iommu/arm-smmu: Report IOMMU_CAP_CACHE_COHERENCY even betterer
[linux-2.6-block.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           16U
37 #define IPMMU_CTX_INVALID       -1
38
39 #define IPMMU_UTLB_MAX          64U
40
41 struct ipmmu_features {
42         bool use_ns_alias_offset;
43         bool has_cache_leaf_nodes;
44         unsigned int number_of_contexts;
45         unsigned int num_utlbs;
46         bool setup_imbuscr;
47         bool twobit_imttbcr_sl0;
48         bool reserved_context;
49         bool cache_snoop;
50         unsigned int ctx_offset_base;
51         unsigned int ctx_offset_stride;
52         unsigned int utlb_offset_base;
53 };
54
55 struct ipmmu_vmsa_device {
56         struct device *dev;
57         void __iomem *base;
58         struct iommu_device iommu;
59         struct ipmmu_vmsa_device *root;
60         const struct ipmmu_features *features;
61         unsigned int num_ctx;
62         spinlock_t lock;                        /* Protects ctx and domains[] */
63         DECLARE_BITMAP(ctx, IPMMU_CTX_MAX);
64         struct ipmmu_vmsa_domain *domains[IPMMU_CTX_MAX];
65         s8 utlb_ctx[IPMMU_UTLB_MAX];
66
67         struct iommu_group *group;
68         struct dma_iommu_mapping *mapping;
69 };
70
71 struct ipmmu_vmsa_domain {
72         struct ipmmu_vmsa_device *mmu;
73         struct iommu_domain io_domain;
74
75         struct io_pgtable_cfg cfg;
76         struct io_pgtable_ops *iop;
77
78         unsigned int context_id;
79         struct mutex mutex;                     /* Protects mappings */
80 };
81
82 static struct ipmmu_vmsa_domain *to_vmsa_domain(struct iommu_domain *dom)
83 {
84         return container_of(dom, struct ipmmu_vmsa_domain, io_domain);
85 }
86
87 static struct ipmmu_vmsa_device *to_ipmmu(struct device *dev)
88 {
89         return dev_iommu_priv_get(dev);
90 }
91
92 #define TLB_LOOP_TIMEOUT                100     /* 100us */
93
94 /* -----------------------------------------------------------------------------
95  * Registers Definition
96  */
97
98 #define IM_NS_ALIAS_OFFSET              0x800
99
100 /* MMU "context" registers */
101 #define IMCTR                           0x0000          /* R-Car Gen2/3 */
102 #define IMCTR_INTEN                     (1 << 2)        /* R-Car Gen2/3 */
103 #define IMCTR_FLUSH                     (1 << 1)        /* R-Car Gen2/3 */
104 #define IMCTR_MMUEN                     (1 << 0)        /* R-Car Gen2/3 */
105
106 #define IMTTBCR                         0x0008          /* R-Car Gen2/3 */
107 #define IMTTBCR_EAE                     (1 << 31)       /* R-Car Gen2/3 */
108 #define IMTTBCR_SH0_INNER_SHAREABLE     (3 << 12)       /* R-Car Gen2 only */
109 #define IMTTBCR_ORGN0_WB_WA             (1 << 10)       /* R-Car Gen2 only */
110 #define IMTTBCR_IRGN0_WB_WA             (1 << 8)        /* R-Car Gen2 only */
111 #define IMTTBCR_SL0_TWOBIT_LVL_1        (2 << 6)        /* R-Car Gen3 only */
112 #define IMTTBCR_SL0_LVL_1               (1 << 4)        /* R-Car Gen2 only */
113
114 #define IMBUSCR                         0x000c          /* R-Car Gen2 only */
115 #define IMBUSCR_DVM                     (1 << 2)        /* R-Car Gen2 only */
116 #define IMBUSCR_BUSSEL_MASK             (3 << 0)        /* R-Car Gen2 only */
117
118 #define IMTTLBR0                        0x0010          /* R-Car Gen2/3 */
119 #define IMTTUBR0                        0x0014          /* R-Car Gen2/3 */
120
121 #define IMSTR                           0x0020          /* R-Car Gen2/3 */
122 #define IMSTR_MHIT                      (1 << 4)        /* R-Car Gen2/3 */
123 #define IMSTR_ABORT                     (1 << 2)        /* R-Car Gen2/3 */
124 #define IMSTR_PF                        (1 << 1)        /* R-Car Gen2/3 */
125 #define IMSTR_TF                        (1 << 0)        /* R-Car Gen2/3 */
126
127 #define IMMAIR0                         0x0028          /* R-Car Gen2/3 */
128
129 #define IMELAR                          0x0030          /* R-Car Gen2/3, IMEAR on R-Car Gen2 */
130 #define IMEUAR                          0x0034          /* R-Car Gen3 only */
131
132 /* uTLB registers */
133 #define IMUCTR(n)                       ((n) < 32 ? IMUCTR0(n) : IMUCTR32(n))
134 #define IMUCTR0(n)                      (0x0300 + ((n) * 16))           /* R-Car Gen2/3 */
135 #define IMUCTR32(n)                     (0x0600 + (((n) - 32) * 16))    /* R-Car Gen3 only */
136 #define IMUCTR_TTSEL_MMU(n)             ((n) << 4)      /* R-Car Gen2/3 */
137 #define IMUCTR_FLUSH                    (1 << 1)        /* R-Car Gen2/3 */
138 #define IMUCTR_MMUEN                    (1 << 0)        /* R-Car Gen2/3 */
139
140 #define IMUASID(n)                      ((n) < 32 ? IMUASID0(n) : IMUASID32(n))
