xref: /openbmc/linux/drivers/iommu/omap-iommu.c (revision babbdf5b)
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 - https://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  * This should be treated as an deprecated API. It is preserved only
70  * to maintain existing functionality for OMAP3 ISP driver.
71  **/
72 void omap_iommu_save_ctx(struct device *dev)
73 {
74 	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
75 	struct omap_iommu *obj;
76 	u32 *p;
77 	int i;
78 
79 	if (!arch_data)
80 		return;
81 
82 	while (arch_data->iommu_dev) {
83 		obj = arch_data->iommu_dev;
84 		p = obj->ctx;
85 		for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
86 			p[i] = iommu_read_reg(obj, i * sizeof(u32));
87 			dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i,
88 				p[i]);
89 		}
90 		arch_data++;
91 	}
92 }
93 EXPORT_SYMBOL_GPL(omap_iommu_save_ctx);
94 
95 /**
96  * omap_iommu_restore_ctx - Restore registers for pm off-mode support
97  * @dev:	client device
98  *
99  * This should be treated as an deprecated API. It is preserved only
100  * to maintain existing functionality for OMAP3 ISP driver.
101  **/
102 void omap_iommu_restore_ctx(struct device *dev)
103 {
104 	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
105 	struct omap_iommu *obj;
106 	u32 *p;
107 	int i;
108 
109 	if (!arch_data)
110 		return;
111 
112 	while (arch_data->iommu_dev) {
113 		obj = arch_data->iommu_dev;
114 		p = obj->ctx;
115 		for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
116 			iommu_write_reg(obj, p[i], i * sizeof(u32));
117 			dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i,
118 				p[i]);
119 		}
120 		arch_data++;
121 	}
122 }
123 EXPORT_SYMBOL_GPL(omap_iommu_restore_ctx);
124 
125 static void dra7_cfg_dspsys_mmu(struct omap_iommu *obj, bool enable)
126 {
127 	u32 val, mask;
128 
129 	if (!obj->syscfg)
130 		return;
131 
132 	mask = (1 << (obj->id * DSP_SYS_MMU_CONFIG_EN_SHIFT));
133 	val = enable ? mask : 0;
134 	regmap_update_bits(obj->syscfg, DSP_SYS_MMU_CONFIG, mask, val);
135 }
136 
137 static void __iommu_set_twl(struct omap_iommu *obj, bool on)
138 {
139 	u32 l = iommu_read_reg(obj, MMU_CNTL);
140 
141 	if (on)
142 		iommu_write_reg(obj, MMU_IRQ_TWL_MASK, MMU_IRQENABLE);
143 	else
144 		iommu_write_reg(obj, MMU_IRQ_TLB_MISS_MASK, MMU_IRQENABLE);
145 
146 	l &= ~MMU_CNTL_MASK;
147 	if (on)
148 		l |= (MMU_CNTL_MMU_EN | MMU_CNTL_TWL_EN);
149 	else
150 		l |= (MMU_CNTL_MMU_EN);
151 
152 	iommu_write_reg(obj, l, MMU_CNTL);
153 }
154 
155 static int omap2_iommu_enable(struct omap_iommu *obj)
156 {
157 	u32 l, pa;
158 
159 	if (!obj->iopgd || !IS_ALIGNED((unsigned long)obj->iopgd,  SZ_16K))
160 		return -EINVAL;
161 
162 	pa = virt_to_phys(obj->iopgd);
163 	if (!IS_ALIGNED(pa, SZ_16K))
164 		return -EINVAL;
165 
166 	l = iommu_read_reg(obj, MMU_REVISION);
167 	dev_info(obj->dev, "%s: version %d.%d\n", obj->name,
168 		 (l >> 4) & 0xf, l & 0xf);
169 
170 	iommu_write_reg(obj, pa, MMU_TTB);
171 
172 	dra7_cfg_dspsys_mmu(obj, true);
173 
174 	if (obj->has_bus_err_back)
175 		iommu_write_reg(obj, MMU_GP_REG_BUS_ERR_BACK_EN, MMU_GP_REG);
176 
177 	__iommu_set_twl(obj, true);
178 
179 	return 0;
180 }
181 
182 static void omap2_iommu_disable(struct omap_iommu *obj)
183 {
184 	u32 l = iommu_read_reg(obj, MMU_CNTL);
185 
186 	l &= ~MMU_CNTL_MASK;
187 	iommu_write_reg(obj, l, MMU_CNTL);
188 	dra7_cfg_dspsys_mmu(obj, false);
189 
190 	dev_dbg(obj->dev, "%s is shutting down\n", obj->name);
191 }
192 
193 static int iommu_enable(struct omap_iommu *obj)
194 {
195 	int ret;
196 
197 	ret = pm_runtime_get_sync(obj->dev);
198 	if (ret < 0)
199 		pm_runtime_put_noidle(obj->dev);
200 
201 	return ret < 0 ? ret : 0;
202 }
203 
204 static void iommu_disable(struct omap_iommu *obj)
205 {
206 	pm_runtime_put_sync(obj->dev);
207 }
208 
209 /*
210  *	TLB operations
211  */
212 static u32 iotlb_cr_to_virt(struct cr_regs *cr)
213 {
214 	u32 page_size = cr->cam & MMU_CAM_PGSZ_MASK;
215 	u32 mask = get_cam_va_mask(cr->cam & page_size);
216 
217 	return cr->cam & mask;
218 }
219 
220 static u32 get_iopte_attr(struct iotlb_entry *e)
221 {
222 	u32 attr;
223 
224 	attr = e->mixed << 5;
225 	attr |= e->endian;
226 	attr |= e->elsz >> 3;
227 	attr <<= (((e->pgsz == MMU_CAM_PGSZ_4K) ||
228 			(e->pgsz == MMU_CAM_PGSZ_64K)) ? 0 : 6);
229 	return attr;
230 }
231 
232 static u32 iommu_report_fault(struct omap_iommu *obj, u32 *da)
233 {
234 	u32 status, fault_addr;
235 
236 	status = iommu_read_reg(obj, MMU_IRQSTATUS);
237 	status &= MMU_IRQ_MASK;
238 	if (!status) {
239 		*da = 0;
240 		return 0;
241 	}
242 
243 	fault_addr = iommu_read_reg(obj, MMU_FAULT_AD);
244 	*da = fault_addr;
245 
246 	iommu_write_reg(obj, status, MMU_IRQSTATUS);
247 
248 	return status;
249 }
250 
251 void iotlb_lock_get(struct omap_iommu *obj, struct iotlb_lock *l)
252 {
253 	u32 val;
254 
255 	val = iommu_read_reg(obj, MMU_LOCK);
256 
257 	l->base = MMU_LOCK_BASE(val);
258 	l->vict = MMU_LOCK_VICT(val);
259 }
260 
261 void iotlb_lock_set(struct omap_iommu *obj, struct iotlb_lock *l)
262 {
263 	u32 val;
264 
265 	val = (l->base << MMU_LOCK_BASE_SHIFT);
266 	val |= (l->vict << MMU_LOCK_VICT_SHIFT);
267 
268 	iommu_write_reg(obj, val, MMU_LOCK);
269 }
270 
271 static void iotlb_read_cr(struct omap_iommu *obj, struct cr_regs *cr)
272 {
273 	cr->cam = iommu_read_reg(obj, MMU_READ_CAM);
274 	cr->ram = iommu_read_reg(obj, MMU_READ_RAM);
275 }
276 
277 static void iotlb_load_cr(struct omap_iommu *obj, struct cr_regs *cr)
278 {
279 	iommu_write_reg(obj, cr->cam | MMU_CAM_V, MMU_CAM);
280 	iommu_write_reg(obj, cr->ram, MMU_RAM);
281 
282 	iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
283 	iommu_write_reg(obj, 1, MMU_LD_TLB);
284 }
285 
286 /* only used in iotlb iteration for-loop */
287 struct cr_regs __iotlb_read_cr(struct omap_iommu *obj, int n)
288 {
289 	struct cr_regs cr;
290 	struct iotlb_lock l;
291 
292 	iotlb_lock_get(obj, &l);
293 	l.vict = n;
294 	iotlb_lock_set(obj, &l);
