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_irq.h>
26#include <linux/of_platform.h>
27#include <linux/regmap.h>
28#include <linux/mfd/syscon.h>
29
30#include <linux/platform_data/iommu-omap.h>
31
32#include "omap-iopgtable.h"
33#include "omap-iommu.h"
34
35static const struct iommu_ops omap_iommu_ops;
36
37#define to_iommu(dev)	((struct omap_iommu *)dev_get_drvdata(dev))
38
39/* bitmap of the page sizes currently supported */
40#define OMAP_IOMMU_PGSIZES	(SZ_4K | SZ_64K | SZ_1M | SZ_16M)
41
42#define MMU_LOCK_BASE_SHIFT	10
43#define MMU_LOCK_BASE_MASK	(0x1f << MMU_LOCK_BASE_SHIFT)
44#define MMU_LOCK_BASE(x)	\
45	((x & MMU_LOCK_BASE_MASK) >> MMU_LOCK_BASE_SHIFT)
46
47#define MMU_LOCK_VICT_SHIFT	4
48#define MMU_LOCK_VICT_MASK	(0x1f << MMU_LOCK_VICT_SHIFT)
49#define MMU_LOCK_VICT(x)	\
50	((x & MMU_LOCK_VICT_MASK) >> MMU_LOCK_VICT_SHIFT)
51
52static struct platform_driver omap_iommu_driver;
53static struct kmem_cache *iopte_cachep;
54
55/**
56 * to_omap_domain - Get struct omap_iommu_domain from generic iommu_domain
57 * @dom:	generic iommu domain handle
58 **/
59static struct omap_iommu_domain *to_omap_domain(struct iommu_domain *dom)
60{
61	return container_of(dom, struct omap_iommu_domain, domain);
62}
63
64/**
65 * omap_iommu_save_ctx - Save registers for pm off-mode support
66 * @dev:	client device
67 *
68 * This should be treated as an deprecated API. It is preserved only
69 * to maintain existing functionality for OMAP3 ISP driver.
70 **/
71void omap_iommu_save_ctx(struct device *dev)
72{
73	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
74	struct omap_iommu *obj;
75	u32 *p;
76	int i;
77
78	if (!arch_data)
79		return;
80
81	while (arch_data->iommu_dev) {
82		obj = arch_data->iommu_dev;
83		p = obj->ctx;
84		for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
85			p[i] = iommu_read_reg(obj, i * sizeof(u32));
86			dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i,
87				p[i]);
88		}
89		arch_data++;
90	}
91}
92EXPORT_SYMBOL_GPL(omap_iommu_save_ctx);
93
94/**
95 * omap_iommu_restore_ctx - Restore registers for pm off-mode support
96 * @dev:	client device
97 *
98 * This should be treated as an deprecated API. It is preserved only
99 * to maintain existing functionality for OMAP3 ISP driver.
100 **/
101void omap_iommu_restore_ctx(struct device *dev)
102{
103	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
104	struct omap_iommu *obj;
105	u32 *p;
106	int i;
107
108	if (!arch_data)
109		return;
110
111	while (arch_data->iommu_dev) {
112		obj = arch_data->iommu_dev;
113		p = obj->ctx;
114		for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
115			iommu_write_reg(obj, p[i], i * sizeof(u32));
116			dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i,
117				p[i]);
118		}
119		arch_data++;
120	}
121}
122EXPORT_SYMBOL_GPL(omap_iommu_restore_ctx);
123
124static void dra7_cfg_dspsys_mmu(struct omap_iommu *obj, bool enable)
125{
126	u32 val, mask;
127
128	if (!obj->syscfg)
129		return;
130
131	mask = (1 << (obj->id * DSP_SYS_MMU_CONFIG_EN_SHIFT));
132	val = enable ? mask : 0;
133	regmap_update_bits(obj->syscfg, DSP_SYS_MMU_CONFIG, mask, val);
134}
135
136static void __iommu_set_twl(struct omap_iommu *obj, bool on)
137{
138	u32 l = iommu_read_reg(obj, MMU_CNTL);
139
140	if (on)
141		iommu_write_reg(obj, MMU_IRQ_TWL_MASK, MMU_IRQENABLE);
142	else
143		iommu_write_reg(obj, MMU_IRQ_TLB_MISS_MASK, MMU_IRQENABLE);
144
145	l &= ~MMU_CNTL_MASK;
146	if (on)
147		l |= (MMU_CNTL_MMU_EN | MMU_CNTL_TWL_EN);
148	else
149		l |= (MMU_CNTL_MMU_EN);
150
151	iommu_write_reg(obj, l, MMU_CNTL);
152}
153
154static int omap2_iommu_enable(struct omap_iommu *obj)
155{
156	u32 l, pa;
157
158	if (!obj->iopgd || !IS_ALIGNED((unsigned long)obj->iopgd,  SZ_16K))
