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/iopoll.h>
18#include <linux/io-pgtable.h>
19#include <linux/iommu.h>
20#include <linux/of.h>
21#include <linux/of_platform.h>
22#include <linux/pci.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#endif
35
36#define IPMMU_CTX_MAX		16U
37#define IPMMU_CTX_INVALID	-1
38
39#define IPMMU_UTLB_MAX		64U
40
41struct 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
55struct 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 dma_iommu_mapping *mapping;
68};
69
70struct ipmmu_vmsa_domain {
71	struct ipmmu_vmsa_device *mmu;
72	struct iommu_domain io_domain;
73
74	struct io_pgtable_cfg cfg;
75	struct io_pgtable_ops *iop;
76
77	unsigned int context_id;
78	struct mutex mutex;			/* Protects mappings */
79};
80
81static struct ipmmu_vmsa_domain *to_vmsa_domain(struct iommu_domain *dom)
82{
83	return container_of(dom, struct ipmmu_vmsa_domain, io_domain);
84}
85
86static struct ipmmu_vmsa_device *to_ipmmu(struct device *dev)
87{
88	return dev_iommu_priv_get(dev);
89}
90
91#define TLB_LOOP_TIMEOUT		100	/* 100us */
92
93/* -----------------------------------------------------------------------------
94 * Registers Definition
95 */
96
97#define IM_NS_ALIAS_OFFSET		0x800
98
99/* MMU "context" registers */
100#define IMCTR				0x0000		/* R-Car Gen2/3 */
101#define IMCTR_INTEN			(1 << 2)	/* R-Car Gen2/3 */
102#define IMCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
103#define IMCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
104
105#define IMTTBCR				0x0008		/* R-Car Gen2/3 */
106#define IMTTBCR_EAE			(1 << 31)	/* R-Car Gen2/3 */
107#define IMTTBCR_SH0_INNER_SHAREABLE	(3 << 12)	/* R-Car Gen2 only */
108#define IMTTBCR_ORGN0_WB_WA		(1 << 10)	/* R-Car Gen2 only */
109#define IMTTBCR_IRGN0_WB_WA		(1 << 8)	/* R-Car Gen2 only */
110#define IMTTBCR_SL0_TWOBIT_LVL_1	(2 << 6)	/* R-Car Gen3 only */
111#define IMTTBCR_SL0_LVL_1		(1 << 4)	/* R-Car Gen2 only */
112
113#define IMBUSCR				0x000c		/* R-Car Gen2 only */
114#define IMBUSCR_DVM			(1 << 2)	/* R-Car Gen2 only */
115#define IMBUSCR_BUSSEL_MASK		(3 << 0)	/* R-Car Gen2 only */
116
117#define IMTTLBR0			0x0010		/* R-Car Gen2/3 */
118#define IMTTUBR0			0x0014		/* R-Car Gen2/3 */
119
120#define IMSTR				0x0020		/* R-Car Gen2/3 */
121#define IMSTR_MHIT			(1 << 4)	/* R-Car Gen2/3 */
122#define IMSTR_ABORT			(1 << 2)	/* R-Car Gen2/3 */
123#define IMSTR_PF			(1 << 1)	/* R-Car Gen2/3 */
124#define IMSTR_TF			(1 << 0)	/* R-Car Gen2/3 */
125
126#define IMMAIR0				0x0028		/* R-Car Gen2/3 */
127
128#define IMELAR				0x0030		/* R-Car Gen2/3, IMEAR on R-Car Gen2 */
129#define IMEUAR				0x0034		/* R-Car Gen3 only */
130
131/* uTLB registers */
132#define IMUCTR(n)			((n) < 32 ? IMUCTR0(n) : IMUCTR32(n))
133#define IMUCTR0(n)			(0x0300 + ((n) * 16))		/* R-Car Gen2/3 */
134#define IMUCTR32(n)			(0x0600 + (((n) - 32) * 16))	/* R-Car Gen3 only */
135#define IMUCTR_TTSEL_MMU(n)		((n) << 4)	/* R-Car Gen2/3 */
136#define IMUCTR_FLUSH			(1 << 1)	/* R-Car Gen2/3 */
137#define IMUCTR_MMUEN			(1 << 0)	/* R-Car Gen2/3 */
138
139#define IMUASID(n)			((n) < 32 ? IMUASID0(n) : IMUASID32(n))
140#define IMUASID0(n)			(0x0308 + ((n) * 16))		/* R-Car Gen2/3 */
141#define IMUASID32(n)			(0x0608 + (((n) - 32) * 16))	/* R-Car Gen3 only */
142
143/* -----------------------------------------------------------------------------
144 * Root device handling
145 */
146
147static struct platform_driver ipmmu_driver;
148
149static bool ipmmu_is_root(struct ipmmu_vmsa_device *mmu)
150{
151	return mmu->root == mmu;
152}
153
154static int __ipmmu_check_device(struct device *dev, void *data)
155{
156	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
157	struct ipmmu_vmsa_device **rootp = data;
158
159	if (ipmmu_is_root(mmu))
160		*rootp = mmu;
161
162	return 0;
163}
164
165static struct ipmmu_vmsa_device *ipmmu_find_root(void)
166{
167	struct ipmmu_vmsa_device *root = NULL;
