xref: /openbmc/linux/drivers/iommu/ipmmu-vmsa.c (revision ee7da21a)
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-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_platform.h>
23 #include <linux/platform_device.h>
24 #include <linux/sizes.h>
25 #include <linux/slab.h>
26 #include <linux/sys_soc.h>
27 
28 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
29 #include <asm/dma-iommu.h>
30 #else
31 #define arm_iommu_create_mapping(...)	NULL
32 #define arm_iommu_attach_device(...)	-ENODEV
33 #define arm_iommu_release_mapping(...)	do {} while (0)
34 #define arm_iommu_detach_device(...)	do {} while (0)
35 #endif
36 
37 #define IPMMU_CTX_MAX		8U
38 #define IPMMU_CTX_INVALID	-1
39 
40 #define IPMMU_UTLB_MAX		48U
41 
42 struct ipmmu_features {
43 	bool use_ns_alias_offset;
44 	bool has_cache_leaf_nodes;
45 	unsigned int number_of_contexts;
46 	unsigned int num_utlbs;
47 	bool setup_imbuscr;
48 	bool twobit_imttbcr_sl0;
49 	bool reserved_context;
50 	bool cache_snoop;
51 	unsigned int ctx_offset_base;
52 	unsigned int ctx_offset_stride;
53 	unsigned int utlb_offset_base;
54 };
55 
56 struct ipmmu_vmsa_device {
57 	struct device *dev;
58 	void __iomem *base;
59 	struct iommu_device iommu;
60 	struct ipmmu_vmsa_device *root;
61 	const struct ipmmu_features *features;
62 	unsigned int num_ctx;
63 	spinlock_t lock;			/* Protects ctx and domains[] */
64 	DECLARE_BITMAP(ctx, IPMMU_CTX_MAX);
65 	struct ipmmu_vmsa_domain *domains[IPMMU_CTX_MAX];
66 	s8 utlb_ctx[IPMMU_UTLB_MAX];
67 
68 	struct iommu_group *group;
69 	struct dma_iommu_mapping *mapping;
70 };
71 
72 struct ipmmu_vmsa_domain {
73 	struct ipmmu_vmsa_device *mmu;
74 	struct iommu_domain io_domain;
75 
76 	struct io_pgtable_cfg cfg;
77 	struct io_pgtable_ops *iop;
78 
79 	unsigned int context_id;
80 	struct mutex mutex;			/* Protects mappings */
81 };
82 
83 static struct ipmmu_vmsa_domain *to_vmsa_domain(struct iommu_domain *dom)
84 {
85 	return container_of(dom, struct ipmmu_vmsa_domain, io_domain);
86 }
87 
88 static struct ipmmu_vmsa_device *to_ipmmu(struct device *dev)
89 {
90 	return dev_iommu_priv_get(dev);
91 }
92 
93 #define TLB_LOOP_TIMEOUT		100	/* 100us */
94 
95 /* -----------------------------------------------------------------------------
96  * Registers Definition
97  */
98 
99 #define IM_NS_ALIAS_OFFSET		0x800
100 
101 /* MMU "context" registers */
102 #define IMCTR				0x0000		/* R-Car Gen2/3 */
103 #define IMCTR_INTEN			(1 << 2)	/* R-Car Gen2/3 */
104 #define IMCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
105 #define IMCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
106 
107 #define IMTTBCR				0x0008		/* R-Car Gen2/3 */
108 #define IMTTBCR_EAE			(1 << 31)	/* R-Car Gen2/3 */
109 #define IMTTBCR_SH0_INNER_SHAREABLE	(3 << 12)	/* R-Car Gen2 only */
110 #define IMTTBCR_ORGN0_WB_WA		(1 << 10)	/* R-Car Gen2 only */
111 #define IMTTBCR_IRGN0_WB_WA		(1 << 8)	/* R-Car Gen2 only */
112 #define IMTTBCR_SL0_TWOBIT_LVL_1	(2 << 6)	/* R-Car Gen3 only */
113 #define IMTTBCR_SL0_LVL_1		(1 << 4)	/* R-Car Gen2 only */
114 
115 #define IMBUSCR				0x000c		/* R-Car Gen2 only */
116 #define IMBUSCR_DVM			(1 << 2)	/* R-Car Gen2 only */
117 #define IMBUSCR_BUSSEL_MASK		(3 << 0)	/* R-Car Gen2 only */
118 
119 #define IMTTLBR0			0x0010		/* R-Car Gen2/3 */
120 #define IMTTUBR0			0x0014		/* R-Car Gen2/3 */
121 
122 #define IMSTR				0x0020		/* R-Car Gen2/3 */
123 #define IMSTR_MHIT			(1 << 4)	/* R-Car Gen2/3 */
124 #define IMSTR_ABORT			(1 << 2)	/* R-Car Gen2/3 */
125 #define IMSTR_PF			(1 << 1)	/* R-Car Gen2/3 */
126 #define IMSTR_TF			(1 << 0)	/* R-Car Gen2/3 */
127 
128 #define IMMAIR0				0x0028		/* R-Car Gen2/3 */
129 
130 #define IMELAR				0x0030		/* R-Car Gen2/3, IMEAR on R-Car Gen2 */
131 #define IMEUAR				0x0034		/* R-Car Gen3 only */
132 
133 /* uTLB registers */
134 #define IMUCTR(n)			((n) < 32 ? IMUCTR0(n) : IMUCTR32(n))
135 #define IMUCTR0(n)			(0x0300 + ((n) * 16))		/* R-Car Gen2/3 */
136 #define IMUCTR32(n)			(0x0600 + (((n) - 32) * 16))	/* R-Car Gen3 only */
137 #define IMUCTR_TTSEL_MMU(n)		((n) << 4)	/* R-Car Gen2/3 */
138 #define IMUCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
139 #define IMUCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