295 	iotlb_read_cr(obj, &cr);
296 
297 	return cr;
298 }
299 
300 #ifdef PREFETCH_IOTLB
301 static struct cr_regs *iotlb_alloc_cr(struct omap_iommu *obj,
302 				      struct iotlb_entry *e)
303 {
304 	struct cr_regs *cr;
305 
306 	if (!e)
307 		return NULL;
308 
309 	if (e->da & ~(get_cam_va_mask(e->pgsz))) {
310 		dev_err(obj->dev, "%s:\twrong alignment: %08x\n", __func__,
311 			e->da);
312 		return ERR_PTR(-EINVAL);
313 	}
314 
315 	cr = kmalloc(sizeof(*cr), GFP_KERNEL);
316 	if (!cr)
317 		return ERR_PTR(-ENOMEM);
318 
319 	cr->cam = (e->da & MMU_CAM_VATAG_MASK) | e->prsvd | e->pgsz | e->valid;
320 	cr->ram = e->pa | e->endian | e->elsz | e->mixed;
321 
322 	return cr;
323 }
324 
325 /**
326  * load_iotlb_entry - Set an iommu tlb entry
327  * @obj:	target iommu
328  * @e:		an iommu tlb entry info
329  **/
330 static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
331 {
332 	int err = 0;
333 	struct iotlb_lock l;
334 	struct cr_regs *cr;
335 
336 	if (!obj || !obj->nr_tlb_entries || !e)
337 		return -EINVAL;
338 
339 	pm_runtime_get_sync(obj->dev);
340 
341 	iotlb_lock_get(obj, &l);
342 	if (l.base == obj->nr_tlb_entries) {
343 		dev_warn(obj->dev, "%s: preserve entries full\n", __func__);
344 		err = -EBUSY;
345 		goto out;
346 	}
347 	if (!e->prsvd) {
348 		int i;
349 		struct cr_regs tmp;
350 
351 		for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, tmp)
352 			if (!iotlb_cr_valid(&tmp))
353 				break;
354 
355 		if (i == obj->nr_tlb_entries) {
356 			dev_dbg(obj->dev, "%s: full: no entry\n", __func__);
357 			err = -EBUSY;
358 			goto out;
359 		}
360 
361 		iotlb_lock_get(obj, &l);
362 	} else {
363 		l.vict = l.base;
364 		iotlb_lock_set(obj, &l);
365 	}
366 
367 	cr = iotlb_alloc_cr(obj, e);
368 	if (IS_ERR(cr)) {
369 		pm_runtime_put_sync(obj->dev);
370 		return PTR_ERR(cr);
371 	}
372 
373 	iotlb_load_cr(obj, cr);
374 	kfree(cr);
375 
376 	if (e->prsvd)
377 		l.base++;
378 	/* increment victim for next tlb load */
379 	if (++l.vict == obj->nr_tlb_entries)
380 		l.vict = l.base;
381 	iotlb_lock_set(obj, &l);
382 out:
383 	pm_runtime_put_sync(obj->dev);
384 	return err;
385 }
386 
387 #else /* !PREFETCH_IOTLB */
388 
389 static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
390 {
391 	return 0;
392 }
393 
394 #endif /* !PREFETCH_IOTLB */
395 
396 static int prefetch_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
397 {
398 	return load_iotlb_entry(obj, e);
399 }
400 
401 /**
402  * flush_iotlb_page - Clear an iommu tlb entry
403  * @obj:	target iommu
404  * @da:		iommu device virtual address
405  *
406  * Clear an iommu tlb entry which includes 'da' address.
407  **/
408 static void flush_iotlb_page(struct omap_iommu *obj, u32 da)
409 {
410 	int i;
411 	struct cr_regs cr;
412 
413 	pm_runtime_get_sync(obj->dev);
414 
415 	for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, cr) {
416 		u32 start;
417 		size_t bytes;
418 
419 		if (!iotlb_cr_valid(&cr))
420 			continue;
421 
422 		start = iotlb_cr_to_virt(&cr);
423 		bytes = iopgsz_to_bytes(cr.cam & 3);
424 
425 		if ((start <= da) && (da < start + bytes)) {
426 			dev_dbg(obj->dev, "%s: %08x<=%08x(%zx)\n",
427 				__func__, start, da, bytes);
428 			iotlb_load_cr(obj, &cr);
429 			iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
430 			break;
431 		}
432 	}
433 	pm_runtime_put_sync(obj->dev);
434 
435 	if (i == obj->nr_tlb_entries)
436 		dev_dbg(obj->dev, "%s: no page for %08x\n", __func__, da);
437 }
438 
439 /**
440  * flush_iotlb_all - Clear all iommu tlb entries
441  * @obj:	target iommu
442  **/
443 static void flush_iotlb_all(struct omap_iommu *obj)
444 {
445 	struct iotlb_lock l;
446 
447 	pm_runtime_get_sync(obj->dev);
448 
449 	l.base = 0;
450 	l.vict = 0;
451 	iotlb_lock_set(obj, &l);
452 
453 	iommu_write_reg(obj, 1, MMU_GFLUSH);
454 
455 	pm_runtime_put_sync(obj->dev);
456 }
457 
458 /*
459  *	H/W pagetable operations
460  */
461 static void flush_iopte_range(struct device *dev, dma_addr_t dma,
462 			      unsigned long offset, int num_entries)
463 {
464 	size_t size = num_entries * sizeof(u32);
465 
466 	dma_sync_single_range_for_device(dev, dma, offset, size, DMA_TO_DEVICE);
467 }
468 
469 static void iopte_free(struct omap_iommu *obj, u32 *iopte, bool dma_valid)
470 {
471 	dma_addr_t pt_dma;
472 
473 	/* Note: freed iopte's must be clean ready for re-use */
474 	if (iopte) {
475 		if (dma_valid) {
476 			pt_dma = virt_to_phys(iopte);
477 			dma_unmap_single(obj->dev, pt_dma, IOPTE_TABLE_SIZE,
478 					 DMA_TO_DEVICE);
479 		}
480 
481 		kmem_cache_free(iopte_cachep, iopte);
482 	}
483 }
484 
485 static u32 *iopte_alloc(struct omap_iommu *obj, u32 *iopgd,
486 			dma_addr_t *pt_dma, u32 da)
487 {
488 	u32 *iopte;
489 	unsigned long offset = iopgd_index(da) * sizeof(da);
490 
491 	/* a table has already existed */
492 	if (*iopgd)
493 		goto pte_ready;
494 
495 	/*
496 	 * do the allocation outside the page table lock
497 	 */
498 	spin_unlock(&obj->page_table_lock);
499 	iopte = kmem_cache_zalloc(iopte_cachep, GFP_KERNEL);
500 	spin_lock(&obj->page_table_lock);
501 
502 	if (!*iopgd) {
503 		if (!iopte)
504 			return ERR_PTR(-ENOMEM);
505 
506 		*pt_dma = dma_map_single(obj->dev, iopte, IOPTE_TABLE_SIZE,
507 					 DMA_TO_DEVICE);
508 		if (dma_mapping_error(obj->dev, *pt_dma)) {
509 			dev_err(obj->dev, "DMA map error for L2 table\n");
510 			iopte_free(obj, iopte, false);
511 			return ERR_PTR(-ENOMEM);
512 		}
513 
514 		/*
515 		 * we rely on dma address and the physical address to be
516 		 * the same for mapping the L2 table
517 		 */
518 		if (WARN_ON(*pt_dma != virt_to_phys(iopte))) {
519 			dev_err(obj->dev, "DMA translation error for L2 table\n");
520 			dma_unmap_single(obj->dev, *pt_dma, IOPTE_TABLE_SIZE,
521 					 DMA_TO_DEVICE);
522 			iopte_free(obj, iopte, false);
523 			return ERR_PTR(-ENOMEM);
524 		}
525 
526 		*iopgd = virt_to_phys(iopte) | IOPGD_TABLE;
527 
528 		flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