168
169	return driver_for_each_device(&ipmmu_driver.driver, NULL, &root,
170				      __ipmmu_check_device) == 0 ? root : NULL;
171}
172
173/* -----------------------------------------------------------------------------
174 * Read/Write Access
175 */
176
177static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
178{
179	return ioread32(mmu->base + offset);
180}
181
182static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
183			u32 data)
184{
185	iowrite32(data, mmu->base + offset);
186}
187
188static unsigned int ipmmu_ctx_reg(struct ipmmu_vmsa_device *mmu,
189				  unsigned int context_id, unsigned int reg)
190{
191	unsigned int base = mmu->features->ctx_offset_base;
192
193	if (context_id > 7)
194		base += 0x800 - 8 * 0x40;
195
196	return base + context_id * mmu->features->ctx_offset_stride + reg;
197}
198
199static u32 ipmmu_ctx_read(struct ipmmu_vmsa_device *mmu,
200			  unsigned int context_id, unsigned int reg)
201{
202	return ipmmu_read(mmu, ipmmu_ctx_reg(mmu, context_id, reg));
203}
204
205static void ipmmu_ctx_write(struct ipmmu_vmsa_device *mmu,
206			    unsigned int context_id, unsigned int reg, u32 data)
207{
208	ipmmu_write(mmu, ipmmu_ctx_reg(mmu, context_id, reg), data);
209}
210
211static u32 ipmmu_ctx_read_root(struct ipmmu_vmsa_domain *domain,
212			       unsigned int reg)
213{
214	return ipmmu_ctx_read(domain->mmu->root, domain->context_id, reg);
215}
216
217static void ipmmu_ctx_write_root(struct ipmmu_vmsa_domain *domain,
218				 unsigned int reg, u32 data)
219{
220	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
221}
222
223static void ipmmu_ctx_write_all(struct ipmmu_vmsa_domain *domain,
224				unsigned int reg, u32 data)
225{
226	if (domain->mmu != domain->mmu->root)
227		ipmmu_ctx_write(domain->mmu, domain->context_id, reg, data);
228
229	ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
230}
231
232static u32 ipmmu_utlb_reg(struct ipmmu_vmsa_device *mmu, unsigned int reg)
233{
234	return mmu->features->utlb_offset_base + reg;
235}
236
237static void ipmmu_imuasid_write(struct ipmmu_vmsa_device *mmu,
238				unsigned int utlb, u32 data)
239{
240	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUASID(utlb)), data);
241}
242
243static void ipmmu_imuctr_write(struct ipmmu_vmsa_device *mmu,
244			       unsigned int utlb, u32 data)
245{
246	ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUCTR(utlb)), data);
247}
248
249/* -----------------------------------------------------------------------------
250 * TLB and microTLB Management
251 */
252
253/* Wait for any pending TLB invalidations to complete */
254static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
255{
256	u32 val;
257
258	if (read_poll_timeout_atomic(ipmmu_ctx_read_root, val,
259				     !(val & IMCTR_FLUSH), 1, TLB_LOOP_TIMEOUT,
260				     false, domain, IMCTR))
261		dev_err_ratelimited(domain->mmu->dev,
262			"TLB sync timed out -- MMU may be deadlocked\n");
263}
264
265static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
266{
267	u32 reg;
268
269	reg = ipmmu_ctx_read_root(domain, IMCTR);
270	reg |= IMCTR_FLUSH;
271	ipmmu_ctx_write_all(domain, IMCTR, reg);
272
273	ipmmu_tlb_sync(domain);
274}
275
276/*
277 * Enable MMU translation for the microTLB.
278 */
279static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
280			      unsigned int utlb)
281{
282	struct ipmmu_vmsa_device *mmu = domain->mmu;
283
284	/*
285	 * TODO: Reference-count the microTLB as several bus masters can be
286	 * connected to the same microTLB.
287	 */
288
289	/* TODO: What should we set the ASID to ? */
290	ipmmu_imuasid_write(mmu, utlb, 0);
291	/* TODO: Do we need to flush the microTLB ? */
292	ipmmu_imuctr_write(mmu, utlb, IMUCTR_TTSEL_MMU(domain->context_id) |
293				      IMUCTR_FLUSH | IMUCTR_MMUEN);
294	mmu->utlb_ctx[utlb] = domain->context_id;
295}
296
297/*
298 * Disable MMU translation for the microTLB.