140 
141 #define IMUASID(n)			((n) < 32 ? IMUASID0(n) : IMUASID32(n))
142 #define IMUASID0(n)			(0x0308 + ((n) * 16))		/* R-Car Gen2/3 */
143 #define IMUASID32(n)			(0x0608 + (((n) - 32) * 16))	/* R-Car Gen3 only */
144 
145 /* -----------------------------------------------------------------------------
146  * Root device handling
147  */
148 
149 static struct platform_driver ipmmu_driver;
150 
151 static bool ipmmu_is_root(struct ipmmu_vmsa_device *mmu)
152 {
153 	return mmu->root == mmu;
154 }
155 
156 static int __ipmmu_check_device(struct device *dev, void *data)
157 {
158 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
159 	struct ipmmu_vmsa_device **rootp = data;
160 
161 	if (ipmmu_is_root(mmu))
162 		*rootp = mmu;
163 
164 	return 0;
165 }
166 
167 static struct ipmmu_vmsa_device *ipmmu_find_root(void)
168 {
169 	struct ipmmu_vmsa_device *root = NULL;
170 
171 	return driver_for_each_device(&ipmmu_driver.driver, NULL, &root,
172 				      __ipmmu_check_device) == 0 ? root : NULL;
173 }
174 
175 /* -----------------------------------------------------------------------------
176  * Read/Write Access
177  */
178 
179 static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
180 {
181 	return ioread32(mmu->base + offset);
182 }
183 
184 static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
185 			u32 data)
186 {
187 	iowrite32(data, mmu->base + offset);
188 }
189 
190 static unsigned int ipmmu_ctx_reg(struct ipmmu_vmsa_device *mmu,
191 				  unsigned int context_id, unsigned int reg)
192 {
193 	return mmu->features->ctx_offset_base +
194 	       context_id * mmu->features->ctx_offset_stride + reg;
195 }
196 
197 static u32 ipmmu_ctx_read(struct ipmmu_vmsa_device *mmu,
198 			  unsigned int context_id, unsigned int reg)
199 {
200 	return ipmmu_read(mmu, ipmmu_ctx_reg(mmu, context_id, reg));
201 }
202 
203 static void ipmmu_ctx_write(struct ipmmu_vmsa_device *mmu,
204 			    unsigned int context_id, unsigned int reg, u32 data)
205 {
206 	ipmmu_write(mmu, ipmmu_ctx_reg(mmu, context_id, reg), data);
207 }
208 
209 static u32 ipmmu_ctx_read_root(struct ipmmu_vmsa_domain *domain,
210 			       unsigned int reg)
211 {
212 	return ipmmu_ctx_read(domain->mmu->root, domain->context_id, reg);
213 }
214 
215 static void ipmmu_ctx_write_root(struct ipmmu_vmsa_domain *domain,
216 				 unsigned int reg, u32 data)
217 {
218 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
219 }
220 
221 static void ipmmu_ctx_write_all(struct ipmmu_vmsa_domain *domain,
222 				unsigned int reg, u32 data)
223 {
224 	if (domain->mmu != domain->mmu->root)
225 		ipmmu_ctx_write(domain->mmu, domain->context_id, reg, data);
226 
227 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
228 }
229 
230 static u32 ipmmu_utlb_reg(struct ipmmu_vmsa_device *mmu, unsigned int reg)
231 {
232 	return mmu->features->utlb_offset_base + reg;
233 }
234 
235 static void ipmmu_imuasid_write(struct ipmmu_vmsa_device *mmu,
236 				unsigned int utlb, u32 data)
237 {
238 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUASID(utlb)), data);
239 }
240 
241 static void ipmmu_imuctr_write(struct ipmmu_vmsa_device *mmu,
242 			       unsigned int utlb, u32 data)
243 {
244 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUCTR(utlb)), data);
245 }
246 
247 /* -----------------------------------------------------------------------------
248  * TLB and microTLB Management
249  */
250 
251 /* Wait for any pending TLB invalidations to complete */
252 static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
253 {
254 	unsigned int count = 0;
255 
256 	while (ipmmu_ctx_read_root(domain, IMCTR) & IMCTR_FLUSH) {
257 		cpu_relax();
258 		if (++count == TLB_LOOP_TIMEOUT) {
259 			dev_err_ratelimited(domain->mmu->dev,
260 			"TLB sync timed out -- MMU may be deadlocked\n");
261 			return;
262 		}
263 		udelay(1);
264 	}
265 }
266 
267 static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
268 {
269 	u32 reg;
270 
271 	reg = ipmmu_ctx_read_root(domain, IMCTR);
272 	reg |= IMCTR_FLUSH;
273 	ipmmu_ctx_write_all(domain, IMCTR, reg);
274 
275 	ipmmu_tlb_sync(domain);
276 }
277 
278 /*
279  * Enable MMU translation for the microTLB.
