xref: /linux/drivers/iommu/omap-iommu.c (revision b8d312aa075f33282565467662c4628dae0a2aff)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * omap iommu: tlb and pagetable primitives
4  *
5  * Copyright (C) 2008-2010 Nokia Corporation
6  * Copyright (C) 2013-2017 Texas Instruments Incorporated - http://www.ti.com/
7  *
8  * Written by Hiroshi DOYU <Hiroshi.DOYU@nokia.com>,
9  *		Paul Mundt and Toshihiro Kobayashi
10  */
11 
12 #include <linux/dma-mapping.h>
13 #include <linux/err.h>
14 #include <linux/slab.h>
15 #include <linux/interrupt.h>
16 #include <linux/ioport.h>
17 #include <linux/platform_device.h>
18 #include <linux/iommu.h>
19 #include <linux/omap-iommu.h>
20 #include <linux/mutex.h>
21 #include <linux/spinlock.h>
22 #include <linux/io.h>
23 #include <linux/pm_runtime.h>
24 #include <linux/of.h>
25 #include <linux/of_iommu.h>
26 #include <linux/of_irq.h>
27 #include <linux/of_platform.h>
28 #include <linux/regmap.h>
29 #include <linux/mfd/syscon.h>
30 
31 #include <linux/platform_data/iommu-omap.h>
32 
33 #include "omap-iopgtable.h"
34 #include "omap-iommu.h"
35 
36 static const struct iommu_ops omap_iommu_ops;
37 
38 #define to_iommu(dev)	((struct omap_iommu *)dev_get_drvdata(dev))
39 
40 /* bitmap of the page sizes currently supported */
41 #define OMAP_IOMMU_PGSIZES	(SZ_4K | SZ_64K | SZ_1M | SZ_16M)
42 
43 #define MMU_LOCK_BASE_SHIFT	10
44 #define MMU_LOCK_BASE_MASK	(0x1f << MMU_LOCK_BASE_SHIFT)
45 #define MMU_LOCK_BASE(x)	\
46 	((x & MMU_LOCK_BASE_MASK) >> MMU_LOCK_BASE_SHIFT)
47 
48 #define MMU_LOCK_VICT_SHIFT	4
49 #define MMU_LOCK_VICT_MASK	(0x1f << MMU_LOCK_VICT_SHIFT)
50 #define MMU_LOCK_VICT(x)	\
51 	((x & MMU_LOCK_VICT_MASK) >> MMU_LOCK_VICT_SHIFT)
52 
53 static struct platform_driver omap_iommu_driver;
54 static struct kmem_cache *iopte_cachep;
55 
56 /**
57  * to_omap_domain - Get struct omap_iommu_domain from generic iommu_domain
58  * @dom:	generic iommu domain handle
59  **/
60 static struct omap_iommu_domain *to_omap_domain(struct iommu_domain *dom)
61 {
62 	return container_of(dom, struct omap_iommu_domain, domain);
63 }
64 
65 /**
66  * omap_iommu_save_ctx - Save registers for pm off-mode support
67  * @dev:	client device
68  **/
69 void omap_iommu_save_ctx(struct device *dev)
70 {
71 	struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
72 	struct omap_iommu *obj;
73 	u32 *p;
74 	int i;
75 
76 	if (!arch_data)
77 		return;
78 
79 	while (arch_data->iommu_dev) {
80 		obj = arch_data->iommu_dev;
81 		p = obj->ctx;
82 		for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
83 			p[i] = iommu_read_reg(obj, i * sizeof(u32));
84 			dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i,
85 				p[i]);
86 		}
87 		arch_data++;
88 	}
89 }
90 EXPORT_SYMBOL_GPL(omap_iommu_save_ctx);
91 
92 /**
93  * omap_iommu_restore_ctx - Restore registers for pm off-mode support
94  * @dev:	client device
95  **/
96 void omap_iommu_restore_ctx(struct device *dev)
97 {
98 	struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
99 	struct omap_iommu *obj;
100 	u32 *p;
101 	int i;
102 
103 	if (!arch_data)
104 		return;
105 
106 	while (arch_data->iommu_dev) {
107 		obj = arch_data->iommu_dev;
108 		p = obj->ctx;
109 		for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
110 			iommu_write_reg(obj, p[i], i * sizeof(u32));
111 			dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i,
112 				p[i]);
113 		}
114 		arch_data++;
115 	}
116 }
117 EXPORT_SYMBOL_GPL(omap_iommu_restore_ctx);
118 
119 static void dra7_cfg_dspsys_mmu(struct omap_iommu *obj, bool enable)
120 {
121 	u32 val, mask;
122 
123 	if (!obj->syscfg)
124 		return;
125 
126 	mask = (1 << (obj->id * DSP_SYS_MMU_CONFIG_EN_SHIFT));
127 	val = enable ? mask : 0;
128 	regmap_update_bits(obj->syscfg, DSP_SYS_MMU_CONFIG, mask, val);
129 }
130 
131 static void __iommu_set_twl(struct omap_iommu *obj, bool on)
132 {
133 	u32 l = iommu_read_reg(obj, MMU_CNTL);
134 
135 	if (on)
136 		iommu_write_reg(obj, MMU_IRQ_TWL_MASK, MMU_IRQENABLE);
137 	else
138 		iommu_write_reg(obj, MMU_IRQ_TLB_MISS_MASK, MMU_IRQENABLE);
139 
140 	l &= ~MMU_CNTL_MASK;
141 	if (on)
142 		l |= (MMU_CNTL_MMU_EN | MMU_CNTL_TWL_EN);
143 	else
144 		l |= (MMU_CNTL_MMU_EN);
145 
146 	iommu_write_reg(obj, l, MMU_CNTL);
147 }
148 
149 static int omap2_iommu_enable(struct omap_iommu *obj)
150 {
151 	u32 l, pa;
152 
153 	if (!obj->iopgd || !IS_ALIGNED((u32)obj->iopgd,  SZ_16K))
154 		return -EINVAL;
155 
156 	pa = virt_to_phys(obj->iopgd);
157 	if (!IS_ALIGNED(pa, SZ_16K))
158 		return -EINVAL;
159 
160 	l = iommu_read_reg(obj, MMU_REVISION);
161 	dev_info(obj->dev, "%s: version %d.%d\n", obj->name,
162 		 (l >> 4) & 0xf, l & 0xf);
163 
164 	iommu_write_reg(obj, pa, MMU_TTB);
165 
166 	dra7_cfg_dspsys_mmu(obj, true);
167 
168 	if (obj->has_bus_err_back)
169 		iommu_write_reg(obj, MMU_GP_REG_BUS_ERR_BACK_EN, MMU_GP_REG);
170 
171 	__iommu_set_twl(obj, true);
172 
173 	return 0;
174 }
175 
176 static void omap2_iommu_disable(struct omap_iommu *obj)
177 {
178 	u32 l = iommu_read_reg(obj, MMU_CNTL);
179 
180 	l &= ~MMU_CNTL_MASK;
181 	iommu_write_reg(obj, l, MMU_CNTL);
182 	dra7_cfg_dspsys_mmu(obj, false);
183 
184 	dev_dbg(obj->dev, "%s is shutting down\n", obj->name);
185 }
186 
187 static int iommu_enable(struct omap_iommu *obj)
188 {
189 	int err;
190 	struct platform_device *pdev = to_platform_device(obj->dev);
191 	struct iommu_platform_data *pdata = dev_get_platdata(&pdev->dev);
192 
193 	if (pdata && pdata->deassert_reset) {