299 */
300static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
301			       unsigned int utlb)
302{
303	struct ipmmu_vmsa_device *mmu = domain->mmu;
304
305	ipmmu_imuctr_write(mmu, utlb, 0);
306	mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
307}
308
309static void ipmmu_tlb_flush_all(void *cookie)
310{
311	struct ipmmu_vmsa_domain *domain = cookie;
312
313	ipmmu_tlb_invalidate(domain);
314}
315
316static void ipmmu_tlb_flush(unsigned long iova, size_t size,
317				size_t granule, void *cookie)
318{
319	ipmmu_tlb_flush_all(cookie);
320}
321
322static const struct iommu_flush_ops ipmmu_flush_ops = {
323	.tlb_flush_all = ipmmu_tlb_flush_all,
324	.tlb_flush_walk = ipmmu_tlb_flush,
325};
326
327/* -----------------------------------------------------------------------------
328 * Domain/Context Management
329 */
330
331static int ipmmu_domain_allocate_context(struct ipmmu_vmsa_device *mmu,
332					 struct ipmmu_vmsa_domain *domain)
333{
334	unsigned long flags;
335	int ret;
336
337	spin_lock_irqsave(&mmu->lock, flags);
338
339	ret = find_first_zero_bit(mmu->ctx, mmu->num_ctx);
340	if (ret != mmu->num_ctx) {
341		mmu->domains[ret] = domain;
342		set_bit(ret, mmu->ctx);
343	} else
344		ret = -EBUSY;
345
346	spin_unlock_irqrestore(&mmu->lock, flags);
347
348	return ret;
349}
350
351static void ipmmu_domain_free_context(struct ipmmu_vmsa_device *mmu,
352				      unsigned int context_id)
353{
354	unsigned long flags;
355
356	spin_lock_irqsave(&mmu->lock, flags);
357
358	clear_bit(context_id, mmu->ctx);
359	mmu->domains[context_id] = NULL;
360
361	spin_unlock_irqrestore(&mmu->lock, flags);
362}
363
364static void ipmmu_domain_setup_context(struct ipmmu_vmsa_domain *domain)
365{
366	u64 ttbr;
367	u32 tmp;
368
369	/* TTBR0 */
370	ttbr = domain->cfg.arm_lpae_s1_cfg.ttbr;
371	ipmmu_ctx_write_root(domain, IMTTLBR0, ttbr);
372	ipmmu_ctx_write_root(domain, IMTTUBR0, ttbr >> 32);
373
374	/*
375	 * TTBCR
376	 * We use long descriptors and allocate the whole 32-bit VA space to
377	 * TTBR0.
378	 */
379	if (domain->mmu->features->twobit_imttbcr_sl0)
380		tmp = IMTTBCR_SL0_TWOBIT_LVL_1;
381	else
382		tmp = IMTTBCR_SL0_LVL_1;
383
384	if (domain->mmu->features->cache_snoop)
385		tmp |= IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
386		       IMTTBCR_IRGN0_WB_WA;
387
388	ipmmu_ctx_write_root(domain, IMTTBCR, IMTTBCR_EAE | tmp);
389
390	/* MAIR0 */
391	ipmmu_ctx_write_root(domain, IMMAIR0,
392			     domain->cfg.arm_lpae_s1_cfg.mair);
393
394	/* IMBUSCR */
395	if (domain->mmu->features->setup_imbuscr)
396		ipmmu_ctx_write_root(domain, IMBUSCR,
397				     ipmmu_ctx_read_root(domain, IMBUSCR) &
398				     ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
399
400	/*
401	 * IMSTR
402	 * Clear all interrupt flags.
403	 */
404	ipmmu_ctx_write_root(domain, IMSTR, ipmmu_ctx_read_root(domain, IMSTR));
405
406	/*
407	 * IMCTR
408	 * Enable the MMU and interrupt generation. The long-descriptor
409	 * translation table format doesn't use TEX remapping. Don't enable AF
410	 * software management as we have no use for it. Flush the TLB as
411	 * required when modifying the context registers.
412	 */
413	ipmmu_ctx_write_all(domain, IMCTR,
414			    IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
415}
416
417static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
418{
419	int ret;
420
421	/*
422	 * Allocate the page table operations.
423	 *
424	 * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory
425	 * access, Long-descriptor format" that the NStable bit being set in a
426	 * table descriptor will result in the NStable and NS bits of all child
427	 * entries being ignored and considered as being set. The IPMMU seems
428	 * not to comply with this, as it generates a secure access page fault
429	 * if any of the NStable and NS bits isn't set when running in
430	 * non-secure mode.