280  */
281 static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
282 			      unsigned int utlb)
283 {
284 	struct ipmmu_vmsa_device *mmu = domain->mmu;
285 
286 	/*
287 	 * TODO: Reference-count the microTLB as several bus masters can be
288 	 * connected to the same microTLB.
289 	 */
290 
291 	/* TODO: What should we set the ASID to ? */
292 	ipmmu_imuasid_write(mmu, utlb, 0);
293 	/* TODO: Do we need to flush the microTLB ? */
294 	ipmmu_imuctr_write(mmu, utlb, IMUCTR_TTSEL_MMU(domain->context_id) |
295 				      IMUCTR_FLUSH | IMUCTR_MMUEN);
296 	mmu->utlb_ctx[utlb] = domain->context_id;
297 }
298 
299 /*
300  * Disable MMU translation for the microTLB.
301  */
302 static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
303 			       unsigned int utlb)
304 {
305 	struct ipmmu_vmsa_device *mmu = domain->mmu;
306 
307 	ipmmu_imuctr_write(mmu, utlb, 0);
308 	mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
309 }
310 
311 static void ipmmu_tlb_flush_all(void *cookie)
312 {
313 	struct ipmmu_vmsa_domain *domain = cookie;
314 
315 	ipmmu_tlb_invalidate(domain);
316 }
317 
318 static void ipmmu_tlb_flush(unsigned long iova, size_t size,
319 				size_t granule, void *cookie)
320 {
321 	ipmmu_tlb_flush_all(cookie);
322 }
323 
324 static const struct iommu_flush_ops ipmmu_flush_ops = {
325 	.tlb_flush_all = ipmmu_tlb_flush_all,
326 	.tlb_flush_walk = ipmmu_tlb_flush,
327 };
328 
329 /* -----------------------------------------------------------------------------
330  * Domain/Context Management
331  */
332 
333 static int ipmmu_domain_allocate_context(struct ipmmu_vmsa_device *mmu,
334 					 struct ipmmu_vmsa_domain *domain)
335 {
336 	unsigned long flags;
337 	int ret;
338 
339 	spin_lock_irqsave(&mmu->lock, flags);
340 
341 	ret = find_first_zero_bit(mmu->ctx, mmu->num_ctx);
342 	if (ret != mmu->num_ctx) {
343 		mmu->domains[ret] = domain;
344 		set_bit(ret, mmu->ctx);
345 	} else
346 		ret = -EBUSY;
347 
348 	spin_unlock_irqrestore(&mmu->lock, flags);
349 
350 	return ret;
351 }
352 
353 static void ipmmu_domain_free_context(struct ipmmu_vmsa_device *mmu,
354 				      unsigned int context_id)
355 {
356 	unsigned long flags;
357 
358 	spin_lock_irqsave(&mmu->lock, flags);
359 
360 	clear_bit(context_id, mmu->ctx);
361 	mmu->domains[context_id] = NULL;
362 
363 	spin_unlock_irqrestore(&mmu->lock, flags);
364 }
365 
366 static void ipmmu_domain_setup_context(struct ipmmu_vmsa_domain *domain)
367 {
368 	u64 ttbr;
369 	u32 tmp;
370 
371 	/* TTBR0 */
372 	ttbr = domain->cfg.arm_lpae_s1_cfg.ttbr;
373 	ipmmu_ctx_write_root(domain, IMTTLBR0, ttbr);
374 	ipmmu_ctx_write_root(domain, IMTTUBR0, ttbr >> 32);
375 
376 	/*
377 	 * TTBCR
378 	 * We use long descriptors and allocate the whole 32-bit VA space to
379 	 * TTBR0.
380 	 */
381 	if (domain->mmu->features->twobit_imttbcr_sl0)
382 		tmp = IMTTBCR_SL0_TWOBIT_LVL_1;
383 	else
384 		tmp = IMTTBCR_SL0_LVL_1;
385 
386 	if (domain->mmu->features->cache_snoop)
387 		tmp |= IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
388 		       IMTTBCR_IRGN0_WB_WA;
389 
390 	ipmmu_ctx_write_root(domain, IMTTBCR, IMTTBCR_EAE | tmp);
391 
392 	/* MAIR0 */
393 	ipmmu_ctx_write_root(domain, IMMAIR0,
394 			     domain->cfg.arm_lpae_s1_cfg.mair);
395 
396 	/* IMBUSCR */
397 	if (domain->mmu->features->setup_imbuscr)
398 		ipmmu_ctx_write_root(domain, IMBUSCR,
399 				     ipmmu_ctx_read_root(domain, IMBUSCR) &
400 				     ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
401 
402 	/*
403 	 * IMSTR
404 	 * Clear all interrupt flags.
405 	 */
406 	ipmmu_ctx_write_root(domain, IMSTR, ipmmu_ctx_read_root(domain, IMSTR));
407 
408 	/*
409 	 * IMCTR
410 	 * Enable the MMU and interrupt generation. The long-descriptor
411 	 * translation table format doesn't use TEX remapping. Don't enable AF
412 	 * software management as we have no use for it. Flush the TLB as
413 	 * required when modifying the context registers.