194 		err = pdata->deassert_reset(pdev, pdata->reset_name);
195 		if (err) {
196 			dev_err(obj->dev, "deassert_reset failed: %d\n", err);
197 			return err;
198 		}
199 	}
200 
201 	pm_runtime_get_sync(obj->dev);
202 
203 	err = omap2_iommu_enable(obj);
204 
205 	return err;
206 }
207 
208 static void iommu_disable(struct omap_iommu *obj)
209 {
210 	struct platform_device *pdev = to_platform_device(obj->dev);
211 	struct iommu_platform_data *pdata = dev_get_platdata(&pdev->dev);
212 
213 	omap2_iommu_disable(obj);
214 
215 	pm_runtime_put_sync(obj->dev);
216 
217 	if (pdata && pdata->assert_reset)
218 		pdata->assert_reset(pdev, pdata->reset_name);
219 }
220 
221 /*
222  *	TLB operations
223  */
224 static u32 iotlb_cr_to_virt(struct cr_regs *cr)
225 {
226 	u32 page_size = cr->cam & MMU_CAM_PGSZ_MASK;
227 	u32 mask = get_cam_va_mask(cr->cam & page_size);
228 
229 	return cr->cam & mask;
230 }
231 
232 static u32 get_iopte_attr(struct iotlb_entry *e)
233 {
234 	u32 attr;
235 
236 	attr = e->mixed << 5;
237 	attr |= e->endian;
238 	attr |= e->elsz >> 3;
239 	attr <<= (((e->pgsz == MMU_CAM_PGSZ_4K) ||
240 			(e->pgsz == MMU_CAM_PGSZ_64K)) ? 0 : 6);
241 	return attr;
242 }
243 
244 static u32 iommu_report_fault(struct omap_iommu *obj, u32 *da)
245 {
246 	u32 status, fault_addr;
247 
248 	status = iommu_read_reg(obj, MMU_IRQSTATUS);
249 	status &= MMU_IRQ_MASK;
250 	if (!status) {
251 		*da = 0;
252 		return 0;
253 	}
254 
255 	fault_addr = iommu_read_reg(obj, MMU_FAULT_AD);
256 	*da = fault_addr;
257 
258 	iommu_write_reg(obj, status, MMU_IRQSTATUS);
259 
260 	return status;
261 }
262 
263 void iotlb_lock_get(struct omap_iommu *obj, struct iotlb_lock *l)
264 {
265 	u32 val;
266 
267 	val = iommu_read_reg(obj, MMU_LOCK);
268 
269 	l->base = MMU_LOCK_BASE(val);
270 	l->vict = MMU_LOCK_VICT(val);
271 }
272 
273 void iotlb_lock_set(struct omap_iommu *obj, struct iotlb_lock *l)
274 {
275 	u32 val;
276 
277 	val = (l->base << MMU_LOCK_BASE_SHIFT);
278 	val |= (l->vict << MMU_LOCK_VICT_SHIFT);
279 
280 	iommu_write_reg(obj, val, MMU_LOCK);
281 }
282 
283 static void iotlb_read_cr(struct omap_iommu *obj, struct cr_regs *cr)
284 {
285 	cr->cam = iommu_read_reg(obj, MMU_READ_CAM);
286 	cr->ram = iommu_read_reg(obj, MMU_READ_RAM);
287 }
288 
289 static void iotlb_load_cr(struct omap_iommu *obj, struct cr_regs *cr)
290 {
291 	iommu_write_reg(obj, cr->cam | MMU_CAM_V, MMU_CAM);
292 	iommu_write_reg(obj, cr->ram, MMU_RAM);
293 
294 	iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
295 	iommu_write_reg(obj, 1, MMU_LD_TLB);
296 }
297 
298 /* only used in iotlb iteration for-loop */
299 struct cr_regs __iotlb_read_cr(struct omap_iommu *obj, int n)
300 {
301 	struct cr_regs cr;
302 	struct iotlb_lock l;
303 
304 	iotlb_lock_get(obj, &l);
305 	l.vict = n;
306 	iotlb_lock_set(obj, &l);
307 	iotlb_read_cr(obj, &cr);
308 
309 	return cr;
310 }
311 
312 #ifdef PREFETCH_IOTLB
313 static struct cr_regs *iotlb_alloc_cr(struct omap_iommu *obj,
314 				      struct iotlb_entry *e)
315 {
316 	struct cr_regs *cr;
317 
318 	if (!e)
319 		return NULL;
320 
321 	if (e->da & ~(get_cam_va_mask(e->pgsz))) {
322 		dev_err(obj->dev, "%s:\twrong alignment: %08x\n", __func__,
323 			e->da);
324 		return ERR_PTR(-EINVAL);
325 	}
326 
327 	cr = kmalloc(sizeof(*cr), GFP_KERNEL);
328 	if (!cr)
329 		return ERR_PTR(-ENOMEM);
330 
331 	cr->cam = (e->da & MMU_CAM_VATAG_MASK) | e->prsvd | e->pgsz | e->valid;
332 	cr->ram = e->pa | e->endian | e->elsz | e->mixed;
333 
334 	return cr;
335 }
336 
337 /**
338  * load_iotlb_entry - Set an iommu tlb entry
339  * @obj:	target iommu
340  * @e:		an iommu tlb entry info
341  **/
342 static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
343 {
344 	int err = 0;
345 	struct iotlb_lock l;
346 	struct cr_regs *cr;
347 
348 	if (!obj || !obj->nr_tlb_entries || !e)
349 		return -EINVAL;
350 
351 	pm_runtime_get_sync(obj->dev);
352 
353 	iotlb_lock_get(obj, &l);
354 	if (l.base == obj->nr_tlb_entries) {
355 		dev_warn(obj->dev, "%s: preserve entries full\n", __func__);
356 		err = -EBUSY;
357 		goto out;
358 	}
359 	if (!e->prsvd) {
360 		int i;
361 		struct cr_regs tmp;
362 
363 		for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, tmp)
364 			if (!iotlb_cr_valid(&tmp))
365 				break;
366 
367 		if (i == obj->nr_tlb_entries) {
368 			dev_dbg(obj->dev, "%s: full: no entry\n", __func__);
369 			err = -EBUSY;
370 			goto out;
371 		}
372 
373 		iotlb_lock_get(obj, &l);
374 	} else {
375 		l.vict = l.base;
376 		iotlb_lock_set(obj, &l);
377 	}
378 
379 	cr = iotlb_alloc_cr(obj, e);
380 	if (IS_ERR(cr)) {
381 		pm_runtime_put_sync(obj->dev);
382 		return PTR_ERR(cr);
383 	}
384 
385 	iotlb_load_cr(obj, cr);
386 	kfree(cr);
387 
388 	if (e->prsvd)
389 		l.base++;
390 	/* increment victim for next tlb load */
391 	if (++l.vict == obj->nr_tlb_entries)
392 		l.vict = l.base;
393 	iotlb_lock_set(obj, &l);
394 out:
395 	pm_runtime_put_sync(obj->dev);
396 	return err;
397 }
398 
399 #else /* !PREFETCH_IOTLB */
400 
401 static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
402 {
403 	return 0;
404 }
405 
406 #endif /* !PREFETCH_IOTLB */
407 
408 static int prefetch_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
409 {
410 	return load_iotlb_entry(obj, e);
411 }
412 
413 /**
414  * flush_iotlb_page - Clear an iommu tlb entry
415  * @obj:	target iommu
416  * @da:		iommu device virtual address
417  *
418  * Clear an iommu tlb entry which includes 'da' address.