159		return -EINVAL;
160
161	pa = virt_to_phys(obj->iopgd);
162	if (!IS_ALIGNED(pa, SZ_16K))
163		return -EINVAL;
164
165	l = iommu_read_reg(obj, MMU_REVISION);
166	dev_info(obj->dev, "%s: version %d.%d\n", obj->name,
167		 (l >> 4) & 0xf, l & 0xf);
168
169	iommu_write_reg(obj, pa, MMU_TTB);
170
171	dra7_cfg_dspsys_mmu(obj, true);
172
173	if (obj->has_bus_err_back)
174		iommu_write_reg(obj, MMU_GP_REG_BUS_ERR_BACK_EN, MMU_GP_REG);
175
176	__iommu_set_twl(obj, true);
177
178	return 0;
179}
180
181static void omap2_iommu_disable(struct omap_iommu *obj)
182{
183	u32 l = iommu_read_reg(obj, MMU_CNTL);
184
185	l &= ~MMU_CNTL_MASK;
186	iommu_write_reg(obj, l, MMU_CNTL);
187	dra7_cfg_dspsys_mmu(obj, false);
188
189	dev_dbg(obj->dev, "%s is shutting down\n", obj->name);
190}
191
192static int iommu_enable(struct omap_iommu *obj)
193{
194	int ret;
195
196	ret = pm_runtime_get_sync(obj->dev);
197	if (ret < 0)
198		pm_runtime_put_noidle(obj->dev);
199
200	return ret < 0 ? ret : 0;
201}
202
203static void iommu_disable(struct omap_iommu *obj)
204{
205	pm_runtime_put_sync(obj->dev);
206}
207
208/*
209 *	TLB operations
210 */
211static u32 iotlb_cr_to_virt(struct cr_regs *cr)
212{
213	u32 page_size = cr->cam & MMU_CAM_PGSZ_MASK;
214	u32 mask = get_cam_va_mask(cr->cam & page_size);
215
216	return cr->cam & mask;
217}
218
219static u32 get_iopte_attr(struct iotlb_entry *e)
220{
221	u32 attr;
222
223	attr = e->mixed << 5;
224	attr |= e->endian;
225	attr |= e->elsz >> 3;
226	attr <<= (((e->pgsz == MMU_CAM_PGSZ_4K) ||
227			(e->pgsz == MMU_CAM_PGSZ_64K)) ? 0 : 6);
228	return attr;
229}
230
231static u32 iommu_report_fault(struct omap_iommu *obj, u32 *da)
232{
233	u32 status, fault_addr;
234
235	status = iommu_read_reg(obj, MMU_IRQSTATUS);
236	status &= MMU_IRQ_MASK;
237	if (!status) {
238		*da = 0;
239		return 0;
240	}
241
242	fault_addr = iommu_read_reg(obj, MMU_FAULT_AD);
243	*da = fault_addr;
244
245	iommu_write_reg(obj, status, MMU_IRQSTATUS);
246
247	return status;
248}
249
250void iotlb_lock_get(struct omap_iommu *obj, struct iotlb_lock *l)
251{
252	u32 val;
253
254	val = iommu_read_reg(obj, MMU_LOCK);
255
256	l->base = MMU_LOCK_BASE(val);
257	l->vict = MMU_LOCK_VICT(val);
258}
259
260void iotlb_lock_set(struct omap_iommu *obj, struct iotlb_lock *l)
261{
262	u32 val;
263
264	val = (l->base << MMU_LOCK_BASE_SHIFT);
265	val |= (l->vict << MMU_LOCK_VICT_SHIFT);
266
267	iommu_write_reg(obj, val, MMU_LOCK);
268}
269
270static void iotlb_read_cr(struct omap_iommu *obj, struct cr_regs *cr)
271{
272	cr->cam = iommu_read_reg(obj, MMU_READ_CAM);
273	cr->ram = iommu_read_reg(obj, MMU_READ_RAM);
274}
275
276static void iotlb_load_cr(struct omap_iommu *obj, struct cr_regs *cr)
277{
278	iommu_write_reg(obj, cr->cam | MMU_CAM_V, MMU_CAM);
279	iommu_write_reg(obj, cr->ram, MMU_RAM);
280
281	iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
282	iommu_write_reg(obj, 1, MMU_LD_TLB);
283}
284
285/* only used in iotlb iteration for-loop */
286struct cr_regs __iotlb_read_cr(struct omap_iommu *obj, int n)
287{
288	struct cr_regs cr;
289	struct iotlb_lock l;
290
291	iotlb_lock_get(obj, &l);
292	l.vict = n;
293	iotlb_lock_set(obj, &l);
294	iotlb_read_cr(obj, &cr);
295
296	return cr;
297}
298
299#ifdef PREFETCH_IOTLB
300static struct cr_regs *iotlb_alloc_cr(struct omap_iommu *obj,
301				      struct iotlb_entry *e)
302{
303	struct cr_regs *cr;
304
305	if (!e)
306		return NULL;
307
308	if (e->da & ~(get_cam_va_mask(e->pgsz))) {
309		dev_err(obj->dev, "%s:\twrong alignment: %08x\n", __func__,
310			e->da);
311		return ERR_PTR(-EINVAL);
312	}
313
314	cr = kmalloc(sizeof(*cr), GFP_KERNEL);
315	if (!cr)
316		return ERR_PTR(-ENOMEM);
317
