xref: /linux/drivers/iommu/ipmmu-vmsa.c (revision 266679ffd867cb247c36717ea4d7998e9304823b)
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) 2014-2020 Renesas Electronics Corporation
7  */
8 
9 #include <linux/bitmap.h>
10 #include <linux/delay.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/err.h>
13 #include <linux/export.h>
14 #include <linux/init.h>
15 #include <linux/interrupt.h>
16 #include <linux/io.h>
17 #include <linux/io-pgtable.h>
18 #include <linux/iommu.h>
19 #include <linux/of.h>
20 #include <linux/of_device.h>
21 #include <linux/of_platform.h>
22 #include <linux/platform_device.h>
23 #include <linux/sizes.h>
24 #include <linux/slab.h>
25 #include <linux/sys_soc.h>
26 
27 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
28 #include <asm/dma-iommu.h>
29 #else
30 #define arm_iommu_create_mapping(...)	NULL
31 #define arm_iommu_attach_device(...)	-ENODEV
32 #define arm_iommu_release_mapping(...)	do {} while (0)
33 #define arm_iommu_detach_device(...)	do {} while (0)
34 #endif
35 
36 #define IPMMU_CTX_MAX		16U
37 #define IPMMU_CTX_INVALID	-1
38 
39 #define IPMMU_UTLB_MAX		64U
40 
41 struct ipmmu_features {
42 	bool use_ns_alias_offset;
43 	bool has_cache_leaf_nodes;
44 	unsigned int number_of_contexts;
45 	unsigned int num_utlbs;
46 	bool setup_imbuscr;
47 	bool twobit_imttbcr_sl0;
48 	bool reserved_context;
49 	bool cache_snoop;
50 	unsigned int ctx_offset_base;
51 	unsigned int ctx_offset_stride;
52 	unsigned int utlb_offset_base;
53 };
54 
55 struct ipmmu_vmsa_device {
56 	struct device *dev;
57 	void __iomem *base;
58 	struct iommu_device iommu;
59 	struct ipmmu_vmsa_device *root;
60 	const struct ipmmu_features *features;
61 	unsigned int num_ctx;
62 	spinlock_t lock;			/* Protects ctx and domains[] */
63 	DECLARE_BITMAP(ctx, IPMMU_CTX_MAX);
64 	struct ipmmu_vmsa_domain *domains[IPMMU_CTX_MAX];
65 	s8 utlb_ctx[IPMMU_UTLB_MAX];
66 
67 	struct iommu_group *group;
68 	struct dma_iommu_mapping *mapping;
69 };
70 
71 struct ipmmu_vmsa_domain {
72 	struct ipmmu_vmsa_device *mmu;
73 	struct iommu_domain io_domain;
74 
75 	struct io_pgtable_cfg cfg;
76 	struct io_pgtable_ops *iop;
77 
78 	unsigned int context_id;
79 	struct mutex mutex;			/* Protects mappings */
80 };
81 
82 static struct ipmmu_vmsa_domain *to_vmsa_domain(struct iommu_domain *dom)
83 {
84 	return container_of(dom, struct ipmmu_vmsa_domain, io_domain);
85 }
86 
87 static struct ipmmu_vmsa_device *to_ipmmu(struct device *dev)
88 {
89 	return dev_iommu_priv_get(dev);
90 }
91 
92 #define TLB_LOOP_TIMEOUT		100	/* 100us */
93 
94 /* -----------------------------------------------------------------------------
95  * Registers Definition
96  */
97 
98 #define IM_NS_ALIAS_OFFSET		0x800
99 
100 /* MMU "context" registers */
101 #define IMCTR				0x0000		/* R-Car Gen2/3 */
102 #define IMCTR_INTEN			(1 << 2)	/* R-Car Gen2/3 */
103 #define IMCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
104 #define IMCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
105 
106 #define IMTTBCR				0x0008		/* R-Car Gen2/3 */
107 #define IMTTBCR_EAE			(1 << 31)	/* R-Car Gen2/3 */
108 #define IMTTBCR_SH0_INNER_SHAREABLE	(3 << 12)	/* R-Car Gen2 only */
109 #define IMTTBCR_ORGN0_WB_WA		(1 << 10)	/* R-Car Gen2 only */
110 #define IMTTBCR_IRGN0_WB_WA		(1 << 8)	/* R-Car Gen2 only */
111 #define IMTTBCR_SL0_TWOBIT_LVL_1	(2 << 6)	/* R-Car Gen3 only */
112 #define IMTTBCR_SL0_LVL_1		(1 << 4)	/* R-Car Gen2 only */
113 
114 #define IMBUSCR				0x000c		/* R-Car Gen2 only */
115 #define IMBUSCR_DVM			(1 << 2)	/* R-Car Gen2 only */
116 #define IMBUSCR_BUSSEL_MASK		(3 << 0)	/* R-Car Gen2 only */
117 
118 #define IMTTLBR0			0x0010		/* R-Car Gen2/3 */
119 #define IMTTUBR0			0x0014		/* R-Car Gen2/3 */
120 
121 #define IMSTR				0x0020		/* R-Car Gen2/3 */
122 #define IMSTR_MHIT			(1 << 4)	/* R-Car Gen2/3 */
123 #define IMSTR_ABORT			(1 << 2)	/* R-Car Gen2/3 */
124 #define IMSTR_PF			(1 << 1)	/* R-Car Gen2/3 */
125 #define IMSTR_TF			(1 << 0)	/* R-Car Gen2/3 */
126 
127 #define IMMAIR0				0x0028		/* R-Car Gen2/3 */
128 
129 #define IMELAR				0x0030		/* R-Car Gen2/3, IMEAR on R-Car Gen2 */
130 #define IMEUAR				0x0034		/* R-Car Gen3 only */
131 
132 /* uTLB registers */
133 #define IMUCTR(n)			((n) < 32 ? IMUCTR0(n) : IMUCTR32(n))
134 #define IMUCTR0(n)			(0x0300 + ((n) * 16))		/* R-Car Gen2/3 */
135 #define IMUCTR32(n)			(0x0600 + (((n) - 32) * 16))	/* R-Car Gen3 only */