419  **/
420 static void flush_iotlb_page(struct omap_iommu *obj, u32 da)
421 {
422 	int i;
423 	struct cr_regs cr;
424 
425 	pm_runtime_get_sync(obj->dev);
426 
427 	for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, cr) {
428 		u32 start;
429 		size_t bytes;
430 
431 		if (!iotlb_cr_valid(&cr))
432 			continue;
433 
434 		start = iotlb_cr_to_virt(&cr);
435 		bytes = iopgsz_to_bytes(cr.cam & 3);
436 
437 		if ((start <= da) && (da < start + bytes)) {
438 			dev_dbg(obj->dev, "%s: %08x<=%08x(%x)\n",
439 				__func__, start, da, bytes);
440 			iotlb_load_cr(obj, &cr);
441 			iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
442 			break;
443 		}
444 	}
445 	pm_runtime_put_sync(obj->dev);
446 
447 	if (i == obj->nr_tlb_entries)
448 		dev_dbg(obj->dev, "%s: no page for %08x\n", __func__, da);
449 }
450 
451 /**
452  * flush_iotlb_all - Clear all iommu tlb entries
453  * @obj:	target iommu
454  **/
455 static void flush_iotlb_all(struct omap_iommu *obj)
456 {
457 	struct iotlb_lock l;
458 
459 	pm_runtime_get_sync(obj->dev);
460 
461 	l.base = 0;
462 	l.vict = 0;
463 	iotlb_lock_set(obj, &l);
464 
465 	iommu_write_reg(obj, 1, MMU_GFLUSH);
466 
467 	pm_runtime_put_sync(obj->dev);
468 }
469 
470 /*
471  *	H/W pagetable operations
472  */
473 static void flush_iopte_range(struct device *dev, dma_addr_t dma,
474 			      unsigned long offset, int num_entries)
475 {
476 	size_t size = num_entries * sizeof(u32);
477 
478 	dma_sync_single_range_for_device(dev, dma, offset, size, DMA_TO_DEVICE);
479 }
480 
481 static void iopte_free(struct omap_iommu *obj, u32 *iopte, bool dma_valid)
482 {
483 	dma_addr_t pt_dma;
484 
485 	/* Note: freed iopte's must be clean ready for re-use */
486 	if (iopte) {
487 		if (dma_valid) {
488 			pt_dma = virt_to_phys(iopte);
489 			dma_unmap_single(obj->dev, pt_dma, IOPTE_TABLE_SIZE,
490 					 DMA_TO_DEVICE);
491 		}
492 
493 		kmem_cache_free(iopte_cachep, iopte);
494 	}
495 }
496 
497 static u32 *iopte_alloc(struct omap_iommu *obj, u32 *iopgd,
498 			dma_addr_t *pt_dma, u32 da)
499 {
500 	u32 *iopte;
501 	unsigned long offset = iopgd_index(da) * sizeof(da);
502 
503 	/* a table has already existed */
504 	if (*iopgd)
505 		goto pte_ready;
506 
507 	/*
508 	 * do the allocation outside the page table lock
509 	 */
510 	spin_unlock(&obj->page_table_lock);
511 	iopte = kmem_cache_zalloc(iopte_cachep, GFP_KERNEL);
512 	spin_lock(&obj->page_table_lock);
513 
514 	if (!*iopgd) {
515 		if (!iopte)
516 			return ERR_PTR(-ENOMEM);
517 
518 		*pt_dma = dma_map_single(obj->dev, iopte, IOPTE_TABLE_SIZE,
519 					 DMA_TO_DEVICE);
520 		if (dma_mapping_error(obj->dev, *pt_dma)) {
521 			dev_err(obj->dev, "DMA map error for L2 table\n");
522 			iopte_free(obj, iopte, false);
523 			return ERR_PTR(-ENOMEM);
524 		}
525 
526 		/*
527 		 * we rely on dma address and the physical address to be
528 		 * the same for mapping the L2 table
529 		 */
530 		if (WARN_ON(*pt_dma != virt_to_phys(iopte))) {
531 			dev_err(obj->dev, "DMA translation error for L2 table\n");
532 			dma_unmap_single(obj->dev, *pt_dma, IOPTE_TABLE_SIZE,
533 					 DMA_TO_DEVICE);
534 			iopte_free(obj, iopte, false);
535 			return ERR_PTR(-ENOMEM);
536 		}
537 
538 		*iopgd = virt_to_phys(iopte) | IOPGD_TABLE;
539 
540 		flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
541 		dev_vdbg(obj->dev, "%s: a new pte:%p\n", __func__, iopte);
542 	} else {
543 		/* We raced, free the reduniovant table */
544 		iopte_free(obj, iopte, false);
545 	}
546 
547 pte_ready:
548 	iopte = iopte_offset(iopgd, da);
549 	*pt_dma = iopgd_page_paddr(iopgd);
550 	dev_vdbg(obj->dev,
551 		 "%s: da:%08x pgd:%p *pgd:%08x pte:%p *pte:%08x\n",
552 		 __func__, da, iopgd, *iopgd, iopte, *iopte);
553 
554 	return iopte;
555 }
556 
557 static int iopgd_alloc_section(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
558 {
559 	u32 *iopgd = iopgd_offset(obj, da);
560 	unsigned long offset = iopgd_index(da) * sizeof(da);
561 
562 	if ((da | pa) & ~IOSECTION_MASK) {
563 		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
564 			__func__, da, pa, IOSECTION_SIZE);
565 		return -EINVAL;
566 	}
567 
568 	*iopgd = (pa & IOSECTION_MASK) | prot | IOPGD_SECTION;
569 	flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
570 	return 0;
571 }
572 
573 static int iopgd_alloc_super(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
574 {
575 	u32 *iopgd = iopgd_offset(obj, da);
576 	unsigned long offset = iopgd_index(da) * sizeof(da);
577 	int i;
578 
579 	if ((da | pa) & ~IOSUPER_MASK) {
580 		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
581 			__func__, da, pa, IOSUPER_SIZE);
582 		return -EINVAL;
583 	}
584 
585 	for (i = 0; i < 16; i++)
586 		*(iopgd + i) = (pa & IOSUPER_MASK) | prot | IOPGD_SUPER;
587 	flush_iopte_range(obj->dev, obj->pd_dma, offset, 16);
588 	return 0;
589 }
590 
591 static int iopte_alloc_page(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
592 {
593 	u32 *iopgd = iopgd_offset(obj, da);
594 	dma_addr_t pt_dma;
595 	u32 *iopte = iopte_alloc(obj, iopgd, &pt_dma, da);
596 	unsigned long offset = iopte_index(da) * sizeof(da);
597 
598 	if (IS_ERR(iopte))