414 	 */
415 	ipmmu_ctx_write_all(domain, IMCTR,
416 			    IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
417 }
418 
419 static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
420 {
421 	int ret;
422 
423 	/*
424 	 * Allocate the page table operations.
425 	 *
426 	 * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory
427 	 * access, Long-descriptor format" that the NStable bit being set in a
428 	 * table descriptor will result in the NStable and NS bits of all child
429 	 * entries being ignored and considered as being set. The IPMMU seems
430 	 * not to comply with this, as it generates a secure access page fault
431 	 * if any of the NStable and NS bits isn't set when running in
432 	 * non-secure mode.
433 	 */
434 	domain->cfg.quirks = IO_PGTABLE_QUIRK_ARM_NS;
435 	domain->cfg.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K;
436 	domain->cfg.ias = 32;
437 	domain->cfg.oas = 40;
438 	domain->cfg.tlb = &ipmmu_flush_ops;
439 	domain->io_domain.geometry.aperture_end = DMA_BIT_MASK(32);
440 	domain->io_domain.geometry.force_aperture = true;
441 	/*
442 	 * TODO: Add support for coherent walk through CCI with DVM and remove
443 	 * cache handling. For now, delegate it to the io-pgtable code.
444 	 */
445 	domain->cfg.coherent_walk = false;
446 	domain->cfg.iommu_dev = domain->mmu->root->dev;
447 
448 	/*
449 	 * Find an unused context.
450 	 */
451 	ret = ipmmu_domain_allocate_context(domain->mmu->root, domain);
452 	if (ret < 0)
453 		return ret;
454 
455 	domain->context_id = ret;
456 
457 	domain->iop = alloc_io_pgtable_ops(ARM_32_LPAE_S1, &domain->cfg,
458 					   domain);
459 	if (!domain->iop) {
460 		ipmmu_domain_free_context(domain->mmu->root,
461 					  domain->context_id);
462 		return -EINVAL;
463 	}
464 
465 	ipmmu_domain_setup_context(domain);
466 	return 0;
467 }
468 
469 static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
470 {
471 	if (!domain->mmu)
472 		return;
473 
474 	/*
475 	 * Disable the context. Flush the TLB as required when modifying the
476 	 * context registers.
477 	 *
478 	 * TODO: Is TLB flush really needed ?
479 	 */
480 	ipmmu_ctx_write_all(domain, IMCTR, IMCTR_FLUSH);
481 	ipmmu_tlb_sync(domain);
482 	ipmmu_domain_free_context(domain->mmu->root, domain->context_id);
483 }
484 
485 /* -----------------------------------------------------------------------------
486  * Fault Handling
487  */
488 
489 static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
490 {
491 	const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
492 	struct ipmmu_vmsa_device *mmu = domain->mmu;
493 	unsigned long iova;
494 	u32 status;
495 
496 	status = ipmmu_ctx_read_root(domain, IMSTR);
497 	if (!(status & err_mask))
498 		return IRQ_NONE;
499 
500 	iova = ipmmu_ctx_read_root(domain, IMELAR);
501 	if (IS_ENABLED(CONFIG_64BIT))
502 		iova |= (u64)ipmmu_ctx_read_root(domain, IMEUAR) << 32;
503 
504 	/*
505 	 * Clear the error status flags. Unlike traditional interrupt flag
506 	 * registers that must be cleared by writing 1, this status register
507 	 * seems to require 0. The error address register must be read before,
508 	 * otherwise its value will be 0.
509 	 */
510 	ipmmu_ctx_write_root(domain, IMSTR, 0);
511 
512 	/* Log fatal errors. */
513 	if (status & IMSTR_MHIT)
514 		dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%lx\n",
515 				    iova);
516 	if (status & IMSTR_ABORT)
517 		dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%lx\n",
518 				    iova);
519 
520 	if (!(status & (IMSTR_PF | IMSTR_TF)))
521 		return IRQ_NONE;
522 
523 	/*
524 	 * Try to handle page faults and translation faults.
525 	 *
526 	 * TODO: We need to look up the faulty device based on the I/O VA. Use
527 	 * the IOMMU device for now.
528 	 */
529 	if (!report_iommu_fault(&domain->io_domain, mmu->dev, iova, 0))
530 		return IRQ_HANDLED;
531 
532 	dev_err_ratelimited(mmu->dev,
533 			    "Unhandled fault: status 0x%08x iova 0x%lx\n",
534 			    status, iova);
535 
536 	return IRQ_HANDLED;
537 }
538 
539 static irqreturn_t ipmmu_irq(int irq, void *dev)
540 {
541 	struct ipmmu_vmsa_device *mmu = dev;
542 	irqreturn_t status = IRQ_NONE;
543 	unsigned int i;
544 	unsigned long flags;
545 
546 	spin_lock_irqsave(&mmu->lock, flags);
547 
548 	/*
549 	 * Check interrupts for all active contexts.
