xref: /openbmc/linux/drivers/iommu/ipmmu-vmsa.c (revision 27ab1c1c)
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_iommu.h>
23 #include <linux/of_platform.h>
24 #include <linux/platform_device.h>
25 #include <linux/sizes.h>
26 #include <linux/slab.h>
27 #include <linux/sys_soc.h>
28 
29 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
30 #include <asm/dma-iommu.h>
31 #else
32 #define arm_iommu_create_mapping(...)	NULL
33 #define arm_iommu_attach_device(...)	-ENODEV
34 #define arm_iommu_release_mapping(...)	do {} while (0)
35 #define arm_iommu_detach_device(...)	do {} while (0)
36 #endif
37 
38 #define IPMMU_CTX_MAX		8U
39 #define IPMMU_CTX_INVALID	-1
40 
41 #define IPMMU_UTLB_MAX		48U
42 
43 struct ipmmu_features {
44 	bool use_ns_alias_offset;
45 	bool has_cache_leaf_nodes;
46 	unsigned int number_of_contexts;
47 	unsigned int num_utlbs;
48 	bool setup_imbuscr;
49 	bool twobit_imttbcr_sl0;
50 	bool reserved_context;
51 	bool cache_snoop;
52 	unsigned int ctx_offset_base;
53 	unsigned int ctx_offset_stride;
54 	unsigned int utlb_offset_base;
55 };
56 
57 struct ipmmu_vmsa_device {
58 	struct device *dev;
59 	void __iomem *base;
60 	struct iommu_device iommu;
61 	struct ipmmu_vmsa_device *root;
62 	const struct ipmmu_features *features;
63 	unsigned int num_ctx;
64 	spinlock_t lock;			/* Protects ctx and domains[] */
65 	DECLARE_BITMAP(ctx, IPMMU_CTX_MAX);
66 	struct ipmmu_vmsa_domain *domains[IPMMU_CTX_MAX];
67 	s8 utlb_ctx[IPMMU_UTLB_MAX];
68 
69 	struct iommu_group *group;
70 	struct dma_iommu_mapping *mapping;
71 };
72 
73 struct ipmmu_vmsa_domain {
74 	struct ipmmu_vmsa_device *mmu;
75 	struct iommu_domain io_domain;
76 
77 	struct io_pgtable_cfg cfg;
78 	struct io_pgtable_ops *iop;
79 
80 	unsigned int context_id;
81 	struct mutex mutex;			/* Protects mappings */
82 };
83 
84 static struct ipmmu_vmsa_domain *to_vmsa_domain(struct iommu_domain *dom)
85 {
86 	return container_of(dom, struct ipmmu_vmsa_domain, io_domain);
87 }
88 
89 static struct ipmmu_vmsa_device *to_ipmmu(struct device *dev)
90 {
91 	return dev_iommu_priv_get(dev);
92 }
93 
94 #define TLB_LOOP_TIMEOUT		100	/* 100us */
95 
96 /* -----------------------------------------------------------------------------
97  * Registers Definition
98  */
99 
100 #define IM_NS_ALIAS_OFFSET		0x800
101 
102 /* MMU "context" registers */
103 #define IMCTR				0x0000		/* R-Car Gen2/3 */
104 #define IMCTR_INTEN			(1 << 2)	/* R-Car Gen2/3 */
105 #define IMCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
106 #define IMCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
107 
108 #define IMTTBCR				0x0008		/* R-Car Gen2/3 */
109 #define IMTTBCR_EAE			(1 << 31)	/* R-Car Gen2/3 */
110 #define IMTTBCR_SH0_INNER_SHAREABLE	(3 << 12)	/* R-Car Gen2 only */
111 #define IMTTBCR_ORGN0_WB_WA		(1 << 10)	/* R-Car Gen2 only */
112 #define IMTTBCR_IRGN0_WB_WA		(1 << 8)	/* R-Car Gen2 only */
113 #define IMTTBCR_SL0_TWOBIT_LVL_1	(2 << 6)	/* R-Car Gen3 only */
114 #define IMTTBCR_SL0_LVL_1		(1 << 4)	/* R-Car Gen2 only */
115 
116 #define IMBUSCR				0x000c		/* R-Car Gen2 only */
117 #define IMBUSCR_DVM			(1 << 2)	/* R-Car Gen2 only */
118 #define IMBUSCR_BUSSEL_MASK		(3 << 0)	/* R-Car Gen2 only */
119 
120 #define IMTTLBR0			0x0010		/* R-Car Gen2/3 */
121 #define IMTTUBR0			0x0014		/* R-Car Gen2/3 */
122 
123 #define IMSTR				0x0020		/* R-Car Gen2/3 */
124 #define IMSTR_MHIT			(1 << 4)	/* R-Car Gen2/3 */
125 #define IMSTR_ABORT			(1 << 2)	/* R-Car Gen2/3 */
126 #define IMSTR_PF			(1 << 1)	/* R-Car Gen2/3 */
127 #define IMSTR_TF			(1 << 0)	/* R-Car Gen2/3 */
128 
129 #define IMMAIR0				0x0028		/* R-Car Gen2/3 */
130 
131 #define IMELAR				0x0030		/* R-Car Gen2/3, IMEAR on R-Car Gen2 */
132 #define IMEUAR				0x0034		/* R-Car Gen3 only */
133 
134 /* uTLB registers */
135 #define IMUCTR(n)			((n) < 32 ? IMUCTR0(n) : IMUCTR32(n))
136 #define IMUCTR0(n)			(0x0300 + ((n) * 16))		/* R-Car Gen2/3 */
137 #define IMUCTR32(n)			(0x0600 + (((n) - 32) * 16))	/* R-Car Gen3 only */
138 #define IMUCTR_TTSEL_MMU(n)		((n) << 4)	/* R-Car Gen2/3 */
139 #define IMUCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
140 #define IMUCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
