Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[platform/kernel/linux-starfive.git] / drivers / iommu / ipmmu-vmsa.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * IOMMU API for Renesas VMSA-compatible IPMMU
4  * Author: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
5  *
6  * Copyright (C) 2014-2020 Renesas Electronics Corporation
7  */
8
9 #include <linux/bitmap.h>
10 #include <linux/delay.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/err.h>
13 #include <linux/export.h>
14 #include <linux/init.h>
15 #include <linux/interrupt.h>
16 #include <linux/io.h>
17 #include <linux/io-pgtable.h>
18 #include <linux/iommu.h>
19 #include <linux/of.h>
20 #include <linux/of_device.h>
21 #include <linux/of_platform.h>
22 #include <linux/platform_device.h>
23 #include <linux/sizes.h>
24 #include <linux/slab.h>
25 #include <linux/sys_soc.h>
26
27 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
28 #include <asm/dma-iommu.h>
29 #else
30 #define arm_iommu_create_mapping(...)   NULL
31 #define arm_iommu_attach_device(...)    -ENODEV
32 #define arm_iommu_release_mapping(...)  do {} while (0)
33 #define arm_iommu_detach_device(...)    do {} while (0)
34 #endif
35
36 #define IPMMU_CTX_MAX           8U
37 #define IPMMU_CTX_INVALID       -1
38
39 #define IPMMU_UTLB_MAX          48U
40
41 struct ipmmu_features {
42         bool use_ns_alias_offset;
43         bool has_cache_leaf_nodes;
44         unsigned int number_of_contexts;
45         unsigned int num_utlbs;
46         bool setup_imbuscr;
47         bool twobit_imttbcr_sl0;
48         bool reserved_context;
49         bool cache_snoop;
50         unsigned int ctx_offset_base;
51         unsigned int ctx_offset_stride;
52         unsigned int utlb_offset_base;
53 };
54
55 struct ipmmu_vmsa_device {
56         struct device *dev;
57         void __iomem *base;
58         struct iommu_device iommu;
59         struct ipmmu_vmsa_device *root;
60         const struct ipmmu_features *features;
61         unsigned int num_ctx;
62         spinlock_t lock;                        /* Protects ctx and domains[] */
63         DECLARE_BITMAP(ctx, IPMMU_CTX_MAX);
64         struct ipmmu_vmsa_domain *domains[IPMMU_CTX_MAX];
65         s8 utlb_ctx[IPMMU_UTLB_MAX];
66
67         struct iommu_group *group;
68         struct dma_iommu_mapping *mapping;
69 };
70
71 struct ipmmu_vmsa_domain {
72         struct ipmmu_vmsa_device *mmu;
73         struct iommu_domain io_domain;
74
75         struct io_pgtable_cfg cfg;
76         struct io_pgtable_ops *iop;
77
78         unsigned int context_id;
79         struct mutex mutex;                     /* Protects mappings */
80 };
81
82 static struct ipmmu_vmsa_domain *to_vmsa_domain(struct iommu_domain *dom)
83 {
84         return container_of(dom, struct ipmmu_vmsa_domain, io_domain);
85 }
86
87 static struct ipmmu_vmsa_device *to_ipmmu(struct device *dev)
88 {
89         return dev_iommu_priv_get(dev);
90 }
91
92 #define TLB_LOOP_TIMEOUT                100     /* 100us */
93
94 /* -----------------------------------------------------------------------------
95  * Registers Definition
96  */
97
98 #define IM_NS_ALIAS_OFFSET              0x800
99
100 /* MMU "context" registers */
101 #define IMCTR                           0x0000          /* R-Car Gen2/3 */
102 #define IMCTR_INTEN                     (1 << 2)        /* R-Car Gen2/3 */
103 #define IMCTR_FLUSH                     (1 << 1)        /* R-Car Gen2/3 */
104 #define IMCTR_MMUEN                     (1 << 0)        /* R-Car Gen2/3 */
105
106 #define IMTTBCR                         0x0008          /* R-Car Gen2/3 */
107 #define IMTTBCR_EAE                     (1 << 31)       /* R-Car Gen2/3 */
108 #define IMTTBCR_SH0_INNER_SHAREABLE     (3 << 12)       /* R-Car Gen2 only */
109 #define IMTTBCR_ORGN0_WB_WA             (1 << 10)       /* R-Car Gen2 only */
110 #define IMTTBCR_IRGN0_WB_WA             (1 << 8)        /* R-Car Gen2 only */
111 #define IMTTBCR_SL0_TWOBIT_LVL_1        (2 << 6)        /* R-Car Gen3 only */
112 #define IMTTBCR_SL0_LVL_1               (1 << 4)        /* R-Car Gen2 only */
113
114 #define IMBUSCR                         0x000c          /* R-Car Gen2 only */
115 #define IMBUSCR_DVM                     (1 << 2)        /* R-Car Gen2 only */
116 #define IMBUSCR_BUSSEL_MASK             (3 << 0)        /* R-Car Gen2 only */
117
118 #define IMTTLBR0                        0x0010          /* R-Car Gen2/3 */
119 #define IMTTUBR0                        0x0014          /* R-Car Gen2/3 */
120
121 #define IMSTR                           0x0020          /* R-Car Gen2/3 */
122 #define IMSTR_MHIT                      (1 << 4)        /* R-Car Gen2/3 */
123 #define IMSTR_ABORT                     (1 << 2)        /* R-Car Gen2/3 */
124 #define IMSTR_PF                        (1 << 1)        /* R-Car Gen2/3 */
125 #define IMSTR_TF                        (1 << 0)        /* R-Car Gen2/3 */
126
127 #define IMMAIR0                         0x0028          /* R-Car Gen2/3 */
128
129 #define IMELAR                          0x0030          /* R-Car Gen2/3, IMEAR on R-Car Gen2 */
130 #define IMEUAR                          0x0034          /* R-Car Gen3 only */
131
132 /* uTLB registers */
133 #define IMUCTR(n)                       ((n) < 32 ? IMUCTR0(n) : IMUCTR32(n))