529 		dev_vdbg(obj->dev, "%s: a new pte:%p\n", __func__, iopte);
530 	} else {
531 		/* We raced, free the reduniovant table */
532 		iopte_free(obj, iopte, false);
533 	}
534 
535 pte_ready:
536 	iopte = iopte_offset(iopgd, da);
537 	*pt_dma = iopgd_page_paddr(iopgd);
538 	dev_vdbg(obj->dev,
539 		 "%s: da:%08x pgd:%p *pgd:%08x pte:%p *pte:%08x\n",
540 		 __func__, da, iopgd, *iopgd, iopte, *iopte);
541 
542 	return iopte;
543 }
544 
545 static int iopgd_alloc_section(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
546 {
547 	u32 *iopgd = iopgd_offset(obj, da);
548 	unsigned long offset = iopgd_index(da) * sizeof(da);
549 
550 	if ((da | pa) & ~IOSECTION_MASK) {
551 		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
552 			__func__, da, pa, IOSECTION_SIZE);
553 		return -EINVAL;
554 	}
555 
556 	*iopgd = (pa & IOSECTION_MASK) | prot | IOPGD_SECTION;
557 	flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
558 	return 0;
559 }
560 
561 static int iopgd_alloc_super(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
562 {
563 	u32 *iopgd = iopgd_offset(obj, da);
564 	unsigned long offset = iopgd_index(da) * sizeof(da);
565 	int i;
566 
567 	if ((da | pa) & ~IOSUPER_MASK) {
568 		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
569 			__func__, da, pa, IOSUPER_SIZE);
570 		return -EINVAL;
571 	}
572 
573 	for (i = 0; i < 16; i++)
574 		*(iopgd + i) = (pa & IOSUPER_MASK) | prot | IOPGD_SUPER;
575 	flush_iopte_range(obj->dev, obj->pd_dma, offset, 16);
576 	return 0;
577 }
578 
579 static int iopte_alloc_page(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
580 {
581 	u32 *iopgd = iopgd_offset(obj, da);
582 	dma_addr_t pt_dma;
583 	u32 *iopte = iopte_alloc(obj, iopgd, &pt_dma, da);
584 	unsigned long offset = iopte_index(da) * sizeof(da);
585 
586 	if (IS_ERR(iopte))
587 		return PTR_ERR(iopte);
588 
589 	*iopte = (pa & IOPAGE_MASK) | prot | IOPTE_SMALL;
590 	flush_iopte_range(obj->dev, pt_dma, offset, 1);
591 
592 	dev_vdbg(obj->dev, "%s: da:%08x pa:%08x pte:%p *pte:%08x\n",
593 		 __func__, da, pa, iopte, *iopte);
594 
595 	return 0;
596 }
597 
598 static int iopte_alloc_large(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
599 {
600 	u32 *iopgd = iopgd_offset(obj, da);
601 	dma_addr_t pt_dma;
602 	u32 *iopte = iopte_alloc(obj, iopgd, &pt_dma, da);
603 	unsigned long offset = iopte_index(da) * sizeof(da);
604 	int i;
605 
606 	if ((da | pa) & ~IOLARGE_MASK) {
607 		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
608 			__func__, da, pa, IOLARGE_SIZE);
609 		return -EINVAL;
610 	}
611 
612 	if (IS_ERR(iopte))
613 		return PTR_ERR(iopte);
614 
615 	for (i = 0; i < 16; i++)
616 		*(iopte + i) = (pa & IOLARGE_MASK) | prot | IOPTE_LARGE;
617 	flush_iopte_range(obj->dev, pt_dma, offset, 16);
618 	return 0;
619 }
620 
621 static int
622 iopgtable_store_entry_core(struct omap_iommu *obj, struct iotlb_entry *e)
623 {
624 	int (*fn)(struct omap_iommu *, u32, u32, u32);
625 	u32 prot;
626 	int err;
627 
628 	if (!obj || !e)
629 		return -EINVAL;
630 
631 	switch (e->pgsz) {
632 	case MMU_CAM_PGSZ_16M:
633 		fn = iopgd_alloc_super;
634 		break;
635 	case MMU_CAM_PGSZ_1M:
636 		fn = iopgd_alloc_section;
637 		break;
638 	case MMU_CAM_PGSZ_64K:
639 		fn = iopte_alloc_large;
640 		break;
641 	case MMU_CAM_PGSZ_4K:
642 		fn = iopte_alloc_page;
643 		break;
644 	default:
645 		fn = NULL;
646 		break;
647 	}
648 
649 	if (WARN_ON(!fn))
650 		return -EINVAL;
651 
652 	prot = get_iopte_attr(e);
653 
654 	spin_lock(&obj->page_table_lock);
655 	err = fn(obj, e->da, e->pa, prot);
656 	spin_unlock(&obj->page_table_lock);
657 
658 	return err;
659 }
660 
661 /**
662  * omap_iopgtable_store_entry - Make an iommu pte entry
663  * @obj:	target iommu
664  * @e:		an iommu tlb entry info
665  **/
666 static int
667 omap_iopgtable_store_entry(struct omap_iommu *obj, struct iotlb_entry *e)
668 {
669 	int err;
670 
671 	flush_iotlb_page(obj, e->da);
672 	err = iopgtable_store_entry_core(obj, e);
673 	if (!err)
674 		prefetch_iotlb_entry(obj, e);
675 	return err;
676 }
677 
678 /**
679  * iopgtable_lookup_entry - Lookup an iommu pte entry
680  * @obj:	target iommu
681  * @da:		iommu device virtual address
682  * @ppgd:	iommu pgd entry pointer to be returned
683  * @ppte:	iommu pte entry pointer to be returned
684  **/
685 static void
686 iopgtable_lookup_entry(struct omap_iommu *obj, u32 da, u32 **ppgd, u32 **ppte)
687 {
688 	u32 *iopgd, *iopte = NULL;
689 
690 	iopgd = iopgd_offset(obj, da);
691 	if (!*iopgd)
692 		goto out;
693 
694 	if (iopgd_is_table(*iopgd))
695 		iopte = iopte_offset(iopgd, da);
696 out:
697 	*ppgd = iopgd;
698 	*ppte = iopte;
699 }
700 
701 static size_t iopgtable_clear_entry_core(struct omap_iommu *obj, u32 da)
702 {
703 	size_t bytes;
704 	u32 *iopgd = iopgd_offset(obj, da);
705 	int nent = 1;
706 	dma_addr_t pt_dma;
707 	unsigned long pd_offset = iopgd_index(da) * sizeof(da);
708 	unsigned long pt_offset = iopte_index(da) * sizeof(da);
709 
710 	if (!*iopgd)
711 		return 0;
712 
713 	if (iopgd_is_table(*iopgd)) {
714 		int i;
715 		u32 *iopte = iopte_offset(iopgd, da);
716 
717 		bytes = IOPTE_SIZE;
718 		if (*iopte & IOPTE_LARGE) {
719 			nent *= 16;
720 			/* rewind to the 1st entry */
721 			iopte = iopte_offset(iopgd, (da & IOLARGE_MASK));
722 		}
723 		bytes *= nent;
724 		memset(iopte, 0, nent * sizeof(*iopte));
725 		pt_dma = iopgd_page_paddr(iopgd);
726 		flush_iopte_range(obj->dev, pt_dma, pt_offset, nent);
727 
728 		/*
729 		 * do table walk to check if this table is necessary or not
730 		 */
731 		iopte = iopte_offset(iopgd, 0);
732 		for (i = 0; i < PTRS_PER_IOPTE; i++)
733 			if (iopte[i])
734 				goto out;
735 
736 		iopte_free(obj, iopte, true);
737 		nent = 1; /* for the next L1 entry */
738 	} else {
739 		bytes = IOPGD_SIZE;
740 		if ((*iopgd & IOPGD_SUPER) == IOPGD_SUPER) {
741 			nent *= 16;
742 			/* rewind to the 1st entry */
743 			iopgd = iopgd_offset(obj, (da & IOSUPER_MASK));
744 		}
745 		bytes *= nent;
746 	}