141 
142 #define IMUASID(n)			((n) < 32 ? IMUASID0(n) : IMUASID32(n))
143 #define IMUASID0(n)			(0x0308 + ((n) * 16))		/* R-Car Gen2/3 */
144 #define IMUASID32(n)			(0x0608 + (((n) - 32) * 16))	/* R-Car Gen3 only */
145 
146 /* -----------------------------------------------------------------------------
147  * Root device handling
148  */
149 
150 static struct platform_driver ipmmu_driver;
151 
152 static bool ipmmu_is_root(struct ipmmu_vmsa_device *mmu)
153 {
154 	return mmu->root == mmu;
155 }
156 
157 static int __ipmmu_check_device(struct device *dev, void *data)
158 {
159 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
160 	struct ipmmu_vmsa_device **rootp = data;
161 
162 	if (ipmmu_is_root(mmu))
163 		*rootp = mmu;
164 
165 	return 0;
166 }
167 
168 static struct ipmmu_vmsa_device *ipmmu_find_root(void)
169 {
170 	struct ipmmu_vmsa_device *root = NULL;
171 
172 	return driver_for_each_device(&ipmmu_driver.driver, NULL, &root,
173 				      __ipmmu_check_device) == 0 ? root : NULL;
174 }
175 
176 /* -----------------------------------------------------------------------------
177  * Read/Write Access
178  */
179 
180 static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
181 {
182 	return ioread32(mmu->base + offset);
183 }
184 
185 static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
186 			u32 data)
187 {
188 	iowrite32(data, mmu->base + offset);
189 }
190 
191 static unsigned int ipmmu_ctx_reg(struct ipmmu_vmsa_device *mmu,
192 				  unsigned int context_id, unsigned int reg)
193 {
194 	return mmu->features->ctx_offset_base +
195 	       context_id * mmu->features->ctx_offset_stride + reg;
196 }
197 
198 static u32 ipmmu_ctx_read(struct ipmmu_vmsa_device *mmu,
199 			  unsigned int context_id, unsigned int reg)
200 {
201 	return ipmmu_read(mmu, ipmmu_ctx_reg(mmu, context_id, reg));
202 }
203 
204 static void ipmmu_ctx_write(struct ipmmu_vmsa_device *mmu,
205 			    unsigned int context_id, unsigned int reg, u32 data)
206 {
207 	ipmmu_write(mmu, ipmmu_ctx_reg(mmu, context_id, reg), data);
208 }
209 
210 static u32 ipmmu_ctx_read_root(struct ipmmu_vmsa_domain *domain,
211 			       unsigned int reg)
212 {
213 	return ipmmu_ctx_read(domain->mmu->root, domain->context_id, reg);
214 }
215 
216 static void ipmmu_ctx_write_root(struct ipmmu_vmsa_domain *domain,
217 				 unsigned int reg, u32 data)
218 {
219 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
220 }
221 
222 static void ipmmu_ctx_write_all(struct ipmmu_vmsa_domain *domain,
223 				unsigned int reg, u32 data)
224 {
225 	if (domain->mmu != domain->mmu->root)
226 		ipmmu_ctx_write(domain->mmu, domain->context_id, reg, data);
227 
228 	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
229 }
230 
231 static u32 ipmmu_utlb_reg(struct ipmmu_vmsa_device *mmu, unsigned int reg)
232 {
233 	return mmu->features->utlb_offset_base + reg;
234 }
235 
236 static void ipmmu_imuasid_write(struct ipmmu_vmsa_device *mmu,
237 				unsigned int utlb, u32 data)
238 {
239 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUASID(utlb)), data);
240 }
241 
242 static void ipmmu_imuctr_write(struct ipmmu_vmsa_device *mmu,
243 			       unsigned int utlb, u32 data)
244 {
245 	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUCTR(utlb)), data);
246 }
247 
248 /* -----------------------------------------------------------------------------
249  * TLB and microTLB Management
250  */
251 
252 /* Wait for any pending TLB invalidations to complete */
253 static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
254 {
255 	unsigned int count = 0;
256 
257 	while (ipmmu_ctx_read_root(domain, IMCTR) & IMCTR_FLUSH) {
258 		cpu_relax();
259 		if (++count == TLB_LOOP_TIMEOUT) {
260 			dev_err_ratelimited(domain->mmu->dev,
261 			"TLB sync timed out -- MMU may be deadlocked\n");
262 			return;
263 		}
264 		udelay(1);
265 	}
266 }
267 
268 static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
269 {
270 	u32 reg;
271 
272 	reg = ipmmu_ctx_read_root(domain, IMCTR);
273 	reg |= IMCTR_FLUSH;
274 	ipmmu_ctx_write_all(domain, IMCTR, reg);
275 
276 	ipmmu_tlb_sync(domain);
277 }
278 
279 /*
280  * Enable MMU translation for the microTLB.