136 #define IMUCTR_TTSEL_MMU(n)		((n) << 4)	/* R-Car Gen2/3 */
137 #define IMUCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
138 #define IMUCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
139 
140 #define IMUASID(n)			((n) < 32 ? IMUASID0(n) : IMUASID32(n))
141 #define IMUASID0(n)			(0x0308 + ((n) * 16))		/* R-Car Gen2/3 */
142 #define IMUASID32(n)			(0x0608 + (((n) - 32) * 16))	/* R-Car Gen3 only */
143 
144 /* -----------------------------------------------------------------------------
145  * Root device handling
146  */
147 
148 static struct platform_driver ipmmu_driver;
149 
150 static bool ipmmu_is_root(struct ipmmu_vmsa_device *mmu)
151 {
152 	return mmu->root == mmu;
153 }
154 
155 static int __ipmmu_check_device(struct device *dev, void *data)
156 {
157 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
158 	struct ipmmu_vmsa_device **rootp = data;
159 
160 	if (ipmmu_is_root(mmu))
161 		*rootp = mmu;
162 
163 	return 0;
164 }
165 
166 static struct ipmmu_vmsa_device *ipmmu_find_root(void)
167 {
168 	struct ipmmu_vmsa_device *root = NULL;
169 
170 	return driver_for_each_device(&ipmmu_driver.driver, NULL, &root,
171 				      __ipmmu_check_device) == 0 ? root : NULL;
172 }
173 
174 /* -----------------------------------------------------------------------------
175  * Read/Write Access
176  */
177 
178 static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
179 {
180 	return ioread32(mmu->base + offset);
181 }
182 
183 static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
184 			u32 data)
185 {
186 	iowrite32(data, mmu->base + offset);
187 }
188 
189 static unsigned int ipmmu_ctx_reg(struct ipmmu_vmsa_device *mmu,
190 				  unsigned int context_id, unsigned int reg)
191 {
192 	unsigned int base = mmu->features->ctx_offset_base;
193 
194 	if (context_id > 7)
195 		base += 0x800 - 8 * 0x40;
196 
197 	return base + context_id * mmu->features->ctx_offset_stride + reg;
198 }
199 
200 static u32 ipmmu_ctx_read(struct ipmmu_vmsa_device *mmu,
201 			  unsigned int context_id, unsigned int reg)
202 {
203 	return ipmmu_read(mmu, ipmmu_ctx_reg(mmu, context_id, reg));
204 }
205 
206 static void ipmmu_ctx_write(struct ipmmu_vmsa_device *mmu,
207 			    unsigned int context_id, unsigned int reg, u32 data)
208 {
209 	ipmmu_write(mmu, ipmmu_ctx_reg(mmu, context_id, reg), data);
210 }
211 
212 static u32 ipmmu_ctx_read_root(struct ipmmu_vmsa_domain *domain,
213 			       unsigned int reg)
214 {
215 	return ipmmu_ctx_read(domain->mmu->root, domain->context_id, reg);
216 }
217 
218 static void ipmmu_ctx_write_root(struct ipmmu_vmsa_domain *domain,
219 				 unsigned int reg, u32 data)
220 {
221 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
222 }
223 
224 static void ipmmu_ctx_write_all(struct ipmmu_vmsa_domain *domain,
225 				unsigned int reg, u32 data)
226 {
227 	if (domain->mmu != domain->mmu->root)
228 		ipmmu_ctx_write(domain->mmu, domain->context_id, reg, data);
229 
230 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
231 }
232 
233 static u32 ipmmu_utlb_reg(struct ipmmu_vmsa_device *mmu, unsigned int reg)
234 {
235 	return mmu->features->utlb_offset_base + reg;
236 }
237 
238 static void ipmmu_imuasid_write(struct ipmmu_vmsa_device *mmu,
239 				unsigned int utlb, u32 data)
240 {
241 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUASID(utlb)), data);
242 }
243 
244 static void ipmmu_imuctr_write(struct ipmmu_vmsa_device *mmu,
245 			       unsigned int utlb, u32 data)
246 {
247 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUCTR(utlb)), data);
248 }
249 
250 /* -----------------------------------------------------------------------------
251  * TLB and microTLB Management
252  */
253 
254 /* Wait for any pending TLB invalidations to complete */
255 static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
256 {
257 	unsigned int count = 0;
258 
259 	while (ipmmu_ctx_read_root(domain, IMCTR) & IMCTR_FLUSH) {
260 		cpu_relax();
261 		if (++count == TLB_LOOP_TIMEOUT) {
262 			dev_err_ratelimited(domain->mmu->dev,
263 			"TLB sync timed out -- MMU may be deadlocked\n");
264 			return;
265 		}
266 		udelay(1);
267 	}
268 }
269 
270 static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
271 {
272 	u32 reg;
273 
274 	reg = ipmmu_ctx_read_root(domain, IMCTR);
275 	reg |= IMCTR_FLUSH;
276 	ipmmu_ctx_write_all(domain, IMCTR, reg);
277 
278 	ipmmu_tlb_sync(domain);
279 }
280 
281 /*
282  * Enable MMU translation for the microTLB.
