xref: /openbmc/linux/drivers/iommu/ipmmu-vmsa.c (revision a1117495)
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 /*
303  * Disable MMU translation for the microTLB.
304  */
305 static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
306 			       unsigned int utlb)
307 {
308 	struct ipmmu_vmsa_device *mmu = domain->mmu;
309 
310 	ipmmu_imuctr_write(mmu, utlb, 0);
311 	mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
312 }
313 
314 static void ipmmu_tlb_flush_all(void *cookie)
315 {
316 	struct ipmmu_vmsa_domain *domain = cookie;
317 
318 	ipmmu_tlb_invalidate(domain);
319 }
320 
321 static void ipmmu_tlb_flush(unsigned long iova, size_t size,
322 				size_t granule, void *cookie)
323 {
324 	ipmmu_tlb_flush_all(cookie);
325 }
326 
327 static const struct iommu_flush_ops ipmmu_flush_ops = {
328 	.tlb_flush_all = ipmmu_tlb_flush_all,
329 	.tlb_flush_walk = 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 	if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
575 		return NULL;
576 
577 	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
578 	if (!domain)
579 		return NULL;
580 
581 	mutex_init(&domain->mutex);
582 
583 	return &domain->io_domain;
584 }
585 
586 static void ipmmu_domain_free(struct iommu_domain *io_domain)
587 {
588 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
589 
590 	/*
591 	 * Free the domain resources. We assume that all devices have already
592 	 * been detached.
593 	 */
594 	ipmmu_domain_destroy_context(domain);
595 	free_io_pgtable_ops(domain->iop);
596 	kfree(domain);
597 }
598 
599 static int ipmmu_attach_device(struct iommu_domain *io_domain,
600 			       struct device *dev)
601 {
602 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
603 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
604 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
605 	unsigned int i;
606 	int ret = 0;
607 
608 	if (!mmu) {
609 		dev_err(dev, "Cannot attach to IPMMU\n");
610 		return -ENXIO;
611 	}
612 
613 	mutex_lock(&domain->mutex);
614 
615 	if (!domain->mmu) {
616 		/* The domain hasn't been used yet, initialize it. */
617 		domain->mmu = mmu;
618 		ret = ipmmu_domain_init_context(domain);
619 		if (ret < 0) {
620 			dev_err(dev, "Unable to initialize IPMMU context\n");
621 			domain->mmu = NULL;
622 		} else {
623 			dev_info(dev, "Using IPMMU context %u\n",
624 				 domain->context_id);
625 		}
626 	} else if (domain->mmu != mmu) {
627 		/*
628 		 * Something is wrong, we can't attach two devices using
629 		 * different IOMMUs to the same domain.
630 		 */
631 		dev_err(dev, "Can't attach IPMMU %s to domain on IPMMU %s\n",
632 			dev_name(mmu->dev), dev_name(domain->mmu->dev));
633 		ret = -EINVAL;
634 	} else
635 		dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
636 
637 	mutex_unlock(&domain->mutex);
638 
639 	if (ret < 0)
640 		return ret;
641 
642 	for (i = 0; i < fwspec->num_ids; ++i)
643 		ipmmu_utlb_enable(domain, fwspec->ids[i]);
644 
645 	return 0;
646 }
647 
648 static void ipmmu_detach_device(struct iommu_domain *io_domain,
649 				struct device *dev)
650 {
651 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
652 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
653 	unsigned int i;
654 
655 	for (i = 0; i < fwspec->num_ids; ++i)
656 		ipmmu_utlb_disable(domain, fwspec->ids[i]);
657 
658 	/*
659 	 * TODO: Optimize by disabling the context when no device is attached.