281  */
282 static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
283 			      unsigned int utlb)
284 {
285 	struct ipmmu_vmsa_device *mmu = domain->mmu;
286 
287 	/*
288 	 * TODO: Reference-count the microTLB as several bus masters can be
289 	 * connected to the same microTLB.
290 	 */
291 
292 	/* TODO: What should we set the ASID to ? */
293 	ipmmu_imuasid_write(mmu, utlb, 0);
294 	/* TODO: Do we need to flush the microTLB ? */
295 	ipmmu_imuctr_write(mmu, utlb, IMUCTR_TTSEL_MMU(domain->context_id) |
296 				      IMUCTR_FLUSH | IMUCTR_MMUEN);
297 	mmu->utlb_ctx[utlb] = domain->context_id;
298 }
299 
300 /*
301  * Disable MMU translation for the microTLB.
302  */
303 static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
304 			       unsigned int utlb)
305 {
306 	struct ipmmu_vmsa_device *mmu = domain->mmu;
307 
308 	ipmmu_imuctr_write(mmu, utlb, 0);
309 	mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
310 }
311 
312 static void ipmmu_tlb_flush_all(void *cookie)
313 {
314 	struct ipmmu_vmsa_domain *domain = cookie;
315 
316 	ipmmu_tlb_invalidate(domain);
317 }
318 
319 static void ipmmu_tlb_flush(unsigned long iova, size_t size,
320 				size_t granule, void *cookie)
321 {
322 	ipmmu_tlb_flush_all(cookie);
323 }
324 
325 static const struct iommu_flush_ops ipmmu_flush_ops = {
326 	.tlb_flush_all = ipmmu_tlb_flush_all,
327 	.tlb_flush_walk = ipmmu_tlb_flush,
328 };
329 
330 /* -----------------------------------------------------------------------------
331  * Domain/Context Management
332  */
333 
334 static int ipmmu_domain_allocate_context(struct ipmmu_vmsa_device *mmu,
335 					 struct ipmmu_vmsa_domain *domain)
336 {
337 	unsigned long flags;
338 	int ret;
339 
340 	spin_lock_irqsave(&mmu->lock, flags);
341 
342 	ret = find_first_zero_bit(mmu->ctx, mmu->num_ctx);
343 	if (ret != mmu->num_ctx) {
344 		mmu->domains[ret] = domain;
345 		set_bit(ret, mmu->ctx);
346 	} else
347 		ret = -EBUSY;
348 
349 	spin_unlock_irqrestore(&mmu->lock, flags);
350 
351 	return ret;
352 }
353 
354 static void ipmmu_domain_free_context(struct ipmmu_vmsa_device *mmu,
355 				      unsigned int context_id)
356 {
357 	unsigned long flags;
358 
359 	spin_lock_irqsave(&mmu->lock, flags);
360 
361 	clear_bit(context_id, mmu->ctx);
362 	mmu->domains[context_id] = NULL;
363 
364 	spin_unlock_irqrestore(&mmu->lock, flags);
365 }
366 
367 static void ipmmu_domain_setup_context(struct ipmmu_vmsa_domain *domain)
368 {
369 	u64 ttbr;
370 	u32 tmp;
371 
372 	/* TTBR0 */
373 	ttbr = domain->cfg.arm_lpae_s1_cfg.ttbr;
374 	ipmmu_ctx_write_root(domain, IMTTLBR0, ttbr);
375 	ipmmu_ctx_write_root(domain, IMTTUBR0, ttbr >> 32);
376 
377 	/*
378 	 * TTBCR
379 	 * We use long descriptors and allocate the whole 32-bit VA space to
380 	 * TTBR0.
381 	 */
382 	if (domain->mmu->features->twobit_imttbcr_sl0)
383 		tmp = IMTTBCR_SL0_TWOBIT_LVL_1;
384 	else
385 		tmp = IMTTBCR_SL0_LVL_1;
386 
387 	if (domain->mmu->features->cache_snoop)
388 		tmp |= IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
389 		       IMTTBCR_IRGN0_WB_WA;
390 
391 	ipmmu_ctx_write_root(domain, IMTTBCR, IMTTBCR_EAE | tmp);
392 
393 	/* MAIR0 */
394 	ipmmu_ctx_write_root(domain, IMMAIR0,
395 			     domain->cfg.arm_lpae_s1_cfg.mair);
396 
397 	/* IMBUSCR */
398 	if (domain->mmu->features->setup_imbuscr)
399 		ipmmu_ctx_write_root(domain, IMBUSCR,
400 				     ipmmu_ctx_read_root(domain, IMBUSCR) &
401 				     ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
402 
403 	/*
404 	 * IMSTR
405 	 * Clear all interrupt flags.
406 	 */
407 	ipmmu_ctx_write_root(domain, IMSTR, ipmmu_ctx_read_root(domain, IMSTR));
408 
409 	/*
410 	 * IMCTR
411 	 * Enable the MMU and interrupt generation. The long-descriptor
412 	 * translation table format doesn't use TEX remapping. Don't enable AF
413 	 * software management as we have no use for it. Flush the TLB as
414 	 * required when modifying the context registers.
415 	 */
416 	ipmmu_ctx_write_all(domain, IMCTR,
417 			    IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
418 }
419 
420 static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
421 {
422 	int ret;
423 
424 	/*
425 	 * Allocate the page table operations.