134 #define IMUCTR0(n)                      (0x0300 + ((n) * 16))           /* R-Car Gen2/3 */
135 #define IMUCTR32(n)                     (0x0600 + (((n) - 32) * 16))    /* R-Car Gen3 only */
136 #define IMUCTR_TTSEL_MMU(n)             ((n) << 4)      /* R-Car Gen2/3 */
137 #define IMUCTR_FLUSH                    (1 << 1)        /* R-Car Gen2/3 */
138 #define IMUCTR_MMUEN                    (1 << 0)        /* R-Car Gen2/3 */
139
140 #define IMUASID(n)                      ((n) < 32 ? IMUASID0(n) : IMUASID32(n))
141 #define IMUASID0(n)                     (0x0308 + ((n) * 16))           /* R-Car Gen2/3 */
142 #define IMUASID32(n)                    (0x0608 + (((n) - 32) * 16))    /* R-Car Gen3 only */
143
144 /* -----------------------------------------------------------------------------
145  * Root device handling
146  */
147
148 static struct platform_driver ipmmu_driver;
149
150 static bool ipmmu_is_root(struct ipmmu_vmsa_device *mmu)
151 {
152         return mmu->root == mmu;
153 }
154
155 static int __ipmmu_check_device(struct device *dev, void *data)
156 {
157         struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
158         struct ipmmu_vmsa_device **rootp = data;
159
160         if (ipmmu_is_root(mmu))
161                 *rootp = mmu;
162
163         return 0;
164 }
165
166 static struct ipmmu_vmsa_device *ipmmu_find_root(void)
167 {
168         struct ipmmu_vmsa_device *root = NULL;
169
170         return driver_for_each_device(&ipmmu_driver.driver, NULL, &root,
171                                       __ipmmu_check_device) == 0 ? root : NULL;
172 }
173
174 /* -----------------------------------------------------------------------------
175  * Read/Write Access
176  */
177
178 static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
179 {
180         return ioread32(mmu->base + offset);
181 }
182
183 static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
184                         u32 data)
185 {
186         iowrite32(data, mmu->base + offset);
187 }
188
189 static unsigned int ipmmu_ctx_reg(struct ipmmu_vmsa_device *mmu,
190                                   unsigned int context_id, unsigned int reg)
191 {
192         return mmu->features->ctx_offset_base +
193                context_id * mmu->features->ctx_offset_stride + reg;
194 }
195
196 static u32 ipmmu_ctx_read(struct ipmmu_vmsa_device *mmu,
197                           unsigned int context_id, unsigned int reg)
198 {
199         return ipmmu_read(mmu, ipmmu_ctx_reg(mmu, context_id, reg));
200 }
201
202 static void ipmmu_ctx_write(struct ipmmu_vmsa_device *mmu,
203                             unsigned int context_id, unsigned int reg, u32 data)
204 {
205         ipmmu_write(mmu, ipmmu_ctx_reg(mmu, context_id, reg), data);
206 }
207
208 static u32 ipmmu_ctx_read_root(struct ipmmu_vmsa_domain *domain,
209                                unsigned int reg)
210 {
211         return ipmmu_ctx_read(domain->mmu->root, domain->context_id, reg);
212 }
213
214 static void ipmmu_ctx_write_root(struct ipmmu_vmsa_domain *domain,
215                                  unsigned int reg, u32 data)
216 {
217         ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
218 }
219
220 static void ipmmu_ctx_write_all(struct ipmmu_vmsa_domain *domain,
221                                 unsigned int reg, u32 data)
222 {
223         if (domain->mmu != domain->mmu->root)
224                 ipmmu_ctx_write(domain->mmu, domain->context_id, reg, data);
225
226         ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
227 }
228
229 static u32 ipmmu_utlb_reg(struct ipmmu_vmsa_device *mmu, unsigned int reg)
230 {
231         return mmu->features->utlb_offset_base + reg;
232 }
233
234 static void ipmmu_imuasid_write(struct ipmmu_vmsa_device *mmu,
235                                 unsigned int utlb, u32 data)
236 {
237         ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUASID(utlb)), data);
238 }
239
240 static void ipmmu_imuctr_write(struct ipmmu_vmsa_device *mmu,
241                                unsigned int utlb, u32 data)
242 {
243         ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUCTR(utlb)), data);
244 }
245
246 /* -----------------------------------------------------------------------------
247  * TLB and microTLB Management
248  */
249
250 /* Wait for any pending TLB invalidations to complete */
251 static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
252 {
253         unsigned int count = 0;
254
255         while (ipmmu_ctx_read_root(domain, IMCTR) & IMCTR_FLUSH) {
256                 cpu_relax();
257                 if (++count == TLB_LOOP_TIMEOUT) {
258                         dev_err_ratelimited(domain->mmu->dev,
259                         "TLB sync timed out -- MMU may be deadlocked\n");
260                         return;
261                 }
262                 udelay(1);
263         }
264 }
265
266 static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
267 {
268         u32 reg;
269
270         reg = ipmmu_ctx_read_root(domain, IMCTR);
271         reg |= IMCTR_FLUSH;
272         ipmmu_ctx_write_all(domain, IMCTR, reg);
273
274         ipmmu_tlb_sync(domain);
275 }
276
277 /*
278  * Enable MMU translation for the microTLB.