747 	memset(iopgd, 0, nent * sizeof(*iopgd));
748 	flush_iopte_range(obj->dev, obj->pd_dma, pd_offset, nent);
749 out:
750 	return bytes;
751 }
752 
753 /**
754  * iopgtable_clear_entry - Remove an iommu pte entry
755  * @obj:	target iommu
756  * @da:		iommu device virtual address
757  **/
758 static size_t iopgtable_clear_entry(struct omap_iommu *obj, u32 da)
759 {
760 	size_t bytes;
761 
762 	spin_lock(&obj->page_table_lock);
763 
764 	bytes = iopgtable_clear_entry_core(obj, da);
765 	flush_iotlb_page(obj, da);
766 
767 	spin_unlock(&obj->page_table_lock);
768 
769 	return bytes;
770 }
771 
772 static void iopgtable_clear_entry_all(struct omap_iommu *obj)
773 {
774 	unsigned long offset;
775 	int i;
776 
777 	spin_lock(&obj->page_table_lock);
778 
779 	for (i = 0; i < PTRS_PER_IOPGD; i++) {
780 		u32 da;
781 		u32 *iopgd;
782 
783 		da = i << IOPGD_SHIFT;
784 		iopgd = iopgd_offset(obj, da);
785 		offset = iopgd_index(da) * sizeof(da);
786 
787 		if (!*iopgd)
788 			continue;
789 
790 		if (iopgd_is_table(*iopgd))
791 			iopte_free(obj, iopte_offset(iopgd, 0), true);
792 
793 		*iopgd = 0;
794 		flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
795 	}
796 
797 	flush_iotlb_all(obj);
798 
799 	spin_unlock(&obj->page_table_lock);
800 }
801 
802 /*
803  *	Device IOMMU generic operations
804  */
805 static irqreturn_t iommu_fault_handler(int irq, void *data)
806 {
807 	u32 da, errs;
808 	u32 *iopgd, *iopte;
809 	struct omap_iommu *obj = data;
810 	struct iommu_domain *domain = obj->domain;
811 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
812 
813 	if (!omap_domain->dev)
814 		return IRQ_NONE;
815 
816 	errs = iommu_report_fault(obj, &da);
817 	if (errs == 0)
818 		return IRQ_HANDLED;
819 
820 	/* Fault callback or TLB/PTE Dynamic loading */
821 	if (!report_iommu_fault(domain, obj->dev, da, 0))
822 		return IRQ_HANDLED;
823 
824 	iommu_write_reg(obj, 0, MMU_IRQENABLE);
825 
826 	iopgd = iopgd_offset(obj, da);
827 
828 	if (!iopgd_is_table(*iopgd)) {
829 		dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:px%08x\n",
830 			obj->name, errs, da, iopgd, *iopgd);
831 		return IRQ_NONE;
832 	}
833 
834 	iopte = iopte_offset(iopgd, da);
835 
836 	dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:0x%08x pte:0x%p *pte:0x%08x\n",
837 		obj->name, errs, da, iopgd, *iopgd, iopte, *iopte);
838 
839 	return IRQ_NONE;
840 }
841 
842 /**
843  * omap_iommu_attach() - attach iommu device to an iommu domain
844  * @obj:	target omap iommu device
845  * @iopgd:	page table
846  **/
847 static int omap_iommu_attach(struct omap_iommu *obj, u32 *iopgd)
848 {
849 	int err;
850 
851 	spin_lock(&obj->iommu_lock);
852 
853 	obj->pd_dma = dma_map_single(obj->dev, iopgd, IOPGD_TABLE_SIZE,
854 				     DMA_TO_DEVICE);
855 	if (dma_mapping_error(obj->dev, obj->pd_dma)) {
856 		dev_err(obj->dev, "DMA map error for L1 table\n");
857 		err = -ENOMEM;
858 		goto out_err;
859 	}
860 
861 	obj->iopgd = iopgd;
862 	err = iommu_enable(obj);
863 	if (err)
864 		goto out_err;
865 	flush_iotlb_all(obj);
866 
867 	spin_unlock(&obj->iommu_lock);
868 
869 	dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
870 
871 	return 0;
872 
873 out_err:
874 	spin_unlock(&obj->iommu_lock);
875 
876 	return err;
877 }
878 
879 /**
880  * omap_iommu_detach - release iommu device
881  * @obj:	target iommu
882  **/
883 static void omap_iommu_detach(struct omap_iommu *obj)
884 {
885 	if (!obj || IS_ERR(obj))
886 		return;
887 
888 	spin_lock(&obj->iommu_lock);
889 
890 	dma_unmap_single(obj->dev, obj->pd_dma, IOPGD_TABLE_SIZE,
891 			 DMA_TO_DEVICE);
892 	obj->pd_dma = 0;
893 	obj->iopgd = NULL;
894 	iommu_disable(obj);
895 
896 	spin_unlock(&obj->iommu_lock);
897 
898 	dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
899 }
900 
901 static void omap_iommu_save_tlb_entries(struct omap_iommu *obj)
902 {
903 	struct iotlb_lock lock;
904 	struct cr_regs cr;
905 	struct cr_regs *tmp;
906 	int i;
907 
908 	/* check if there are any locked tlbs to save */
909 	iotlb_lock_get(obj, &lock);
910 	obj->num_cr_ctx = lock.base;
911 	if (!obj->num_cr_ctx)
912 		return;
913 
914 	tmp = obj->cr_ctx;
915 	for_each_iotlb_cr(obj, obj->num_cr_ctx, i, cr)
916 		* tmp++ = cr;
917 }
918 
919 static void omap_iommu_restore_tlb_entries(struct omap_iommu *obj)
920 {
921 	struct iotlb_lock l;
922 	struct cr_regs *tmp;
923 	int i;
924 
925 	/* no locked tlbs to restore */
926 	if (!obj->num_cr_ctx)
927 		return;
928 
929 	l.base = 0;
930 	tmp = obj->cr_ctx;
931 	for (i = 0; i < obj->num_cr_ctx; i++, tmp++) {
932 		l.vict = i;
933 		iotlb_lock_set(obj, &l);
934 		iotlb_load_cr(obj, tmp);
935 	}
936 	l.base = obj->num_cr_ctx;
937 	l.vict = i;
938 	iotlb_lock_set(obj, &l);
939 }
940 
941 /**
942  * omap_iommu_domain_deactivate - deactivate attached iommu devices
943  * @domain: iommu domain attached to the target iommu device
944  *
945  * This API allows the client devices of IOMMU devices to suspend
946  * the IOMMUs they control at runtime, after they are idled and
947  * suspended all activity. System Suspend will leverage the PM
948  * driver late callbacks.
949  **/
950 int omap_iommu_domain_deactivate(struct iommu_domain *domain)
951 {
952 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
953 	struct omap_iommu_device *iommu;
954 	struct omap_iommu *oiommu;
955 	int i;
956 
957 	if (!omap_domain->dev)
958 		return 0;
959 
960 	iommu = omap_domain->iommus;
961 	iommu += (omap_domain->num_iommus - 1);
962 	for (i = 0; i < omap_domain->num_iommus; i++, iommu--) {
963 		oiommu = iommu->iommu_dev;
964 		pm_runtime_put_sync(oiommu->dev);
965 	}
966 
967 	return 0;
968 }
969 EXPORT_SYMBOL_GPL(omap_iommu_domain_deactivate);
970 
971 /**
972  * omap_iommu_domain_activate - activate attached iommu devices
973  * @domain: iommu domain attached to the target iommu device
974  *
975  * This API allows the client devices of IOMMU devices to resume the
976  * IOMMUs they control at runtime, before they can resume operations.