660 	 */
661 }
662 
663 static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
664 		     phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
665 {
666 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
667 
668 	if (!domain)
669 		return -ENODEV;
670 
671 	return domain->iop->map(domain->iop, iova, paddr, size, prot, gfp);
672 }
673 
674 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
675 			  size_t size, struct iommu_iotlb_gather *gather)
676 {
677 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
678 
679 	return domain->iop->unmap(domain->iop, iova, size, gather);
680 }
681 
682 static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
683 {
684 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
685 
686 	if (domain->mmu)
687 		ipmmu_tlb_flush_all(domain);
688 }
689 
690 static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
691 			     struct iommu_iotlb_gather *gather)
692 {
693 	ipmmu_flush_iotlb_all(io_domain);
694 }
695 
696 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
697 				      dma_addr_t iova)
698 {
699 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
700 
701 	/* TODO: Is locking needed ? */
702 
703 	return domain->iop->iova_to_phys(domain->iop, iova);
704 }
705 
706 static int ipmmu_init_platform_device(struct device *dev,
707 				      struct of_phandle_args *args)
708 {
709 	struct platform_device *ipmmu_pdev;
710 
711 	ipmmu_pdev = of_find_device_by_node(args->np);
712 	if (!ipmmu_pdev)
713 		return -ENODEV;
714 
715 	dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
716 
717 	return 0;
718 }
719 
720 static const struct soc_device_attribute soc_needs_opt_in[] = {
721 	{ .family = "R-Car Gen3", },
722 	{ .family = "RZ/G2", },
723 	{ /* sentinel */ }
724 };
725 
726 static const struct soc_device_attribute soc_denylist[] = {
727 	{ .soc_id = "r8a774a1", },
728 	{ .soc_id = "r8a7795", .revision = "ES1.*" },
729 	{ .soc_id = "r8a7795", .revision = "ES2.*" },
730 	{ .soc_id = "r8a7796", },
731 	{ /* sentinel */ }
732 };
733 
734 static const char * const devices_allowlist[] = {
735 	"ee100000.mmc",
736 	"ee120000.mmc",
737 	"ee140000.mmc",
738 	"ee160000.mmc"
739 };
740 
741 static bool ipmmu_device_is_allowed(struct device *dev)
742 {
743 	unsigned int i;
744 
745 	/*
746 	 * R-Car Gen3 and RZ/G2 use the allow list to opt-in devices.
747 	 * For Other SoCs, this returns true anyway.
748 	 */
749 	if (!soc_device_match(soc_needs_opt_in))
750 		return true;
751 
752 	/* Check whether this SoC can use the IPMMU correctly or not */
753 	if (soc_device_match(soc_denylist))
754 		return false;
755 
756 	/* Check whether this device can work with the IPMMU */
757 	for (i = 0; i < ARRAY_SIZE(devices_allowlist); i++) {
758 		if (!strcmp(dev_name(dev), devices_allowlist[i]))
759 			return true;
760 	}
761 
762 	/* Otherwise, do not allow use of IPMMU */
763 	return false;
764 }
765 
766 static int ipmmu_of_xlate(struct device *dev,
767 			  struct of_phandle_args *spec)
768 {
769 	if (!ipmmu_device_is_allowed(dev))
770 		return -ENODEV;
771 
772 	iommu_fwspec_add_ids(dev, spec->args, 1);
773 
774 	/* Initialize once - xlate() will call multiple times */
775 	if (to_ipmmu(dev))
776 		return 0;
777 
778 	return ipmmu_init_platform_device(dev, spec);
779 }
780 
781 static int ipmmu_init_arm_mapping(struct device *dev)
782 {
783 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
784 	int ret;
785 
786 	/*
787 	 * Create the ARM mapping, used by the ARM DMA mapping core to allocate
788 	 * VAs. This will allocate a corresponding IOMMU domain.
789 	 *
790 	 * TODO:
791 	 * - Create one mapping per context (TLB).
792 	 * - Make the mapping size configurable ? We currently use a 2GB mapping
793 	 *   at a 1GB offset to ensure that NULL VAs will fault.