279  */
280 static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
281                               unsigned int utlb)
282 {
283         struct ipmmu_vmsa_device *mmu = domain->mmu;
284
285         /*
286          * TODO: Reference-count the microTLB as several bus masters can be
287          * connected to the same microTLB.
288          */
289
290         /* TODO: What should we set the ASID to ? */
291         ipmmu_imuasid_write(mmu, utlb, 0);
292         /* TODO: Do we need to flush the microTLB ? */
293         ipmmu_imuctr_write(mmu, utlb, IMUCTR_TTSEL_MMU(domain->context_id) |
294                                       IMUCTR_FLUSH | IMUCTR_MMUEN);
295         mmu->utlb_ctx[utlb] = domain->context_id;
296 }
297
298 /*
299  * Disable MMU translation for the microTLB.
300  */
301 static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
302                                unsigned int utlb)
303 {
304         struct ipmmu_vmsa_device *mmu = domain->mmu;
305
306         ipmmu_imuctr_write(mmu, utlb, 0);
307         mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
308 }
309
310 static void ipmmu_tlb_flush_all(void *cookie)
311 {
312         struct ipmmu_vmsa_domain *domain = cookie;
313
314         ipmmu_tlb_invalidate(domain);
315 }
316
317 static void ipmmu_tlb_flush(unsigned long iova, size_t size,
318                                 size_t granule, void *cookie)
319 {
320         ipmmu_tlb_flush_all(cookie);
321 }
322
323 static const struct iommu_flush_ops ipmmu_flush_ops = {
324         .tlb_flush_all = ipmmu_tlb_flush_all,
325         .tlb_flush_walk = ipmmu_tlb_flush,
326 };
327
328 /* -----------------------------------------------------------------------------
329  * Domain/Context Management
330  */
331
332 static int ipmmu_domain_allocate_context(struct ipmmu_vmsa_device *mmu,
333                                          struct ipmmu_vmsa_domain *domain)
334 {
335         unsigned long flags;
336         int ret;
337
338         spin_lock_irqsave(&mmu->lock, flags);
339
340         ret = find_first_zero_bit(mmu->ctx, mmu->num_ctx);
341         if (ret != mmu->num_ctx) {
342                 mmu->domains[ret] = domain;
343                 set_bit(ret, mmu->ctx);
344         } else
345                 ret = -EBUSY;
346
347         spin_unlock_irqrestore(&mmu->lock, flags);
348
349         return ret;
350 }
351
352 static void ipmmu_domain_free_context(struct ipmmu_vmsa_device *mmu,
353                                       unsigned int context_id)
354 {
355         unsigned long flags;
356
357         spin_lock_irqsave(&mmu->lock, flags);
358
359         clear_bit(context_id, mmu->ctx);
360         mmu->domains[context_id] = NULL;
361
362         spin_unlock_irqrestore(&mmu->lock, flags);
363 }
364
365 static void ipmmu_domain_setup_context(struct ipmmu_vmsa_domain *domain)
366 {
367         u64 ttbr;
368         u32 tmp;
369
370         /* TTBR0 */
371         ttbr = domain->cfg.arm_lpae_s1_cfg.ttbr;
372         ipmmu_ctx_write_root(domain, IMTTLBR0, ttbr);
373         ipmmu_ctx_write_root(domain, IMTTUBR0, ttbr >> 32);
374
375         /*
376          * TTBCR
377          * We use long descriptors and allocate the whole 32-bit VA space to
378          * TTBR0.