977  * System Resume will leverage the PM driver late callbacks.
978  **/
979 int omap_iommu_domain_activate(struct iommu_domain *domain)
980 {
981 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
982 	struct omap_iommu_device *iommu;
983 	struct omap_iommu *oiommu;
984 	int i;
985 
986 	if (!omap_domain->dev)
987 		return 0;
988 
989 	iommu = omap_domain->iommus;
990 	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
991 		oiommu = iommu->iommu_dev;
992 		pm_runtime_get_sync(oiommu->dev);
993 	}
994 
995 	return 0;
996 }
997 EXPORT_SYMBOL_GPL(omap_iommu_domain_activate);
998 
999 /**
1000  * omap_iommu_runtime_suspend - disable an iommu device
1001  * @dev:	iommu device
1002  *
1003  * This function performs all that is necessary to disable an
1004  * IOMMU device, either during final detachment from a client
1005  * device, or during system/runtime suspend of the device. This
1006  * includes programming all the appropriate IOMMU registers, and
1007  * managing the associated omap_hwmod's state and the device's
1008  * reset line. This function also saves the context of any
1009  * locked TLBs if suspending.
1010  **/
1011 static __maybe_unused int omap_iommu_runtime_suspend(struct device *dev)
1012 {
1013 	struct platform_device *pdev = to_platform_device(dev);
1014 	struct iommu_platform_data *pdata = dev_get_platdata(dev);
1015 	struct omap_iommu *obj = to_iommu(dev);
1016 	int ret;
1017 
1018 	/* save the TLBs only during suspend, and not for power down */
1019 	if (obj->domain && obj->iopgd)
1020 		omap_iommu_save_tlb_entries(obj);
1021 
1022 	omap2_iommu_disable(obj);
1023 
1024 	if (pdata && pdata->device_idle)
1025 		pdata->device_idle(pdev);
1026 
1027 	if (pdata && pdata->assert_reset)
1028 		pdata->assert_reset(pdev, pdata->reset_name);
1029 
1030 	if (pdata && pdata->set_pwrdm_constraint) {
1031 		ret = pdata->set_pwrdm_constraint(pdev, false, &obj->pwrst);
1032 		if (ret) {
1033 			dev_warn(obj->dev, "pwrdm_constraint failed to be reset, status = %d\n",
1034 				 ret);
1035 		}
1036 	}
1037 
1038 	return 0;
1039 }
1040 
1041 /**
1042  * omap_iommu_runtime_resume - enable an iommu device
1043  * @dev:	iommu device
1044  *
1045  * This function performs all that is necessary to enable an
1046  * IOMMU device, either during initial attachment to a client
1047  * device, or during system/runtime resume of the device. This
1048  * includes programming all the appropriate IOMMU registers, and
1049  * managing the associated omap_hwmod's state and the device's
1050  * reset line. The function also restores any locked TLBs if
1051  * resuming after a suspend.
1052  **/
1053 static __maybe_unused int omap_iommu_runtime_resume(struct device *dev)
1054 {
1055 	struct platform_device *pdev = to_platform_device(dev);
1056 	struct iommu_platform_data *pdata = dev_get_platdata(dev);
1057 	struct omap_iommu *obj = to_iommu(dev);
1058 	int ret = 0;
1059 
1060 	if (pdata && pdata->set_pwrdm_constraint) {
1061 		ret = pdata->set_pwrdm_constraint(pdev, true, &obj->pwrst);
1062 		if (ret) {
1063 			dev_warn(obj->dev, "pwrdm_constraint failed to be set, status = %d\n",
1064 				 ret);
1065 		}
1066 	}
1067 
1068 	if (pdata && pdata->deassert_reset) {
1069 		ret = pdata->deassert_reset(pdev, pdata->reset_name);
1070 		if (ret) {
1071 			dev_err(dev, "deassert_reset failed: %d\n", ret);
1072 			return ret;
1073 		}
1074 	}
1075 
1076 	if (pdata && pdata->device_enable)
1077 		pdata->device_enable(pdev);
1078 
1079 	/* restore the TLBs only during resume, and not for power up */
1080 	if (obj->domain)
1081 		omap_iommu_restore_tlb_entries(obj);
1082 
1083 	ret = omap2_iommu_enable(obj);
1084 
1085 	return ret;
1086 }
1087 
1088 /**
1089  * omap_iommu_suspend_prepare - prepare() dev_pm_ops implementation
1090  * @dev:	iommu device
1091  *
1092  * This function performs the necessary checks to determine if the IOMMU
1093  * device needs suspending or not. The function checks if the runtime_pm
1094  * status of the device is suspended, and returns 1 in that case. This
1095  * results in the PM core to skip invoking any of the Sleep PM callbacks
1096  * (suspend, suspend_late, resume, resume_early etc).
1097  */
1098 static int omap_iommu_prepare(struct device *dev)
1099 {
1100 	if (pm_runtime_status_suspended(dev))
1101 		return 1;
1102 	return 0;
1103 }
1104 
1105 static bool omap_iommu_can_register(struct platform_device *pdev)
1106 {
1107 	struct device_node *np = pdev->dev.of_node;
1108 
1109 	if (!of_device_is_compatible(np, "ti,dra7-dsp-iommu"))
1110 		return true;
1111 
1112 	/*
1113 	 * restrict IOMMU core registration only for processor-port MDMA MMUs
1114 	 * on DRA7 DSPs
1115 	 */
1116 	if ((!strcmp(dev_name(&pdev->dev), "40d01000.mmu")) ||
1117 	    (!strcmp(dev_name(&pdev->dev), "41501000.mmu")))
1118 		return true;
1119 
1120 	return false;
1121 }
1122 
1123 static int omap_iommu_dra7_get_dsp_system_cfg(struct platform_device *pdev,
1124 					      struct omap_iommu *obj)
1125 {
1126 	struct device_node *np = pdev->dev.of_node;
1127 	int ret;
1128 
1129 	if (!of_device_is_compatible(np, "ti,dra7-dsp-iommu"))
1130 		return 0;
1131 
1132 	if (!of_property_read_bool(np, "ti,syscon-mmuconfig")) {
1133 		dev_err(&pdev->dev, "ti,syscon-mmuconfig property is missing\n");
1134 		return -EINVAL;
1135 	}
1136 
1137 	obj->syscfg =
1138 		syscon_regmap_lookup_by_phandle(np, "ti,syscon-mmuconfig");
1139 	if (IS_ERR(obj->syscfg)) {
1140 		/* can fail with -EPROBE_DEFER */
1141 		ret = PTR_ERR(obj->syscfg);
1142 		return ret;
1143 	}
1144 
1145 	if (of_property_read_u32_index(np, "ti,syscon-mmuconfig", 1,
1146 				       &obj->id)) {
1147 		dev_err(&pdev->dev, "couldn't get the IOMMU instance id within subsystem\n");
1148 		return -EINVAL;
1149 	}
1150 
1151 	if (obj->id != 0 && obj->id != 1) {
1152 		dev_err(&pdev->dev, "invalid IOMMU instance id\n");
1153 		return -EINVAL;
1154 	}
1155 
1156 	return 0;
1157 }
1158 
1159 /*
1160  *	OMAP Device MMU(IOMMU) detection
1161  */
1162 static int omap_iommu_probe(struct platform_device *pdev)
1163 {
1164 	int err = -ENODEV;
1165 	int irq;
1166 	struct omap_iommu *obj;
1167 	struct resource *res;
1168 	struct device_node *of = pdev->dev.of_node;
1169 
1170 	if (!of) {
1171 		pr_err("%s: only DT-based devices are supported\n", __func__);
1172 		return -ENODEV;
1173 	}
1174 
1175 	obj = devm_kzalloc(&pdev->dev, sizeof(*obj) + MMU_REG_SIZE, GFP_KERNEL);
1176 	if (!obj)