283  */
284 static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
285 			      unsigned int utlb)
286 {
287 	struct ipmmu_vmsa_device *mmu = domain->mmu;
288 
289 	/*
290 	 * TODO: Reference-count the microTLB as several bus masters can be
291 	 * connected to the same microTLB.
292 	 */
293 
294 	/* TODO: What should we set the ASID to ? */
295 	ipmmu_imuasid_write(mmu, utlb, 0);
296 	/* TODO: Do we need to flush the microTLB ? */
297 	ipmmu_imuctr_write(mmu, utlb, IMUCTR_TTSEL_MMU(domain->context_id) |
298 				      IMUCTR_FLUSH | IMUCTR_MMUEN);
299 	mmu->utlb_ctx[utlb] = domain->context_id;
300 }
301 
302 static void ipmmu_tlb_flush_all(void *cookie)
303 {
304 	struct ipmmu_vmsa_domain *domain = cookie;
305 
306 	ipmmu_tlb_invalidate(domain);
307 }
308 
309 static void ipmmu_tlb_flush(unsigned long iova, size_t size,
310 				size_t granule, void *cookie)
311 {
312 	ipmmu_tlb_flush_all(cookie);
313 }
314 
315 static const struct iommu_flush_ops ipmmu_flush_ops = {
316 	.tlb_flush_all = ipmmu_tlb_flush_all,
317 	.tlb_flush_walk = ipmmu_tlb_flush,
318 };
319 
320 /* -----------------------------------------------------------------------------
321  * Domain/Context Management
322  */
323 
324 static int ipmmu_domain_allocate_context(struct ipmmu_vmsa_device *mmu,
325 					 struct ipmmu_vmsa_domain *domain)
326 {
327 	unsigned long flags;
328 	int ret;
329 
330 	spin_lock_irqsave(&mmu->lock, flags);
331 
332 	ret = find_first_zero_bit(mmu->ctx, mmu->num_ctx);
333 	if (ret != mmu->num_ctx) {
334 		mmu->domains[ret] = domain;
335 		set_bit(ret, mmu->ctx);
336 	} else
337 		ret = -EBUSY;
338 
339 	spin_unlock_irqrestore(&mmu->lock, flags);
340 
341 	return ret;
342 }
343 
344 static void ipmmu_domain_free_context(struct ipmmu_vmsa_device *mmu,
345 				      unsigned int context_id)
346 {
347 	unsigned long flags;
348 
349 	spin_lock_irqsave(&mmu->lock, flags);
350 
351 	clear_bit(context_id, mmu->ctx);
352 	mmu->domains[context_id] = NULL;
353 
354 	spin_unlock_irqrestore(&mmu->lock, flags);
355 }
356 
357 static void ipmmu_domain_setup_context(struct ipmmu_vmsa_domain *domain)
358 {
359 	u64 ttbr;
360 	u32 tmp;
361 
362 	/* TTBR0 */
363 	ttbr = domain->cfg.arm_lpae_s1_cfg.ttbr;
364 	ipmmu_ctx_write_root(domain, IMTTLBR0, ttbr);
365 	ipmmu_ctx_write_root(domain, IMTTUBR0, ttbr >> 32);
366 
367 	/*
368 	 * TTBCR
369 	 * We use long descriptors and allocate the whole 32-bit VA space to
370 	 * TTBR0.
371 	 */
372 	if (domain->mmu->features->twobit_imttbcr_sl0)
373 		tmp = IMTTBCR_SL0_TWOBIT_LVL_1;
374 	else
375 		tmp = IMTTBCR_SL0_LVL_1;
376 
377 	if (domain->mmu->features->cache_snoop)
378 		tmp |= IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
379 		       IMTTBCR_IRGN0_WB_WA;
380 
381 	ipmmu_ctx_write_root(domain, IMTTBCR, IMTTBCR_EAE | tmp);
382 
383 	/* MAIR0 */
384 	ipmmu_ctx_write_root(domain, IMMAIR0,
385 			     domain->cfg.arm_lpae_s1_cfg.mair);
386 
387 	/* IMBUSCR */
388 	if (domain->mmu->features->setup_imbuscr)
389 		ipmmu_ctx_write_root(domain, IMBUSCR,
390 				     ipmmu_ctx_read_root(domain, IMBUSCR) &
391 				     ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
392 
393 	/*
394 	 * IMSTR
395 	 * Clear all interrupt flags.
