
584 {
585 	struct iommu_domain *io_domain = NULL;
586 
587 	switch (type) {
588 	case IOMMU_DOMAIN_UNMANAGED:
589 		io_domain = __ipmmu_domain_alloc(type);
590 		break;
591 
592 	case IOMMU_DOMAIN_DMA:
593 		io_domain = __ipmmu_domain_alloc(type);
594 		if (io_domain && iommu_get_dma_cookie(io_domain)) {
595 			kfree(io_domain);
596 			io_domain = NULL;
597 		}
598 		break;
599 	}
600 
601 	return io_domain;
602 }
603 
604 static void ipmmu_domain_free(struct iommu_domain *io_domain)
605 {
606 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
607 
608 	/*
609 	 * Free the domain resources. We assume that all devices have already
610 	 * been detached.
611 	 */
612 	iommu_put_dma_cookie(io_domain);
613 	ipmmu_domain_destroy_context(domain);
614 	free_io_pgtable_ops(domain->iop);
615 	kfree(domain);
616 }
617 
618 static int ipmmu_attach_device(struct iommu_domain *io_domain,
619 			       struct device *dev)
620 {
621 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
622 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
623 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
624 	unsigned int i;
625 	int ret = 0;
626 
627 	if (!mmu) {
628 		dev_err(dev, "Cannot attach to IPMMU\n");
629 		return -ENXIO;
630 	}
631 
632 	mutex_lock(&domain->mutex);
633 
634 	if (!domain->mmu) {
635 		/* The domain hasn't been used yet, initialize it. */
636 		domain->mmu = mmu;
637 		ret = ipmmu_domain_init_context(domain);
638 		if (ret < 0) {
639 			dev_err(dev, "Unable to initialize IPMMU context\n");
640 			domain->mmu = NULL;
641 		} else {
642 			dev_info(dev, "Using IPMMU context %u\n",
643 				 domain->context_id);
644 		}
645 	} else if (domain->mmu != mmu) {
646 		/*
647 		 * Something is wrong, we can't attach two devices using
648 		 * different IOMMUs to the same domain.
649 		 */
650 		dev_err(dev, "Can't attach IPMMU %s to domain on IPMMU %s\n",
651 			dev_name(mmu->dev), dev_name(domain->mmu->dev));
652 		ret = -EINVAL;
653 	} else
654 		dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
655 
656 	mutex_unlock(&domain->mutex);
657 
658 	if (ret < 0)
659 		return ret;
660 
661 	for (i = 0; i < fwspec->num_ids; ++i)
662 		ipmmu_utlb_enable(domain, fwspec->ids[i]);
663 
664 	return 0;
665 }
666 
667 static void ipmmu_detach_device(struct iommu_domain *io_domain,
668 				struct device *dev)
669 {
670 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
671 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
672 	unsigned int i;
673 
674 	for (i = 0; i < fwspec->num_ids; ++i)
675 		ipmmu_utlb_disable(domain, fwspec->ids[i]);
676 
677 	/*
678 	 * TODO: Optimize by disabling the context when no device is attached.