1177 		return -ENOMEM;
1178 
1179 	/*
1180 	 * self-manage the ordering dependencies between omap_device_enable/idle
1181 	 * and omap_device_assert/deassert_hardreset API
1182 	 */
1183 	if (pdev->dev.pm_domain) {
1184 		dev_dbg(&pdev->dev, "device pm_domain is being reset\n");
1185 		pdev->dev.pm_domain = NULL;
1186 	}
1187 
1188 	obj->name = dev_name(&pdev->dev);
1189 	obj->nr_tlb_entries = 32;
1190 	err = of_property_read_u32(of, "ti,#tlb-entries", &obj->nr_tlb_entries);
1191 	if (err && err != -EINVAL)
1192 		return err;
1193 	if (obj->nr_tlb_entries != 32 && obj->nr_tlb_entries != 8)
1194 		return -EINVAL;
1195 	if (of_find_property(of, "ti,iommu-bus-err-back", NULL))
1196 		obj->has_bus_err_back = MMU_GP_REG_BUS_ERR_BACK_EN;
1197 
1198 	obj->dev = &pdev->dev;
1199 	obj->ctx = (void *)obj + sizeof(*obj);
1200 	obj->cr_ctx = devm_kzalloc(&pdev->dev,
1201 				   sizeof(*obj->cr_ctx) * obj->nr_tlb_entries,
1202 				   GFP_KERNEL);
1203 	if (!obj->cr_ctx)
1204 		return -ENOMEM;
1205 
1206 	spin_lock_init(&obj->iommu_lock);
1207 	spin_lock_init(&obj->page_table_lock);
1208 
1209 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1210 	obj->regbase = devm_ioremap_resource(obj->dev, res);
1211 	if (IS_ERR(obj->regbase))
1212 		return PTR_ERR(obj->regbase);
1213 
1214 	err = omap_iommu_dra7_get_dsp_system_cfg(pdev, obj);
1215 	if (err)
1216 		return err;
1217 
1218 	irq = platform_get_irq(pdev, 0);
1219 	if (irq < 0)
1220 		return -ENODEV;
1221 
1222 	err = devm_request_irq(obj->dev, irq, iommu_fault_handler, IRQF_SHARED,
1223 			       dev_name(obj->dev), obj);
1224 	if (err < 0)
1225 		return err;
1226 	platform_set_drvdata(pdev, obj);
1227 
1228 	if (omap_iommu_can_register(pdev)) {
1229 		obj->group = iommu_group_alloc();
1230 		if (IS_ERR(obj->group))
1231 			return PTR_ERR(obj->group);
1232 
1233 		err = iommu_device_sysfs_add(&obj->iommu, obj->dev, NULL,
1234 					     obj->name);
1235 		if (err)
1236 			goto out_group;
1237 
1238 		err = iommu_device_register(&obj->iommu, &omap_iommu_ops, &pdev->dev);
1239 		if (err)
1240 			goto out_sysfs;
1241 	}
1242 
1243 	pm_runtime_enable(obj->dev);
1244 
1245 	omap_iommu_debugfs_add(obj);
1246 
1247 	dev_info(&pdev->dev, "%s registered\n", obj->name);
1248 
1249 	/* Re-probe bus to probe device attached to this IOMMU */
1250 	bus_iommu_probe(&platform_bus_type);
1251 
1252 	return 0;
1253 
1254 out_sysfs:
1255 	iommu_device_sysfs_remove(&obj->iommu);
1256 out_group:
1257 	iommu_group_put(obj->group);
1258 	return err;
1259 }
1260 
1261 static int omap_iommu_remove(struct platform_device *pdev)
1262 {
1263 	struct omap_iommu *obj = platform_get_drvdata(pdev);
1264 
1265 	if (obj->group) {
1266 		iommu_group_put(obj->group);
1267 		obj->group = NULL;
1268 
1269 		iommu_device_sysfs_remove(&obj->iommu);
1270 		iommu_device_unregister(&obj->iommu);
1271 	}
1272 
1273 	omap_iommu_debugfs_remove(obj);
1274 
1275 	pm_runtime_disable(obj->dev);
1276 
1277 	dev_info(&pdev->dev, "%s removed\n", obj->name);
1278 	return 0;
1279 }
1280 
1281 static const struct dev_pm_ops omap_iommu_pm_ops = {
1282 	.prepare = omap_iommu_prepare,
1283 	SET_LATE_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1284 				     pm_runtime_force_resume)
1285 	SET_RUNTIME_PM_OPS(omap_iommu_runtime_suspend,
1286 			   omap_iommu_runtime_resume, NULL)
1287 };
1288 
1289 static const struct of_device_id omap_iommu_of_match[] = {
1290 	{ .compatible = "ti,omap2-iommu" },
1291 	{ .compatible = "ti,omap4-iommu" },
1292 	{ .compatible = "ti,dra7-iommu"	},
1293 	{ .compatible = "ti,dra7-dsp-iommu" },
1294 	{},
1295 };
1296 
1297 static struct platform_driver omap_iommu_driver = {
1298 	.probe	= omap_iommu_probe,
1299 	.remove	= omap_iommu_remove,
1300 	.driver	= {
1301 		.name	= "omap-iommu",
1302 		.pm	= &omap_iommu_pm_ops,
1303 		.of_match_table = of_match_ptr(omap_iommu_of_match),
1304 	},
1305 };
1306 
1307 static u32 iotlb_init_entry(struct iotlb_entry *e, u32 da, u32 pa, int pgsz)
1308 {
1309 	memset(e, 0, sizeof(*e));
1310 
1311 	e->da		= da;
1312 	e->pa		= pa;
1313 	e->valid	= MMU_CAM_V;
1314 	e->pgsz		= pgsz;
1315 	e->endian	= MMU_RAM_ENDIAN_LITTLE;
1316 	e->elsz		= MMU_RAM_ELSZ_8;
1317 	e->mixed	= 0;
1318 
1319 	return iopgsz_to_bytes(e->pgsz);
1320 }
1321 
1322 static int omap_iommu_map(struct iommu_domain *domain, unsigned long da,
1323 			  phys_addr_t pa, size_t bytes, int prot, gfp_t gfp)
1324 {
1325 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1326 	struct device *dev = omap_domain->dev;
1327 	struct omap_iommu_device *iommu;
1328 	struct omap_iommu *oiommu;
1329 	struct iotlb_entry e;
1330 	int omap_pgsz;
1331 	u32 ret = -EINVAL;
1332 	int i;
1333 
1334 	omap_pgsz = bytes_to_iopgsz(bytes);
1335 	if (omap_pgsz < 0) {
1336 		dev_err(dev, "invalid size to map: %zu\n", bytes);
1337 		return -EINVAL;
1338 	}
1339 
1340 	dev_dbg(dev, "mapping da 0x%lx to pa %pa size 0x%zx\n", da, &pa, bytes);
1341 
1342 	iotlb_init_entry(&e, da, pa, omap_pgsz);
1343 
1344 	iommu = omap_domain->iommus;
1345 	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
1346 		oiommu = iommu->iommu_dev;
1347 		ret = omap_iopgtable_store_entry(oiommu, &e);
1348 		if (ret) {
1349 			dev_err(dev, "omap_iopgtable_store_entry failed: %d\n",
1350 				ret);
1351 			break;
1352 		}
1353 	}
1354 
1355 	if (ret) {
1356 		while (i--) {
1357 			iommu--;
1358 			oiommu = iommu->iommu_dev;
1359 			iopgtable_clear_entry(oiommu, da);
1360 		}
1361 	}
1362 
1363 	return ret;
1364 }
1365 
1366 static size_t omap_iommu_unmap(struct iommu_domain *domain, unsigned long da,
1367 			       size_t size, struct iommu_iotlb_gather *gather)
1368 {
1369 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1370 	struct device *dev = omap_domain->dev;
1371 	struct omap_iommu_device *iommu;
1372 	struct omap_iommu *oiommu;
1373 	bool error = false;
1374 	size_t bytes = 0;
1375 	int i;
1376 
1377 	dev_dbg(dev, "unmapping da 0x%lx size %zu\n", da, size);
1378 
1379 	iommu = omap_domain->iommus;
1380 	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
1381 		oiommu = iommu->iommu_dev;
1382 		bytes = iopgtable_clear_entry(oiommu, da);
1383 		if (!bytes)
1384 			error = true;
1385 	}
1386 
1387 	/*
1388 	 * simplify return - we are only checking if any of the iommus
1389 	 * reported an error, but not if all of them are unmapping the
1390 	 * same number of entries. This should not occur due to the
1391 	 * mirror programming.