431	 */
432	domain->cfg.quirks = IO_PGTABLE_QUIRK_ARM_NS;
433	domain->cfg.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K;
434	domain->cfg.ias = 32;
435	domain->cfg.oas = 40;
436	domain->cfg.tlb = &ipmmu_flush_ops;
437	domain->io_domain.geometry.aperture_end = DMA_BIT_MASK(32);
438	domain->io_domain.geometry.force_aperture = true;
439	/*
440	 * TODO: Add support for coherent walk through CCI with DVM and remove
441	 * cache handling. For now, delegate it to the io-pgtable code.
442	 */
443	domain->cfg.coherent_walk = false;
444	domain->cfg.iommu_dev = domain->mmu->root->dev;
445
446	/*
447	 * Find an unused context.
448	 */
449	ret = ipmmu_domain_allocate_context(domain->mmu->root, domain);
450	if (ret < 0)
451		return ret;
452
453	domain->context_id = ret;
454
455	domain->iop = alloc_io_pgtable_ops(ARM_32_LPAE_S1, &domain->cfg,
456					   domain);
457	if (!domain->iop) {
458		ipmmu_domain_free_context(domain->mmu->root,
459					  domain->context_id);
460		return -EINVAL;
461	}
462
463	ipmmu_domain_setup_context(domain);
464	return 0;
465}
466
467static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
468{
469	if (!domain->mmu)
470		return;
471
472	/*
473	 * Disable the context. Flush the TLB as required when modifying the
474	 * context registers.
475	 *
476	 * TODO: Is TLB flush really needed ?
477	 */
478	ipmmu_ctx_write_all(domain, IMCTR, IMCTR_FLUSH);
479	ipmmu_tlb_sync(domain);
480	ipmmu_domain_free_context(domain->mmu->root, domain->context_id);
481}
482
483/* -----------------------------------------------------------------------------
484 * Fault Handling
485 */
486
487static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
488{
489	const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
490	struct ipmmu_vmsa_device *mmu = domain->mmu;
491	unsigned long iova;
492	u32 status;
493
494	status = ipmmu_ctx_read_root(domain, IMSTR);
495	if (!(status & err_mask))
496		return IRQ_NONE;
497
498	iova = ipmmu_ctx_read_root(domain, IMELAR);
499	if (IS_ENABLED(CONFIG_64BIT))
500		iova |= (u64)ipmmu_ctx_read_root(domain, IMEUAR) << 32;
501
502	/*
503	 * Clear the error status flags. Unlike traditional interrupt flag
504	 * registers that must be cleared by writing 1, this status register
505	 * seems to require 0. The error address register must be read before,
506	 * otherwise its value will be 0.
507	 */
508	ipmmu_ctx_write_root(domain, IMSTR, 0);
509
510	/* Log fatal errors. */
511	if (status & IMSTR_MHIT)
512		dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%lx\n",
513				    iova);
514	if (status & IMSTR_ABORT)
515		dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%lx\n",
516				    iova);
517
518	if (!(status & (IMSTR_PF | IMSTR_TF)))
519		return IRQ_NONE;
520
521	/*
522	 * Try to handle page faults and translation faults.
523	 *
524	 * TODO: We need to look up the faulty device based on the I/O VA. Use
525	 * the IOMMU device for now.
526	 */
527	if (!report_iommu_fault(&domain->io_domain, mmu->dev, iova, 0))
528		return IRQ_HANDLED;
529
530	dev_err_ratelimited(mmu->dev,
531			    "Unhandled fault: status 0x%08x iova 0x%lx\n",
532			    status, iova);
533
534	return IRQ_HANDLED;
535}
536
537static irqreturn_t ipmmu_irq(int irq, void *dev)
538{
539	struct ipmmu_vmsa_device *mmu = dev;
540	irqreturn_t status = IRQ_NONE;
541	unsigned int i;
542	unsigned long flags;
543
544	spin_lock_irqsave(&mmu->lock, flags);
545
546	/*
547	 * Check interrupts for all active contexts.
548	 */
549	for (i = 0; i < mmu->num_ctx; i++) {
550		if (!mmu->domains[i])
551			continue;
552		if (ipmmu_domain_irq(mmu->domains[i]) == IRQ_HANDLED)
553			status = IRQ_HANDLED;
554	}
555
556	spin_unlock_irqrestore(&mmu->lock, flags);
557
558	return status;
559}
560
561/* -----------------------------------------------------------------------------
562 * IOMMU Operations
563 */
564
565static struct iommu_domain *ipmmu_domain_alloc_paging(struct device *dev)
566{
567	struct ipmmu_vmsa_domain *domain;
568
569	domain = kzalloc(sizeof(*domain), GFP_KERNEL);
570	if (!domain)
571		return NULL;
572
573	mutex_init(&domain->mutex);
574
575	return &domain->io_domain;
576}
577
578static void ipmmu_domain_free(struct iommu_domain *io_domain)
579{
580	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
581
582	/*
583	 * Free the domain resources. We assume that all devices have already
584	 * been detached.