550 	 */
551 	for (i = 0; i < mmu->num_ctx; i++) {
552 		if (!mmu->domains[i])
553 			continue;
554 		if (ipmmu_domain_irq(mmu->domains[i]) == IRQ_HANDLED)
555 			status = IRQ_HANDLED;
556 	}
557 
558 	spin_unlock_irqrestore(&mmu->lock, flags);
559 
560 	return status;
561 }
562 
563 /* -----------------------------------------------------------------------------
564  * IOMMU Operations
565  */
566 
567 static struct iommu_domain *__ipmmu_domain_alloc(unsigned type)
568 {
569 	struct ipmmu_vmsa_domain *domain;
570 
571 	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
572 	if (!domain)
573 		return NULL;
574 
575 	mutex_init(&domain->mutex);
576 
577 	return &domain->io_domain;
578 }
579 
580 static struct iommu_domain *ipmmu_domain_alloc(unsigned type)
581 {
582 	struct iommu_domain *io_domain = NULL;
583 
584 	switch (type) {
585 	case IOMMU_DOMAIN_UNMANAGED:
586 		io_domain = __ipmmu_domain_alloc(type);
587 		break;
588 
589 	case IOMMU_DOMAIN_DMA:
590 		io_domain = __ipmmu_domain_alloc(type);
591 		if (io_domain && iommu_get_dma_cookie(io_domain)) {
592 			kfree(io_domain);
593 			io_domain = NULL;
594 		}
595 		break;
596 	}
597 
598 	return io_domain;
599 }
600 
601 static void ipmmu_domain_free(struct iommu_domain *io_domain)
602 {
603 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
604 
605 	/*
606 	 * Free the domain resources. We assume that all devices have already
607 	 * been detached.
608 	 */
609 	iommu_put_dma_cookie(io_domain);
610 	ipmmu_domain_destroy_context(domain);
611 	free_io_pgtable_ops(domain->iop);
612 	kfree(domain);
613 }
614 
615 static int ipmmu_attach_device(struct iommu_domain *io_domain,
616 			       struct device *dev)
617 {
618 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
619 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
620 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
621 	unsigned int i;
622 	int ret = 0;
623 
624 	if (!mmu) {
625 		dev_err(dev, "Cannot attach to IPMMU\n");
626 		return -ENXIO;
627 	}
628 
629 	mutex_lock(&domain->mutex);
630 
631 	if (!domain->mmu) {
632 		/* The domain hasn't been used yet, initialize it. */
633 		domain->mmu = mmu;
634 		ret = ipmmu_domain_init_context(domain);
635 		if (ret < 0) {
636 			dev_err(dev, "Unable to initialize IPMMU context\n");
637 			domain->mmu = NULL;
638 		} else {
639 			dev_info(dev, "Using IPMMU context %u\n",
640 				 domain->context_id);
641 		}
642 	} else if (domain->mmu != mmu) {
643 		/*
644 		 * Something is wrong, we can't attach two devices using
645 		 * different IOMMUs to the same domain.
646 		 */
647 		dev_err(dev, "Can't attach IPMMU %s to domain on IPMMU %s\n",
648 			dev_name(mmu->dev), dev_name(domain->mmu->dev));
649 		ret = -EINVAL;
650 	} else
651 		dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
652 
653 	mutex_unlock(&domain->mutex);
654 
655 	if (ret < 0)
656 		return ret;
657 
658 	for (i = 0; i < fwspec->num_ids; ++i)
659 		ipmmu_utlb_enable(domain, fwspec->ids[i]);
660 
661 	return 0;
662 }
663 
664 static void ipmmu_detach_device(struct iommu_domain *io_domain,
665 				struct device *dev)
666 {
667 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
668 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
669 	unsigned int i;
670 
671 	for (i = 0; i < fwspec->num_ids; ++i)
672 		ipmmu_utlb_disable(domain, fwspec->ids[i]);
673 
674 	/*
675 	 * TODO: Optimize by disabling the context when no device is attached.
676 	 */
677 }
678 
679 static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
680 		     phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
681 {
682 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
683 
684 	if (!domain)
685 		return -ENODEV;
686 
687 	return domain->iop->map(domain->iop, iova, paddr, size, prot, gfp);
688 }
689 
690 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
691 			  size_t size, struct iommu_iotlb_gather *gather)
692 {
693 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
694 
695 	return domain->iop->unmap(domain->iop, iova, size, gather);
696 }
697 
698 static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
699 {
700 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
701 
702 	if (domain->mmu)
703 		ipmmu_tlb_flush_all(domain);
704 }
705 
706 static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
707 			     struct iommu_iotlb_gather *gather)
708 {
709 	ipmmu_flush_iotlb_all(io_domain);
710 }
711 
712 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
713 				      dma_addr_t iova)
714 {
715 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
716 
717 	/* TODO: Is locking needed ? */
718 
719 	return domain->iop->iova_to_phys(domain->iop, iova);
720 }
721 
722 static int ipmmu_init_platform_device(struct device *dev,
723 				      struct of_phandle_args *args)
724 {
725 	struct platform_device *ipmmu_pdev;
726 
727 	ipmmu_pdev = of_find_device_by_node(args->np);
728 	if (!ipmmu_pdev)
729 		return -ENODEV;
730 
731 	dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
732 
733 	return 0;
734 }
735 
736 static const struct soc_device_attribute soc_needs_opt_in[] = {
737 	{ .family = "R-Car Gen3", },
738 	{ .family = "RZ/G2", },
739 	{ /* sentinel */ }
740 };
741 
742 static const struct soc_device_attribute soc_denylist[] = {
743 	{ .soc_id = "r8a774a1", },
744 	{ .soc_id = "r8a7795", .revision = "ES1.*" },
745 	{ .soc_id = "r8a7795", .revision = "ES2.*" },
746 	{ .soc_id = "r8a7796", },
747 	{ /* sentinel */ }
748 };
749 
750 static const char * const devices_allowlist[] = {
751 	"ee100000.mmc",
752 	"ee120000.mmc",
753 	"ee140000.mmc",
754 	"ee160000.mmc"
755 };
756 
757 static bool ipmmu_device_is_allowed(struct device *dev)
758 {
759 	unsigned int i;
760 
761 	/*
762 	 * R-Car Gen3 and RZ/G2 use the allow list to opt-in devices.