679 	 */
680 }
681 
682 static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
683 		     phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
684 {
685 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
686 
687 	if (!domain)
688 		return -ENODEV;
689 
690 	return domain->iop->map(domain->iop, iova, paddr, size, prot);
691 }
692 
693 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
694 			  size_t size, struct iommu_iotlb_gather *gather)
695 {
696 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
697 
698 	return domain->iop->unmap(domain->iop, iova, size, gather);
699 }
700 
701 static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
702 {
703 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
704 
705 	if (domain->mmu)
706 		ipmmu_tlb_flush_all(domain);
707 }
708 
709 static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
710 			     struct iommu_iotlb_gather *gather)
711 {
712 	ipmmu_flush_iotlb_all(io_domain);
713 }
714 
715 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
716 				      dma_addr_t iova)
717 {
718 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
719 
720 	/* TODO: Is locking needed ? */
721 
722 	return domain->iop->iova_to_phys(domain->iop, iova);
723 }
724 
725 static int ipmmu_init_platform_device(struct device *dev,
726 				      struct of_phandle_args *args)
727 {
728 	struct platform_device *ipmmu_pdev;
729 
730 	ipmmu_pdev = of_find_device_by_node(args->np);
731 	if (!ipmmu_pdev)
732 		return -ENODEV;
733 
734 	dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
735 
736 	return 0;
737 }
738 
739 static const struct soc_device_attribute soc_rcar_gen3[] = {
740 	{ .soc_id = "r8a774a1", },
741 	{ .soc_id = "r8a774b1", },
742 	{ .soc_id = "r8a774c0", },
743 	{ .soc_id = "r8a7795", },
744 	{ .soc_id = "r8a7796", },
745 	{ .soc_id = "r8a77965", },
746 	{ .soc_id = "r8a77970", },
747 	{ .soc_id = "r8a77990", },
748 	{ .soc_id = "r8a77995", },
749 	{ /* sentinel */ }
750 };
751 
752 static const struct soc_device_attribute soc_rcar_gen3_whitelist[] = {
753 	{ .soc_id = "r8a774b1", },
754 	{ .soc_id = "r8a774c0", },
755 	{ .soc_id = "r8a7795", .revision = "ES3.*" },
756 	{ .soc_id = "r8a77965", },
757 	{ .soc_id = "r8a77990", },
758 	{ .soc_id = "r8a77995", },
759 	{ /* sentinel */ }
760 };
761 
762 static const char * const rcar_gen3_slave_whitelist[] = {
763 };
764 
765 static bool ipmmu_slave_whitelist(struct device *dev)
766 {
767 	unsigned int i;
768 
769 	/*
770 	 * For R-Car Gen3 use a white list to opt-in slave devices.
771 	 * For Other SoCs, this returns true anyway.
772 	 */
773 	if (!soc_device_match(soc_rcar_gen3))
774 		return true;
775 
776 	/* Check whether this R-Car Gen3 can use the IPMMU correctly or not */
777 	if (!soc_device_match(soc_rcar_gen3_whitelist))
778 		return false;
779 
780 	/* Check whether this slave device can work with the IPMMU */
781 	for (i = 0; i < ARRAY_SIZE(rcar_gen3_slave_whitelist); i++) {
782 		if (!strcmp(dev_name(dev), rcar_gen3_slave_whitelist[i]))
783 			return true;
784 	}
785 
786 	/* Otherwise, do not allow use of IPMMU */
787 	return false;
788 }
789 
790 static int ipmmu_of_xlate(struct device *dev,
791 			  struct of_phandle_args *spec)
792 {
793 	if (!ipmmu_slave_whitelist(dev))
794 		return -ENODEV;
795 
796 	iommu_fwspec_add_ids(dev, spec->args, 1);
797 
798 	/* Initialize once - xlate() will call multiple times */
799 	if (to_ipmmu(dev))
800 		return 0;
801 
802 	return ipmmu_init_platform_device(dev, spec);
803 }
804 
805 static int ipmmu_init_arm_mapping(struct device *dev)
806 {
807 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
808 	struct iommu_group *group;
809 	int ret;
810 
811 	/* Create a device group and add the device to it. */
812 	group = iommu_group_alloc();
813 	if (IS_ERR(group)) {
814 		dev_err(dev, "Failed to allocate IOMMU group\n");
815 		return PTR_ERR(group);
816 	}
817 
818 	ret = iommu_group_add_device(group, dev);
819 	iommu_group_put(group);
820 
821 	if (ret < 0) {
822 		dev_err(dev, "Failed to add device to IPMMU group\n");
823 		return ret;
824 	}
825 
826 	/*
827 	 * Create the ARM mapping, used by the ARM DMA mapping core to allocate
828 	 * VAs. This will allocate a corresponding IOMMU domain.
829 	 *
830 	 * TODO:
831 	 * - Create one mapping per context (TLB).
832 	 * - Make the mapping size configurable ? We currently use a 2GB mapping
833 	 *   at a 1GB offset to ensure that NULL VAs will fault.