585	 */
586	ipmmu_domain_destroy_context(domain);
587	free_io_pgtable_ops(domain->iop);
588	kfree(domain);
589}
590
591static int ipmmu_attach_device(struct iommu_domain *io_domain,
592			       struct device *dev)
593{
594	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
595	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
596	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
597	unsigned int i;
598	int ret = 0;
599
600	if (!mmu) {
601		dev_err(dev, "Cannot attach to IPMMU\n");
602		return -ENXIO;
603	}
604
605	mutex_lock(&domain->mutex);
606
607	if (!domain->mmu) {
608		/* The domain hasn't been used yet, initialize it. */
609		domain->mmu = mmu;
610		ret = ipmmu_domain_init_context(domain);
611		if (ret < 0) {
612			dev_err(dev, "Unable to initialize IPMMU context\n");
613			domain->mmu = NULL;
614		} else {
615			dev_info(dev, "Using IPMMU context %u\n",
616				 domain->context_id);
617		}
618	} else if (domain->mmu != mmu) {
619		/*
620		 * Something is wrong, we can't attach two devices using
621		 * different IOMMUs to the same domain.
622		 */
623		ret = -EINVAL;
624	} else
625		dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
626
627	mutex_unlock(&domain->mutex);
628
629	if (ret < 0)
630		return ret;
631
632	for (i = 0; i < fwspec->num_ids; ++i)
633		ipmmu_utlb_enable(domain, fwspec->ids[i]);
634
635	return 0;
636}
637
638static int ipmmu_iommu_identity_attach(struct iommu_domain *identity_domain,
639				       struct device *dev)
640{
641	struct iommu_domain *io_domain = iommu_get_domain_for_dev(dev);
642	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
643	struct ipmmu_vmsa_domain *domain;
644	unsigned int i;
645
646	if (io_domain == identity_domain || !io_domain)
647		return 0;
648
649	domain = to_vmsa_domain(io_domain);
650	for (i = 0; i < fwspec->num_ids; ++i)
651		ipmmu_utlb_disable(domain, fwspec->ids[i]);
652
653	/*
654	 * TODO: Optimize by disabling the context when no device is attached.
655	 */
656	return 0;
657}
658
659static struct iommu_domain_ops ipmmu_iommu_identity_ops = {
660	.attach_dev = ipmmu_iommu_identity_attach,
661};
662
663static struct iommu_domain ipmmu_iommu_identity_domain = {
664	.type = IOMMU_DOMAIN_IDENTITY,
665	.ops = &ipmmu_iommu_identity_ops,
666};
667
668static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
669		     phys_addr_t paddr, size_t pgsize, size_t pgcount,
670		     int prot, gfp_t gfp, size_t *mapped)
671{
672	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
673
674	return domain->iop->map_pages(domain->iop, iova, paddr, pgsize, pgcount,
675				      prot, gfp, mapped);
676}
677
678static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
679			  size_t pgsize, size_t pgcount,
680			  struct iommu_iotlb_gather *gather)
681{
682	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
683
684	return domain->iop->unmap_pages(domain->iop, iova, pgsize, pgcount, gather);
685}
686
687static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
688{
689	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
690
691	if (domain->mmu)
692		ipmmu_tlb_flush_all(domain);
693}
694
695static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
696			     struct iommu_iotlb_gather *gather)
697{
698	ipmmu_flush_iotlb_all(io_domain);
699}
700
701static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
702				      dma_addr_t iova)
703{
704	struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
705
706	/* TODO: Is locking needed ? */
707
708	return domain->iop->iova_to_phys(domain->iop, iova);
709}
710
711static int ipmmu_init_platform_device(struct device *dev,
712				      const struct of_phandle_args *args)
713{
714	struct platform_device *ipmmu_pdev;
715
716	ipmmu_pdev = of_find_device_by_node(args->np);
717	if (!ipmmu_pdev)
718		return -ENODEV;
719
720	dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
721
722	return 0;
723}
724
725static const struct soc_device_attribute soc_needs_opt_in[] = {
726	{ .family = "R-Car Gen3", },
727	{ .family = "R-Car Gen4", },
728	{ .family = "RZ/G2", },
729	{ /* sentinel */ }
730};
731
732static const struct soc_device_attribute soc_denylist[] = {
733	{ .soc_id = "r8a774a1", },
734	{ .soc_id = "r8a7795", .revision = "ES2.*" },
735	{ .soc_id = "r8a7796", },
736	{ /* sentinel */ }
737};
738
739static const char * const devices_allowlist[] = {
740	"ee100000.mmc",
741	"ee120000.mmc",
742	"ee140000.mmc",
743	"ee160000.mmc"
744};
745
746static bool ipmmu_device_is_allowed(struct device *dev)
747{
748	unsigned int i;
749
750	/*
751	 * R-Car Gen3/4 and RZ/G2 use the allow list to opt-in devices.