379          */
380         if (domain->mmu->features->twobit_imttbcr_sl0)
381                 tmp = IMTTBCR_SL0_TWOBIT_LVL_1;
382         else
383                 tmp = IMTTBCR_SL0_LVL_1;
384
385         if (domain->mmu->features->cache_snoop)
386                 tmp |= IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
387                        IMTTBCR_IRGN0_WB_WA;
388
389         ipmmu_ctx_write_root(domain, IMTTBCR, IMTTBCR_EAE | tmp);
390
391         /* MAIR0 */
392         ipmmu_ctx_write_root(domain, IMMAIR0,
393                              domain->cfg.arm_lpae_s1_cfg.mair);
394
395         /* IMBUSCR */
396         if (domain->mmu->features->setup_imbuscr)
397                 ipmmu_ctx_write_root(domain, IMBUSCR,
398                                      ipmmu_ctx_read_root(domain, IMBUSCR) &
399                                      ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
400
401         /*
402          * IMSTR
403          * Clear all interrupt flags.
404          */
405         ipmmu_ctx_write_root(domain, IMSTR, ipmmu_ctx_read_root(domain, IMSTR));
406
407         /*
408          * IMCTR
409          * Enable the MMU and interrupt generation. The long-descriptor
410          * translation table format doesn't use TEX remapping. Don't enable AF
411          * software management as we have no use for it. Flush the TLB as
412          * required when modifying the context registers.
413          */
414         ipmmu_ctx_write_all(domain, IMCTR,
415                             IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
416 }
417
418 static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
419 {
420         int ret;
421
422         /*
423          * Allocate the page table operations.
424          *
425          * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory
426          * access, Long-descriptor format" that the NStable bit being set in a
427          * table descriptor will result in the NStable and NS bits of all child
428          * entries being ignored and considered as being set. The IPMMU seems
429          * not to comply with this, as it generates a secure access page fault
430          * if any of the NStable and NS bits isn't set when running in
431          * non-secure mode.
432          */
433         domain->cfg.quirks = IO_PGTABLE_QUIRK_ARM_NS;
434         domain->cfg.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K;
435         domain->cfg.ias = 32;
436         domain->cfg.oas = 40;
437         domain->cfg.tlb = &ipmmu_flush_ops;
438         domain->io_domain.geometry.aperture_end = DMA_BIT_MASK(32);
439         domain->io_domain.geometry.force_aperture = true;
440         /*
441          * TODO: Add support for coherent walk through CCI with DVM and remove
442          * cache handling. For now, delegate it to the io-pgtable code.
443          */
444         domain->cfg.coherent_walk = false;
445         domain->cfg.iommu_dev = domain->mmu->root->dev;
446
447         /*
448          * Find an unused context.
449          */
450         ret = ipmmu_domain_allocate_context(domain->mmu->root, domain);
451         if (ret < 0)
452                 return ret;
453
454         domain->context_id = ret;
455
456         domain->iop = alloc_io_pgtable_ops(ARM_32_LPAE_S1, &domain->cfg,
457                                            domain);
458         if (!domain->iop) {
459                 ipmmu_domain_free_context(domain->mmu->root,
460                                           domain->context_id);
461                 return -EINVAL;
462         }
463
464         ipmmu_domain_setup_context(domain);
465         return 0;
466 }
467
468 static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
469 {
470         if (!domain->mmu)
471                 return;
472
473         /*
474          * Disable the context. Flush the TLB as required when modifying the
475          * context registers.
476          *
477          * TODO: Is TLB flush really needed ?
478          */
479         ipmmu_ctx_write_all(domain, IMCTR, IMCTR_FLUSH);
480         ipmmu_tlb_sync(domain);
481         ipmmu_domain_free_context(domain->mmu->root, domain->context_id);
482 }
483
484 /* -----------------------------------------------------------------------------
485  * Fault Handling
486  */
487
488 static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
489 {
490         const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
491         struct ipmmu_vmsa_device *mmu = domain->mmu;
492         unsigned long iova;
493         u32 status;
494
495         status = ipmmu_ctx_read_root(domain, IMSTR);
496         if (!(status & err_mask))
497                 return IRQ_NONE;
498
499         iova = ipmmu_ctx_read_root(domain, IMELAR);
500         if (IS_ENABLED(CONFIG_64BIT))
501                 iova |= (u64)ipmmu_ctx_read_root(domain, IMEUAR) << 32;
502
503         /*
504          * Clear the error status flags. Unlike traditional interrupt flag
505          * registers that must be cleared by writing 1, this status register
506          * seems to require 0. The error address register must be read before,
507          * otherwise its value will be 0.
508          */
509         ipmmu_ctx_write_root(domain, IMSTR, 0);
510
511         /* Log fatal errors. */
512         if (status & IMSTR_MHIT)
513                 dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%lx\n",
514                                     iova);
515         if (status & IMSTR_ABORT)
516                 dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%lx\n",
517                                     iova);
518
519         if (!(status & (IMSTR_PF | IMSTR_TF)))
520                 return IRQ_NONE;
521
522         /*
523          * Try to handle page faults and translation faults.
524          *
525          * TODO: We need to look up the faulty device based on the I/O VA. Use
526          * the IOMMU device for now.