599 		return PTR_ERR(iopte);
600 
601 	*iopte = (pa & IOPAGE_MASK) | prot | IOPTE_SMALL;
602 	flush_iopte_range(obj->dev, pt_dma, offset, 1);
603 
604 	dev_vdbg(obj->dev, "%s: da:%08x pa:%08x pte:%p *pte:%08x\n",
605 		 __func__, da, pa, iopte, *iopte);
606 
607 	return 0;
608 }
609 
610 static int iopte_alloc_large(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
611 {
612 	u32 *iopgd = iopgd_offset(obj, da);
613 	dma_addr_t pt_dma;
614 	u32 *iopte = iopte_alloc(obj, iopgd, &pt_dma, da);
615 	unsigned long offset = iopte_index(da) * sizeof(da);
616 	int i;
617 
618 	if ((da | pa) & ~IOLARGE_MASK) {
619 		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
620 			__func__, da, pa, IOLARGE_SIZE);
621 		return -EINVAL;
622 	}
623 
624 	if (IS_ERR(iopte))
625 		return PTR_ERR(iopte);
626 
627 	for (i = 0; i < 16; i++)
628 		*(iopte + i) = (pa & IOLARGE_MASK) | prot | IOPTE_LARGE;
629 	flush_iopte_range(obj->dev, pt_dma, offset, 16);
630 	return 0;
631 }
632 
633 static int
634 iopgtable_store_entry_core(struct omap_iommu *obj, struct iotlb_entry *e)
635 {
636 	int (*fn)(struct omap_iommu *, u32, u32, u32);
637 	u32 prot;
638 	int err;
639 
640 	if (!obj || !e)
641 		return -EINVAL;
642 
643 	switch (e->pgsz) {
644 	case MMU_CAM_PGSZ_16M:
645 		fn = iopgd_alloc_super;
646 		break;
647 	case MMU_CAM_PGSZ_1M:
648 		fn = iopgd_alloc_section;
649 		break;
650 	case MMU_CAM_PGSZ_64K:
651 		fn = iopte_alloc_large;
652 		break;
653 	case MMU_CAM_PGSZ_4K:
654 		fn = iopte_alloc_page;
655 		break;
656 	default:
657 		fn = NULL;
658 		break;
659 	}
660 
661 	if (WARN_ON(!fn))
662 		return -EINVAL;
663 
664 	prot = get_iopte_attr(e);
665 
666 	spin_lock(&obj->page_table_lock);
667 	err = fn(obj, e->da, e->pa, prot);
668 	spin_unlock(&obj->page_table_lock);
669 
670 	return err;
671 }
672 
673 /**
674  * omap_iopgtable_store_entry - Make an iommu pte entry
675  * @obj:	target iommu
676  * @e:		an iommu tlb entry info
677  **/
678 static int
679 omap_iopgtable_store_entry(struct omap_iommu *obj, struct iotlb_entry *e)
680 {
681 	int err;
682 
683 	flush_iotlb_page(obj, e->da);
684 	err = iopgtable_store_entry_core(obj, e);
685 	if (!err)
686 		prefetch_iotlb_entry(obj, e);
687 	return err;
688 }
689 
690 /**
691  * iopgtable_lookup_entry - Lookup an iommu pte entry
692  * @obj:	target iommu
693  * @da:		iommu device virtual address
694  * @ppgd:	iommu pgd entry pointer to be returned
695  * @ppte:	iommu pte entry pointer to be returned
696  **/
697 static void
698 iopgtable_lookup_entry(struct omap_iommu *obj, u32 da, u32 **ppgd, u32 **ppte)
699 {
700 	u32 *iopgd, *iopte = NULL;
701 
702 	iopgd = iopgd_offset(obj, da);
703 	if (!*iopgd)
704 		goto out;
705 
706 	if (iopgd_is_table(*iopgd))
707 		iopte = iopte_offset(iopgd, da);
708 out:
709 	*ppgd = iopgd;
710 	*ppte = iopte;
711 }
712 
713 static size_t iopgtable_clear_entry_core(struct omap_iommu *obj, u32 da)
714 {
715 	size_t bytes;
716 	u32 *iopgd = iopgd_offset(obj, da);
717 	int nent = 1;
718 	dma_addr_t pt_dma;
719 	unsigned long pd_offset = iopgd_index(da) * sizeof(da);
720 	unsigned long pt_offset = iopte_index(da) * sizeof(da);
721 
722 	if (!*iopgd)
723 		return 0;
724 
725 	if (iopgd_is_table(*iopgd)) {
726 		int i;
727 		u32 *iopte = iopte_offset(iopgd, da);
728 
729 		bytes = IOPTE_SIZE;
730 		if (*iopte & IOPTE_LARGE) {
731 			nent *= 16;
732 			/* rewind to the 1st entry */
733 			iopte = iopte_offset(iopgd, (da & IOLARGE_MASK));
734 		}
735 		bytes *= nent;
736 		memset(iopte, 0, nent * sizeof(*iopte));
737 		pt_dma = iopgd_page_paddr(iopgd);
738 		flush_iopte_range(obj->dev, pt_dma, pt_offset, nent);
739 
740 		/*
741 		 * do table walk to check if this table is necessary or not
742 		 */
743 		iopte = iopte_offset(iopgd, 0);
744 		for (i = 0; i < PTRS_PER_IOPTE; i++)
745 			if (iopte[i])
746 				goto out;
747 
748 		iopte_free(obj, iopte, true);
749 		nent = 1; /* for the next L1 entry */
750 	} else {
751 		bytes = IOPGD_SIZE;
752 		if ((*iopgd & IOPGD_SUPER) == IOPGD_SUPER) {
753 			nent *= 16;
754 			/* rewind to the 1st entry */
755 			iopgd = iopgd_offset(obj, (da & IOSUPER_MASK));
756 		}
757 		bytes *= nent;
758 	}
759 	memset(iopgd, 0, nent * sizeof(*iopgd));
760 	flush_iopte_range(obj->dev, obj->pd_dma, pd_offset, nent);
761 out:
762 	return bytes;
763 }
764 
765 /**
766  * iopgtable_clear_entry - Remove an iommu pte entry
767  * @obj:	target iommu
768  * @da:		iommu device virtual address
769  **/
770 static size_t iopgtable_clear_entry(struct omap_iommu *obj, u32 da)
771 {
772 	size_t bytes;
773 
774 	spin_lock(&obj->page_table_lock);
775 
776 	bytes = iopgtable_clear_entry_core(obj, da);
777 	flush_iotlb_page(obj, da);
778 
779 	spin_unlock(&obj->page_table_lock);
780 
781 	return bytes;
782 }
783 
784 static void iopgtable_clear_entry_all(struct omap_iommu *obj)
785 {
786 	unsigned long offset;
787 	int i;
788 
789 	spin_lock(&obj->page_table_lock);
790 
791 	for (i = 0; i < PTRS_PER_IOPGD; i++) {
792 		u32 da;
793 		u32 *iopgd;
794 
795 		da = i << IOPGD_SHIFT;
796 		iopgd = iopgd_offset(obj, da);
797 		offset = iopgd_index(da) * sizeof(da);
798 
799 		if (!*iopgd)
800 			continue;