426 	 *
427 	 * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory
428 	 * access, Long-descriptor format" that the NStable bit being set in a
429 	 * table descriptor will result in the NStable and NS bits of all child
430 	 * entries being ignored and considered as being set. The IPMMU seems
431 	 * not to comply with this, as it generates a secure access page fault
432 	 * if any of the NStable and NS bits isn't set when running in
433 	 * non-secure mode.
434 	 */
435 	domain->cfg.quirks = IO_PGTABLE_QUIRK_ARM_NS;
436 	domain->cfg.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K;
437 	domain->cfg.ias = 32;
438 	domain->cfg.oas = 40;
439 	domain->cfg.tlb = &ipmmu_flush_ops;
440 	domain->io_domain.geometry.aperture_end = DMA_BIT_MASK(32);
441 	domain->io_domain.geometry.force_aperture = true;
442 	/*
443 	 * TODO: Add support for coherent walk through CCI with DVM and remove
444 	 * cache handling. For now, delegate it to the io-pgtable code.
445 	 */
446 	domain->cfg.coherent_walk = false;
447 	domain->cfg.iommu_dev = domain->mmu->root->dev;
448 
449 	/*
450 	 * Find an unused context.
451 	 */
452 	ret = ipmmu_domain_allocate_context(domain->mmu->root, domain);
453 	if (ret < 0)
454 		return ret;
455 
456 	domain->context_id = ret;
457 
458 	domain->iop = alloc_io_pgtable_ops(ARM_32_LPAE_S1, &domain->cfg,
459 					   domain);
460 	if (!domain->iop) {
461 		ipmmu_domain_free_context(domain->mmu->root,
462 					  domain->context_id);
463 		return -EINVAL;
464 	}
465 
466 	ipmmu_domain_setup_context(domain);
467 	return 0;
468 }
469 
470 static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
471 {
472 	if (!domain->mmu)
473 		return;
474 
475 	/*
476 	 * Disable the context. Flush the TLB as required when modifying the
477 	 * context registers.
478 	 *
479 	 * TODO: Is TLB flush really needed ?
480 	 */
481 	ipmmu_ctx_write_all(domain, IMCTR, IMCTR_FLUSH);
482 	ipmmu_tlb_sync(domain);
483 	ipmmu_domain_free_context(domain->mmu->root, domain->context_id);
484 }
485 
486 /* -----------------------------------------------------------------------------
487  * Fault Handling
488  */
489 
490 static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
491 {
492 	const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
493 	struct ipmmu_vmsa_device *mmu = domain->mmu;
494 	unsigned long iova;
495 	u32 status;
496 
497 	status = ipmmu_ctx_read_root(domain, IMSTR);
498 	if (!(status & err_mask))
499 		return IRQ_NONE;
500 
501 	iova = ipmmu_ctx_read_root(domain, IMELAR);
502 	if (IS_ENABLED(CONFIG_64BIT))
503 		iova |= (u64)ipmmu_ctx_read_root(domain, IMEUAR) << 32;
504 
505 	/*
506 	 * Clear the error status flags. Unlike traditional interrupt flag
507 	 * registers that must be cleared by writing 1, this status register
508 	 * seems to require 0. The error address register must be read before,
509 	 * otherwise its value will be 0.
510 	 */
511 	ipmmu_ctx_write_root(domain, IMSTR, 0);
512 
513 	/* Log fatal errors. */
514 	if (status & IMSTR_MHIT)
515 		dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%lx\n",
516 				    iova);
517 	if (status & IMSTR_ABORT)
518 		dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%lx\n",
519 				    iova);
520 
521 	if (!(status & (IMSTR_PF | IMSTR_TF)))
522 		return IRQ_NONE;
523 
524 	/*
525 	 * Try to handle page faults and translation faults.
526 	 *
527 	 * TODO: We need to look up the faulty device based on the I/O VA. Use
528 	 * the IOMMU device for now.
529 	 */
530 	if (!report_iommu_fault(&domain->io_domain, mmu->dev, iova, 0))
531 		return IRQ_HANDLED;
532 
533 	dev_err_ratelimited(mmu->dev,
534 			    "Unhandled fault: status 0x%08x iova 0x%lx\n",
535 			    status, iova);
536 
537 	return IRQ_HANDLED;
538 }
539 
540 static irqreturn_t ipmmu_irq(int irq, void *dev)
541 {
542 	struct ipmmu_vmsa_device *mmu = dev;
543 	irqreturn_t status = IRQ_NONE;
544 	unsigned int i;
545 	unsigned long flags;
546 
547 	spin_lock_irqsave(&mmu->lock, flags);
548 
549 	/*
550 	 * Check interrupts for all active contexts.