141 #define IMUASID0(n)                     (0x0308 + ((n) * 16))           /* R-Car Gen2/3 */
142 #define IMUASID32(n)                    (0x0608 + (((n) - 32) * 16))    /* R-Car Gen3 only */
143
144 /* -----------------------------------------------------------------------------
145  * Root device handling
146  */
147
148 static struct platform_driver ipmmu_driver;
149
150 static bool ipmmu_is_root(struct ipmmu_vmsa_device *mmu)
151 {
152         return mmu->root == mmu;
153 }
154
155 static int __ipmmu_check_device(struct device *dev, void *data)
156 {
157         struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
158         struct ipmmu_vmsa_device **rootp = data;
159
160         if (ipmmu_is_root(mmu))
161                 *rootp = mmu;
162
163         return 0;
164 }
165
166 static struct ipmmu_vmsa_device *ipmmu_find_root(void)
167 {
168         struct ipmmu_vmsa_device *root = NULL;
169
170         return driver_for_each_device(&ipmmu_driver.driver, NULL, &root,
171                                       __ipmmu_check_device) == 0 ? root : NULL;
172 }
173
174 /* -----------------------------------------------------------------------------
175  * Read/Write Access
176  */
177
178 static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
179 {
180         return ioread32(mmu->base + offset);
181 }
182
183 static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
184                         u32 data)
185 {
186         iowrite32(data, mmu->base + offset);
187 }
188
189 static unsigned int ipmmu_ctx_reg(struct ipmmu_vmsa_device *mmu,
190                                   unsigned int context_id, unsigned int reg)
191 {
192         unsigned int base = mmu->features->ctx_offset_base;
193
194         if (context_id > 7)
195                 base += 0x800 - 8 * 0x40;
196
197         return base + context_id * mmu->features->ctx_offset_stride + reg;
198 }
199
200 static u32 ipmmu_ctx_read(struct ipmmu_vmsa_device *mmu,
201                           unsigned int context_id, unsigned int reg)
202 {
203         return ipmmu_read(mmu, ipmmu_ctx_reg(mmu, context_id, reg));
204 }
205
206 static void ipmmu_ctx_write(struct ipmmu_vmsa_device *mmu,
207                             unsigned int context_id, unsigned int reg, u32 data)
208 {
209         ipmmu_write(mmu, ipmmu_ctx_reg(mmu, context_id, reg), data);
210 }
211
212 static u32 ipmmu_ctx_read_root(struct ipmmu_vmsa_domain *domain,
213                                unsigned int reg)
214 {
215         return ipmmu_ctx_read(domain->mmu->root, domain->context_id, reg);
216 }
217
218 static void ipmmu_ctx_write_root(struct ipmmu_vmsa_domain *domain,
219                                  unsigned int reg, u32 data)
220 {
221         ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
222 }
223
224 static void ipmmu_ctx_write_all(struct ipmmu_vmsa_domain *domain,
225                                 unsigned int reg, u32 data)
226 {
227         if (domain->mmu != domain->mmu->root)
228                 ipmmu_ctx_write(domain->mmu, domain->context_id, reg, data);
229
230         ipmmu_ctx_write(domain->mmu->root, domain->context_id, reg, data);
231 }
232
233 static u32 ipmmu_utlb_reg(struct ipmmu_vmsa_device *mmu, unsigned int reg)
234 {
235         return mmu->features->utlb_offset_base + reg;
236 }
237
238 static void ipmmu_imuasid_write(struct ipmmu_vmsa_device *mmu,
239                                 unsigned int utlb, u32 data)
240 {
241         ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUASID(utlb)), data);
242 }
243
244 static void ipmmu_imuctr_write(struct ipmmu_vmsa_device *mmu,
245                                unsigned int utlb, u32 data)
246 {
247         ipmmu_write(mmu, ipmmu_utlb_reg(mmu, IMUCTR(utlb)), data);
248 }
249
250 /* -----------------------------------------------------------------------------
251  * TLB and microTLB Management
252  */
253
254 /* Wait for any pending TLB invalidations to complete */
255 static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
256 {
257         unsigned int count = 0;
258
259         while (ipmmu_ctx_read_root(domain, IMCTR) & IMCTR_FLUSH) {
260                 cpu_relax();
261                 if (++count == TLB_LOOP_TIMEOUT) {
262                         dev_err_ratelimited(domain->mmu->dev,
263                         "TLB sync timed out -- MMU may be deadlocked\n");
264                         return;
265                 }
266                 udelay(1);
267         }
268 }
269
270 static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
271 {
272         u32 reg;
273
274         reg = ipmmu_ctx_read_root(domain, IMCTR);
275         reg |= IMCTR_FLUSH;
276         ipmmu_ctx_write_all(domain, IMCTR, reg);
277
278         ipmmu_tlb_sync(domain);
279 }
280
281 /*
282  * Enable MMU translation for the microTLB.