1392 	 */
1393 	return error ? 0 : bytes;
1394 }
1395 
1396 static int omap_iommu_count(struct device *dev)
1397 {
1398 	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1399 	int count = 0;
1400 
1401 	while (arch_data->iommu_dev) {
1402 		count++;
1403 		arch_data++;
1404 	}
1405 
1406 	return count;
1407 }
1408 
1409 /* caller should call cleanup if this function fails */
1410 static int omap_iommu_attach_init(struct device *dev,
1411 				  struct omap_iommu_domain *odomain)
1412 {
1413 	struct omap_iommu_device *iommu;
1414 	int i;
1415 
1416 	odomain->num_iommus = omap_iommu_count(dev);
1417 	if (!odomain->num_iommus)
1418 		return -EINVAL;
1419 
1420 	odomain->iommus = kcalloc(odomain->num_iommus, sizeof(*iommu),
1421 				  GFP_ATOMIC);
1422 	if (!odomain->iommus)
1423 		return -ENOMEM;
1424 
1425 	iommu = odomain->iommus;
1426 	for (i = 0; i < odomain->num_iommus; i++, iommu++) {
1427 		iommu->pgtable = kzalloc(IOPGD_TABLE_SIZE, GFP_ATOMIC);
1428 		if (!iommu->pgtable)
1429 			return -ENOMEM;
1430 
1431 		/*
1432 		 * should never fail, but please keep this around to ensure
1433 		 * we keep the hardware happy
1434 		 */
1435 		if (WARN_ON(!IS_ALIGNED((long)iommu->pgtable,
1436 					IOPGD_TABLE_SIZE)))
1437 			return -EINVAL;
1438 	}
1439 
1440 	return 0;
1441 }
1442 
1443 static void omap_iommu_detach_fini(struct omap_iommu_domain *odomain)
1444 {
1445 	int i;
1446 	struct omap_iommu_device *iommu = odomain->iommus;
1447 
1448 	for (i = 0; iommu && i < odomain->num_iommus; i++, iommu++)
1449 		kfree(iommu->pgtable);
1450 
1451 	kfree(odomain->iommus);
1452 	odomain->num_iommus = 0;
1453 	odomain->iommus = NULL;
1454 }
1455 
1456 static int
1457 omap_iommu_attach_dev(struct iommu_domain *domain, struct device *dev)
1458 {
1459 	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1460 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1461 	struct omap_iommu_device *iommu;
1462 	struct omap_iommu *oiommu;
1463 	int ret = 0;
1464 	int i;
1465 
1466 	if (!arch_data || !arch_data->iommu_dev) {
1467 		dev_err(dev, "device doesn't have an associated iommu\n");
1468 		return -EINVAL;
1469 	}
1470 
1471 	spin_lock(&omap_domain->lock);
1472 
1473 	/* only a single client device can be attached to a domain */
1474 	if (omap_domain->dev) {
1475 		dev_err(dev, "iommu domain is already attached\n");
1476 		ret = -EBUSY;
1477 		goto out;
1478 	}
1479 
1480 	ret = omap_iommu_attach_init(dev, omap_domain);
1481 	if (ret) {
1482 		dev_err(dev, "failed to allocate required iommu data %d\n",
1483 			ret);
1484 		goto init_fail;
1485 	}
1486 
1487 	iommu = omap_domain->iommus;
1488 	for (i = 0; i < omap_domain->num_iommus; i++, iommu++, arch_data++) {
1489 		/* configure and enable the omap iommu */
1490 		oiommu = arch_data->iommu_dev;
1491 		ret = omap_iommu_attach(oiommu, iommu->pgtable);
1492 		if (ret) {
1493 			dev_err(dev, "can't get omap iommu: %d\n", ret);
1494 			goto attach_fail;
1495 		}
1496 
1497 		oiommu->domain = domain;
1498 		iommu->iommu_dev = oiommu;
1499 	}
1500 
1501 	omap_domain->dev = dev;
1502 
1503 	goto out;
1504 
1505 attach_fail:
1506 	while (i--) {
1507 		iommu--;
1508 		arch_data--;
1509 		oiommu = iommu->iommu_dev;
1510 		omap_iommu_detach(oiommu);
1511 		iommu->iommu_dev = NULL;
1512 		oiommu->domain = NULL;
1513 	}
1514 init_fail:
1515 	omap_iommu_detach_fini(omap_domain);
1516 out:
1517 	spin_unlock(&omap_domain->lock);
1518 	return ret;
1519 }
1520 
1521 static void _omap_iommu_detach_dev(struct omap_iommu_domain *omap_domain,
1522 				   struct device *dev)
1523 {
1524 	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1525 	struct omap_iommu_device *iommu = omap_domain->iommus;
1526 	struct omap_iommu *oiommu;
1527 	int i;
1528 
1529 	if (!omap_domain->dev) {
1530 		dev_err(dev, "domain has no attached device\n");
1531 		return;
1532 	}
1533 
1534 	/* only a single device is supported per domain for now */
1535 	if (omap_domain->dev != dev) {
1536 		dev_err(dev, "invalid attached device\n");
1537 		return;
1538 	}
1539 
1540 	/*
1541 	 * cleanup in the reverse order of attachment - this addresses
1542 	 * any h/w dependencies between multiple instances, if any
1543 	 */
1544 	iommu += (omap_domain->num_iommus - 1);
1545 	arch_data += (omap_domain->num_iommus - 1);
1546 	for (i = 0; i < omap_domain->num_iommus; i++, iommu--, arch_data--) {
1547 		oiommu = iommu->iommu_dev;
1548 		iopgtable_clear_entry_all(oiommu);
1549 
1550 		omap_iommu_detach(oiommu);
1551 		iommu->iommu_dev = NULL;
1552 		oiommu->domain = NULL;
1553 	}
1554 
1555 	omap_iommu_detach_fini(omap_domain);
1556 
1557 	omap_domain->dev = NULL;
1558 }
1559 
1560 static void omap_iommu_detach_dev(struct iommu_domain *domain,
1561 				  struct device *dev)
1562 {
1563 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1564 
1565 	spin_lock(&omap_domain->lock);
1566 	_omap_iommu_detach_dev(omap_domain, dev);
1567 	spin_unlock(&omap_domain->lock);
1568 }
1569 
1570 static struct iommu_domain *omap_iommu_domain_alloc(unsigned type)
1571 {
1572 	struct omap_iommu_domain *omap_domain;
1573 
1574 	if (type != IOMMU_DOMAIN_UNMANAGED)
1575 		return NULL;
1576 
1577 	omap_domain = kzalloc(sizeof(*omap_domain), GFP_KERNEL);
1578 	if (!omap_domain)
1579 		return NULL;
1580 
1581 	spin_lock_init(&omap_domain->lock);
1582 
1583 	omap_domain->domain.geometry.aperture_start = 0;
1584 	omap_domain->domain.geometry.aperture_end   = (1ULL << 32) - 1;
1585 	omap_domain->domain.geometry.force_aperture = true;
1586 
1587 	return &omap_domain->domain;
1588 }
1589 
1590 static void omap_iommu_domain_free(struct iommu_domain *domain)
1591 {
1592 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1593 
1594 	/*
1595 	 * An iommu device is still attached
1596 	 * (currently, only one device can be attached) ?