396 	 */
397 	ipmmu_ctx_write_root(domain, IMSTR, ipmmu_ctx_read_root(domain, IMSTR));
398 
399 	/*
400 	 * IMCTR
401 	 * Enable the MMU and interrupt generation. The long-descriptor
402 	 * translation table format doesn't use TEX remapping. Don't enable AF
403 	 * software management as we have no use for it. Flush the TLB as
404 	 * required when modifying the context registers.
405 	 */
406 	ipmmu_ctx_write_all(domain, IMCTR,
407 			    IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
408 }
409 
410 static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
411 {
412 	int ret;
413 
414 	/*
415 	 * Allocate the page table operations.
416 	 *
417 	 * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory
418 	 * access, Long-descriptor format" that the NStable bit being set in a
419 	 * table descriptor will result in the NStable and NS bits of all child
420 	 * entries being ignored and considered as being set. The IPMMU seems
421 	 * not to comply with this, as it generates a secure access page fault
422 	 * if any of the NStable and NS bits isn't set when running in
423 	 * non-secure mode.
424 	 */
425 	domain->cfg.quirks = IO_PGTABLE_QUIRK_ARM_NS;
426 	domain->cfg.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K;
427 	domain->cfg.ias = 32;
428 	domain->cfg.oas = 40;
429 	domain->cfg.tlb = &ipmmu_flush_ops;
430 	domain->io_domain.geometry.aperture_end = DMA_BIT_MASK(32);
431 	domain->io_domain.geometry.force_aperture = true;
432 	/*
433 	 * TODO: Add support for coherent walk through CCI with DVM and remove
434 	 * cache handling. For now, delegate it to the io-pgtable code.
435 	 */
436 	domain->cfg.coherent_walk = false;
437 	domain->cfg.iommu_dev = domain->mmu->root->dev;
438 
439 	/*
440 	 * Find an unused context.
441 	 */
442 	ret = ipmmu_domain_allocate_context(domain->mmu->root, domain);
443 	if (ret < 0)
444 		return ret;
445 
446 	domain->context_id = ret;
447 
448 	domain->iop = alloc_io_pgtable_ops(ARM_32_LPAE_S1, &domain->cfg,
449 					   domain);
450 	if (!domain->iop) {
451 		ipmmu_domain_free_context(domain->mmu->root,
452 					  domain->context_id);
453 		return -EINVAL;
454 	}
455 
456 	ipmmu_domain_setup_context(domain);
457 	return 0;
458 }
459 
460 static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
461 {
462 	if (!domain->mmu)
463 		return;
464 
465 	/*
466 	 * Disable the context. Flush the TLB as required when modifying the
467 	 * context registers.
468 	 *
469 	 * TODO: Is TLB flush really needed ?
470 	 */
471 	ipmmu_ctx_write_all(domain, IMCTR, IMCTR_FLUSH);
472 	ipmmu_tlb_sync(domain);
473 	ipmmu_domain_free_context(domain->mmu->root, domain->context_id);
474 }
475 
476 /* -----------------------------------------------------------------------------
477  * Fault Handling
478  */
479 
480 static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
481 {
482 	const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
483 	struct ipmmu_vmsa_device *mmu = domain->mmu;
484 	unsigned long iova;
485 	u32 status;
486 
487 	status = ipmmu_ctx_read_root(domain, IMSTR);
488 	if (!(status & err_mask))
489 		return IRQ_NONE;
490 
491 	iova = ipmmu_ctx_read_root(domain, IMELAR);
492 	if (IS_ENABLED(CONFIG_64BIT))
493 		iova |= (u64)ipmmu_ctx_read_root(domain, IMEUAR) << 32;
494 
495 	/*
496 	 * Clear the error status flags. Unlike traditional interrupt flag
497 	 * registers that must be cleared by writing 1, this status register
498 	 * seems to require 0. The error address register must be read before,
499 	 * otherwise its value will be 0.
500 	 */
501 	ipmmu_ctx_write_root(domain, IMSTR, 0);
502 
503 	/* Log fatal errors. */
504 	if (status & IMSTR_MHIT)
505 		dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%lx\n",
506 				    iova);
507 	if (status & IMSTR_ABORT)
508 		dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%lx\n",
509 				    iova);
510 
511 	if (!(status & (IMSTR_PF | IMSTR_TF)))
512 		return IRQ_NONE;
513 
514 	/*
515 	 * Try to handle page faults and translation faults.
516 	 *
517 	 * TODO: We need to look up the faulty device based on the I/O VA. Use
518 	 * the IOMMU device for now.
519 	 */
520 	if (!report_iommu_fault(&domain->io_domain, mmu->dev, iova, 0))
521 		return IRQ_HANDLED;
522 
523 	dev_err_ratelimited(mmu->dev,
524 			    "Unhandled fault: status 0x%08x iova 0x%lx\n",
525 			    status, iova);
526 
527 	return IRQ_HANDLED;
528 }
529 
530 static irqreturn_t ipmmu_irq(int irq, void *dev)
531 {
532 	struct ipmmu_vmsa_device *mmu = dev;
533 	irqreturn_t status = IRQ_NONE;
534 	unsigned int i;
535 	unsigned long flags;
536 
537 	spin_lock_irqsave(&mmu->lock, flags);
538 
539 	/*
540 	 * Check interrupts for all active contexts.