283  */
284 static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
285                               unsigned int utlb)
286 {
287         struct ipmmu_vmsa_device *mmu = domain->mmu;
288
289         /*
290          * TODO: Reference-count the microTLB as several bus masters can be
291          * connected to the same microTLB.
292          */
293
294         /* TODO: What should we set the ASID to ? */
295         ipmmu_imuasid_write(mmu, utlb, 0);
296         /* TODO: Do we need to flush the microTLB ? */
297         ipmmu_imuctr_write(mmu, utlb, IMUCTR_TTSEL_MMU(domain->context_id) |
298                                       IMUCTR_FLUSH | IMUCTR_MMUEN);
299         mmu->utlb_ctx[utlb] = domain->context_id;
300 }
301
302 /*
303  * Disable MMU translation for the microTLB.
304  */
305 static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
306                                unsigned int utlb)
307 {
308         struct ipmmu_vmsa_device *mmu = domain->mmu;
309
310         ipmmu_imuctr_write(mmu, utlb, 0);
311         mmu->utlb_ctx[utlb] = IPMMU_CTX_INVALID;
312 }
313
314 static void ipmmu_tlb_flush_all(void *cookie)
315 {
316         struct ipmmu_vmsa_domain *domain = cookie;
317
318         ipmmu_tlb_invalidate(domain);
319 }
320
321 static void ipmmu_tlb_flush(unsigned long iova, size_t size,
322                                 size_t granule, void *cookie)
323 {
324         ipmmu_tlb_flush_all(cookie);
325 }
326
327 static const struct iommu_flush_ops ipmmu_flush_ops = {
328         .tlb_flush_all = ipmmu_tlb_flush_all,
329         .tlb_flush_walk = ipmmu_tlb_flush,
330 };
331
332 /* -----------------------------------------------------------------------------
333  * Domain/Context Management
334  */
335
336 static int ipmmu_domain_allocate_context(struct ipmmu_vmsa_device *mmu,
337                                          struct ipmmu_vmsa_domain *domain)
338 {
339         unsigned long flags;
340         int ret;
341
342         spin_lock_irqsave(&mmu->lock, flags);
343
344         ret = find_first_zero_bit(mmu->ctx, mmu->num_ctx);
345         if (ret != mmu->num_ctx) {
346                 mmu->domains[ret] = domain;
347                 set_bit(ret, mmu->ctx);
348         } else
349                 ret = -EBUSY;
350
351         spin_unlock_irqrestore(&mmu->lock, flags);
352
353         return ret;
354 }
355
356 static void ipmmu_domain_free_context(struct ipmmu_vmsa_device *mmu,
357                                       unsigned int context_id)
358 {
359         unsigned long flags;
360
361         spin_lock_irqsave(&mmu->lock, flags);
362
363         clear_bit(context_id, mmu->ctx);
364         mmu->domains[context_id] = NULL;
365
366         spin_unlock_irqrestore(&mmu->lock, flags);
367 }
368
369 static void ipmmu_domain_setup_context(struct ipmmu_vmsa_domain *domain)
370 {
371         u64 ttbr;
372         u32 tmp;
373
374         /* TTBR0 */
375         ttbr = domain->cfg.arm_lpae_s1_cfg.ttbr;
376         ipmmu_ctx_write_root(domain, IMTTLBR0, ttbr);
377         ipmmu_ctx_write_root(domain, IMTTUBR0, ttbr >> 32);
378
379         /*
380          * TTBCR
381          * We use long descriptors and allocate the whole 32-bit VA space to
382          * TTBR0.