1597 	 */
1598 	if (omap_domain->dev)
1599 		_omap_iommu_detach_dev(omap_domain, omap_domain->dev);
1600 
1601 	kfree(omap_domain);
1602 }
1603 
1604 static phys_addr_t omap_iommu_iova_to_phys(struct iommu_domain *domain,
1605 					   dma_addr_t da)
1606 {
1607 	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1608 	struct omap_iommu_device *iommu = omap_domain->iommus;
1609 	struct omap_iommu *oiommu = iommu->iommu_dev;
1610 	struct device *dev = oiommu->dev;
1611 	u32 *pgd, *pte;
1612 	phys_addr_t ret = 0;
1613 
1614 	/*
1615 	 * all the iommus within the domain will have identical programming,
1616 	 * so perform the lookup using just the first iommu
1617 	 */
1618 	iopgtable_lookup_entry(oiommu, da, &pgd, &pte);
1619 
1620 	if (pte) {
1621 		if (iopte_is_small(*pte))
1622 			ret = omap_iommu_translate(*pte, da, IOPTE_MASK);
1623 		else if (iopte_is_large(*pte))
1624 			ret = omap_iommu_translate(*pte, da, IOLARGE_MASK);
1625 		else
1626 			dev_err(dev, "bogus pte 0x%x, da 0x%llx", *pte,
1627 				(unsigned long long)da);
1628 	} else {
1629 		if (iopgd_is_section(*pgd))
1630 			ret = omap_iommu_translate(*pgd, da, IOSECTION_MASK);
1631 		else if (iopgd_is_super(*pgd))
1632 			ret = omap_iommu_translate(*pgd, da, IOSUPER_MASK);
1633 		else
1634 			dev_err(dev, "bogus pgd 0x%x, da 0x%llx", *pgd,
1635 				(unsigned long long)da);
1636 	}
1637 
1638 	return ret;
1639 }
1640 
1641 static struct iommu_device *omap_iommu_probe_device(struct device *dev)
1642 {
1643 	struct omap_iommu_arch_data *arch_data, *tmp;
1644 	struct platform_device *pdev;
1645 	struct omap_iommu *oiommu;
1646 	struct device_node *np;
1647 	int num_iommus, i;
1648 
1649 	/*
1650 	 * Allocate the per-device iommu structure for DT-based devices.
1651 	 *
1652 	 * TODO: Simplify this when removing non-DT support completely from the
1653 	 * IOMMU users.
1654 	 */
1655 	if (!dev->of_node)
1656 		return ERR_PTR(-ENODEV);
1657 
1658 	/*
1659 	 * retrieve the count of IOMMU nodes using phandle size as element size
1660 	 * since #iommu-cells = 0 for OMAP
1661 	 */
1662 	num_iommus = of_property_count_elems_of_size(dev->of_node, "iommus",
1663 						     sizeof(phandle));
1664 	if (num_iommus < 0)
1665 		return 0;
1666 
1667 	arch_data = kcalloc(num_iommus + 1, sizeof(*arch_data), GFP_KERNEL);
1668 	if (!arch_data)
1669 		return ERR_PTR(-ENOMEM);
1670 
1671 	for (i = 0, tmp = arch_data; i < num_iommus; i++, tmp++) {
1672 		np = of_parse_phandle(dev->of_node, "iommus", i);
1673 		if (!np) {
1674 			kfree(arch_data);
1675 			return ERR_PTR(-EINVAL);
1676 		}
1677 
1678 		pdev = of_find_device_by_node(np);
1679 		if (!pdev) {
1680 			of_node_put(np);
1681 			kfree(arch_data);
1682 			return ERR_PTR(-ENODEV);
1683 		}
1684 
1685 		oiommu = platform_get_drvdata(pdev);
1686 		if (!oiommu) {
1687 			of_node_put(np);
1688 			kfree(arch_data);
1689 			return ERR_PTR(-EINVAL);
1690 		}
1691 
1692 		tmp->iommu_dev = oiommu;
1693 		tmp->dev = &pdev->dev;
1694 
1695 		of_node_put(np);
1696 	}
1697 
1698 	dev_iommu_priv_set(dev, arch_data);
1699 
1700 	/*
1701 	 * use the first IOMMU alone for the sysfs device linking.
1702 	 * TODO: Evaluate if a single iommu_group needs to be
1703 	 * maintained for both IOMMUs
1704 	 */
1705 	oiommu = arch_data->iommu_dev;
1706 
1707 	return &oiommu->iommu;
1708 }
1709 
1710 static void omap_iommu_release_device(struct device *dev)
1711 {
1712 	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1713 
1714 	if (!dev->of_node || !arch_data)
1715 		return;
1716 
1717 	dev_iommu_priv_set(dev, NULL);
1718 	kfree(arch_data);
1719 
1720 }
1721 
1722 static struct iommu_group *omap_iommu_device_group(struct device *dev)
1723 {
1724 	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1725 	struct iommu_group *group = ERR_PTR(-EINVAL);
1726 
1727 	if (!arch_data)
1728 		return ERR_PTR(-ENODEV);
1729 
1730 	if (arch_data->iommu_dev)
1731 		group = iommu_group_ref_get(arch_data->iommu_dev->group);
1732 
1733 	return group;
1734 }
1735 
1736 static const struct iommu_ops omap_iommu_ops = {
1737 	.domain_alloc	= omap_iommu_domain_alloc,
1738 	.domain_free	= omap_iommu_domain_free,
1739 	.attach_dev	= omap_iommu_attach_dev,
1740 	.detach_dev	= omap_iommu_detach_dev,
1741 	.map		= omap_iommu_map,
1742 	.unmap		= omap_iommu_unmap,
1743 	.iova_to_phys	= omap_iommu_iova_to_phys,
1744 	.probe_device	= omap_iommu_probe_device,
1745 	.release_device	= omap_iommu_release_device,
1746 	.device_group	= omap_iommu_device_group,
1747 	.pgsize_bitmap	= OMAP_IOMMU_PGSIZES,
1748 };
1749 
1750 static int __init omap_iommu_init(void)
1751 {
1752 	struct kmem_cache *p;
1753 	const slab_flags_t flags = SLAB_HWCACHE_ALIGN;
1754 	size_t align = 1 << 10; /* L2 pagetable alignement */
1755 	struct device_node *np;
1756 	int ret;
1757 
1758 	np = of_find_matching_node(NULL, omap_iommu_of_match);
1759 	if (!np)
1760 		return 0;
1761 
1762 	of_node_put(np);
1763 
1764 	p = kmem_cache_create("iopte_cache", IOPTE_TABLE_SIZE, align, flags,
1765 			      NULL);
1766 	if (!p)
1767 		return -ENOMEM;
1768 	iopte_cachep = p;
1769 
1770 	omap_iommu_debugfs_init();
1771 
1772 	ret = platform_driver_register(&omap_iommu_driver);
1773 	if (ret) {
1774 		pr_err("%s: failed to register driver\n", __func__);
1775 		goto fail_driver;
1776 	}
1777 
1778 	ret = bus_set_iommu(&platform_bus_type, &omap_iommu_ops);
1779 	if (ret)
1780 		goto fail_bus;
1781 
1782 	return 0;
1783 
1784 fail_bus:
1785 	platform_driver_unregister(&omap_iommu_driver);
1786 fail_driver:
1787 	kmem_cache_destroy(iopte_cachep);
1788 	return ret;
1789 }
1790 subsys_initcall(omap_iommu_init);
1791 /* must be ready before omap3isp is probed */
1792