801 
802 		if (iopgd_is_table(*iopgd))
803 			iopte_free(obj, iopte_offset(iopgd, 0), true);
804 
805 		*iopgd = 0;
806 		flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
807 	}
808 
809 	flush_iotlb_all(obj);
810 
811 	spin_unlock(&obj->page_table_lock);
812 }
813 
814 /*
815  *	Device IOMMU generic operations
816  */
817 static irqreturn_t iommu_fault_handler(int irq, void *data)
818 {
819 	u32 da, errs;
820 	u32 *iopgd, *iopte;
821 	struct omap_iommu *obj = data;
822 	struct iommu_domain *domain = obj->domain;
823 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
824 
825 	if (!omap_domain->dev)
826 		return IRQ_NONE;
827 
828 	errs = iommu_report_fault(obj, &da);
829 	if (errs == 0)
830 		return IRQ_HANDLED;
831 
832 	/* Fault callback or TLB/PTE Dynamic loading */
833 	if (!report_iommu_fault(domain, obj->dev, da, 0))
834 		return IRQ_HANDLED;
835 
836 	iommu_write_reg(obj, 0, MMU_IRQENABLE);
837 
838 	iopgd = iopgd_offset(obj, da);
839 
840 	if (!iopgd_is_table(*iopgd)) {
841 		dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:px%08x\n",
842 			obj->name, errs, da, iopgd, *iopgd);
843 		return IRQ_NONE;
844 	}
845 
846 	iopte = iopte_offset(iopgd, da);
847 
848 	dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:0x%08x pte:0x%p *pte:0x%08x\n",
849 		obj->name, errs, da, iopgd, *iopgd, iopte, *iopte);
850 
851 	return IRQ_NONE;
852 }
853 
854 /**
855  * omap_iommu_attach() - attach iommu device to an iommu domain
856  * @obj:	target omap iommu device
857  * @iopgd:	page table
858  **/
859 static int omap_iommu_attach(struct omap_iommu *obj, u32 *iopgd)
860 {
861 	int err;
862 
863 	spin_lock(&obj->iommu_lock);
864 
865 	obj->pd_dma = dma_map_single(obj->dev, iopgd, IOPGD_TABLE_SIZE,
866 				     DMA_TO_DEVICE);
867 	if (dma_mapping_error(obj->dev, obj->pd_dma)) {
868 		dev_err(obj->dev, "DMA map error for L1 table\n");
869 		err = -ENOMEM;
870 		goto out_err;
871 	}
872 
873 	obj->iopgd = iopgd;
874 	err = iommu_enable(obj);
875 	if (err)
876 		goto out_err;
877 	flush_iotlb_all(obj);
878 
879 	spin_unlock(&obj->iommu_lock);
880 
881 	dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
882 
883 	return 0;
884 
885 out_err:
886 	spin_unlock(&obj->iommu_lock);
887 
888 	return err;
889 }
890 
891 /**
892  * omap_iommu_detach - release iommu device
893  * @obj:	target iommu
894  **/
895 static void omap_iommu_detach(struct omap_iommu *obj)
896 {
897 	if (!obj || IS_ERR(obj))
898 		return;
899 
900 	spin_lock(&obj->iommu_lock);
901 
902 	dma_unmap_single(obj->dev, obj->pd_dma, IOPGD_TABLE_SIZE,
903 			 DMA_TO_DEVICE);
904 	iommu_disable(obj);
905 	obj->pd_dma = 0;
906 	obj->iopgd = NULL;
907 
908 	spin_unlock(&obj->iommu_lock);
909 
910 	dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
911 }
912 
913 static bool omap_iommu_can_register(struct platform_device *pdev)
914 {
915 	struct device_node *np = pdev->dev.of_node;
916 
917 	if (!of_device_is_compatible(np, "ti,dra7-dsp-iommu"))
918 		return true;
919 
920 	/*
921 	 * restrict IOMMU core registration only for processor-port MDMA MMUs
922 	 * on DRA7 DSPs
923 	 */
924 	if ((!strcmp(dev_name(&pdev->dev), "40d01000.mmu")) ||
925 	    (!strcmp(dev_name(&pdev->dev), "41501000.mmu")))
926 		return true;
927 
928 	return false;
929 }
930 
931 static int omap_iommu_dra7_get_dsp_system_cfg(struct platform_device *pdev,
932 					      struct omap_iommu *obj)
933 {
934 	struct device_node *np = pdev->dev.of_node;
935 	int ret;
936 
937 	if (!of_device_is_compatible(np, "ti,dra7-dsp-iommu"))
938 		return 0;
939 
940 	if (!of_property_read_bool(np, "ti,syscon-mmuconfig")) {
941 		dev_err(&pdev->dev, "ti,syscon-mmuconfig property is missing\n");
942 		return -EINVAL;
943 	}
944 
945 	obj->syscfg =
946 		syscon_regmap_lookup_by_phandle(np, "ti,syscon-mmuconfig");
947 	if (IS_ERR(obj->syscfg)) {
948 		/* can fail with -EPROBE_DEFER */
949 		ret = PTR_ERR(obj->syscfg);
950 		return ret;
951 	}
952 
953 	if (of_property_read_u32_index(np, "ti,syscon-mmuconfig", 1,
954 				       &obj->id)) {
955 		dev_err(&pdev->dev, "couldn't get the IOMMU instance id within subsystem\n");
956 		return -EINVAL;
957 	}
958 
959 	if (obj->id != 0 && obj->id != 1) {
960 		dev_err(&pdev->dev, "invalid IOMMU instance id\n");
961 		return -EINVAL;
962 	}
963 
964 	return 0;
965 }
966 
967 /*
968  *	OMAP Device MMU(IOMMU) detection
969  */
970 static int omap_iommu_probe(struct platform_device *pdev)
971 {
972 	int err = -ENODEV;
973 	int irq;
974 	struct omap_iommu *obj;
975 	struct resource *res;
976 	struct device_node *of = pdev->dev.of_node;
977 
978 	if (!of) {
979 		pr_err("%s: only DT-based devices are supported\n", __func__);
980 		return -ENODEV;
981 	}
982 
983 	obj = devm_kzalloc(&pdev->dev, sizeof(*obj) + MMU_REG_SIZE, GFP_KERNEL);
984 	if (!obj)
985 		return -ENOMEM;
986 
987 	obj->name = dev_name(&pdev->dev);
988 	obj->nr_tlb_entries = 32;
989 	err = of_property_read_u32(of, "ti,#tlb-entries", &obj->nr_tlb_entries);
990 	if (err && err != -EINVAL)
991 		return err;
992 	if (obj->nr_tlb_entries != 32 && obj->nr_tlb_entries != 8)
993 		return -EINVAL;
994 	if (of_find_property(of, "ti,iommu-bus-err-back", NULL))