527          */
528         if (!report_iommu_fault(&domain->io_domain, mmu->dev, iova, 0))
529                 return IRQ_HANDLED;
530
531         dev_err_ratelimited(mmu->dev,
532                             "Unhandled fault: status 0x%08x iova 0x%lx\n",
533                             status, iova);
534
535         return IRQ_HANDLED;
536 }
537
538 static irqreturn_t ipmmu_irq(int irq, void *dev)
539 {
540         struct ipmmu_vmsa_device *mmu = dev;
541         irqreturn_t status = IRQ_NONE;
542         unsigned int i;
543         unsigned long flags;
544
545         spin_lock_irqsave(&mmu->lock, flags);
546
547         /*
548          * Check interrupts for all active contexts.
549          */
550         for (i = 0; i < mmu->num_ctx; i++) {
551                 if (!mmu->domains[i])
552                         continue;
553                 if (ipmmu_domain_irq(mmu->domains[i]) == IRQ_HANDLED)
554                         status = IRQ_HANDLED;
555         }
556
557         spin_unlock_irqrestore(&mmu->lock, flags);
558
559         return status;
560 }
561
562 /* -----------------------------------------------------------------------------
563  * IOMMU Operations
564  */
565
566 static struct iommu_domain *ipmmu_domain_alloc(unsigned type)
567 {
568         struct ipmmu_vmsa_domain *domain;
569
570         if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
571                 return NULL;
572
573         domain = kzalloc(sizeof(*domain), GFP_KERNEL);
574         if (!domain)
575                 return NULL;
576
577         mutex_init(&domain->mutex);
578
579         return &domain->io_domain;
580 }
581
582 static void ipmmu_domain_free(struct iommu_domain *io_domain)
583 {
584         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
585
586         /*
587          * Free the domain resources. We assume that all devices have already
588          * been detached.
589          */
590         ipmmu_domain_destroy_context(domain);
591         free_io_pgtable_ops(domain->iop);
592         kfree(domain);
593 }
594
595 static int ipmmu_attach_device(struct iommu_domain *io_domain,
596                                struct device *dev)
597 {
598         struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
599         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
600         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
601         unsigned int i;
602         int ret = 0;
603
604         if (!mmu) {
605                 dev_err(dev, "Cannot attach to IPMMU\n");
606                 return -ENXIO;
607         }
608
609         mutex_lock(&domain->mutex);
610
611         if (!domain->mmu) {
612                 /* The domain hasn't been used yet, initialize it. */
613                 domain->mmu = mmu;
614                 ret = ipmmu_domain_init_context(domain);
615                 if (ret < 0) {
616                         dev_err(dev, "Unable to initialize IPMMU context\n");
617                         domain->mmu = NULL;
618                 } else {
619                         dev_info(dev, "Using IPMMU context %u\n",
620                                  domain->context_id);
621                 }
622         } else if (domain->mmu != mmu) {
623                 /*
624                  * Something is wrong, we can't attach two devices using
625                  * different IOMMUs to the same domain.
626                  */
627                 dev_err(dev, "Can't attach IPMMU %s to domain on IPMMU %s\n",
628                         dev_name(mmu->dev), dev_name(domain->mmu->dev));
629                 ret = -EINVAL;
630         } else
631                 dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
632
633         mutex_unlock(&domain->mutex);
634
635         if (ret < 0)
636                 return ret;
637
638         for (i = 0; i < fwspec->num_ids; ++i)
639                 ipmmu_utlb_enable(domain, fwspec->ids[i]);
640
641         return 0;
642 }
643
644 static void ipmmu_detach_device(struct iommu_domain *io_domain,
645                                 struct device *dev)
646 {
647         struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
648         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
649         unsigned int i;
650
651         for (i = 0; i < fwspec->num_ids; ++i)
652                 ipmmu_utlb_disable(domain, fwspec->ids[i]);
653
654         /*
655          * TODO: Optimize by disabling the context when no device is attached.
656          */
657 }
658
659 static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
660                      phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
661 {
662         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
663
664         if (!domain)
665                 return -ENODEV;
666
667         return domain->iop->map(domain->iop, iova, paddr, size, prot, gfp);
668 }
669
670 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
671                           size_t size, struct iommu_iotlb_gather *gather)
672 {
673         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
674
675         return domain->iop->unmap(domain->iop, iova, size, gather);
676 }
677
678 static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
679 {
680         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
681
682         if (domain->mmu)
683                 ipmmu_tlb_flush_all(domain);
684 }
685
686 static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
687                              struct iommu_iotlb_gather *gather)
688 {
689         ipmmu_flush_iotlb_all(io_domain);
690 }
691
692 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
693                                       dma_addr_t iova)
694 {
695         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
696
697         /* TODO: Is locking needed ? */
698
699         return domain->iop->iova_to_phys(domain->iop, iova);
700 }
701
702 static int ipmmu_init_platform_device(struct device *dev,
703                                       struct of_phandle_args *args)
704 {
705         struct platform_device *ipmmu_pdev;
706
707         ipmmu_pdev = of_find_device_by_node(args->np);
708         if (!ipmmu_pdev)
709                 return -ENODEV;
710
711         dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
712
713         return 0;
714 }
715
716 static const struct soc_device_attribute soc_needs_opt_in[] = {
717         { .family = "R-Car Gen3", },
718         { .family = "RZ/G2", },
719         { /* sentinel */ }
720 };
721
722 static const struct soc_device_attribute soc_denylist[] = {
723         { .soc_id = "r8a774a1", },
724         { .soc_id = "r8a7795", .revision = "ES1.*" },
725         { .soc_id = "r8a7795", .revision = "ES2.*" },
726         { .soc_id = "r8a7796", },
727         { /* sentinel */ }
728 };
729
730 static const char * const devices_allowlist[] = {
731         "ee100000.mmc",
732         "ee120000.mmc",
733         "ee140000.mmc",
734         "ee160000.mmc"
735 };
736
737 static bool ipmmu_device_is_allowed(struct device *dev)
738 {
739         unsigned int i;
740
741         /*
742          * R-Car Gen3 and RZ/G2 use the allow list to opt-in devices.