763 	 * For Other SoCs, this returns true anyway.
764 	 */
765 	if (!soc_device_match(soc_needs_opt_in))
766 		return true;
767 
768 	/* Check whether this SoC can use the IPMMU correctly or not */
769 	if (soc_device_match(soc_denylist))
770 		return false;
771 
772 	/* Check whether this device can work with the IPMMU */
773 	for (i = 0; i < ARRAY_SIZE(devices_allowlist); i++) {
774 		if (!strcmp(dev_name(dev), devices_allowlist[i]))
775 			return true;
776 	}
777 
778 	/* Otherwise, do not allow use of IPMMU */
779 	return false;
780 }
781 
782 static int ipmmu_of_xlate(struct device *dev,
783 			  struct of_phandle_args *spec)
784 {
785 	if (!ipmmu_device_is_allowed(dev))
786 		return -ENODEV;
787 
788 	iommu_fwspec_add_ids(dev, spec->args, 1);
789 
790 	/* Initialize once - xlate() will call multiple times */
791 	if (to_ipmmu(dev))
792 		return 0;
793 
794 	return ipmmu_init_platform_device(dev, spec);
795 }
796 
797 static int ipmmu_init_arm_mapping(struct device *dev)
798 {
799 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
800 	int ret;
801 
802 	/*
803 	 * Create the ARM mapping, used by the ARM DMA mapping core to allocate
804 	 * VAs. This will allocate a corresponding IOMMU domain.
805 	 *
806 	 * TODO:
807 	 * - Create one mapping per context (TLB).
808 	 * - Make the mapping size configurable ? We currently use a 2GB mapping
809 	 *   at a 1GB offset to ensure that NULL VAs will fault.
810 	 */
811 	if (!mmu->mapping) {
812 		struct dma_iommu_mapping *mapping;
813 
814 		mapping = arm_iommu_create_mapping(&platform_bus_type,
815 						   SZ_1G, SZ_2G);
816 		if (IS_ERR(mapping)) {
817 			dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
818 			ret = PTR_ERR(mapping);
819 			goto error;
820 		}
821 
822 		mmu->mapping = mapping;
823 	}
824 
825 	/* Attach the ARM VA mapping to the device. */
826 	ret = arm_iommu_attach_device(dev, mmu->mapping);
827 	if (ret < 0) {
828 		dev_err(dev, "Failed to attach device to VA mapping\n");
829 		goto error;
830 	}
831 
832 	return 0;
833 
834 error:
835 	if (mmu->mapping)
836 		arm_iommu_release_mapping(mmu->mapping);
837 
838 	return ret;
839 }
840 
841 static struct iommu_device *ipmmu_probe_device(struct device *dev)
842 {
843 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
844 
845 	/*
846 	 * Only let through devices that have been verified in xlate()
847 	 */
848 	if (!mmu)
849 		return ERR_PTR(-ENODEV);
850 
851 	return &mmu->iommu;
852 }
853 
854 static void ipmmu_probe_finalize(struct device *dev)
855 {
856 	int ret = 0;
857 
858 	if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA))
859 		ret = ipmmu_init_arm_mapping(dev);
860 
861 	if (ret)
862 		dev_err(dev, "Can't create IOMMU mapping - DMA-OPS will not work\n");
863 }
864 
865 static void ipmmu_release_device(struct device *dev)
866 {
867 	arm_iommu_detach_device(dev);
868 }
869 
870 static struct iommu_group *ipmmu_find_group(struct device *dev)
871 {
872 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
873 	struct iommu_group *group;
874 
875 	if (mmu->group)
876 		return iommu_group_ref_get(mmu->group);
877 
878 	group = iommu_group_alloc();
879 	if (!IS_ERR(group))
880 		mmu->group = group;
881 
882 	return group;
883 }
884 
885 static const struct iommu_ops ipmmu_ops = {
886 	.domain_alloc = ipmmu_domain_alloc,
887 	.domain_free = ipmmu_domain_free,
888 	.attach_dev = ipmmu_attach_device,
889 	.detach_dev = ipmmu_detach_device,
890 	.map = ipmmu_map,
891 	.unmap = ipmmu_unmap,
892 	.flush_iotlb_all = ipmmu_flush_iotlb_all,
893 	.iotlb_sync = ipmmu_iotlb_sync,
894 	.iova_to_phys = ipmmu_iova_to_phys,
895 	.probe_device = ipmmu_probe_device,
896 	.release_device = ipmmu_release_device,
897 	.probe_finalize = ipmmu_probe_finalize,
898 	.device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
899 			? generic_device_group : ipmmu_find_group,
900 	.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
901 	.of_xlate = ipmmu_of_xlate,
902 };
903 
904 /* -----------------------------------------------------------------------------
905  * Probe/remove and init
906  */
907 
908 static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