995 		obj->has_bus_err_back = MMU_GP_REG_BUS_ERR_BACK_EN;
996 
997 	obj->dev = &pdev->dev;
998 	obj->ctx = (void *)obj + sizeof(*obj);
999 
1000 	spin_lock_init(&obj->iommu_lock);
1001 	spin_lock_init(&obj->page_table_lock);
1002 
1003 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1004 	obj->regbase = devm_ioremap_resource(obj->dev, res);
1005 	if (IS_ERR(obj->regbase))
1006 		return PTR_ERR(obj->regbase);
1007 
1008 	err = omap_iommu_dra7_get_dsp_system_cfg(pdev, obj);
1009 	if (err)
1010 		return err;
1011 
1012 	irq = platform_get_irq(pdev, 0);
1013 	if (irq < 0)
1014 		return -ENODEV;
1015 
1016 	err = devm_request_irq(obj->dev, irq, iommu_fault_handler, IRQF_SHARED,
1017 			       dev_name(obj->dev), obj);
1018 	if (err < 0)
1019 		return err;
1020 	platform_set_drvdata(pdev, obj);
1021 
1022 	if (omap_iommu_can_register(pdev)) {
1023 		obj->group = iommu_group_alloc();
1024 		if (IS_ERR(obj->group))
1025 			return PTR_ERR(obj->group);
1026 
1027 		err = iommu_device_sysfs_add(&obj->iommu, obj->dev, NULL,
1028 					     obj->name);
1029 		if (err)
1030 			goto out_group;
1031 
1032 		iommu_device_set_ops(&obj->iommu, &omap_iommu_ops);
1033 
1034 		err = iommu_device_register(&obj->iommu);
1035 		if (err)
1036 			goto out_sysfs;
1037 	}
1038 
1039 	pm_runtime_irq_safe(obj->dev);
1040 	pm_runtime_enable(obj->dev);
1041 
1042 	omap_iommu_debugfs_add(obj);
1043 
1044 	dev_info(&pdev->dev, "%s registered\n", obj->name);
1045 
1046 	return 0;
1047 
1048 out_sysfs:
1049 	iommu_device_sysfs_remove(&obj->iommu);
1050 out_group:
1051 	iommu_group_put(obj->group);
1052 	return err;
1053 }
1054 
1055 static int omap_iommu_remove(struct platform_device *pdev)
1056 {
1057 	struct omap_iommu *obj = platform_get_drvdata(pdev);
1058 
1059 	if (obj->group) {
1060 		iommu_group_put(obj->group);
1061 		obj->group = NULL;
1062 
1063 		iommu_device_sysfs_remove(&obj->iommu);
1064 		iommu_device_unregister(&obj->iommu);
1065 	}
1066 
1067 	omap_iommu_debugfs_remove(obj);
1068 
1069 	pm_runtime_disable(obj->dev);
1070 
1071 	dev_info(&pdev->dev, "%s removed\n", obj->name);
1072 	return 0;
1073 }
1074 
1075 static const struct of_device_id omap_iommu_of_match[] = {
1076 	{ .compatible = "ti,omap2-iommu" },
1077 	{ .compatible = "ti,omap4-iommu" },
1078 	{ .compatible = "ti,dra7-iommu"	},
1079 	{ .compatible = "ti,dra7-dsp-iommu" },
1080 	{},
1081 };
1082 
1083 static struct platform_driver omap_iommu_driver = {
1084 	.probe	= omap_iommu_probe,
1085 	.remove	= omap_iommu_remove,
1086 	.driver	= {
1087 		.name	= "omap-iommu",
1088 		.of_match_table = of_match_ptr(omap_iommu_of_match),
1089 	},
1090 };
1091 
1092 static u32 iotlb_init_entry(struct iotlb_entry *e, u32 da, u32 pa, int pgsz)
1093 {
1094 	memset(e, 0, sizeof(*e));
1095 
1096 	e->da		= da;
1097 	e->pa		= pa;
1098 	e->valid	= MMU_CAM_V;
1099 	e->pgsz		= pgsz;
1100 	e->endian	= MMU_RAM_ENDIAN_LITTLE;
1101 	e->elsz		= MMU_RAM_ELSZ_8;
1102 	e->mixed	= 0;
1103 
1104 	return iopgsz_to_bytes(e->pgsz);
1105 }
1106 
1107 static int omap_iommu_map(struct iommu_domain *domain, unsigned long da,
1108 			  phys_addr_t pa, size_t bytes, int prot)
1109 {
1110 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1111 	struct device *dev = omap_domain->dev;
1112 	struct omap_iommu_device *iommu;
1113 	struct omap_iommu *oiommu;
1114 	struct iotlb_entry e;
1115 	int omap_pgsz;
1116 	u32 ret = -EINVAL;
1117 	int i;
1118 
1119 	omap_pgsz = bytes_to_iopgsz(bytes);
1120 	if (omap_pgsz < 0) {
1121 		dev_err(dev, "invalid size to map: %d\n", bytes);
1122 		return -EINVAL;
1123 	}
1124 
1125 	dev_dbg(dev, "mapping da 0x%lx to pa %pa size 0x%x\n", da, &pa, bytes);
1126 
1127 	iotlb_init_entry(&e, da, pa, omap_pgsz);
1128 
1129 	iommu = omap_domain->iommus;
1130 	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
1131 		oiommu = iommu->iommu_dev;
1132 		ret = omap_iopgtable_store_entry(oiommu, &e);
1133 		if (ret) {
1134 			dev_err(dev, "omap_iopgtable_store_entry failed: %d\n",
1135 				ret);
1136 			break;
1137 		}
1138 	}
1139 
1140 	if (ret) {
1141 		while (i--) {
1142 			iommu--;
1143 			oiommu = iommu->iommu_dev;
1144 			iopgtable_clear_entry(oiommu, da);
1145 		}
1146 	}
1147 
1148 	return ret;
1149 }
1150 
1151 static size_t omap_iommu_unmap(struct iommu_domain *domain, unsigned long da,
1152 			       size_t size)
1153 {
1154 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1155 	struct device *dev = omap_domain->dev;
1156 	struct omap_iommu_device *iommu;
1157 	struct omap_iommu *oiommu;
1158 	bool error = false;
1159 	size_t bytes = 0;
1160 	int i;
1161 
1162 	dev_dbg(dev, "unmapping da 0x%lx size %u\n", da, size);
1163 
1164 	iommu = omap_domain->iommus;
1165 	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
1166 		oiommu = iommu->iommu_dev;
1167 		bytes = iopgtable_clear_entry(oiommu, da);
1168 		if (!bytes)
1169 			error = true;
1170 	}
1171 
1172 	/*
1173 	 * simplify return - we are only checking if any of the iommus
1174 	 * reported an error, but not if all of them are unmapping the
1175 	 * same number of entries. This should not occur due to the
1176 	 * mirror programming.