541 	 */
542 	for (i = 0; i < mmu->num_ctx; i++) {
543 		if (!mmu->domains[i])
544 			continue;
545 		if (ipmmu_domain_irq(mmu->domains[i]) == IRQ_HANDLED)
546 			status = IRQ_HANDLED;
547 	}
548 
549 	spin_unlock_irqrestore(&mmu->lock, flags);
550 
551 	return status;
552 }
553 
554 /* -----------------------------------------------------------------------------
555  * IOMMU Operations
556  */
557 
558 static struct iommu_domain *ipmmu_domain_alloc(unsigned type)
559 {
560 	struct ipmmu_vmsa_domain *domain;
561 
562 	if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
563 		return NULL;
564 
565 	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
566 	if (!domain)
567 		return NULL;
568 
569 	mutex_init(&domain->mutex);
570 
571 	return &domain->io_domain;
572 }
573 
574 static void ipmmu_domain_free(struct iommu_domain *io_domain)
575 {
576 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
577 
578 	/*
579 	 * Free the domain resources. We assume that all devices have already
580 	 * been detached.
581 	 */
582 	ipmmu_domain_destroy_context(domain);
583 	free_io_pgtable_ops(domain->iop);
584 	kfree(domain);
585 }
586 
587 static int ipmmu_attach_device(struct iommu_domain *io_domain,
588 			       struct device *dev)
589 {
590 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
591 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
592 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
593 	unsigned int i;
594 	int ret = 0;
595 
596 	if (!mmu) {
597 		dev_err(dev, "Cannot attach to IPMMU\n");
598 		return -ENXIO;
599 	}
600 
601 	mutex_lock(&domain->mutex);
602 
603 	if (!domain->mmu) {
604 		/* The domain hasn't been used yet, initialize it. */
605 		domain->mmu = mmu;
606 		ret = ipmmu_domain_init_context(domain);
607 		if (ret < 0) {
608 			dev_err(dev, "Unable to initialize IPMMU context\n");
609 			domain->mmu = NULL;
610 		} else {
611 			dev_info(dev, "Using IPMMU context %u\n",
612 				 domain->context_id);
613 		}
614 	} else if (domain->mmu != mmu) {
615 		/*
616 		 * Something is wrong, we can't attach two devices using
617 		 * different IOMMUs to the same domain.
618 		 */
619 		ret = -EINVAL;
620 	} else
621 		dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
622 
623 	mutex_unlock(&domain->mutex);
624 
625 	if (ret < 0)
626 		return ret;
627 
628 	for (i = 0; i < fwspec->num_ids; ++i)
629 		ipmmu_utlb_enable(domain, fwspec->ids[i]);
630 
631 	return 0;
632 }
633 
634 static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
635 		     phys_addr_t paddr, size_t pgsize, size_t pgcount,
636 		     int prot, gfp_t gfp, size_t *mapped)
637 {
638 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
639 
640 	return domain->iop->map_pages(domain->iop, iova, paddr, pgsize, pgcount,
641 				      prot, gfp, mapped);
642 }
643 
644 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
645 			  size_t pgsize, size_t pgcount,
646 			  struct iommu_iotlb_gather *gather)
647 {
648 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
649 
650 	return domain->iop->unmap_pages(domain->iop, iova, pgsize, pgcount, gather);
651 }
652 
653 static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
654 {
655 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
656 
657 	if (domain->mmu)
658 		ipmmu_tlb_flush_all(domain);
659 }
660 
661 static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
662 			     struct iommu_iotlb_gather *gather)
663 {
664 	ipmmu_flush_iotlb_all(io_domain);
665 }
666 
667 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
668 				      dma_addr_t iova)
669 {
670 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
671 
672 	/* TODO: Is locking needed ? */
673 
674 	return domain->iop->iova_to_phys(domain->iop, iova);
675 }
676 
677 static int ipmmu_init_platform_device(struct device *dev,
678 				      struct of_phandle_args *args)
679 {
680 	struct platform_device *ipmmu_pdev;
681 
682 	ipmmu_pdev = of_find_device_by_node(args->np);
683 	if (!ipmmu_pdev)
684 		return -ENODEV;
685 
686 	dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
687 
688 	return 0;
689 }
690 
691 static const struct soc_device_attribute soc_needs_opt_in[] = {
692 	{ .family = "R-Car Gen3", },
693 	{ .family = "R-Car Gen4", },
694 	{ .family = "RZ/G2", },
695 	{ /* sentinel */ }
696 };
697 
698 static const struct soc_device_attribute soc_denylist[] = {
699 	{ .soc_id = "r8a774a1", },
700 	{ .soc_id = "r8a7795", .revision = "ES1.*" },
701 	{ .soc_id = "r8a7795", .revision = "ES2.*" },
702 	{ .soc_id = "r8a7796", },
703 	{ /* sentinel */ }
704 };
705 
706 static const char * const devices_allowlist[] = {
707 	"ee100000.mmc",
708 	"ee120000.mmc",
709 	"ee140000.mmc",
710 	"ee160000.mmc"
711 };
712 
713 static bool ipmmu_device_is_allowed(struct device *dev)
714 {
715 	unsigned int i;
716 
717 	/*
718 	 * R-Car Gen3/4 and RZ/G2 use the allow list to opt-in devices.