909 {
910 	unsigned int i;
911 
912 	/* Disable all contexts. */
913 	for (i = 0; i < mmu->num_ctx; ++i)
914 		ipmmu_ctx_write(mmu, i, IMCTR, 0);
915 }
916 
917 static const struct ipmmu_features ipmmu_features_default = {
918 	.use_ns_alias_offset = true,
919 	.has_cache_leaf_nodes = false,
920 	.number_of_contexts = 1, /* software only tested with one context */
921 	.num_utlbs = 32,
922 	.setup_imbuscr = true,
923 	.twobit_imttbcr_sl0 = false,
924 	.reserved_context = false,
925 	.cache_snoop = true,
926 	.ctx_offset_base = 0,
927 	.ctx_offset_stride = 0x40,
928 	.utlb_offset_base = 0,
929 };
930 
931 static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
932 	.use_ns_alias_offset = false,
933 	.has_cache_leaf_nodes = true,
934 	.number_of_contexts = 8,
935 	.num_utlbs = 48,
936 	.setup_imbuscr = false,
937 	.twobit_imttbcr_sl0 = true,
938 	.reserved_context = true,
939 	.cache_snoop = false,
940 	.ctx_offset_base = 0,
941 	.ctx_offset_stride = 0x40,
942 	.utlb_offset_base = 0,
943 };
944 
945 static const struct of_device_id ipmmu_of_ids[] = {
946 	{
947 		.compatible = "renesas,ipmmu-vmsa",
948 		.data = &ipmmu_features_default,
949 	}, {
950 		.compatible = "renesas,ipmmu-r8a774a1",
951 		.data = &ipmmu_features_rcar_gen3,
952 	}, {
953 		.compatible = "renesas,ipmmu-r8a774b1",
954 		.data = &ipmmu_features_rcar_gen3,
955 	}, {
956 		.compatible = "renesas,ipmmu-r8a774c0",
957 		.data = &ipmmu_features_rcar_gen3,
958 	}, {
959 		.compatible = "renesas,ipmmu-r8a774e1",
960 		.data = &ipmmu_features_rcar_gen3,
961 	}, {
962 		.compatible = "renesas,ipmmu-r8a7795",
963 		.data = &ipmmu_features_rcar_gen3,
964 	}, {
965 		.compatible = "renesas,ipmmu-r8a7796",
966 		.data = &ipmmu_features_rcar_gen3,
967 	}, {
968 		.compatible = "renesas,ipmmu-r8a77961",
969 		.data = &ipmmu_features_rcar_gen3,
970 	}, {
971 		.compatible = "renesas,ipmmu-r8a77965",
972 		.data = &ipmmu_features_rcar_gen3,
973 	}, {
974 		.compatible = "renesas,ipmmu-r8a77970",
975 		.data = &ipmmu_features_rcar_gen3,
976 	}, {
977 		.compatible = "renesas,ipmmu-r8a77990",
978 		.data = &ipmmu_features_rcar_gen3,
979 	}, {
980 		.compatible = "renesas,ipmmu-r8a77995",
981 		.data = &ipmmu_features_rcar_gen3,
982 	}, {
983 		/* Terminator */
984 	},
985 };
986 
987 static int ipmmu_probe(struct platform_device *pdev)
988 {
989 	struct ipmmu_vmsa_device *mmu;
990 	struct resource *res;
991 	int irq;
992 	int ret;
993 
994 	mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
995 	if (!mmu) {
996 		dev_err(&pdev->dev, "cannot allocate device data\n");
997 		return -ENOMEM;
998 	}
999 
1000 	mmu->dev = &pdev->dev;
1001 	spin_lock_init(&mmu->lock);
1002 	bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
1003 	mmu->features = of_device_get_match_data(&pdev->dev);
1004 	memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
1005 	dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
1006 
1007 	/* Map I/O memory and request IRQ. */
1008 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1009 	mmu->base = devm_ioremap_resource(&pdev->dev, res);
1010 	if (IS_ERR(mmu->base))
1011 		return PTR_ERR(mmu->base);
1012 
1013 	/*
1014 	 * The IPMMU has two register banks, for secure and non-secure modes.
1015 	 * The bank mapped at the beginning of the IPMMU address space
1016 	 * corresponds to the running mode of the CPU. When running in secure
1017 	 * mode the non-secure register bank is also available at an offset.
1018 	 *
1019 	 * Secure mode operation isn't clearly documented and is thus currently
1020 	 * not implemented in the driver. Furthermore, preliminary tests of
1021 	 * non-secure operation with the main register bank were not successful.
1022 	 * Offset the registers base unconditionally to point to the non-secure
1023 	 * alias space for now.
1024 	 */
1025 	if (mmu->features->use_ns_alias_offset)
1026 		mmu->base += IM_NS_ALIAS_OFFSET;
1027 
1028 	mmu->num_ctx = min(IPMMU_CTX_MAX, mmu->features->number_of_contexts);
1029 
1030 	/*
1031 	 * Determine if this IPMMU instance is a root device by checking for
1032 	 * the lack of has_cache_leaf_nodes flag or renesas,ipmmu-main property.