1177 	 */
1178 	return error ? 0 : bytes;
1179 }
1180 
1181 static int omap_iommu_count(struct device *dev)
1182 {
1183 	struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
1184 	int count = 0;
1185 
1186 	while (arch_data->iommu_dev) {
1187 		count++;
1188 		arch_data++;
1189 	}
1190 
1191 	return count;
1192 }
1193 
1194 /* caller should call cleanup if this function fails */
1195 static int omap_iommu_attach_init(struct device *dev,
1196 				  struct omap_iommu_domain *odomain)
1197 {
1198 	struct omap_iommu_device *iommu;
1199 	int i;
1200 
1201 	odomain->num_iommus = omap_iommu_count(dev);
1202 	if (!odomain->num_iommus)
1203 		return -EINVAL;
1204 
1205 	odomain->iommus = kcalloc(odomain->num_iommus, sizeof(*iommu),
1206 				  GFP_ATOMIC);
1207 	if (!odomain->iommus)
1208 		return -ENOMEM;
1209 
1210 	iommu = odomain->iommus;
1211 	for (i = 0; i < odomain->num_iommus; i++, iommu++) {
1212 		iommu->pgtable = kzalloc(IOPGD_TABLE_SIZE, GFP_ATOMIC);
1213 		if (!iommu->pgtable)
1214 			return -ENOMEM;
1215 
1216 		/*
1217 		 * should never fail, but please keep this around to ensure
1218 		 * we keep the hardware happy
1219 		 */
1220 		if (WARN_ON(!IS_ALIGNED((long)iommu->pgtable,
1221 					IOPGD_TABLE_SIZE)))
1222 			return -EINVAL;
1223 	}
1224 
1225 	return 0;
1226 }
1227 
1228 static void omap_iommu_detach_fini(struct omap_iommu_domain *odomain)
1229 {
1230 	int i;
1231 	struct omap_iommu_device *iommu = odomain->iommus;
1232 
1233 	for (i = 0; iommu && i < odomain->num_iommus; i++, iommu++)
1234 		kfree(iommu->pgtable);
1235 
1236 	kfree(odomain->iommus);
1237 	odomain->num_iommus = 0;
1238 	odomain->iommus = NULL;
1239 }
1240 
1241 static int
1242 omap_iommu_attach_dev(struct iommu_domain *domain, struct device *dev)
1243 {
1244 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1245 	struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
1246 	struct omap_iommu_device *iommu;
1247 	struct omap_iommu *oiommu;
1248 	int ret = 0;
1249 	int i;
1250 
1251 	if (!arch_data || !arch_data->iommu_dev) {
1252 		dev_err(dev, "device doesn't have an associated iommu\n");
1253 		return -EINVAL;
1254 	}
1255 
1256 	spin_lock(&omap_domain->lock);
1257 
1258 	/* only a single client device can be attached to a domain */
1259 	if (omap_domain->dev) {
1260 		dev_err(dev, "iommu domain is already attached\n");
1261 		ret = -EBUSY;
1262 		goto out;
1263 	}
1264 
1265 	ret = omap_iommu_attach_init(dev, omap_domain);
1266 	if (ret) {
1267 		dev_err(dev, "failed to allocate required iommu data %d\n",
1268 			ret);
1269 		goto init_fail;
1270 	}
1271 
1272 	iommu = omap_domain->iommus;
1273 	for (i = 0; i < omap_domain->num_iommus; i++, iommu++, arch_data++) {
1274 		/* configure and enable the omap iommu */
1275 		oiommu = arch_data->iommu_dev;
1276 		ret = omap_iommu_attach(oiommu, iommu->pgtable);
1277 		if (ret) {
1278 			dev_err(dev, "can't get omap iommu: %d\n", ret);
1279 			goto attach_fail;
1280 		}
1281 
1282 		oiommu->domain = domain;
1283 		iommu->iommu_dev = oiommu;
1284 	}
1285 
1286 	omap_domain->dev = dev;
1287 
1288 	goto out;
1289 
1290 attach_fail:
1291 	while (i--) {
1292 		iommu--;
1293 		arch_data--;
1294 		oiommu = iommu->iommu_dev;
1295 		omap_iommu_detach(oiommu);
1296 		iommu->iommu_dev = NULL;
1297 		oiommu->domain = NULL;
1298 	}
1299 init_fail:
1300 	omap_iommu_detach_fini(omap_domain);
1301 out:
1302 	spin_unlock(&omap_domain->lock);
1303 	return ret;
1304 }
1305 
1306 static void _omap_iommu_detach_dev(struct omap_iommu_domain *omap_domain,
1307 				   struct device *dev)
1308 {
1309 	struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
1310 	struct omap_iommu_device *iommu = omap_domain->iommus;
1311 	struct omap_iommu *oiommu;
1312 	int i;
1313 
1314 	if (!omap_domain->dev) {
1315 		dev_err(dev, "domain has no attached device\n");
1316 		return;
1317 	}
1318 
1319 	/* only a single device is supported per domain for now */
1320 	if (omap_domain->dev != dev) {
1321 		dev_err(dev, "invalid attached device\n");
1322 		return;
1323 	}
1324 
1325 	/*
1326 	 * cleanup in the reverse order of attachment - this addresses
1327 	 * any h/w dependencies between multiple instances, if any
1328 	 */
1329 	iommu += (omap_domain->num_iommus - 1);
1330 	arch_data += (omap_domain->num_iommus - 1);
1331 	for (i = 0; i < omap_domain->num_iommus; i++, iommu--, arch_data--) {
1332 		oiommu = iommu->iommu_dev;
1333 		iopgtable_clear_entry_all(oiommu);
1334 
1335 		omap_iommu_detach(oiommu);
1336 		iommu->iommu_dev = NULL;
1337 		oiommu->domain = NULL;
1338 	}
1339 
1340 	omap_iommu_detach_fini(omap_domain);
1341 
1342 	omap_domain->dev = NULL;
1343 }
1344 
1345 static void omap_iommu_detach_dev(struct iommu_domain *domain,
1346 				  struct device *dev)
1347 {
1348 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1349 
1350 	spin_lock(&omap_domain->lock);
1351 	_omap_iommu_detach_dev(omap_domain, dev);
1352 	spin_unlock(&omap_domain->lock);
1353 }
1354 
1355 static struct iommu_domain *omap_iommu_domain_alloc(unsigned type)
1356 {
1357 	struct omap_iommu_domain *omap_domain;
1358 
1359 	if (type != IOMMU_DOMAIN_UNMANAGED)
1360 		return NULL;
1361 
1362 	omap_domain = kzalloc(sizeof(*omap_domain), GFP_KERNEL);
1363 	if (!omap_domain)
1364 		return NULL;
1365 
1366 	spin_lock_init(&omap_domain->lock);
1367 
1368 	omap_domain->domain.geometry.aperture_start = 0;
1369 	omap_domain->domain.geometry.aperture_end   = (1ULL << 32) - 1;
1370 	omap_domain->domain.geometry.force_aperture = true;
1371 
1372 	return &omap_domain->domain;
1373 }
1374 
1375 static void omap_iommu_domain_free(struct iommu_domain *domain)
1376 {
1377 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1378 
1379 	/*
1380 	 * An iommu device is still attached
1381 	 * (currently, only one device can be attached) ?