551 	 */
552 	for (i = 0; i < mmu->num_ctx; i++) {
553 		if (!mmu->domains[i])
554 			continue;
555 		if (ipmmu_domain_irq(mmu->domains[i]) == IRQ_HANDLED)
556 			status = IRQ_HANDLED;
557 	}
558 
559 	spin_unlock_irqrestore(&mmu->lock, flags);
560 
561 	return status;
562 }
563 
564 /* -----------------------------------------------------------------------------
565  * IOMMU Operations
566  */
567 
568 static struct iommu_domain *__ipmmu_domain_alloc(unsigned type)
569 {
570 	struct ipmmu_vmsa_domain *domain;
571 
572 	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
573 	if (!domain)
574 		return NULL;
575 
576 	mutex_init(&domain->mutex);
577 
578 	return &domain->io_domain;
579 }
580 
581 static struct iommu_domain *ipmmu_domain_alloc(unsigned type)
582 {
583 	struct iommu_domain *io_domain = NULL;
584 
585 	switch (type) {
586 	case IOMMU_DOMAIN_UNMANAGED:
587 		io_domain = __ipmmu_domain_alloc(type);
588 		break;
589 
590 	case IOMMU_DOMAIN_DMA:
591 		io_domain = __ipmmu_domain_alloc(type);
592 		if (io_domain && iommu_get_dma_cookie(io_domain)) {
593 			kfree(io_domain);
594 			io_domain = NULL;
595 		}
596 		break;
597 	}
598 
599 	return io_domain;
600 }
601 
602 static void ipmmu_domain_free(struct iommu_domain *io_domain)
603 {
604 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
605 
606 	/*
607 	 * Free the domain resources. We assume that all devices have already
608 	 * been detached.
609 	 */
610 	iommu_put_dma_cookie(io_domain);
611 	ipmmu_domain_destroy_context(domain);
612 	free_io_pgtable_ops(domain->iop);
613 	kfree(domain);
614 }
615 
616 static int ipmmu_attach_device(struct iommu_domain *io_domain,
617 			       struct device *dev)
618 {
619 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
620 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
621 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
622 	unsigned int i;
623 	int ret = 0;
624 
625 	if (!mmu) {
626 		dev_err(dev, "Cannot attach to IPMMU\n");
627 		return -ENXIO;
628 	}
629 
630 	mutex_lock(&domain->mutex);
631 
632 	if (!domain->mmu) {
633 		/* The domain hasn't been used yet, initialize it. */
634 		domain->mmu = mmu;
635 		ret = ipmmu_domain_init_context(domain);
636 		if (ret < 0) {
637 			dev_err(dev, "Unable to initialize IPMMU context\n");
638 			domain->mmu = NULL;
639 		} else {
640 			dev_info(dev, "Using IPMMU context %u\n",
641 				 domain->context_id);
642 		}
643 	} else if (domain->mmu != mmu) {
644 		/*
645 		 * Something is wrong, we can't attach two devices using
646 		 * different IOMMUs to the same domain.
647 		 */
648 		dev_err(dev, "Can't attach IPMMU %s to domain on IPMMU %s\n",
649 			dev_name(mmu->dev), dev_name(domain->mmu->dev));
650 		ret = -EINVAL;
651 	} else
652 		dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
653 
654 	mutex_unlock(&domain->mutex);
655 
656 	if (ret < 0)
657 		return ret;
658 
659 	for (i = 0; i < fwspec->num_ids; ++i)
660 		ipmmu_utlb_enable(domain, fwspec->ids[i]);
661 
662 	return 0;
663 }
664 
665 static void ipmmu_detach_device(struct iommu_domain *io_domain,
666 				struct device *dev)
667 {
668 	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
669 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
670 	unsigned int i;
671 
672 	for (i = 0; i < fwspec->num_ids; ++i)
673 		ipmmu_utlb_disable(domain, fwspec->ids[i]);
674 
675 	/*
676 	 * TODO: Optimize by disabling the context when no device is attached.
677 	 */
678 }
679 
680 static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
681 		     phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
682 {
683 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
684 
685 	if (!domain)
686 		return -ENODEV;
687 
688 	return domain->iop->map(domain->iop, iova, paddr, size, prot, gfp);
689 }
690 
691 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
692 			  size_t size, struct iommu_iotlb_gather *gather)
693 {
694 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
695 
696 	return domain->iop->unmap(domain->iop, iova, size, gather);
697 }
698 
699 static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
700 {
701 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
702 
703 	if (domain->mmu)
704 		ipmmu_tlb_flush_all(domain);
705 }
706 
707 static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
708 			     struct iommu_iotlb_gather *gather)
709 {
710 	ipmmu_flush_iotlb_all(io_domain);
711 }
712 
713 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
714 				      dma_addr_t iova)
715 {
716 	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
717 
718 	/* TODO: Is locking needed ? */
719 
720 	return domain->iop->iova_to_phys(domain->iop, iova);
721 }
722 
723 static int ipmmu_init_platform_device(struct device *dev,
724 				      struct of_phandle_args *args)
725 {
726 	struct platform_device *ipmmu_pdev;
727 
728 	ipmmu_pdev = of_find_device_by_node(args->np);
729 	if (!ipmmu_pdev)
730 		return -ENODEV;
731 
732 	dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
733 
734 	return 0;
735 }
736 
737 static const struct soc_device_attribute soc_rcar_gen3[] = {
738 	{ .soc_id = "r8a774a1", },
739 	{ .soc_id = "r8a774b1", },
740 	{ .soc_id = "r8a774c0", },
741 	{ .soc_id = "r8a774e1", },
742 	{ .soc_id = "r8a7795", },
743 	{ .soc_id = "r8a77961", },
744 	{ .soc_id = "r8a7796", },
745 	{ .soc_id = "r8a77965", },
746 	{ .soc_id = "r8a77970", },
747 	{ .soc_id = "r8a77990", },
748 	{ .soc_id = "r8a77995", },
749 	{ /* sentinel */ }
750 };
751 
752 static const struct soc_device_attribute soc_rcar_gen3_whitelist[] = {
753 	{ .soc_id = "r8a774b1", },
754 	{ .soc_id = "r8a774c0", },
755 	{ .soc_id = "r8a774e1", },
756 	{ .soc_id = "r8a7795", .revision = "ES3.*" },
757 	{ .soc_id = "r8a77961", },
758 	{ .soc_id = "r8a77965", },
759 	{ .soc_id = "r8a77990", },
760 	{ .soc_id = "r8a77995", },
761 	{ /* sentinel */ }
762 };
763 
764 static const char * const rcar_gen3_slave_whitelist[] = {
765 };
766 
767 static bool ipmmu_slave_whitelist(struct device *dev)
768 {
769 	unsigned int i;
770 
771 	/*
772 	 * For R-Car Gen3 use a white list to opt-in slave devices.