834 	 */
835 	if (!mmu->mapping) {
836 		struct dma_iommu_mapping *mapping;
837 
838 		mapping = arm_iommu_create_mapping(&platform_bus_type,
839 						   SZ_1G, SZ_2G);
840 		if (IS_ERR(mapping)) {
841 			dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
842 			ret = PTR_ERR(mapping);
843 			goto error;
844 		}
845 
846 		mmu->mapping = mapping;
847 	}
848 
849 	/* Attach the ARM VA mapping to the device. */
850 	ret = arm_iommu_attach_device(dev, mmu->mapping);
851 	if (ret < 0) {
852 		dev_err(dev, "Failed to attach device to VA mapping\n");
853 		goto error;
854 	}
855 
856 	return 0;
857 
858 error:
859 	iommu_group_remove_device(dev);
860 	if (mmu->mapping)
861 		arm_iommu_release_mapping(mmu->mapping);
862 
863 	return ret;
864 }
865 
866 static int ipmmu_add_device(struct device *dev)
867 {
868 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
869 	struct iommu_group *group;
870 	int ret;
871 
872 	/*
873 	 * Only let through devices that have been verified in xlate()
874 	 */
875 	if (!mmu)
876 		return -ENODEV;
877 
878 	if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)) {
879 		ret = ipmmu_init_arm_mapping(dev);
880 		if (ret)
881 			return ret;
882 	} else {
883 		group = iommu_group_get_for_dev(dev);
884 		if (IS_ERR(group))
885 			return PTR_ERR(group);
886 
887 		iommu_group_put(group);
888 	}
889 
890 	iommu_device_link(&mmu->iommu, dev);
891 	return 0;
892 }
893 
894 static void ipmmu_remove_device(struct device *dev)
895 {
896 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
897 
898 	iommu_device_unlink(&mmu->iommu, dev);
899 	arm_iommu_detach_device(dev);
900 	iommu_group_remove_device(dev);
901 }
902 
903 static struct iommu_group *ipmmu_find_group(struct device *dev)
904 {
905 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
906 	struct iommu_group *group;
907 
908 	if (mmu->group)
909 		return iommu_group_ref_get(mmu->group);
910 
911 	group = iommu_group_alloc();
912 	if (!IS_ERR(group))
913 		mmu->group = group;
914 
915 	return group;
916 }
917 
918 static const struct iommu_ops ipmmu_ops = {
919 	.domain_alloc = ipmmu_domain_alloc,
920 	.domain_free = ipmmu_domain_free,
921 	.attach_dev = ipmmu_attach_device,
922 	.detach_dev = ipmmu_detach_device,
923 	.map = ipmmu_map,
924 	.unmap = ipmmu_unmap,
925 	.flush_iotlb_all = ipmmu_flush_iotlb_all,
926 	.iotlb_sync = ipmmu_iotlb_sync,
927 	.iova_to_phys = ipmmu_iova_to_phys,
928 	.add_device = ipmmu_add_device,
929 	.remove_device = ipmmu_remove_device,
930 	.device_group = ipmmu_find_group,
931 	.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
932 	.of_xlate = ipmmu_of_xlate,
933 };
934 
935 /* -----------------------------------------------------------------------------
936  * Probe/remove and init
937  */
938 
939 static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
940 {
941 	unsigned int i;
942 
943 	/* Disable all contexts. */
944 	for (i = 0; i < mmu->num_ctx; ++i)
945 		ipmmu_ctx_write(mmu, i, IMCTR, 0);
946 }
947 
948 static const struct ipmmu_features ipmmu_features_default = {
949 	.use_ns_alias_offset = true,
950 	.has_cache_leaf_nodes = false,
951 	.number_of_contexts = 1, /* software only tested with one context */
952 	.num_utlbs = 32,
953 	.setup_imbuscr = true,
954 	.twobit_imttbcr_sl0 = false,
955 	.reserved_context = false,
956 	.cache_snoop = true,
957 	.ctx_offset_base = 0,
958 	.ctx_offset_stride = 0x40,
959 	.utlb_offset_base = 0,
960 };
961 
962 static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
963 	.use_ns_alias_offset = false,