719 	 * For Other SoCs, this returns true anyway.
720 	 */
721 	if (!soc_device_match(soc_needs_opt_in))
722 		return true;
723 
724 	/* Check whether this SoC can use the IPMMU correctly or not */
725 	if (soc_device_match(soc_denylist))
726 		return false;
727 
728 	/* Check whether this device can work with the IPMMU */
729 	for (i = 0; i < ARRAY_SIZE(devices_allowlist); i++) {
730 		if (!strcmp(dev_name(dev), devices_allowlist[i]))
731 			return true;
732 	}
733 
734 	/* Otherwise, do not allow use of IPMMU */
735 	return false;
736 }
737 
738 static int ipmmu_of_xlate(struct device *dev,
739 			  struct of_phandle_args *spec)
740 {
741 	if (!ipmmu_device_is_allowed(dev))
742 		return -ENODEV;
743 
744 	iommu_fwspec_add_ids(dev, spec->args, 1);
745 
746 	/* Initialize once - xlate() will call multiple times */
747 	if (to_ipmmu(dev))
748 		return 0;
749 
750 	return ipmmu_init_platform_device(dev, spec);
751 }
752 
753 static int ipmmu_init_arm_mapping(struct device *dev)
754 {
755 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
756 	int ret;
757 
758 	/*
759 	 * Create the ARM mapping, used by the ARM DMA mapping core to allocate
760 	 * VAs. This will allocate a corresponding IOMMU domain.
761 	 *
762 	 * TODO:
763 	 * - Create one mapping per context (TLB).
764 	 * - Make the mapping size configurable ? We currently use a 2GB mapping
765 	 *   at a 1GB offset to ensure that NULL VAs will fault.
766 	 */
767 	if (!mmu->mapping) {
768 		struct dma_iommu_mapping *mapping;
769 
770 		mapping = arm_iommu_create_mapping(&platform_bus_type,
771 						   SZ_1G, SZ_2G);
772 		if (IS_ERR(mapping)) {
773 			dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
774 			ret = PTR_ERR(mapping);
775 			goto error;
776 		}
777 
778 		mmu->mapping = mapping;
779 	}
780 
781 	/* Attach the ARM VA mapping to the device. */
782 	ret = arm_iommu_attach_device(dev, mmu->mapping);
783 	if (ret < 0) {
784 		dev_err(dev, "Failed to attach device to VA mapping\n");
785 		goto error;
786 	}
787 
788 	return 0;
789 
790 error:
791 	if (mmu->mapping)
792 		arm_iommu_release_mapping(mmu->mapping);
793 
794 	return ret;
795 }
796 
797 static struct iommu_device *ipmmu_probe_device(struct device *dev)
798 {
799 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
800 
801 	/*
802 	 * Only let through devices that have been verified in xlate()
803 	 */
804 	if (!mmu)
805 		return ERR_PTR(-ENODEV);
806 
807 	return &mmu->iommu;
808 }
809 
810 static void ipmmu_probe_finalize(struct device *dev)
811 {
812 	int ret = 0;
813 
814 	if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA))
815 		ret = ipmmu_init_arm_mapping(dev);
816 
817 	if (ret)
818 		dev_err(dev, "Can't create IOMMU mapping - DMA-OPS will not work\n");
819 }
820 
821 static void ipmmu_release_device(struct device *dev)
822 {
823 	arm_iommu_detach_device(dev);
824 }
825 
826 static struct iommu_group *ipmmu_find_group(struct device *dev)
827 {
828 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
829 	struct iommu_group *group;
830 
831 	if (mmu->group)
832 		return iommu_group_ref_get(mmu->group);
833 
834 	group = iommu_group_alloc();
835 	if (!IS_ERR(group))
836 		mmu->group = group;
837 
838 	return group;
839 }
840 
841 static const struct iommu_ops ipmmu_ops = {
842 	.domain_alloc = ipmmu_domain_alloc,
843 	.probe_device = ipmmu_probe_device,
844 	.release_device = ipmmu_release_device,
845 	.probe_finalize = ipmmu_probe_finalize,
846 	.device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
847 			? generic_device_group : ipmmu_find_group,
848 	.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
849 	.of_xlate = ipmmu_of_xlate,
850 	.default_domain_ops = &(const struct iommu_domain_ops) {
851 		.attach_dev	= ipmmu_attach_device,
852 		.map_pages	= ipmmu_map,
853 		.unmap_pages	= ipmmu_unmap,
854 		.flush_iotlb_all = ipmmu_flush_iotlb_all,
855 		.iotlb_sync	= ipmmu_iotlb_sync,
856 		.iova_to_phys	= ipmmu_iova_to_phys,
857 		.free		= ipmmu_domain_free,
858 	}
859 };
860 
861 /* -----------------------------------------------------------------------------
862  * Probe/remove and init
863  */
864 
865 static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