773 	 * For Other SoCs, this returns true anyway.
774 	 */
775 	if (!soc_device_match(soc_rcar_gen3))
776 		return true;
777 
778 	/* Check whether this R-Car Gen3 can use the IPMMU correctly or not */
779 	if (!soc_device_match(soc_rcar_gen3_whitelist))
780 		return false;
781 
782 	/* Check whether this slave device can work with the IPMMU */
783 	for (i = 0; i < ARRAY_SIZE(rcar_gen3_slave_whitelist); i++) {
784 		if (!strcmp(dev_name(dev), rcar_gen3_slave_whitelist[i]))
785 			return true;
786 	}
787 
788 	/* Otherwise, do not allow use of IPMMU */
789 	return false;
790 }
791 
792 static int ipmmu_of_xlate(struct device *dev,
793 			  struct of_phandle_args *spec)
794 {
795 	if (!ipmmu_slave_whitelist(dev))
796 		return -ENODEV;
797 
798 	iommu_fwspec_add_ids(dev, spec->args, 1);
799 
800 	/* Initialize once - xlate() will call multiple times */
801 	if (to_ipmmu(dev))
802 		return 0;
803 
804 	return ipmmu_init_platform_device(dev, spec);
805 }
806 
807 static int ipmmu_init_arm_mapping(struct device *dev)
808 {
809 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
810 	int ret;
811 
812 	/*
813 	 * Create the ARM mapping, used by the ARM DMA mapping core to allocate
814 	 * VAs. This will allocate a corresponding IOMMU domain.
815 	 *
816 	 * TODO:
817 	 * - Create one mapping per context (TLB).
818 	 * - Make the mapping size configurable ? We currently use a 2GB mapping
819 	 *   at a 1GB offset to ensure that NULL VAs will fault.
820 	 */
821 	if (!mmu->mapping) {
822 		struct dma_iommu_mapping *mapping;
823 
824 		mapping = arm_iommu_create_mapping(&platform_bus_type,
825 						   SZ_1G, SZ_2G);
826 		if (IS_ERR(mapping)) {
827 			dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
828 			ret = PTR_ERR(mapping);
829 			goto error;
830 		}
831 
832 		mmu->mapping = mapping;
833 	}
834 
835 	/* Attach the ARM VA mapping to the device. */
836 	ret = arm_iommu_attach_device(dev, mmu->mapping);
837 	if (ret < 0) {
838 		dev_err(dev, "Failed to attach device to VA mapping\n");
839 		goto error;
840 	}
841 
842 	return 0;
843 
844 error:
845 	if (mmu->mapping)
846 		arm_iommu_release_mapping(mmu->mapping);
847 
848 	return ret;
849 }
850 
851 static struct iommu_device *ipmmu_probe_device(struct device *dev)
852 {
853 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
854 
855 	/*
856 	 * Only let through devices that have been verified in xlate()
857 	 */
858 	if (!mmu)
859 		return ERR_PTR(-ENODEV);
860 
861 	return &mmu->iommu;
862 }
863 
864 static void ipmmu_probe_finalize(struct device *dev)
865 {
866 	int ret = 0;
867 
868 	if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA))
869 		ret = ipmmu_init_arm_mapping(dev);
870 
871 	if (ret)
872 		dev_err(dev, "Can't create IOMMU mapping - DMA-OPS will not work\n");
873 }
874 
875 static void ipmmu_release_device(struct device *dev)
876 {
877 	arm_iommu_detach_device(dev);
878 }
879 
880 static struct iommu_group *ipmmu_find_group(struct device *dev)
881 {
882 	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
883 	struct iommu_group *group;
884 
885 	if (mmu->group)
886 		return iommu_group_ref_get(mmu->group);
887 
888 	group = iommu_group_alloc();
889 	if (!IS_ERR(group))
890 		mmu->group = group;
891 
892 	return group;
893 }
894 
895 static const struct iommu_ops ipmmu_ops = {
896 	.domain_alloc = ipmmu_domain_alloc,
897 	.domain_free = ipmmu_domain_free,
898 	.attach_dev = ipmmu_attach_device,
899 	.detach_dev = ipmmu_detach_device,
900 	.map = ipmmu_map,
901 	.unmap = ipmmu_unmap,
902 	.flush_iotlb_all = ipmmu_flush_iotlb_all,
903 	.iotlb_sync = ipmmu_iotlb_sync,
904 	.iova_to_phys = ipmmu_iova_to_phys,
905 	.probe_device = ipmmu_probe_device,
906 	.release_device = ipmmu_release_device,
907 	.probe_finalize = ipmmu_probe_finalize,
908 	.device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
909 			? generic_device_group : ipmmu_find_group,
910 	.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
911 	.of_xlate = ipmmu_of_xlate,
912 };
913 
914 /* -----------------------------------------------------------------------------
915  * Probe/remove and init
916  */
917 
918 static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