1033 	 */
1034 	if (!mmu->features->has_cache_leaf_nodes ||
1035 	    !of_find_property(pdev->dev.of_node, "renesas,ipmmu-main", NULL))
1036 		mmu->root = mmu;
1037 	else
1038 		mmu->root = ipmmu_find_root();
1039 
1040 	/*
1041 	 * Wait until the root device has been registered for sure.
1042 	 */
1043 	if (!mmu->root)
1044 		return -EPROBE_DEFER;
1045 
1046 	/* Root devices have mandatory IRQs */
1047 	if (ipmmu_is_root(mmu)) {
1048 		irq = platform_get_irq(pdev, 0);
1049 		if (irq < 0)
1050 			return irq;
1051 
1052 		ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
1053 				       dev_name(&pdev->dev), mmu);
1054 		if (ret < 0) {
1055 			dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
1056 			return ret;
1057 		}
1058 
1059 		ipmmu_device_reset(mmu);
1060 
1061 		if (mmu->features->reserved_context) {
1062 			dev_info(&pdev->dev, "IPMMU context 0 is reserved\n");
1063 			set_bit(0, mmu->ctx);
1064 		}
1065 	}
1066 
1067 	/*
1068 	 * Register the IPMMU to the IOMMU subsystem in the following cases:
1069 	 * - R-Car Gen2 IPMMU (all devices registered)
1070 	 * - R-Car Gen3 IPMMU (leaf devices only - skip root IPMMU-MM device)
1071 	 */
1072 	if (!mmu->features->has_cache_leaf_nodes || !ipmmu_is_root(mmu)) {
1073 		ret = iommu_device_sysfs_add(&mmu->iommu, &pdev->dev, NULL,
1074 					     dev_name(&pdev->dev));
1075 		if (ret)
1076 			return ret;
1077 
1078 		ret = iommu_device_register(&mmu->iommu, &ipmmu_ops, &pdev->dev);
1079 		if (ret)
1080 			return ret;
1081 
1082 #if defined(CONFIG_IOMMU_DMA)
1083 		if (!iommu_present(&platform_bus_type))
1084 			bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1085 #endif
1086 	}
1087 
1088 	/*
1089 	 * We can't create the ARM mapping here as it requires the bus to have
1090 	 * an IOMMU, which only happens when bus_set_iommu() is called in
1091 	 * ipmmu_init() after the probe function returns.
1092 	 */
1093 
1094 	platform_set_drvdata(pdev, mmu);
1095 
1096 	return 0;
1097 }
1098 
1099 static int ipmmu_remove(struct platform_device *pdev)
1100 {
1101 	struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1102 
1103 	iommu_device_sysfs_remove(&mmu->iommu);
1104 	iommu_device_unregister(&mmu->iommu);
1105 
1106 	arm_iommu_release_mapping(mmu->mapping);
1107 
1108 	ipmmu_device_reset(mmu);
1109 
1110 	return 0;
1111 }
1112 
1113 #ifdef CONFIG_PM_SLEEP
1114 static int ipmmu_resume_noirq(struct device *dev)
1115 {
1116 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1117 	unsigned int i;
1118 
1119 	/* Reset root MMU and restore contexts */
1120 	if (ipmmu_is_root(mmu)) {
1121 		ipmmu_device_reset(mmu);
1122 
1123 		for (i = 0; i < mmu->num_ctx; i++) {
1124 			if (!mmu->domains[i])
1125 				continue;
1126 
1127 			ipmmu_domain_setup_context(mmu->domains[i]);
1128 		}
1129 	}
1130 
1131 	/* Re-enable active micro-TLBs */
1132 	for (i = 0; i < mmu->features->num_utlbs; i++) {
1133 		if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1134 			continue;
1135 
1136 		ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1137 	}
1138 
1139 	return 0;
1140 }
1141 
1142 static const struct dev_pm_ops ipmmu_pm  = {
1143 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1144 };
1145 #define DEV_PM_OPS	&ipmmu_pm
1146 #else
1147 #define DEV_PM_OPS	NULL
1148 #endif /* CONFIG_PM_SLEEP */
1149 
1150 static struct platform_driver ipmmu_driver = {
1151 	.driver = {
1152 		.name = "ipmmu-vmsa",
1153 		.of_match_table = of_match_ptr(ipmmu_of_ids),
1154 		.pm = DEV_PM_OPS,
1155 	},
1156 	.probe = ipmmu_probe,
1157 	.remove	= ipmmu_remove,
1158 };
1159 
1160 static int __init ipmmu_init(void)
1161 {
1162 	struct device_node *np;
1163 	static bool setup_done;
1164 	int ret;
1165 
1166 	if (setup_done)
1167 		return 0;
1168 
1169 	np = of_find_matching_node(NULL, ipmmu_of_ids);
1170 	if (!np)
1171 		return 0;
1172 
1173 	of_node_put(np);
1174 
1175 	ret = platform_driver_register(&ipmmu_driver);
1176 	if (ret < 0)
1177 		return ret;
1178 
1179 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
1180 	if (!iommu_present(&platform_bus_type))
1181 		bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1182 #endif
1183 
1184 	setup_done = true;
1185 	return 0;
1186 }
1187 subsys_initcall(ipmmu_init);
1188