866 {
867 	unsigned int i;
868 
869 	/* Disable all contexts. */
870 	for (i = 0; i < mmu->num_ctx; ++i)
871 		ipmmu_ctx_write(mmu, i, IMCTR, 0);
872 }
873 
874 static const struct ipmmu_features ipmmu_features_default = {
875 	.use_ns_alias_offset = true,
876 	.has_cache_leaf_nodes = false,
877 	.number_of_contexts = 1, /* software only tested with one context */
878 	.num_utlbs = 32,
879 	.setup_imbuscr = true,
880 	.twobit_imttbcr_sl0 = false,
881 	.reserved_context = false,
882 	.cache_snoop = true,
883 	.ctx_offset_base = 0,
884 	.ctx_offset_stride = 0x40,
885 	.utlb_offset_base = 0,
886 };
887 
888 static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
889 	.use_ns_alias_offset = false,
890 	.has_cache_leaf_nodes = true,
891 	.number_of_contexts = 8,
892 	.num_utlbs = 48,
893 	.setup_imbuscr = false,
894 	.twobit_imttbcr_sl0 = true,
895 	.reserved_context = true,
896 	.cache_snoop = false,
897 	.ctx_offset_base = 0,
898 	.ctx_offset_stride = 0x40,
899 	.utlb_offset_base = 0,
900 };
901 
902 static const struct ipmmu_features ipmmu_features_rcar_gen4 = {
903 	.use_ns_alias_offset = false,
904 	.has_cache_leaf_nodes = true,
905 	.number_of_contexts = 16,
906 	.num_utlbs = 64,
907 	.setup_imbuscr = false,
908 	.twobit_imttbcr_sl0 = true,
909 	.reserved_context = true,
910 	.cache_snoop = false,
911 	.ctx_offset_base = 0x10000,
912 	.ctx_offset_stride = 0x1040,
913 	.utlb_offset_base = 0x3000,
914 };
915 
916 static const struct of_device_id ipmmu_of_ids[] = {
917 	{
918 		.compatible = "renesas,ipmmu-vmsa",
919 		.data = &ipmmu_features_default,
920 	}, {
921 		.compatible = "renesas,ipmmu-r8a774a1",
922 		.data = &ipmmu_features_rcar_gen3,
923 	}, {
924 		.compatible = "renesas,ipmmu-r8a774b1",
925 		.data = &ipmmu_features_rcar_gen3,
926 	}, {
927 		.compatible = "renesas,ipmmu-r8a774c0",
928 		.data = &ipmmu_features_rcar_gen3,
929 	}, {
930 		.compatible = "renesas,ipmmu-r8a774e1",
931 		.data = &ipmmu_features_rcar_gen3,
932 	}, {
933 		.compatible = "renesas,ipmmu-r8a7795",
934 		.data = &ipmmu_features_rcar_gen3,
935 	}, {
936 		.compatible = "renesas,ipmmu-r8a7796",
937 		.data = &ipmmu_features_rcar_gen3,
938 	}, {
939 		.compatible = "renesas,ipmmu-r8a77961",
940 		.data = &ipmmu_features_rcar_gen3,
941 	}, {
942 		.compatible = "renesas,ipmmu-r8a77965",
943 		.data = &ipmmu_features_rcar_gen3,
944 	}, {
945 		.compatible = "renesas,ipmmu-r8a77970",
946 		.data = &ipmmu_features_rcar_gen3,
947 	}, {
948 		.compatible = "renesas,ipmmu-r8a77980",
949 		.data = &ipmmu_features_rcar_gen3,
950 	}, {
951 		.compatible = "renesas,ipmmu-r8a77990",
952 		.data = &ipmmu_features_rcar_gen3,
953 	}, {
954 		.compatible = "renesas,ipmmu-r8a77995",
955 		.data = &ipmmu_features_rcar_gen3,
956 	}, {
957 		.compatible = "renesas,ipmmu-r8a779a0",
958 		.data = &ipmmu_features_rcar_gen4,
959 	}, {
960 		.compatible = "renesas,rcar-gen4-ipmmu-vmsa",
961 		.data = &ipmmu_features_rcar_gen4,
962 	}, {
963 		/* Terminator */
964 	},
965 };
966 
967 static int ipmmu_probe(struct platform_device *pdev)
968 {
969 	struct ipmmu_vmsa_device *mmu;
970 	struct resource *res;
971 	int irq;
972 	int ret;
973 
974 	mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
975 	if (!mmu) {
976 		dev_err(&pdev->dev, "cannot allocate device data\n");
977 		return -ENOMEM;
978 	}
979 
980 	mmu->dev = &pdev->dev;
981 	spin_lock_init(&mmu->lock);
982 	bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
983 	mmu->features = of_device_get_match_data(&pdev->dev);
984 	memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
985 	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
986 	if (ret)
987 		return ret;
988 
989 	/* Map I/O memory and request IRQ. */
990 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
991 	mmu->base = devm_ioremap_resource(&pdev->dev, res);
992 	if (IS_ERR(mmu->base))
993 		return PTR_ERR(mmu->base);
994 
995 	/*
996 	 * The IPMMU has two register banks, for secure and non-secure modes.
997 	 * The bank mapped at the beginning of the IPMMU address space
998 	 * corresponds to the running mode of the CPU. When running in secure
999 	 * mode the non-secure register bank is also available at an offset.