919 {
920 	unsigned int i;
921 
922 	/* Disable all contexts. */
923 	for (i = 0; i < mmu->num_ctx; ++i)
924 		ipmmu_ctx_write(mmu, i, IMCTR, 0);
925 }
926 
927 static const struct ipmmu_features ipmmu_features_default = {
928 	.use_ns_alias_offset = true,
929 	.has_cache_leaf_nodes = false,
930 	.number_of_contexts = 1, /* software only tested with one context */
931 	.num_utlbs = 32,
932 	.setup_imbuscr = true,
933 	.twobit_imttbcr_sl0 = false,
934 	.reserved_context = false,
935 	.cache_snoop = true,
936 	.ctx_offset_base = 0,
937 	.ctx_offset_stride = 0x40,
938 	.utlb_offset_base = 0,
939 };
940 
941 static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
942 	.use_ns_alias_offset = false,
943 	.has_cache_leaf_nodes = true,
944 	.number_of_contexts = 8,
945 	.num_utlbs = 48,
946 	.setup_imbuscr = false,
947 	.twobit_imttbcr_sl0 = true,
948 	.reserved_context = true,
949 	.cache_snoop = false,
950 	.ctx_offset_base = 0,
951 	.ctx_offset_stride = 0x40,
952 	.utlb_offset_base = 0,
953 };
954 
955 static const struct of_device_id ipmmu_of_ids[] = {
956 	{
957 		.compatible = "renesas,ipmmu-vmsa",
958 		.data = &ipmmu_features_default,
959 	}, {
960 		.compatible = "renesas,ipmmu-r8a774a1",
961 		.data = &ipmmu_features_rcar_gen3,
962 	}, {
963 		.compatible = "renesas,ipmmu-r8a774b1",
964 		.data = &ipmmu_features_rcar_gen3,
965 	}, {
966 		.compatible = "renesas,ipmmu-r8a774c0",
967 		.data = &ipmmu_features_rcar_gen3,
968 	}, {
969 		.compatible = "renesas,ipmmu-r8a774e1",
970 		.data = &ipmmu_features_rcar_gen3,
971 	}, {
972 		.compatible = "renesas,ipmmu-r8a7795",
973 		.data = &ipmmu_features_rcar_gen3,
974 	}, {
975 		.compatible = "renesas,ipmmu-r8a7796",
976 		.data = &ipmmu_features_rcar_gen3,
977 	}, {
978 		.compatible = "renesas,ipmmu-r8a77961",
979 		.data = &ipmmu_features_rcar_gen3,
980 	}, {
981 		.compatible = "renesas,ipmmu-r8a77965",
982 		.data = &ipmmu_features_rcar_gen3,
983 	}, {
984 		.compatible = "renesas,ipmmu-r8a77970",
985 		.data = &ipmmu_features_rcar_gen3,
986 	}, {
987 		.compatible = "renesas,ipmmu-r8a77990",
988 		.data = &ipmmu_features_rcar_gen3,
989 	}, {
990 		.compatible = "renesas,ipmmu-r8a77995",
991 		.data = &ipmmu_features_rcar_gen3,
992 	}, {
993 		/* Terminator */
994 	},
995 };
996 
997 static int ipmmu_probe(struct platform_device *pdev)
998 {
999 	struct ipmmu_vmsa_device *mmu;
1000 	struct resource *res;
1001 	int irq;
1002 	int ret;
1003 
1004 	mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
1005 	if (!mmu) {
1006 		dev_err(&pdev->dev, "cannot allocate device data\n");
1007 		return -ENOMEM;
1008 	}
1009 
1010 	mmu->dev = &pdev->dev;
1011 	spin_lock_init(&mmu->lock);
1012 	bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
1013 	mmu->features = of_device_get_match_data(&pdev->dev);
1014 	memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
1015 	dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
1016 
1017 	/* Map I/O memory and request IRQ. */
1018 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1019 	mmu->base = devm_ioremap_resource(&pdev->dev, res);
1020 	if (IS_ERR(mmu->base))
1021 		return PTR_ERR(mmu->base);
1022 
1023 	/*
1024 	 * The IPMMU has two register banks, for secure and non-secure modes.
1025 	 * The bank mapped at the beginning of the IPMMU address space
1026 	 * corresponds to the running mode of the CPU. When running in secure
1027 	 * mode the non-secure register bank is also available at an offset.
1028 	 *
1029 	 * Secure mode operation isn't clearly documented and is thus currently
1030 	 * not implemented in the driver. Furthermore, preliminary tests of
1031 	 * non-secure operation with the main register bank were not successful.
1032 	 * Offset the registers base unconditionally to point to the non-secure
1033 	 * alias space for now.
1034 	 */
1035 	if (mmu->features->use_ns_alias_offset)
1036 		mmu->base += IM_NS_ALIAS_OFFSET;
1037 
1038 	mmu->num_ctx = min(IPMMU_CTX_MAX, mmu->features->number_of_contexts);
1039 
1040 	/*
1041 	 * Determine if this IPMMU instance is a root device by checking for
1042 	 * the lack of has_cache_leaf_nodes flag or renesas,ipmmu-main property.