964 	.has_cache_leaf_nodes = true,
965 	.number_of_contexts = 8,
966 	.num_utlbs = 48,
967 	.setup_imbuscr = false,
968 	.twobit_imttbcr_sl0 = true,
969 	.reserved_context = true,
970 	.cache_snoop = false,
971 	.ctx_offset_base = 0,
972 	.ctx_offset_stride = 0x40,
973 	.utlb_offset_base = 0,
974 };
975 
976 static const struct of_device_id ipmmu_of_ids[] = {
977 	{
978 		.compatible = "renesas,ipmmu-vmsa",
979 		.data = &ipmmu_features_default,
980 	}, {
981 		.compatible = "renesas,ipmmu-r8a774a1",
982 		.data = &ipmmu_features_rcar_gen3,
983 	}, {
984 		.compatible = "renesas,ipmmu-r8a774b1",
985 		.data = &ipmmu_features_rcar_gen3,
986 	}, {
987 		.compatible = "renesas,ipmmu-r8a774c0",
988 		.data = &ipmmu_features_rcar_gen3,
989 	}, {
990 		.compatible = "renesas,ipmmu-r8a7795",
991 		.data = &ipmmu_features_rcar_gen3,
992 	}, {
993 		.compatible = "renesas,ipmmu-r8a7796",
994 		.data = &ipmmu_features_rcar_gen3,
995 	}, {
996 		.compatible = "renesas,ipmmu-r8a77965",
997 		.data = &ipmmu_features_rcar_gen3,
998 	}, {
999 		.compatible = "renesas,ipmmu-r8a77970",
1000 		.data = &ipmmu_features_rcar_gen3,
1001 	}, {
1002 		.compatible = "renesas,ipmmu-r8a77990",
1003 		.data = &ipmmu_features_rcar_gen3,
1004 	}, {
1005 		.compatible = "renesas,ipmmu-r8a77995",
1006 		.data = &ipmmu_features_rcar_gen3,
1007 	}, {
1008 		/* Terminator */
1009 	},
1010 };
1011 
1012 static int ipmmu_probe(struct platform_device *pdev)
1013 {
1014 	struct ipmmu_vmsa_device *mmu;
1015 	struct resource *res;
1016 	int irq;
1017 	int ret;
1018 
1019 	mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
1020 	if (!mmu) {
1021 		dev_err(&pdev->dev, "cannot allocate device data\n");
1022 		return -ENOMEM;
1023 	}
1024 
1025 	mmu->dev = &pdev->dev;
1026 	spin_lock_init(&mmu->lock);
1027 	bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
1028 	mmu->features = of_device_get_match_data(&pdev->dev);
1029 	memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
1030 	dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
1031 
1032 	/* Map I/O memory and request IRQ. */
1033 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1034 	mmu->base = devm_ioremap_resource(&pdev->dev, res);
1035 	if (IS_ERR(mmu->base))
1036 		return PTR_ERR(mmu->base);
1037 
1038 	/*
1039 	 * The IPMMU has two register banks, for secure and non-secure modes.
1040 	 * The bank mapped at the beginning of the IPMMU address space
1041 	 * corresponds to the running mode of the CPU. When running in secure
1042 	 * mode the non-secure register bank is also available at an offset.
1043 	 *
1044 	 * Secure mode operation isn't clearly documented and is thus currently
1045 	 * not implemented in the driver. Furthermore, preliminary tests of
1046 	 * non-secure operation with the main register bank were not successful.
1047 	 * Offset the registers base unconditionally to point to the non-secure
1048 	 * alias space for now.
1049 	 */
1050 	if (mmu->features->use_ns_alias_offset)
1051 		mmu->base += IM_NS_ALIAS_OFFSET;
1052 
1053 	mmu->num_ctx = min(IPMMU_CTX_MAX, mmu->features->number_of_contexts);
1054 
1055 	/*
1056 	 * Determine if this IPMMU instance is a root device by checking for
1057 	 * the lack of has_cache_leaf_nodes flag or renesas,ipmmu-main property.
1058 	 */
1059 	if (!mmu->features->has_cache_leaf_nodes ||
1060 	    !of_find_property(pdev->dev.of_node, "renesas,ipmmu-main", NULL))
1061 		mmu->root = mmu;
1062 	else
1063 		mmu->root = ipmmu_find_root();
1064 
1065 	/*
1066 	 * Wait until the root device has been registered for sure.