383          */
384         if (domain->mmu->features->twobit_imttbcr_sl0)
385                 tmp = IMTTBCR_SL0_TWOBIT_LVL_1;
386         else
387                 tmp = IMTTBCR_SL0_LVL_1;
388
389         if (domain->mmu->features->cache_snoop)
390                 tmp |= IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
391                        IMTTBCR_IRGN0_WB_WA;
392
393         ipmmu_ctx_write_root(domain, IMTTBCR, IMTTBCR_EAE | tmp);
394
395         /* MAIR0 */
396         ipmmu_ctx_write_root(domain, IMMAIR0,
397                              domain->cfg.arm_lpae_s1_cfg.mair);
398
399         /* IMBUSCR */
400         if (domain->mmu->features->setup_imbuscr)
401                 ipmmu_ctx_write_root(domain, IMBUSCR,
402                                      ipmmu_ctx_read_root(domain, IMBUSCR) &
403                                      ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
404
405         /*
406          * IMSTR
407          * Clear all interrupt flags.
408          */
409         ipmmu_ctx_write_root(domain, IMSTR, ipmmu_ctx_read_root(domain, IMSTR));
410
411         /*
412          * IMCTR
413          * Enable the MMU and interrupt generation. The long-descriptor
414          * translation table format doesn't use TEX remapping. Don't enable AF
415          * software management as we have no use for it. Flush the TLB as
416          * required when modifying the context registers.
417          */
418         ipmmu_ctx_write_all(domain, IMCTR,
419                             IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
420 }
421
422 static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
423 {
424         int ret;
425
426         /*
427          * Allocate the page table operations.
428          *
429          * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory
430          * access, Long-descriptor format" that the NStable bit being set in a
431          * table descriptor will result in the NStable and NS bits of all child
432          * entries being ignored and considered as being set. The IPMMU seems
433          * not to comply with this, as it generates a secure access page fault
434          * if any of the NStable and NS bits isn't set when running in
435          * non-secure mode.
436          */
437         domain->cfg.quirks = IO_PGTABLE_QUIRK_ARM_NS;
438         domain->cfg.pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K;
439         domain->cfg.ias = 32;
440         domain->cfg.oas = 40;
441         domain->cfg.tlb = &ipmmu_flush_ops;
442         domain->io_domain.geometry.aperture_end = DMA_BIT_MASK(32);
443         domain->io_domain.geometry.force_aperture = true;
444         /*
445          * TODO: Add support for coherent walk through CCI with DVM and remove
446          * cache handling. For now, delegate it to the io-pgtable code.
447          */
448         domain->cfg.coherent_walk = false;
449         domain->cfg.iommu_dev = domain->mmu->root->dev;
450
451         /*
452          * Find an unused context.
453          */
454         ret = ipmmu_domain_allocate_context(domain->mmu->root, domain);
455         if (ret < 0)
456                 return ret;
457
458         domain->context_id = ret;
459
460         domain->iop = alloc_io_pgtable_ops(ARM_32_LPAE_S1, &domain->cfg,
461                                            domain);
462         if (!domain->iop) {
463                 ipmmu_domain_free_context(domain->mmu->root,
464                                           domain->context_id);
465                 return -EINVAL;
466         }
467
468         ipmmu_domain_setup_context(domain);
469         return 0;
470 }
471
472 static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
473 {
474         if (!domain->mmu)
475                 return;
476
477         /*
478          * Disable the context. Flush the TLB as required when modifying the
479          * context registers.
480          *
481          * TODO: Is TLB flush really needed ?
482          */
483         ipmmu_ctx_write_all(domain, IMCTR, IMCTR_FLUSH);
484         ipmmu_tlb_sync(domain);
485         ipmmu_domain_free_context(domain->mmu->root, domain->context_id);
486 }
487
488 /* -----------------------------------------------------------------------------
489  * Fault Handling
490  */
491
492 static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
493 {
494         const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
495         struct ipmmu_vmsa_device *mmu = domain->mmu;
496         unsigned long iova;
497         u32 status;
498
499         status = ipmmu_ctx_read_root(domain, IMSTR);
500         if (!(status & err_mask))
501                 return IRQ_NONE;
502
503         iova = ipmmu_ctx_read_root(domain, IMELAR);
504         if (IS_ENABLED(CONFIG_64BIT))
505                 iova |= (u64)ipmmu_ctx_read_root(domain, IMEUAR) << 32;
506
507         /*
508          * Clear the error status flags. Unlike traditional interrupt flag
509          * registers that must be cleared by writing 1, this status register
510          * seems to require 0. The error address register must be read before,
511          * otherwise its value will be 0.
512          */
513         ipmmu_ctx_write_root(domain, IMSTR, 0);
514
515         /* Log fatal errors. */
516         if (status & IMSTR_MHIT)
517                 dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%lx\n",
518                                     iova);
519         if (status & IMSTR_ABORT)
520                 dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%lx\n",
521                                     iova);
522
523         if (!(status & (IMSTR_PF | IMSTR_TF)))
524                 return IRQ_NONE;
525
526         /*
527          * Try to handle page faults and translation faults.