743          * For Other SoCs, this returns true anyway.
744          */
745         if (!soc_device_match(soc_needs_opt_in))
746                 return true;
747
748         /* Check whether this SoC can use the IPMMU correctly or not */
749         if (soc_device_match(soc_denylist))
750                 return false;
751
752         /* Check whether this device can work with the IPMMU */
753         for (i = 0; i < ARRAY_SIZE(devices_allowlist); i++) {
754                 if (!strcmp(dev_name(dev), devices_allowlist[i]))
755                         return true;
756         }
757
758         /* Otherwise, do not allow use of IPMMU */
759         return false;
760 }
761
762 static int ipmmu_of_xlate(struct device *dev,
763                           struct of_phandle_args *spec)
764 {
765         if (!ipmmu_device_is_allowed(dev))
766                 return -ENODEV;
767
768         iommu_fwspec_add_ids(dev, spec->args, 1);
769
770         /* Initialize once - xlate() will call multiple times */
771         if (to_ipmmu(dev))
772                 return 0;
773
774         return ipmmu_init_platform_device(dev, spec);
775 }
776
777 static int ipmmu_init_arm_mapping(struct device *dev)
778 {
779         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
780         int ret;
781
782         /*
783          * Create the ARM mapping, used by the ARM DMA mapping core to allocate
784          * VAs. This will allocate a corresponding IOMMU domain.
785          *
786          * TODO:
787          * - Create one mapping per context (TLB).
788          * - Make the mapping size configurable ? We currently use a 2GB mapping
789          *   at a 1GB offset to ensure that NULL VAs will fault.
790          */
791         if (!mmu->mapping) {
792                 struct dma_iommu_mapping *mapping;
793
794                 mapping = arm_iommu_create_mapping(&platform_bus_type,
795                                                    SZ_1G, SZ_2G);
796                 if (IS_ERR(mapping)) {
797                         dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
798                         ret = PTR_ERR(mapping);
799                         goto error;
800                 }
801
802                 mmu->mapping = mapping;
803         }
804
805         /* Attach the ARM VA mapping to the device. */
806         ret = arm_iommu_attach_device(dev, mmu->mapping);
807         if (ret < 0) {
808                 dev_err(dev, "Failed to attach device to VA mapping\n");
809                 goto error;
810         }
811
812         return 0;
813
814 error:
815         if (mmu->mapping)
816                 arm_iommu_release_mapping(mmu->mapping);
817
818         return ret;
819 }
820
821 static struct iommu_device *ipmmu_probe_device(struct device *dev)
822 {
823         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
824
825         /*
826          * Only let through devices that have been verified in xlate()
827          */
828         if (!mmu)
829                 return ERR_PTR(-ENODEV);
830
831         return &mmu->iommu;
832 }
833
834 static void ipmmu_probe_finalize(struct device *dev)
835 {
836         int ret = 0;
837
838         if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA))
839                 ret = ipmmu_init_arm_mapping(dev);
840
841         if (ret)
842                 dev_err(dev, "Can't create IOMMU mapping - DMA-OPS will not work\n");
843 }
844
845 static void ipmmu_release_device(struct device *dev)
846 {
847         arm_iommu_detach_device(dev);
848 }
849
850 static struct iommu_group *ipmmu_find_group(struct device *dev)
851 {
852         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
853         struct iommu_group *group;
854
855         if (mmu->group)
856                 return iommu_group_ref_get(mmu->group);
857
858         group = iommu_group_alloc();
859         if (!IS_ERR(group))
860                 mmu->group = group;
861
862         return group;
863 }
864
865 static const struct iommu_ops ipmmu_ops = {
866         .domain_alloc = ipmmu_domain_alloc,
867         .domain_free = ipmmu_domain_free,
868         .attach_dev = ipmmu_attach_device,
869         .detach_dev = ipmmu_detach_device,
870         .map = ipmmu_map,
871         .unmap = ipmmu_unmap,
872         .flush_iotlb_all = ipmmu_flush_iotlb_all,
873         .iotlb_sync = ipmmu_iotlb_sync,
874         .iova_to_phys = ipmmu_iova_to_phys,
875         .probe_device = ipmmu_probe_device,
876         .release_device = ipmmu_release_device,
877         .probe_finalize = ipmmu_probe_finalize,
878         .device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
879                         ? generic_device_group : ipmmu_find_group,
880         .pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
881         .of_xlate = ipmmu_of_xlate,
882 };
883
884 /* -----------------------------------------------------------------------------
885  * Probe/remove and init
886  */
887
888 static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
889 {