1000 	 *
1001 	 * Secure mode operation isn't clearly documented and is thus currently
1002 	 * not implemented in the driver. Furthermore, preliminary tests of
1003 	 * non-secure operation with the main register bank were not successful.
1004 	 * Offset the registers base unconditionally to point to the non-secure
1005 	 * alias space for now.
1006 	 */
1007 	if (mmu->features->use_ns_alias_offset)
1008 		mmu->base += IM_NS_ALIAS_OFFSET;
1009 
1010 	mmu->num_ctx = min(IPMMU_CTX_MAX, mmu->features->number_of_contexts);
1011 
1012 	/*
1013 	 * Determine if this IPMMU instance is a root device by checking for
1014 	 * the lack of has_cache_leaf_nodes flag or renesas,ipmmu-main property.
1015 	 */
1016 	if (!mmu->features->has_cache_leaf_nodes ||
1017 	    !of_find_property(pdev->dev.of_node, "renesas,ipmmu-main", NULL))
1018 		mmu->root = mmu;
1019 	else
1020 		mmu->root = ipmmu_find_root();
1021 
1022 	/*
1023 	 * Wait until the root device has been registered for sure.
1024 	 */
1025 	if (!mmu->root)
1026 		return -EPROBE_DEFER;
1027 
1028 	/* Root devices have mandatory IRQs */
1029 	if (ipmmu_is_root(mmu)) {
1030 		irq = platform_get_irq(pdev, 0);
1031 		if (irq < 0)
1032 			return irq;
1033 
1034 		ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
1035 				       dev_name(&pdev->dev), mmu);
1036 		if (ret < 0) {
1037 			dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
1038 			return ret;
1039 		}
1040 
1041 		ipmmu_device_reset(mmu);
1042 
1043 		if (mmu->features->reserved_context) {
1044 			dev_info(&pdev->dev, "IPMMU context 0 is reserved\n");
1045 			set_bit(0, mmu->ctx);
1046 		}
1047 	}
1048 
1049 	/*
1050 	 * Register the IPMMU to the IOMMU subsystem in the following cases:
1051 	 * - R-Car Gen2 IPMMU (all devices registered)
1052 	 * - R-Car Gen3 IPMMU (leaf devices only - skip root IPMMU-MM device)
1053 	 */
1054 	if (!mmu->features->has_cache_leaf_nodes || !ipmmu_is_root(mmu)) {
1055 		ret = iommu_device_sysfs_add(&mmu->iommu, &pdev->dev, NULL,
1056 					     dev_name(&pdev->dev));
1057 		if (ret)
1058 			return ret;
1059 
1060 		ret = iommu_device_register(&mmu->iommu, &ipmmu_ops, &pdev->dev);
1061 		if (ret)
1062 			return ret;
1063 	}
1064 
1065 	/*
1066 	 * We can't create the ARM mapping here as it requires the bus to have
1067 	 * an IOMMU, which only happens when bus_set_iommu() is called in
1068 	 * ipmmu_init() after the probe function returns.
1069 	 */
1070 
1071 	platform_set_drvdata(pdev, mmu);
1072 
1073 	return 0;
1074 }
1075 
1076 static int ipmmu_remove(struct platform_device *pdev)
1077 {
1078 	struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1079 
1080 	iommu_device_sysfs_remove(&mmu->iommu);
1081 	iommu_device_unregister(&mmu->iommu);
1082 
1083 	arm_iommu_release_mapping(mmu->mapping);
1084 
1085 	ipmmu_device_reset(mmu);
1086 
1087 	return 0;
1088 }
1089 
1090 #ifdef CONFIG_PM_SLEEP
1091 static int ipmmu_resume_noirq(struct device *dev)
1092 {
1093 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1094 	unsigned int i;
1095 
1096 	/* Reset root MMU and restore contexts */
1097 	if (ipmmu_is_root(mmu)) {
1098 		ipmmu_device_reset(mmu);
1099 
1100 		for (i = 0; i < mmu->num_ctx; i++) {
1101 			if (!mmu->domains[i])
1102 				continue;
1103 
1104 			ipmmu_domain_setup_context(mmu->domains[i]);
1105 		}
1106 	}
1107 
1108 	/* Re-enable active micro-TLBs */
1109 	for (i = 0; i < mmu->features->num_utlbs; i++) {
1110 		if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1111 			continue;
1112 
1113 		ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1114 	}
1115 
1116 	return 0;
1117 }
1118 
1119 static const struct dev_pm_ops ipmmu_pm  = {
1120 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1121 };
1122 #define DEV_PM_OPS	&ipmmu_pm
1123 #else
1124 #define DEV_PM_OPS	NULL
1125 #endif /* CONFIG_PM_SLEEP */
1126 
1127 static struct platform_driver ipmmu_driver = {
1128 	.driver = {
1129 		.name = "ipmmu-vmsa",
1130 		.of_match_table = of_match_ptr(ipmmu_of_ids),
1131 		.pm = DEV_PM_OPS,
1132 	},
1133 	.probe = ipmmu_probe,
1134 	.remove	= ipmmu_remove,
1135 };
1136 builtin_platform_driver(ipmmu_driver);
1137