528          *
529          * TODO: We need to look up the faulty device based on the I/O VA. Use
530          * the IOMMU device for now.
531          */
532         if (!report_iommu_fault(&domain->io_domain, mmu->dev, iova, 0))
533                 return IRQ_HANDLED;
534
535         dev_err_ratelimited(mmu->dev,
536                             "Unhandled fault: status 0x%08x iova 0x%lx\n",
537                             status, iova);
538
539         return IRQ_HANDLED;
540 }
541
542 static irqreturn_t ipmmu_irq(int irq, void *dev)
543 {
544         struct ipmmu_vmsa_device *mmu = dev;
545         irqreturn_t status = IRQ_NONE;
546         unsigned int i;
547         unsigned long flags;
548
549         spin_lock_irqsave(&mmu->lock, flags);
550
551         /*
552          * Check interrupts for all active contexts.
553          */
554         for (i = 0; i < mmu->num_ctx; i++) {
555                 if (!mmu->domains[i])
556                         continue;
557                 if (ipmmu_domain_irq(mmu->domains[i]) == IRQ_HANDLED)
558                         status = IRQ_HANDLED;
559         }
560
561         spin_unlock_irqrestore(&mmu->lock, flags);
562
563         return status;
564 }
565
566 /* -----------------------------------------------------------------------------
567  * IOMMU Operations
568  */
569
570 static struct iommu_domain *ipmmu_domain_alloc(unsigned type)
571 {
572         struct ipmmu_vmsa_domain *domain;
573
574         if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
575                 return NULL;
576
577         domain = kzalloc(sizeof(*domain), GFP_KERNEL);
578         if (!domain)
579                 return NULL;
580
581         mutex_init(&domain->mutex);
582
583         return &domain->io_domain;
584 }
585
586 static void ipmmu_domain_free(struct iommu_domain *io_domain)
587 {
588         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
589
590         /*
591          * Free the domain resources. We assume that all devices have already
592          * been detached.
593          */
594         ipmmu_domain_destroy_context(domain);
595         free_io_pgtable_ops(domain->iop);
596         kfree(domain);
597 }
598
599 static int ipmmu_attach_device(struct iommu_domain *io_domain,
600                                struct device *dev)
601 {
602         struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
603         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
604         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
605         unsigned int i;
606         int ret = 0;
607
608         if (!mmu) {
609                 dev_err(dev, "Cannot attach to IPMMU\n");
610                 return -ENXIO;
611         }
612
613         mutex_lock(&domain->mutex);
614
615         if (!domain->mmu) {
616                 /* The domain hasn't been used yet, initialize it. */
617                 domain->mmu = mmu;
618                 ret = ipmmu_domain_init_context(domain);
619                 if (ret < 0) {
620                         dev_err(dev, "Unable to initialize IPMMU context\n");
621                         domain->mmu = NULL;
622                 } else {
623                         dev_info(dev, "Using IPMMU context %u\n",
624                                  domain->context_id);
625                 }
626         } else if (domain->mmu != mmu) {
627                 /*
628                  * Something is wrong, we can't attach two devices using
629                  * different IOMMUs to the same domain.
630                  */
631                 ret = -EINVAL;
632         } else
633                 dev_info(dev, "Reusing IPMMU context %u\n", domain->context_id);
634
635         mutex_unlock(&domain->mutex);
636
637         if (ret < 0)
638                 return ret;
639
640         for (i = 0; i < fwspec->num_ids; ++i)
641                 ipmmu_utlb_enable(domain, fwspec->ids[i]);
642
643         return 0;
644 }
645
646 static void ipmmu_detach_device(struct iommu_domain *io_domain,
647                                 struct device *dev)
648 {
649         struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
650         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
651         unsigned int i;
652
653         for (i = 0; i < fwspec->num_ids; ++i)
654                 ipmmu_utlb_disable(domain, fwspec->ids[i]);
655
656         /*
657          * TODO: Optimize by disabling the context when no device is attached.