890         unsigned int i;
891
892         /* Disable all contexts. */
893         for (i = 0; i < mmu->num_ctx; ++i)
894                 ipmmu_ctx_write(mmu, i, IMCTR, 0);
895 }
896
897 static const struct ipmmu_features ipmmu_features_default = {
898         .use_ns_alias_offset = true,
899         .has_cache_leaf_nodes = false,
900         .number_of_contexts = 1, /* software only tested with one context */
901         .num_utlbs = 32,
902         .setup_imbuscr = true,
903         .twobit_imttbcr_sl0 = false,
904         .reserved_context = false,
905         .cache_snoop = true,
906         .ctx_offset_base = 0,
907         .ctx_offset_stride = 0x40,
908         .utlb_offset_base = 0,
909 };
910
911 static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
912         .use_ns_alias_offset = false,
913         .has_cache_leaf_nodes = true,
914         .number_of_contexts = 8,
915         .num_utlbs = 48,
916         .setup_imbuscr = false,
917         .twobit_imttbcr_sl0 = true,
918         .reserved_context = true,
919         .cache_snoop = false,
920         .ctx_offset_base = 0,
921         .ctx_offset_stride = 0x40,
922         .utlb_offset_base = 0,
923 };
924
925 static const struct of_device_id ipmmu_of_ids[] = {
926         {
927                 .compatible = "renesas,ipmmu-vmsa",
928                 .data = &ipmmu_features_default,
929         }, {
930                 .compatible = "renesas,ipmmu-r8a774a1",
931                 .data = &ipmmu_features_rcar_gen3,
932         }, {
933                 .compatible = "renesas,ipmmu-r8a774b1",
934                 .data = &ipmmu_features_rcar_gen3,
935         }, {
936                 .compatible = "renesas,ipmmu-r8a774c0",
937                 .data = &ipmmu_features_rcar_gen3,
938         }, {
939                 .compatible = "renesas,ipmmu-r8a774e1",
940                 .data = &ipmmu_features_rcar_gen3,
941         }, {
942                 .compatible = "renesas,ipmmu-r8a7795",
943                 .data = &ipmmu_features_rcar_gen3,
944         }, {
945                 .compatible = "renesas,ipmmu-r8a7796",
946                 .data = &ipmmu_features_rcar_gen3,
947         }, {
948                 .compatible = "renesas,ipmmu-r8a77961",
949                 .data = &ipmmu_features_rcar_gen3,
950         }, {
951                 .compatible = "renesas,ipmmu-r8a77965",
952                 .data = &ipmmu_features_rcar_gen3,
953         }, {
954                 .compatible = "renesas,ipmmu-r8a77970",
955                 .data = &ipmmu_features_rcar_gen3,
956         }, {
957                 .compatible = "renesas,ipmmu-r8a77990",
958                 .data = &ipmmu_features_rcar_gen3,
959         }, {
960                 .compatible = "renesas,ipmmu-r8a77995",
961                 .data = &ipmmu_features_rcar_gen3,
962         }, {
963                 /* Terminator */
964         },
965 };
966
967 static int ipmmu_probe(struct platform_device *pdev)
968 {
969         struct ipmmu_vmsa_device *mmu;
970         struct resource *res;
971         int irq;
972         int ret;
973
974         mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
975         if (!mmu) {
976                 dev_err(&pdev->dev, "cannot allocate device data\n");
977                 return -ENOMEM;
978         }
979
980         mmu->dev = &pdev->dev;
981         spin_lock_init(&mmu->lock);
982         bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
983         mmu->features = of_device_get_match_data(&pdev->dev);
984         memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
985         dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
986
987         /* Map I/O memory and request IRQ. */
988         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
989         mmu->base = devm_ioremap_resource(&pdev->dev, res);
990         if (IS_ERR(mmu->base))
991                 return PTR_ERR(mmu->base);
992
993         /*
994          * The IPMMU has two register banks, for secure and non-secure modes.
995          * The bank mapped at the beginning of the IPMMU address space
996          * corresponds to the running mode of the CPU. When running in secure
997          * mode the non-secure register bank is also available at an offset.
998          *
999          * Secure mode operation isn't clearly documented and is thus currently
1000          * not implemented in the driver. Furthermore, preliminary tests of
1001          * non-secure operation with the main register bank were not successful.
1002          * Offset the registers base unconditionally to point to the non-secure
1003          * alias space for now.
1004          */
1005         if (mmu->features->use_ns_alias_offset)
1006                 mmu->base += IM_NS_ALIAS_OFFSET;
1007
1008         mmu->num_ctx = min(IPMMU_CTX_MAX, mmu->features->number_of_contexts);
1009
1010         /*
1011          * Determine if this IPMMU instance is a root device by checking for
1012          * the lack of has_cache_leaf_nodes flag or renesas,ipmmu-main property.