752	 * For Other SoCs, this returns true anyway.
753	 */
754	if (!soc_device_match(soc_needs_opt_in))
755		return true;
756
757	/* Check whether this SoC can use the IPMMU correctly or not */
758	if (soc_device_match(soc_denylist))
759		return false;
760
761	/* Check whether this device is a PCI device */
762	if (dev_is_pci(dev))
763		return true;
764
765	/* Check whether this device can work with the IPMMU */
766	for (i = 0; i < ARRAY_SIZE(devices_allowlist); i++) {
767		if (!strcmp(dev_name(dev), devices_allowlist[i]))
768			return true;
769	}
770
771	/* Otherwise, do not allow use of IPMMU */
772	return false;
773}
774
775static int ipmmu_of_xlate(struct device *dev,
776			  const struct of_phandle_args *spec)
777{
778	if (!ipmmu_device_is_allowed(dev))
779		return -ENODEV;
780
781	iommu_fwspec_add_ids(dev, spec->args, 1);
782
783	/* Initialize once - xlate() will call multiple times */
784	if (to_ipmmu(dev))
785		return 0;
786
787	return ipmmu_init_platform_device(dev, spec);
788}
789
790static int ipmmu_init_arm_mapping(struct device *dev)
791{
792	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
793	int ret;
794
795	/*
796	 * Create the ARM mapping, used by the ARM DMA mapping core to allocate
797	 * VAs. This will allocate a corresponding IOMMU domain.
798	 *
799	 * TODO:
800	 * - Create one mapping per context (TLB).
801	 * - Make the mapping size configurable ? We currently use a 2GB mapping
802	 *   at a 1GB offset to ensure that NULL VAs will fault.
803	 */
804	if (!mmu->mapping) {
805		struct dma_iommu_mapping *mapping;
806
807		mapping = arm_iommu_create_mapping(&platform_bus_type,
808						   SZ_1G, SZ_2G);
809		if (IS_ERR(mapping)) {
810			dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
811			ret = PTR_ERR(mapping);
812			goto error;
813		}
814
815		mmu->mapping = mapping;
816	}
817
818	/* Attach the ARM VA mapping to the device. */
819	ret = arm_iommu_attach_device(dev, mmu->mapping);
820	if (ret < 0) {
821		dev_err(dev, "Failed to attach device to VA mapping\n");
822		goto error;
823	}
824
825	return 0;
826
827error:
828	if (mmu->mapping)
829		arm_iommu_release_mapping(mmu->mapping);
830
831	return ret;
832}
833
834static struct iommu_device *ipmmu_probe_device(struct device *dev)
835{
836	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
837
838	/*
839	 * Only let through devices that have been verified in xlate()
840	 */
841	if (!mmu)
842		return ERR_PTR(-ENODEV);
843
844	return &mmu->iommu;
845}
846
847static void ipmmu_probe_finalize(struct device *dev)
848{
849	int ret = 0;
850
851	if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA))
852		ret = ipmmu_init_arm_mapping(dev);
853
854	if (ret)
855		dev_err(dev, "Can't create IOMMU mapping - DMA-OPS will not work\n");
856}
857
858static void ipmmu_release_device(struct device *dev)
859{
860	struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
861	struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
862	unsigned int i;
863
864	for (i = 0; i < fwspec->num_ids; ++i) {
865		unsigned int utlb = fwspec->ids[i];
866
867		ipmmu_imuctr_write(mmu, utlb, 0);
868		mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
869	}
870
871	arm_iommu_release_mapping(mmu->mapping);
872}
873
874static const struct iommu_ops ipmmu_ops = {
875	.identity_domain = &ipmmu_iommu_identity_domain,
876	.domain_alloc_paging = ipmmu_domain_alloc_paging,
877	.probe_device = ipmmu_probe_device,
878	.release_device = ipmmu_release_device,
879	.probe_finalize = ipmmu_probe_finalize,
880	/*
881	 * FIXME: The device grouping is a fixed property of the hardware's
882	 * ability to isolate and control DMA, it should not depend on kconfig.