794 	 */
795 	if (!mmu->mapping) {
796 		struct dma_iommu_mapping *mapping;
797 
798 		mapping = arm_iommu_create_mapping(&platform_bus_type,
799 						   SZ_1G, SZ_2G);
800 		if (IS_ERR(mapping)) {
801 			dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
802 			ret = PTR_ERR(mapping);
803 			goto error;
804 		}
805 
806 		mmu->mapping = mapping;
807 	}
808 
809 	/* Attach the ARM VA mapping to the device. */
810 	ret = arm_iommu_attach_device(dev, mmu->mapping);
811 	if (ret < 0) {
812 		dev_err(dev, "Failed to attach device to VA mapping\n");
813 		goto error;
814 	}
815 
816 	return 0;
817 
818 error:
819 	if (mmu->mapping)
820 		arm_iommu_release_mapping(mmu->mapping);
821 
822 	return ret;
823 }
824 
825 static struct iommu_device *ipmmu_probe_device(struct device *dev)
826 {
827 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
828 
829 	/*
830 	 * Only let through devices that have been verified in xlate()
831 	 */
832 	if (!mmu)
833 		return ERR_PTR(-ENODEV);
834 
835 	return &mmu->iommu;
836 }
837 
838 static void ipmmu_probe_finalize(struct device *dev)
839 {
840 	int ret = 0;
841 
842 	if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA))
843 		ret = ipmmu_init_arm_mapping(dev);
844 
845 	if (ret)
846 		dev_err(dev, "Can't create IOMMU mapping - DMA-OPS will not work\n");
847 }
848 
849 static void ipmmu_release_device(struct device *dev)
850 {
851 	arm_iommu_detach_device(dev);
852 }
853 
854 static struct iommu_group *ipmmu_find_group(struct device *dev)
855 {
856 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
857 	struct iommu_group *group;
858 
859 	if (mmu->group)
860 		return iommu_group_ref_get(mmu->group);
861 
862 	group = iommu_group_alloc();
863 	if (!IS_ERR(group))
864 		mmu->group = group;
865 
866 	return group;
867 }
868 
869 static const struct iommu_ops ipmmu_ops = {
870 	.domain_alloc = ipmmu_domain_alloc,
871 	.domain_free = ipmmu_domain_free,
872 	.attach_dev = ipmmu_attach_device,
873 	.detach_dev = ipmmu_detach_device,
874 	.map = ipmmu_map,
875 	.unmap = ipmmu_unmap,
876 	.flush_iotlb_all = ipmmu_flush_iotlb_all,
877 	.iotlb_sync = ipmmu_iotlb_sync,
878 	.iova_to_phys = ipmmu_iova_to_phys,
879 	.probe_device = ipmmu_probe_device,
880 	.release_device = ipmmu_release_device,
881 	.probe_finalize = ipmmu_probe_finalize,
882 	.device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
883 			? generic_device_group : ipmmu_find_group,
884 	.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
885 	.of_xlate = ipmmu_of_xlate,
886 };
887 
888 /* -----------------------------------------------------------------------------
889  * Probe/remove and init
890  */
891 
892 static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
893 {
894 	unsigned int i;
895 
896 	/* Disable all contexts. */
897 	for (i = 0; i < mmu->num_ctx; ++i)
898 		ipmmu_ctx_write(mmu, i, IMCTR, 0);
899 }
900 
901 static const struct ipmmu_features ipmmu_features_default = {
902 	.use_ns_alias_offset = true,
903 	.has_cache_leaf_nodes = false,
904 	.number_of_contexts = 1, /* software only tested with one context */
905 	.num_utlbs = 32,
906 	.setup_imbuscr = true,
907 	.twobit_imttbcr_sl0 = false,
908 	.reserved_context = false,
909 	.cache_snoop = true,
910 	.ctx_offset_base = 0,
911 	.ctx_offset_stride = 0x40,
912 	.utlb_offset_base = 0,
913 };
914 
915 static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
916 	.use_ns_alias_offset = false,
917 	.has_cache_leaf_nodes = true,
918 	.number_of_contexts = 8,
919 	.num_utlbs = 48,
920 	.setup_imbuscr = false,
921 	.twobit_imttbcr_sl0 = true,
922 	.reserved_context = true,
923 	.cache_snoop = false,
924 	.ctx_offset_base = 0,
925 	.ctx_offset_stride = 0x40,
926 	.utlb_offset_base = 0,
927 };
928 
929 static const struct ipmmu_features ipmmu_features_r8a779a0 = {
930 	.use_ns_alias_offset = false,
931 	.has_cache_leaf_nodes = true,
932 	.number_of_contexts = 16,