1013          */
1014         if (!mmu->features->has_cache_leaf_nodes ||
1015             !of_find_property(pdev->dev.of_node, "renesas,ipmmu-main", NULL))
1016                 mmu->root = mmu;
1017         else
1018                 mmu->root = ipmmu_find_root();
1019
1020         /*
1021          * Wait until the root device has been registered for sure.
1022          */
1023         if (!mmu->root)
1024                 return -EPROBE_DEFER;
1025
1026         /* Root devices have mandatory IRQs */
1027         if (ipmmu_is_root(mmu)) {
1028                 irq = platform_get_irq(pdev, 0);
1029                 if (irq < 0)
1030                         return irq;
1031
1032                 ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
1033                                        dev_name(&pdev->dev), mmu);
1034                 if (ret < 0) {
1035                         dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
1036                         return ret;
1037                 }
1038
1039                 ipmmu_device_reset(mmu);
1040
1041                 if (mmu->features->reserved_context) {
1042                         dev_info(&pdev->dev, "IPMMU context 0 is reserved\n");
1043                         set_bit(0, mmu->ctx);
1044                 }
1045         }
1046
1047         /*
1048          * Register the IPMMU to the IOMMU subsystem in the following cases:
1049          * - R-Car Gen2 IPMMU (all devices registered)
1050          * - R-Car Gen3 IPMMU (leaf devices only - skip root IPMMU-MM device)
1051          */
1052         if (!mmu->features->has_cache_leaf_nodes || !ipmmu_is_root(mmu)) {
1053                 ret = iommu_device_sysfs_add(&mmu->iommu, &pdev->dev, NULL,
1054                                              dev_name(&pdev->dev));
1055                 if (ret)
1056                         return ret;
1057
1058                 ret = iommu_device_register(&mmu->iommu, &ipmmu_ops, &pdev->dev);
1059                 if (ret)
1060                         return ret;
1061
1062 #if defined(CONFIG_IOMMU_DMA)
1063                 if (!iommu_present(&platform_bus_type))
1064                         bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1065 #endif
1066         }
1067
1068         /*
1069          * We can't create the ARM mapping here as it requires the bus to have
1070          * an IOMMU, which only happens when bus_set_iommu() is called in
1071          * ipmmu_init() after the probe function returns.
1072          */
1073
1074         platform_set_drvdata(pdev, mmu);
1075
1076         return 0;
1077 }
1078
1079 static int ipmmu_remove(struct platform_device *pdev)
1080 {
1081         struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1082
1083         iommu_device_sysfs_remove(&mmu->iommu);
1084         iommu_device_unregister(&mmu->iommu);
1085
1086         arm_iommu_release_mapping(mmu->mapping);
1087
1088         ipmmu_device_reset(mmu);
1089
1090         return 0;
1091 }
1092
1093 #ifdef CONFIG_PM_SLEEP
1094 static int ipmmu_resume_noirq(struct device *dev)
1095 {
1096         struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1097         unsigned int i;
1098
1099         /* Reset root MMU and restore contexts */
1100         if (ipmmu_is_root(mmu)) {
1101                 ipmmu_device_reset(mmu);
1102
1103                 for (i = 0; i < mmu->num_ctx; i++) {
1104                         if (!mmu->domains[i])
1105                                 continue;
1106
1107                         ipmmu_domain_setup_context(mmu->domains[i]);
1108                 }
1109         }
1110
1111         /* Re-enable active micro-TLBs */
1112         for (i = 0; i < mmu->features->num_utlbs; i++) {
1113                 if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1114                         continue;
1115
1116                 ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1117         }
1118
1119         return 0;
1120 }
1121
1122 static const struct dev_pm_ops ipmmu_pm  = {
1123         SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1124 };
1125 #define DEV_PM_OPS      &ipmmu_pm
1126 #else
1127 #define DEV_PM_OPS      NULL
1128 #endif /* CONFIG_PM_SLEEP */
1129
1130 static struct platform_driver ipmmu_driver = {
1131         .driver = {
1132                 .name = "ipmmu-vmsa",
1133                 .of_match_table = of_match_ptr(ipmmu_of_ids),
1134                 .pm = DEV_PM_OPS,
1135         },
1136         .probe = ipmmu_probe,
1137         .remove = ipmmu_remove,
1138 };
1139
1140 static int __init ipmmu_init(void)
1141 {
1142         struct device_node *np;
1143         static bool setup_done;
1144         int ret;
1145
1146         if (setup_done)
1147                 return 0;
1148
1149         np = of_find_matching_node(NULL, ipmmu_of_ids);
1150         if (!np)
1151                 return 0;
1152
1153         of_node_put(np);
1154
1155         ret = platform_driver_register(&ipmmu_driver);
1156         if (ret < 0)
1157                 return ret;
1158
1159 #if defined(CONFIG_ARM) && !defined(CONFIG_IOMMU_DMA)
1160         if (!iommu_present(&platform_bus_type))
1161                 bus_set_iommu(&platform_bus_type, &ipmmu_ops);
1162 #endif
1163
1164         setup_done = true;
1165         return 0;
1166 }
1167 subsys_initcall(ipmmu_init);