658          */
659 }
660
661 static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
662                      phys_addr_t paddr, size_t pgsize, size_t pgcount,
663                      int prot, gfp_t gfp, size_t *mapped)
664 {
665         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
666
667         return domain->iop->map_pages(domain->iop, iova, paddr, pgsize, pgcount,
668                                       prot, gfp, mapped);
669 }
670
671 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
672                           size_t pgsize, size_t pgcount,
673                           struct iommu_iotlb_gather *gather)
674 {
675         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
676
677         return domain->iop->unmap_pages(domain->iop, iova, pgsize, pgcount, gather);
678 }
679
680 static void ipmmu_flush_iotlb_all(struct iommu_domain *io_domain)
681 {
682         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
683
684         if (domain->mmu)
685                 ipmmu_tlb_flush_all(domain);
686 }
687
688 static void ipmmu_iotlb_sync(struct iommu_domain *io_domain,
689                              struct iommu_iotlb_gather *gather)
690 {
691         ipmmu_flush_iotlb_all(io_domain);
692 }
693
694 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
695                                       dma_addr_t iova)
696 {
697         struct ipmmu_vmsa_domain *domain = to_vmsa_domain(io_domain);
698
699         /* TODO: Is locking needed ? */
700
701         return domain->iop->iova_to_phys(domain->iop, iova);
702 }
703
704 static int ipmmu_init_platform_device(struct device *dev,
705                                       struct of_phandle_args *args)
706 {
707         struct platform_device *ipmmu_pdev;
708
709         ipmmu_pdev = of_find_device_by_node(args->np);
710         if (!ipmmu_pdev)
711                 return -ENODEV;
712
713         dev_iommu_priv_set(dev, platform_get_drvdata(ipmmu_pdev));
714
715         return 0;
716 }
717
718 static const struct soc_device_attribute soc_needs_opt_in[] = {
719         { .family = "R-Car Gen3", },
720         { .family = "R-Car Gen4", },
721         { .family = "RZ/G2", },
722         { /* sentinel */ }
723 };
724
725 static const struct soc_device_attribute soc_denylist[] = {
726         { .soc_id = "r8a774a1", },
727         { .soc_id = "r8a7795", .revision = "ES1.*" },
728         { .soc_id = "r8a7795", .revision = "ES2.*" },
729         { .soc_id = "r8a7796", },
730         { /* sentinel */ }
731 };
732
733 static const char * const devices_allowlist[] = {
734         "ee100000.mmc",
735         "ee120000.mmc",
736         "ee140000.mmc",
737         "ee160000.mmc"
738 };
739
740 static bool ipmmu_device_is_allowed(struct device *dev)
741 {
742         unsigned int i;
743
744         /*
745          * R-Car Gen3/4 and RZ/G2 use the allow list to opt-in devices.
746          * For Other SoCs, this returns true anyway.
747          */
748         if (!soc_device_match(soc_needs_opt_in))
749                 return true;
750
751         /* Check whether this SoC can use the IPMMU correctly or not */
752         if (soc_device_match(soc_denylist))
753                 return false;
754
755         /* Check whether this device can work with the IPMMU */
756         for (i = 0; i < ARRAY_SIZE(devices_allowlist); i++) {
757                 if (!strcmp(dev_name(dev), devices_allowlist[i]))
758                         return true;
759         }
760
761         /* Otherwise, do not allow use of IPMMU */
762         return false;
763 }
764
765 static int ipmmu_of_xlate(struct device *dev,
766                           struct of_phandle_args *spec)
767 {
768         if (!ipmmu_device_is_allowed(dev))
769                 return -ENODEV;
770
771         iommu_fwspec_add_ids(dev, spec->args, 1);
772
773         /* Initialize once - xlate() will call multiple times */
774         if (to_ipmmu(dev))
775                 return 0;
776
777         return ipmmu_init_platform_device(dev, spec);
778 }
779
780 static int ipmmu_init_arm_mapping(struct device *dev)
781 {
782         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
783         int ret;
784
785         /*
786          * Create the ARM mapping, used by the ARM DMA mapping core to allocate
787          * VAs. This will allocate a corresponding IOMMU domain.
788          *
789          * TODO:
790          * - Create one mapping per context (TLB).
791          * - Make the mapping size configurable ? We currently use a 2GB mapping
792          *   at a 1GB offset to ensure that NULL VAs will fault.