1043 	 */
1044 	if (!mmu->features->has_cache_leaf_nodes ||
1045 	    !of_find_property(pdev->dev.of_node, "renesas,ipmmu-main", NULL))
1046 		mmu->root = mmu;
1047 	else
1048 		mmu->root = ipmmu_find_root();
1049 
1050 	/*
1051 	 * Wait until the root device has been registered for sure.
1052 	 */
1053 	if (!mmu->root)
1054 		return -EPROBE_DEFER;
1055 
1056 	/* Root devices have mandatory IRQs */
1057 	if (ipmmu_is_root(mmu)) {
1058 		irq = platform_get_irq(pdev, 0);
1059 		if (irq < 0)
1060 			return irq;
1061 
1062 		ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
1063 				       dev_name(&pdev->dev), mmu);
1064 		if (ret < 0) {
1065 			dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
1066 			return ret;
1067 		}
1068 
1069 		ipmmu_device_reset(mmu);
1070 
1071 		if (mmu->features->reserved_context) {
1072 			dev_info(&pdev->dev, "IPMMU context 0 is reserved\n");
1073 			set_bit(0, mmu->ctx);
1074 		}
1075 	}
1076 
1077 	/*
1078 	 * Register the IPMMU to the IOMMU subsystem in the following cases:
1079 	 * - R-Car Gen2 IPMMU (all devices registered)
1080 	 * - R-Car Gen3 IPMMU (leaf devices only - skip root IPMMU-MM device)
1081 	 */
1082 	if (!mmu->features->has_cache_leaf_nodes || !ipmmu_is_root(mmu)) {
1083 		ret = iommu_device_sysfs_add(&mmu->iommu, &pdev->dev, NULL,
1084 					     dev_name(&pdev->dev));
1085 		if (ret)
1086 			return ret;
1087 
1088 		iommu_device_set_ops(&mmu->iommu, &ipmmu_ops);
1089 		iommu_device_set_fwnode(&mmu->iommu,
1090 					&pdev->dev.of_node->fwnode);
1091 
1092 		ret = iommu_device_register(&mmu->iommu);
1093 		if (ret)
1094 			return ret;
1095 
1096 #if defined(CONFIG_IOMMU_DMA)
1097 		if (!iommu_present(&platform_bus_type))
1098 			bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1099 #endif
1100 	}
1101 
1102 	/*
1103 	 * We can't create the ARM mapping here as it requires the bus to have
1104 	 * an IOMMU, which only happens when bus_set_iommu() is called in
1105 	 * ipmmu_init() after the probe function returns.
1106 	 */
1107 
1108 	platform_set_drvdata(pdev, mmu);
1109 
1110 	return 0;
1111 }
1112 
1113 static int ipmmu_remove(struct platform_device *pdev)
1114 {
1115 	struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1116 
1117 	iommu_device_sysfs_remove(&mmu->iommu);
1118 	iommu_device_unregister(&mmu->iommu);
1119 
1120 	arm_iommu_release_mapping(mmu->mapping);
1121 
1122 	ipmmu_device_reset(mmu);
1123 
1124 	return 0;
1125 }
1126 
1127 #ifdef CONFIG_PM_SLEEP
1128 static int ipmmu_resume_noirq(struct device *dev)
1129 {
1130 	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1131 	unsigned int i;
1132 
1133 	/* Reset root MMU and restore contexts */
1134 	if (ipmmu_is_root(mmu)) {
1135 		ipmmu_device_reset(mmu);
1136 
1137 		for (i = 0; i < mmu->num_ctx; i++) {
1138 			if (!mmu->domains[i])
1139 				continue;
1140 
1141 			ipmmu_domain_setup_context(mmu->domains[i]);
1142 		}
1143 	}
1144 
1145 	/* Re-enable active micro-TLBs */
1146 	for (i = 0; i < mmu->features->num_utlbs; i++) {
1147 		if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1148 			continue;
1149 
1150 		ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1151 	}
1152 
1153 	return 0;
1154 }
1155 
1156 static const struct dev_pm_ops ipmmu_pm  = {
1157 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1158 };
1159 #define DEV_PM_OPS	&ipmmu_pm
1160 #else
1161 #define DEV_PM_OPS	NULL
1162 #endif /* CONFIG_PM_SLEEP */
1163 
1164 static struct platform_driver ipmmu_driver = {
1165 	.driver = {
1166 		.name = "ipmmu-vmsa",
1167 		.of_match_table = of_match_ptr(ipmmu_of_ids),
1168 		.pm = DEV_PM_OPS,
1169 	},
1170 	.probe = ipmmu_probe,
1171 	.remove	= ipmmu_remove,
1172 };
1173 
1174 static int __init ipmmu_init(void)
1175 {
1176 	struct device_node *np;
1177 	static bool setup_done;
1178 	int ret;
1179 
1180 	if (setup_done)
1181 		return 0;
1182 
1183 	np = of_find_matching_node(NULL, ipmmu_of_ids);
1184 	if (!np)
1185 		return 0;
1186 
1187 	of_node_put(np);
1188 
1189 	ret = platform_driver_register(&ipmmu_driver);
1190 	if (ret < 0)
1191 		return ret;
1192 
1193 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
1194 	if (!iommu_present(&platform_bus_type))
1195 		bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1196 #endif
1197 
1198 	setup_done = true;
1199 	return 0;
1200 }
1201 subsys_initcall(ipmmu_init);
1202