1382 	 */
1383 	if (omap_domain->dev)
1384 		_omap_iommu_detach_dev(omap_domain, omap_domain->dev);
1385 
1386 	kfree(omap_domain);
1387 }
1388 
1389 static phys_addr_t omap_iommu_iova_to_phys(struct iommu_domain *domain,
1390 					   dma_addr_t da)
1391 {
1392 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1393 	struct omap_iommu_device *iommu = omap_domain->iommus;
1394 	struct omap_iommu *oiommu = iommu->iommu_dev;
1395 	struct device *dev = oiommu->dev;
1396 	u32 *pgd, *pte;
1397 	phys_addr_t ret = 0;
1398 
1399 	/*
1400 	 * all the iommus within the domain will have identical programming,
1401 	 * so perform the lookup using just the first iommu
1402 	 */
1403 	iopgtable_lookup_entry(oiommu, da, &pgd, &pte);
1404 
1405 	if (pte) {
1406 		if (iopte_is_small(*pte))
1407 			ret = omap_iommu_translate(*pte, da, IOPTE_MASK);
1408 		else if (iopte_is_large(*pte))
1409 			ret = omap_iommu_translate(*pte, da, IOLARGE_MASK);
1410 		else
1411 			dev_err(dev, "bogus pte 0x%x, da 0x%llx", *pte,
1412 				(unsigned long long)da);
1413 	} else {
1414 		if (iopgd_is_section(*pgd))
1415 			ret = omap_iommu_translate(*pgd, da, IOSECTION_MASK);
1416 		else if (iopgd_is_super(*pgd))
1417 			ret = omap_iommu_translate(*pgd, da, IOSUPER_MASK);
1418 		else
1419 			dev_err(dev, "bogus pgd 0x%x, da 0x%llx", *pgd,
1420 				(unsigned long long)da);
1421 	}
1422 
1423 	return ret;
1424 }
1425 
1426 static int omap_iommu_add_device(struct device *dev)
1427 {
1428 	struct omap_iommu_arch_data *arch_data, *tmp;
1429 	struct omap_iommu *oiommu;
1430 	struct iommu_group *group;
1431 	struct device_node *np;
1432 	struct platform_device *pdev;
1433 	int num_iommus, i;
1434 	int ret;
1435 
1436 	/*
1437 	 * Allocate the archdata iommu structure for DT-based devices.
1438 	 *
1439 	 * TODO: Simplify this when removing non-DT support completely from the
1440 	 * IOMMU users.
1441 	 */
1442 	if (!dev->of_node)
1443 		return 0;
1444 
1445 	/*
1446 	 * retrieve the count of IOMMU nodes using phandle size as element size
1447 	 * since #iommu-cells = 0 for OMAP
1448 	 */
1449 	num_iommus = of_property_count_elems_of_size(dev->of_node, "iommus",
1450 						     sizeof(phandle));
1451 	if (num_iommus < 0)
1452 		return 0;
1453 
1454 	arch_data = kcalloc(num_iommus + 1, sizeof(*arch_data), GFP_KERNEL);
1455 	if (!arch_data)
1456 		return -ENOMEM;
1457 
1458 	for (i = 0, tmp = arch_data; i < num_iommus; i++, tmp++) {
1459 		np = of_parse_phandle(dev->of_node, "iommus", i);
1460 		if (!np) {
1461 			kfree(arch_data);
1462 			return -EINVAL;
1463 		}
1464 
1465 		pdev = of_find_device_by_node(np);
1466 		if (WARN_ON(!pdev)) {
1467 			of_node_put(np);
1468 			kfree(arch_data);
1469 			return -EINVAL;
1470 		}
1471 
1472 		oiommu = platform_get_drvdata(pdev);
1473 		if (!oiommu) {
1474 			of_node_put(np);
1475 			kfree(arch_data);
1476 			return -EINVAL;
1477 		}
1478 
1479 		tmp->iommu_dev = oiommu;
1480 
1481 		of_node_put(np);
1482 	}
1483 
1484 	/*
1485 	 * use the first IOMMU alone for the sysfs device linking.
1486 	 * TODO: Evaluate if a single iommu_group needs to be
1487 	 * maintained for both IOMMUs
1488 	 */
1489 	oiommu = arch_data->iommu_dev;
1490 	ret = iommu_device_link(&oiommu->iommu, dev);
1491 	if (ret) {
1492 		kfree(arch_data);
1493 		return ret;
1494 	}
1495 
1496 	dev->archdata.iommu = arch_data;
1497 
1498 	/*
1499 	 * IOMMU group initialization calls into omap_iommu_device_group, which
1500 	 * needs a valid dev->archdata.iommu pointer
1501 	 */
1502 	group = iommu_group_get_for_dev(dev);
1503 	if (IS_ERR(group)) {
1504 		iommu_device_unlink(&oiommu->iommu, dev);
1505 		dev->archdata.iommu = NULL;
1506 		kfree(arch_data);
1507 		return PTR_ERR(group);
1508 	}
1509 	iommu_group_put(group);
1510 
1511 	return 0;
1512 }
1513 
1514 static void omap_iommu_remove_device(struct device *dev)
1515 {
1516 	struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
1517 
1518 	if (!dev->of_node || !arch_data)
1519 		return;
1520 
1521 	iommu_device_unlink(&arch_data->iommu_dev->iommu, dev);
1522 	iommu_group_remove_device(dev);
1523 
1524 	dev->archdata.iommu = NULL;
1525 	kfree(arch_data);
1526 
1527 }
1528 
1529 static struct iommu_group *omap_iommu_device_group(struct device *dev)
1530 {
1531 	struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
1532 	struct iommu_group *group = ERR_PTR(-EINVAL);
1533 
1534 	if (arch_data->iommu_dev)
1535 		group = iommu_group_ref_get(arch_data->iommu_dev->group);
1536 
1537 	return group;
1538 }
1539 
1540 static const struct iommu_ops omap_iommu_ops = {
1541 	.domain_alloc	= omap_iommu_domain_alloc,
1542 	.domain_free	= omap_iommu_domain_free,
1543 	.attach_dev	= omap_iommu_attach_dev,
1544 	.detach_dev	= omap_iommu_detach_dev,
1545 	.map		= omap_iommu_map,
1546 	.unmap		= omap_iommu_unmap,
1547 	.iova_to_phys	= omap_iommu_iova_to_phys,
1548 	.add_device	= omap_iommu_add_device,
1549 	.remove_device	= omap_iommu_remove_device,
1550 	.device_group	= omap_iommu_device_group,
1551 	.pgsize_bitmap	= OMAP_IOMMU_PGSIZES,
1552 };
1553 
1554 static int __init omap_iommu_init(void)
1555 {
1556 	struct kmem_cache *p;
1557 	const unsigned long flags = SLAB_HWCACHE_ALIGN;
1558 	size_t align = 1 << 10; /* L2 pagetable alignement */
1559 	struct device_node *np;
1560 	int ret;
1561 
1562 	np = of_find_matching_node(NULL, omap_iommu_of_match);
1563 	if (!np)
1564 		return 0;
1565 
1566 	of_node_put(np);
1567 
1568 	p = kmem_cache_create("iopte_cache", IOPTE_TABLE_SIZE, align, flags,
1569 			      NULL);
1570 	if (!p)
1571 		return -ENOMEM;
1572 	iopte_cachep = p;
1573 
1574 	omap_iommu_debugfs_init();
1575 
1576 	ret = platform_driver_register(&omap_iommu_driver);
1577 	if (ret) {
1578 		pr_err("%s: failed to register driver\n", __func__);
1579 		goto fail_driver;
1580 	}
1581 
1582 	ret = bus_set_iommu(&platform_bus_type, &omap_iommu_ops);
1583 	if (ret)
1584 		goto fail_bus;
1585 
1586 	return 0;
1587 
1588 fail_bus:
1589 	platform_driver_unregister(&omap_iommu_driver);
1590 fail_driver:
1591 	kmem_cache_destroy(iopte_cachep);
1592 	return ret;
1593 }
1594 subsys_initcall(omap_iommu_init);
1595 /* must be ready before omap3isp is probed */
1596