793          */
794         if (!mmu->mapping) {
795                 struct dma_iommu_mapping *mapping;
796
797                 mapping = arm_iommu_create_mapping(&platform_bus_type,
798                                                    SZ_1G, SZ_2G);
799                 if (IS_ERR(mapping)) {
800                         dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
801                         ret = PTR_ERR(mapping);
802                         goto error;
803                 }
804
805                 mmu->mapping = mapping;
806         }
807
808         /* Attach the ARM VA mapping to the device. */
809         ret = arm_iommu_attach_device(dev, mmu->mapping);
810         if (ret < 0) {
811                 dev_err(dev, "Failed to attach device to VA mapping\n");
812                 goto error;
813         }
814
815         return 0;
816
817 error:
818         if (mmu->mapping)
819                 arm_iommu_release_mapping(mmu->mapping);
820
821         return ret;
822 }
823
824 static struct iommu_device *ipmmu_probe_device(struct device *dev)
825 {
826         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
827
828         /*
829          * Only let through devices that have been verified in xlate()
830          */
831         if (!mmu)
832                 return ERR_PTR(-ENODEV);
833
834         return &mmu->iommu;
835 }
836
837 static void ipmmu_probe_finalize(struct device *dev)
838 {
839         int ret = 0;
840
841         if (IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA))
842                 ret = ipmmu_init_arm_mapping(dev);
843
844         if (ret)
845                 dev_err(dev, "Can't create IOMMU mapping - DMA-OPS will not work\n");
846 }
847
848 static void ipmmu_release_device(struct device *dev)
849 {
850         arm_iommu_detach_device(dev);
851 }
852
853 static struct iommu_group *ipmmu_find_group(struct device *dev)
854 {
855         struct ipmmu_vmsa_device *mmu = to_ipmmu(dev);
856         struct iommu_group *group;
857
858         if (mmu->group)
859                 return iommu_group_ref_get(mmu->group);
860
861         group = iommu_group_alloc();
862         if (!IS_ERR(group))
863                 mmu->group = group;
864
865         return group;
866 }
867
868 static const struct iommu_ops ipmmu_ops = {
869         .domain_alloc = ipmmu_domain_alloc,
870         .probe_device = ipmmu_probe_device,
871         .release_device = ipmmu_release_device,
872         .probe_finalize = ipmmu_probe_finalize,
873         .device_group = IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_IOMMU_DMA)
874                         ? generic_device_group : ipmmu_find_group,
875         .pgsize_bitmap = SZ_1G | SZ_2M | SZ_4K,
876         .of_xlate = ipmmu_of_xlate,
877         .default_domain_ops = &(const struct iommu_domain_ops) {
878                 .attach_dev     = ipmmu_attach_device,
879                 .detach_dev     = ipmmu_detach_device,
880                 .map_pages      = ipmmu_map,
881                 .unmap_pages    = ipmmu_unmap,
882                 .flush_iotlb_all = ipmmu_flush_iotlb_all,
883                 .iotlb_sync     = ipmmu_iotlb_sync,
884                 .iova_to_phys   = ipmmu_iova_to_phys,
885                 .free           = ipmmu_domain_free,
886         }
887 };
888
889 /* -----------------------------------------------------------------------------
890  * Probe/remove and init
891  */
892
893 static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
894 {
895         unsigned int i;
896
897         /* Disable all contexts. */
898         for (i = 0; i < mmu->num_ctx; ++i)
899                 ipmmu_ctx_write(mmu, i, IMCTR, 0);
900 }
901
902 static const struct ipmmu_features ipmmu_features_default = {
903         .use_ns_alias_offset = true,
904         .has_cache_leaf_nodes = false,
905         .number_of_contexts = 1, /* software only tested with one context */
906         .num_utlbs = 32,
907         .setup_imbuscr = true,
908         .twobit_imttbcr_sl0 = false,
909         .reserved_context = false,
910         .cache_snoop = true,
911         .ctx_offset_base = 0,
912         .ctx_offset_stride = 0x40,
913         .utlb_offset_base = 0,
914 };
915
916 static const struct ipmmu_features ipmmu_features_rcar_gen3 = {
917         .use_ns_alias_offset = false,
918         .has_cache_leaf_nodes = true,
919         .number_of_contexts = 8,
920         .num_utlbs = 48,
921         .setup_imbuscr = false,
922         .twobit_imttbcr_sl0 = true,
923         .reserved_context = true,
924         .cache_snoop = false,
925         .ctx_offset_base = 0,
926         .ctx_offset_stride = 0x40,
927         .utlb_offset_base = 0,
928 };
929
930 static const struct ipmmu_features ipmmu_features_rcar_gen4 = {
931         .use_ns_alias_offset = false,
932         .has_cache_leaf_nodes = true,
933         .number_of_contexts = 16,
934         .num_utlbs = 64,
935         .setup_imbuscr = false,
936         .twobit_imttbcr_sl0 = true,
937         .reserved_context = true,
938         .cache_snoop = false,
939         .ctx_offset_base = 0x10000,
940         .ctx_offset_stride = 0x1040,
941         .utlb_offset_base = 0x3000,
942 };
943
944 static const struct of_device_id ipmmu_of_ids[] = {
945         {
946                 .compatible = "renesas,ipmmu-vmsa",
947                 .data = &ipmmu_features_default,