1067 	 */
1068 	if (!mmu->root)
1069 		return -EPROBE_DEFER;
1070 
1071 	/* Root devices have mandatory IRQs */
1072 	if (ipmmu_is_root(mmu)) {
1073 		irq = platform_get_irq(pdev, 0);
1074 		if (irq < 0)
1075 			return irq;
1076 
1077 		ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
1078 				       dev_name(&pdev->dev), mmu);
1079 		if (ret < 0) {
1080 			dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
1081 			return ret;
1082 		}
1083 
1084 		ipmmu_device_reset(mmu);
1085 
1086 		if (mmu->features->reserved_context) {
1087 			dev_info(&pdev->dev, "IPMMU context 0 is reserved\n");
1088 			set_bit(0, mmu->ctx);
1089 		}
1090 	}
1091 
1092 	/*
1093 	 * Register the IPMMU to the IOMMU subsystem in the following cases:
1094 	 * - R-Car Gen2 IPMMU (all devices registered)
1095 	 * - R-Car Gen3 IPMMU (leaf devices only - skip root IPMMU-MM device)
1096 	 */
1097 	if (!mmu->features->has_cache_leaf_nodes || !ipmmu_is_root(mmu)) {
1098 		ret = iommu_device_sysfs_add(&mmu->iommu, &pdev->dev, NULL,
1099 					     dev_name(&pdev->dev));
1100 		if (ret)
1101 			return ret;
1102 
1103 		iommu_device_set_ops(&mmu->iommu, &ipmmu_ops);
1104 		iommu_device_set_fwnode(&mmu->iommu,
1105 					&pdev->dev.of_node->fwnode);
1106 
1107 		ret = iommu_device_register(&mmu->iommu);
1108 		if (ret)
1109 			return ret;
1110 
1111 #if defined(CONFIG_IOMMU_DMA)
1112 		if (!iommu_present(&platform_bus_type))
1113 			bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1114 #endif
1115 	}
1116 
1117 	/*
1118 	 * We can't create the ARM mapping here as it requires the bus to have
1119 	 * an IOMMU, which only happens when bus_set_iommu() is called in
1120 	 * ipmmu_init() after the probe function returns.
1121 	 */
1122 
1123 	platform_set_drvdata(pdev, mmu);
1124 
1125 	return 0;
1126 }
1127 
1128 static int ipmmu_remove(struct platform_device *pdev)
1129 {
1130 	struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1131 
1132 	iommu_device_sysfs_remove(&mmu->iommu);
1133 	iommu_device_unregister(&mmu->iommu);
1134 
1135 	arm_iommu_release_mapping(mmu->mapping);
1136 
1137 	ipmmu_device_reset(mmu);
1138 
1139 	return 0;
1140 }
1141 
1142 #ifdef CONFIG_PM_SLEEP
1143 static int ipmmu_resume_noirq(struct device *dev)
1144 {
1145 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1146 	unsigned int i;
1147 
1148 	/* Reset root MMU and restore contexts */
1149 	if (ipmmu_is_root(mmu)) {
1150 		ipmmu_device_reset(mmu);
1151 
1152 		for (i = 0; i < mmu->num_ctx; i++) {
1153 			if (!mmu->domains[i])
1154 				continue;
1155 
1156 			ipmmu_domain_setup_context(mmu->domains[i]);
1157 		}
1158 	}
1159 
1160 	/* Re-enable active micro-TLBs */
1161 	for (i = 0; i < mmu->features->num_utlbs; i++) {
1162 		if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1163 			continue;
1164 
1165 		ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1166 	}
1167 
1168 	return 0;
1169 }
1170 
1171 static const struct dev_pm_ops ipmmu_pm  = {
1172 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1173 };
1174 #define DEV_PM_OPS	&ipmmu_pm
1175 #else
1176 #define DEV_PM_OPS	NULL
1177 #endif /* CONFIG_PM_SLEEP */
1178 
1179 static struct platform_driver ipmmu_driver = {
1180 	.driver = {
1181 		.name = "ipmmu-vmsa",
1182 		.of_match_table = of_match_ptr(ipmmu_of_ids),
1183 		.pm = DEV_PM_OPS,
1184 	},
1185 	.probe = ipmmu_probe,
1186 	.remove	= ipmmu_remove,
1187 };
1188 
1189 static int __init ipmmu_init(void)
1190 {
1191 	struct device_node *np;
1192 	static bool setup_done;
1193 	int ret;
1194 
1195 	if (setup_done)
1196 		return 0;
1197 
1198 	np = of_find_matching_node(NULL, ipmmu_of_ids);
1199 	if (!np)
1200 		return 0;
1201 
1202 	of_node_put(np);
1203 
1204 	ret = platform_driver_register(&ipmmu_driver);
1205 	if (ret < 0)
1206 		return ret;
1207 
1208 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
1209 	if (!iommu_present(&platform_bus_type))
1210 		bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1211 #endif
1212 
1213 	setup_done = true;
1214 	return 0;
1215 }
1216 subsys_initcall(ipmmu_init);
1217