933 	.num_utlbs = 64,
934 	.setup_imbuscr = false,
935 	.twobit_imttbcr_sl0 = true,
936 	.reserved_context = true,
937 	.cache_snoop = false,
938 	.ctx_offset_base = 0x10000,
939 	.ctx_offset_stride = 0x1040,
940 	.utlb_offset_base = 0x3000,
941 };
942 
943 static const struct of_device_id ipmmu_of_ids[] = {
944 	{
945 		.compatible = "renesas,ipmmu-vmsa",
946 		.data = &ipmmu_features_default,
947 	}, {
948 		.compatible = "renesas,ipmmu-r8a774a1",
949 		.data = &ipmmu_features_rcar_gen3,
950 	}, {
951 		.compatible = "renesas,ipmmu-r8a774b1",
952 		.data = &ipmmu_features_rcar_gen3,
953 	}, {
954 		.compatible = "renesas,ipmmu-r8a774c0",
955 		.data = &ipmmu_features_rcar_gen3,
956 	}, {
957 		.compatible = "renesas,ipmmu-r8a774e1",
958 		.data = &ipmmu_features_rcar_gen3,
959 	}, {
960 		.compatible = "renesas,ipmmu-r8a7795",
961 		.data = &ipmmu_features_rcar_gen3,
962 	}, {
963 		.compatible = "renesas,ipmmu-r8a7796",
964 		.data = &ipmmu_features_rcar_gen3,
965 	}, {
966 		.compatible = "renesas,ipmmu-r8a77961",
967 		.data = &ipmmu_features_rcar_gen3,
968 	}, {
969 		.compatible = "renesas,ipmmu-r8a77965",
970 		.data = &ipmmu_features_rcar_gen3,
971 	}, {
972 		.compatible = "renesas,ipmmu-r8a77970",
973 		.data = &ipmmu_features_rcar_gen3,
974 	}, {
975 		.compatible = "renesas,ipmmu-r8a77980",
976 		.data = &ipmmu_features_rcar_gen3,
977 	}, {
978 		.compatible = "renesas,ipmmu-r8a77990",
979 		.data = &ipmmu_features_rcar_gen3,
980 	}, {
981 		.compatible = "renesas,ipmmu-r8a77995",
982 		.data = &ipmmu_features_rcar_gen3,
983 	}, {
984 		.compatible = "renesas,ipmmu-r8a779a0",
985 		.data = &ipmmu_features_r8a779a0,
986 	}, {
987 		/* Terminator */
988 	},
989 };
990 
991 static int ipmmu_probe(struct platform_device *pdev)
992 {
993 	struct ipmmu_vmsa_device *mmu;
994 	struct resource *res;
995 	int irq;
996 	int ret;
997 
998 	mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
999 	if (!mmu) {
1000 		dev_err(&pdev->dev, "cannot allocate device data\n");
1001 		return -ENOMEM;
1002 	}
1003 
1004 	mmu->dev = &pdev->dev;
1005 	spin_lock_init(&mmu->lock);
1006 	bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
1007 	mmu->features = of_device_get_match_data(&pdev->dev);
1008 	memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
1009 	dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
1010 
1011 	/* Map I/O memory and request IRQ. */
1012 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1013 	mmu->base = devm_ioremap_resource(&pdev->dev, res);
1014 	if (IS_ERR(mmu->base))
1015 		return PTR_ERR(mmu->base);
1016 
1017 	/*
1018 	 * The IPMMU has two register banks, for secure and non-secure modes.
1019 	 * The bank mapped at the beginning of the IPMMU address space
1020 	 * corresponds to the running mode of the CPU. When running in secure
1021 	 * mode the non-secure register bank is also available at an offset.
1022 	 *
1023 	 * Secure mode operation isn't clearly documented and is thus currently
1024 	 * not implemented in the driver. Furthermore, preliminary tests of
1025 	 * non-secure operation with the main register bank were not successful.
1026 	 * Offset the registers base unconditionally to point to the non-secure
1027 	 * alias space for now.
1028 	 */
1029 	if (mmu->features->use_ns_alias_offset)
1030 		mmu->base += IM_NS_ALIAS_OFFSET;
1031 
1032 	mmu->num_ctx = min(IPMMU_CTX_MAX, mmu->features->number_of_contexts);
1033 
1034 	/*
1035 	 * Determine if this IPMMU instance is a root device by checking for
1036 	 * the lack of has_cache_leaf_nodes flag or renesas,ipmmu-main property.
1037 	 */
1038 	if (!mmu->features->has_cache_leaf_nodes ||
1039 	    !of_find_property(pdev->dev.of_node, "renesas,ipmmu-main", NULL))
1040 		mmu->root = mmu;
1041 	else
1042 		mmu->root = ipmmu_find_root();
1043 
1044 	/*
1045 	 * Wait until the root device has been registered for sure.