883	 */
884	.device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
885			? generic_device_group : generic_single_device_group,
886	.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
887	.of_xlate = ipmmu_of_xlate,
888	.default_domain_ops = &(const struct iommu_domain_ops) {
889		.attach_dev	= ipmmu_attach_device,
890		.map_pages	= ipmmu_map,
891		.unmap_pages	= 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		.free		= ipmmu_domain_free,
896	}
897};
898
899/* -----------------------------------------------------------------------------
900 * Probe/remove and init
901 */
902
903static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
904{
905	unsigned int i;
906
907	/* Disable all contexts. */
908	for (i = 0; i < mmu->num_ctx; ++i)
909		ipmmu_ctx_write(mmu, i, IMCTR, 0);
910}
911
912static const struct ipmmu_features ipmmu_features_default = {
913	.use_ns_alias_offset = true,
914	.has_cache_leaf_nodes = false,
915	.number_of_contexts = 1, /* software only tested with one context */
916	.num_utlbs = 32,
917	.setup_imbuscr = true,
918	.twobit_imttbcr_sl0 = false,
919	.reserved_context = false,
920	.cache_snoop = true,
921	.ctx_offset_base = 0,
922	.ctx_offset_stride = 0x40,
923	.utlb_offset_base = 0,
924};
925
926static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
927	.use_ns_alias_offset = false,
928	.has_cache_leaf_nodes = true,
929	.number_of_contexts = 8,
930	.num_utlbs = 48,
931	.setup_imbuscr = false,
932	.twobit_imttbcr_sl0 = true,
933	.reserved_context = true,
934	.cache_snoop = false,
935	.ctx_offset_base = 0,
936	.ctx_offset_stride = 0x40,
937	.utlb_offset_base = 0,
938};
939
940static const struct ipmmu_features ipmmu_features_rcar_gen4 = {
941	.use_ns_alias_offset = false,
942	.has_cache_leaf_nodes = true,
943	.number_of_contexts = 16,
944	.num_utlbs = 64,
945	.setup_imbuscr = false,
946	.twobit_imttbcr_sl0 = true,
947	.reserved_context = true,
948	.cache_snoop = false,
949	.ctx_offset_base = 0x10000,
950	.ctx_offset_stride = 0x1040,
951	.utlb_offset_base = 0x3000,
952};
953
954static const struct of_device_id ipmmu_of_ids[] = {
955	{
956		.compatible = "renesas,ipmmu-vmsa",
957		.data = &ipmmu_features_default,
958	}, {
959		.compatible = "renesas,ipmmu-r8a774a1",
960		.data = &ipmmu_features_rcar_gen3,
961	}, {
962		.compatible = "renesas,ipmmu-r8a774b1",
963		.data = &ipmmu_features_rcar_gen3,
964	}, {
965		.compatible = "renesas,ipmmu-r8a774c0",
966		.data = &ipmmu_features_rcar_gen3,
967	}, {
968		.compatible = "renesas,ipmmu-r8a774e1",
969		.data = &ipmmu_features_rcar_gen3,
970	}, {
971		.compatible = "renesas,ipmmu-r8a7795",
972		.data = &ipmmu_features_rcar_gen3,
973	}, {
974		.compatible = "renesas,ipmmu-r8a7796",
975		.data = &ipmmu_features_rcar_gen3,
976	}, {
977		.compatible = "renesas,ipmmu-r8a77961",
978		.data = &ipmmu_features_rcar_gen3,
979	}, {
980		.compatible = "renesas,ipmmu-r8a77965",
981		.data = &ipmmu_features_rcar_gen3,
982	}, {
983		.compatible = "renesas,ipmmu-r8a77970",
984		.data = &ipmmu_features_rcar_gen3,
985	}, {
986		.compatible = "renesas,ipmmu-r8a77980",
987		.data = &ipmmu_features_rcar_gen3,
988	}, {
989		.compatible = "renesas,ipmmu-r8a77990",
990		.data = &ipmmu_features_rcar_gen3,
991	}, {
992		.compatible = "renesas,ipmmu-r8a77995",
993		.data = &ipmmu_features_rcar_gen3,
994	}, {
995		.compatible = "renesas,ipmmu-r8a779a0",
996		.data = &ipmmu_features_rcar_gen4,
997	}, {
998		.compatible = "renesas,rcar-gen4-ipmmu-vmsa",
999		.data = &ipmmu_features_rcar_gen4,
1000	}, {
1001		/* Terminator */
1002	},
1003};
1004
1005static int ipmmu_probe(struct platform_device *pdev)
1006{
1007	struct ipmmu_vmsa_device *mmu;
1008	int irq;
1009	int ret;
1010
1011	mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
1012	if (!mmu) {
1013		dev_err(&pdev->dev, "cannot allocate device data\n");
1014		return -ENOMEM;
1015	}
1016
1017	mmu->dev = &pdev->dev;
1018	spin_lock_init(&mmu->lock);
1019	bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
1020	mmu->features = of_device_get_match_data(&pdev->dev);
1021	memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
1022	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
1023	if (ret)
1024		return ret;
1025
1026	/* Map I/O memory and request IRQ. */
1027	mmu->base = devm_platform_ioremap_resource(pdev, 0);
1028	if (IS_ERR(mmu->base))
1029		return PTR_ERR(mmu->base);
1030
1031	/*
1032	 * The IPMMU has two register banks, for secure and non-secure modes.