948         }, {
949                 .compatible = "renesas,ipmmu-r8a774a1",
950                 .data = &ipmmu_features_rcar_gen3,
951         }, {
952                 .compatible = "renesas,ipmmu-r8a774b1",
953                 .data = &ipmmu_features_rcar_gen3,
954         }, {
955                 .compatible = "renesas,ipmmu-r8a774c0",
956                 .data = &ipmmu_features_rcar_gen3,
957         }, {
958                 .compatible = "renesas,ipmmu-r8a774e1",
959                 .data = &ipmmu_features_rcar_gen3,
960         }, {
961                 .compatible = "renesas,ipmmu-r8a7795",
962                 .data = &ipmmu_features_rcar_gen3,
963         }, {
964                 .compatible = "renesas,ipmmu-r8a7796",
965                 .data = &ipmmu_features_rcar_gen3,
966         }, {
967                 .compatible = "renesas,ipmmu-r8a77961",
968                 .data = &ipmmu_features_rcar_gen3,
969         }, {
970                 .compatible = "renesas,ipmmu-r8a77965",
971                 .data = &ipmmu_features_rcar_gen3,
972         }, {
973                 .compatible = "renesas,ipmmu-r8a77970",
974                 .data = &ipmmu_features_rcar_gen3,
975         }, {
976                 .compatible = "renesas,ipmmu-r8a77980",
977                 .data = &ipmmu_features_rcar_gen3,
978         }, {
979                 .compatible = "renesas,ipmmu-r8a77990",
980                 .data = &ipmmu_features_rcar_gen3,
981         }, {
982                 .compatible = "renesas,ipmmu-r8a77995",
983                 .data = &ipmmu_features_rcar_gen3,
984         }, {
985                 .compatible = "renesas,ipmmu-r8a779a0",
986                 .data = &ipmmu_features_rcar_gen4,
987         }, {
988                 .compatible = "renesas,rcar-gen4-ipmmu-vmsa",
989                 .data = &ipmmu_features_rcar_gen4,
990         }, {
991                 /* Terminator */
992         },
993 };
994
995 static int ipmmu_probe(struct platform_device *pdev)
996 {
997         struct ipmmu_vmsa_device *mmu;
998         struct resource *res;
999         int irq;
1000         int ret;
1001
1002         mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
1003         if (!mmu) {
1004                 dev_err(&pdev->dev, "cannot allocate device data\n");
1005                 return -ENOMEM;
1006         }
1007
1008         mmu->dev = &pdev->dev;
1009         spin_lock_init(&mmu->lock);
1010         bitmap_zero(mmu->ctx, IPMMU_CTX_MAX);
1011         mmu->features = of_device_get_match_data(&pdev->dev);
1012         memset(mmu->utlb_ctx, IPMMU_CTX_INVALID, mmu->features->num_utlbs);
1013         ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(40));
1014         if (ret)
1015                 return ret;
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                 ret = iommu_device_register(&mmu->iommu, &ipmmu_ops, &pdev->dev);
1089                 if (ret)
1090                         return ret;
1091         }
1092
1093         /*
1094          * We can't create the ARM mapping here as it requires the bus to have
1095          * an IOMMU, which only happens when bus_set_iommu() is called in
1096          * ipmmu_init() after the probe function returns.
1097          */
1098
1099         platform_set_drvdata(pdev, mmu);
1100
1101         return 0;
1102 }
1103
1104 static int ipmmu_remove(struct platform_device *pdev)
1105 {
1106         struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
1107
1108         iommu_device_sysfs_remove(&mmu->iommu);
1109         iommu_device_unregister(&mmu->iommu);
1110
1111         arm_iommu_release_mapping(mmu->mapping);
1112
1113         ipmmu_device_reset(mmu);
1114
1115         return 0;
1116 }
1117
1118 #ifdef CONFIG_PM_SLEEP
1119 static int ipmmu_resume_noirq(struct device *dev)
1120 {
1121         struct ipmmu_vmsa_device *mmu = dev_get_drvdata(dev);
1122         unsigned int i;
1123
1124         /* Reset root MMU and restore contexts */
1125         if (ipmmu_is_root(mmu)) {
1126                 ipmmu_device_reset(mmu);
1127
1128                 for (i = 0; i < mmu->num_ctx; i++) {
1129                         if (!mmu->domains[i])
1130                                 continue;
1131
1132                         ipmmu_domain_setup_context(mmu->domains[i]);
1133                 }
1134         }
1135
1136         /* Re-enable active micro-TLBs */
1137         for (i = 0; i < mmu->features->num_utlbs; i++) {
1138                 if (mmu->utlb_ctx[i] == IPMMU_CTX_INVALID)
1139                         continue;
1140
1141                 ipmmu_utlb_enable(mmu->root->domains[mmu->utlb_ctx[i]], i);
1142         }
1143
1144         return 0;
1145 }
1146
1147 static const struct dev_pm_ops ipmmu_pm  = {
1148         SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(NULL, ipmmu_resume_noirq)
1149 };
1150 #define DEV_PM_OPS      &ipmmu_pm
1151 #else
1152 #define DEV_PM_OPS      NULL
1153 #endif /* CONFIG_PM_SLEEP */
1154
1155 static struct platform_driver ipmmu_driver = {
1156         .driver = {
1157                 .name = "ipmmu-vmsa",
1158                 .of_match_table = of_match_ptr(ipmmu_of_ids),
1159                 .pm = DEV_PM_OPS,
1160         },
1161         .probe = ipmmu_probe,
1162         .remove = ipmmu_remove,
1163 };
1164 builtin_platform_driver(ipmmu_driver);