1046 	 */
1047 	if (!mmu->root)
1048 		return -EPROBE_DEFER;
1049 
1050 	/* Root devices have mandatory IRQs */
1051 	if (ipmmu_is_root(mmu)) {
1052 		irq = platform_get_irq(pdev, 0);
1053 		if (irq < 0)
1054 			return irq;
1055 
1056 		ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
1057 				       dev_name(&pdev->dev), mmu);
1058 		if (ret < 0) {
1059 			dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
1060 			return ret;
1061 		}
1062 
1063 		ipmmu_device_reset(mmu);
1064 
1065 		if (mmu->features->reserved_context) {
1066 			dev_info(&pdev->dev, "IPMMU context 0 is reserved\n");
1067 			set_bit(0, mmu->ctx);
1068 		}
1069 	}
1070 
1071 	/*
1072 	 * Register the IPMMU to the IOMMU subsystem in the following cases:
1073 	 * - R-Car Gen2 IPMMU (all devices registered)
1074 	 * - R-Car Gen3 IPMMU (leaf devices only - skip root IPMMU-MM device)
1075 	 */
1076 	if (!mmu->features->has_cache_leaf_nodes || !ipmmu_is_root(mmu)) {
1077 		ret = iommu_device_sysfs_add(&mmu->iommu, &pdev->dev, NULL,
1078 					     dev_name(&pdev->dev));
1079 		if (ret)
1080 			return ret;
1081 
1082 		ret = iommu_device_register(&mmu->iommu, &ipmmu_ops, &pdev->dev);
1083 		if (ret)
1084 			return ret;
1085 
1086 #if defined(CONFIG_IOMMU_DMA)
1087 		if (!iommu_present(&platform_bus_type))
1088 			bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1089 #endif
1090 	}
1091 
1092 	/*
1093 	 * We can't create the ARM mapping here as it requires the bus to have
1094 	 * an IOMMU, which only happens when bus_set_iommu() is called in
1095 	 * ipmmu_init() after the probe function returns.
1096 	 */
1097 
1098 	platform_set_drvdata(pdev, mmu);
1099 
1100 	return 0;
1101 }
1102 
1103 static int ipmmu_remove(struct platform_device *pdev)
1104 {
1105 	struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1106 
1107 	iommu_device_sysfs_remove(&mmu->iommu);
1108 	iommu_device_unregister(&mmu->iommu);
1109 
1110 	arm_iommu_release_mapping(mmu->mapping);
1111 
1112 	ipmmu_device_reset(mmu);
1113 
1114 	return 0;
1115 }
1116 
1117 #ifdef CONFIG_PM_SLEEP
1118 static int ipmmu_resume_noirq(struct device *dev)
1119 {
1120 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1121 	unsigned int i;
1122 
1123 	/* Reset root MMU and restore contexts */
1124 	if (ipmmu_is_root(mmu)) {
1125 		ipmmu_device_reset(mmu);
1126 
1127 		for (i = 0; i < mmu->num_ctx; i++) {
1128 			if (!mmu->domains[i])
1129 				continue;
1130 
1131 			ipmmu_domain_setup_context(mmu->domains[i]);
1132 		}
1133 	}
1134 
1135 	/* Re-enable active micro-TLBs */
1136 	for (i = 0; i < mmu->features->num_utlbs; i++) {
1137 		if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1138 			continue;
1139 
1140 		ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1141 	}
1142 
1143 	return 0;
1144 }
1145 
1146 static const struct dev_pm_ops ipmmu_pm  = {
1147 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1148 };
1149 #define DEV_PM_OPS	&ipmmu_pm
1150 #else
1151 #define DEV_PM_OPS	NULL
1152 #endif /* CONFIG_PM_SLEEP */
1153 
1154 static struct platform_driver ipmmu_driver = {
1155 	.driver = {
1156 		.name = "ipmmu-vmsa",
1157 		.of_match_table = of_match_ptr(ipmmu_of_ids),
1158 		.pm = DEV_PM_OPS,
1159 	},
1160 	.probe = ipmmu_probe,
1161 	.remove	= ipmmu_remove,
1162 };
1163 
1164 static int __init ipmmu_init(void)
1165 {
1166 	struct device_node *np;
1167 	static bool setup_done;
1168 	int ret;
1169 
1170 	if (setup_done)
1171 		return 0;
1172 
1173 	np = of_find_matching_node(NULL, ipmmu_of_ids);
1174 	if (!np)
1175 		return 0;
1176 
1177 	of_node_put(np);
1178 
1179 	ret = platform_driver_register(&ipmmu_driver);
1180 	if (ret < 0)
1181 		return ret;
1182 
1183 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
1184 	if (!iommu_present(&platform_bus_type))
1185 		bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1186 #endif
1187 
1188 	setup_done = true;
1189 	return 0;
1190 }
1191 subsys_initcall(ipmmu_init);
1192