1033	 * The bank mapped at the beginning of the IPMMU address space
1034	 * corresponds to the running mode of the CPU. When running in secure
1035	 * mode the non-secure register bank is also available at an offset.
1036	 *
1037	 * Secure mode operation isn't clearly documented and is thus currently
1038	 * not implemented in the driver. Furthermore, preliminary tests of
1039	 * non-secure operation with the main register bank were not successful.
1040	 * Offset the registers base unconditionally to point to the non-secure
1041	 * alias space for now.
1042	 */
1043	if (mmu->features->use_ns_alias_offset)
1044		mmu->base += IM_NS_ALIAS_OFFSET;
1045
1046	mmu->num_ctx = min(IPMMU_CTX_MAX, mmu->features->number_of_contexts);
1047
1048	/*
1049	 * Determine if this IPMMU instance is a root device by checking for
1050	 * the lack of has_cache_leaf_nodes flag or renesas,ipmmu-main property.
1051	 */
1052	if (!mmu->features->has_cache_leaf_nodes ||
1053	    !of_property_present(pdev->dev.of_node, "renesas,ipmmu-main"))
1054		mmu->root = mmu;
1055	else
1056		mmu->root = ipmmu_find_root();
1057
1058	/*
1059	 * Wait until the root device has been registered for sure.
1060	 */
1061	if (!mmu->root)
1062		return -EPROBE_DEFER;
1063
1064	/* Root devices have mandatory IRQs */
1065	if (ipmmu_is_root(mmu)) {
1066		irq = platform_get_irq(pdev, 0);
1067		if (irq < 0)
1068			return irq;
1069
1070		ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
1071				       dev_name(&pdev->dev), mmu);
1072		if (ret < 0) {
1073			dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
1074			return ret;
1075		}
1076
1077		ipmmu_device_reset(mmu);
1078
1079		if (mmu->features->reserved_context) {
1080			dev_info(&pdev->dev, "IPMMU context 0 is reserved\n");
1081			set_bit(0, mmu->ctx);
1082		}
1083	}
1084
1085	/*
1086	 * Register the IPMMU to the IOMMU subsystem in the following cases:
1087	 * - R-Car Gen2 IPMMU (all devices registered)
1088	 * - R-Car Gen3 IPMMU (leaf devices only - skip root IPMMU-MM device)
1089	 */
1090	if (!mmu->features->has_cache_leaf_nodes || !ipmmu_is_root(mmu)) {
1091		ret = iommu_device_sysfs_add(&mmu->iommu, &pdev->dev, NULL,
1092					     dev_name(&pdev->dev));
1093		if (ret)
1094			return ret;
1095
1096		ret = iommu_device_register(&mmu->iommu, &ipmmu_ops, &pdev->dev);
1097		if (ret)
1098			return ret;
1099	}
1100
1101	/*
1102	 * We can't create the ARM mapping here as it requires the bus to have
1103	 * an IOMMU, which only happens when bus_set_iommu() is called in
1104	 * ipmmu_init() after the probe function returns.
1105	 */
1106
1107	platform_set_drvdata(pdev, mmu);
1108
1109	return 0;
1110}
1111
1112static void ipmmu_remove(struct platform_device *pdev)
1113{
1114	struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1115
1116	iommu_device_sysfs_remove(&mmu->iommu);
1117	iommu_device_unregister(&mmu->iommu);
1118
1119	arm_iommu_release_mapping(mmu->mapping);
1120
1121	ipmmu_device_reset(mmu);
1122}
1123
1124static int ipmmu_resume_noirq(struct device *dev)
1125{
1126	struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1127	unsigned int i;
1128
1129	/* Reset root MMU and restore contexts */
1130	if (ipmmu_is_root(mmu)) {
1131		ipmmu_device_reset(mmu);
1132
1133		for (i = 0; i < mmu->num_ctx; i++) {
1134			if (!mmu->domains[i])
1135				continue;
1136
1137			ipmmu_domain_setup_context(mmu->domains[i]);
1138		}
1139	}
1140
1141	/* Re-enable active micro-TLBs */
1142	for (i = 0; i < mmu->features->num_utlbs; i++) {
1143		if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1144			continue;
1145
1146		ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1147	}
1148
1149	return 0;
1150}
1151
1152static const struct dev_pm_ops ipmmu_pm  = {
1153	NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1154};
1155
1156static struct platform_driver ipmmu_driver = {
1157	.driver = {
1158		.name = "ipmmu-vmsa",
1159		.of_match_table = ipmmu_of_ids,
1160		.pm = pm_sleep_ptr(&ipmmu_pm),
1161	},
1162	.probe = ipmmu_probe,
1163	.remove_new = ipmmu_remove,
1164};
1165builtin_platform_driver(ipmmu_driver);
1166