318	cr->cam = (e->da & MMU_CAM_VATAG_MASK) | e->prsvd | e->pgsz | e->valid;
319	cr->ram = e->pa | e->endian | e->elsz | e->mixed;
320
321	return cr;
322}
323
324/**
325 * load_iotlb_entry - Set an iommu tlb entry
326 * @obj:	target iommu
327 * @e:		an iommu tlb entry info
328 **/
329static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
330{
331	int err = 0;
332	struct iotlb_lock l;
333	struct cr_regs *cr;
334
335	if (!obj || !obj->nr_tlb_entries || !e)
336		return -EINVAL;
337
338	pm_runtime_get_sync(obj->dev);
339
340	iotlb_lock_get(obj, &l);
341	if (l.base == obj->nr_tlb_entries) {
342		dev_warn(obj->dev, "%s: preserve entries full\n", __func__);
343		err = -EBUSY;
344		goto out;
345	}
346	if (!e->prsvd) {
347		int i;
348		struct cr_regs tmp;
349
350		for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, tmp)
351			if (!iotlb_cr_valid(&tmp))
352				break;
353
354		if (i == obj->nr_tlb_entries) {
355			dev_dbg(obj->dev, "%s: full: no entry\n", __func__);
356			err = -EBUSY;
357			goto out;
358		}
359
360		iotlb_lock_get(obj, &l);
361	} else {
362		l.vict = l.base;
363		iotlb_lock_set(obj, &l);
364	}
365
366	cr = iotlb_alloc_cr(obj, e);
367	if (IS_ERR(cr)) {
368		pm_runtime_put_sync(obj->dev);
369		return PTR_ERR(cr);
370	}
371
372	iotlb_load_cr(obj, cr);
373	kfree(cr);
374
375	if (e->prsvd)
376		l.base++;
377	/* increment victim for next tlb load */
378	if (++l.vict == obj->nr_tlb_entries)
379		l.vict = l.base;
380	iotlb_lock_set(obj, &l);
381out:
382	pm_runtime_put_sync(obj->dev);
383	return err;
384}
385
386#else /* !PREFETCH_IOTLB */
387
388static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
389{
390	return 0;
391}
392
393#endif /* !PREFETCH_IOTLB */
394
395static int prefetch_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
396{
397	return load_iotlb_entry(obj, e);
398}
399
400/**
401 * flush_iotlb_page - Clear an iommu tlb entry
402 * @obj:	target iommu
403 * @da:		iommu device virtual address
404 *
405 * Clear an iommu tlb entry which includes 'da' address.
406 **/
407static void flush_iotlb_page(struct omap_iommu *obj, u32 da)
408{
409	int i;
410	struct cr_regs cr;
411
412	pm_runtime_get_sync(obj->dev);
413
414	for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, cr) {
415		u32 start;
416		size_t bytes;
417
418		if (!iotlb_cr_valid(&cr))
419			continue;
420
421		start = iotlb_cr_to_virt(&cr);
422		bytes = iopgsz_to_bytes(cr.cam & 3);
423
424		if ((start <= da) && (da < start + bytes)) {
425			dev_dbg(obj->dev, "%s: %08x<=%08x(%zx)\n",
426				__func__, start, da, bytes);
427			iotlb_load_cr(obj, &cr);
428			iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
429			break;
430		}
431	}
432	pm_runtime_put_sync(obj->dev);
433
434	if (i == obj->nr_tlb_entries)
435		dev_dbg(obj->dev, "%s: no page for %08x\n", __func__, da);
436}
437
438/**
439 * flush_iotlb_all - Clear all iommu tlb entries
440 * @obj:	target iommu
441 **/
442static void flush_iotlb_all(struct omap_iommu *obj)
443{
444	struct iotlb_lock l;
445
446	pm_runtime_get_sync(obj->dev);
447
448	l.base = 0;
449	l.vict = 0;
450	iotlb_lock_set(obj, &l);
451
452	iommu_write_reg(obj, 1, MMU_GFLUSH);
453
454	pm_runtime_put_sync(obj->dev);
455}
456
457/*
458 *	H/W pagetable operations
459 */
460static void flush_iopte_range(struct device *dev, dma_addr_t dma,
461			      unsigned long offset, int num_entries)
462{
463	size_t size = num_entries * sizeof(u32);
464
465	dma_sync_single_range_for_device(dev, dma, offset, size, DMA_TO_DEVICE);
466}
467
468static void iopte_free(struct omap_iommu *obj, u32 *iopte, bool dma_valid)
469{
470	dma_addr_t pt_dma;
471
472	/* Note: freed iopte's must be clean ready for re-use */
473	if (iopte) {
474		if (dma_valid) {
475			pt_dma = virt_to_phys(iopte);
476			dma_unmap_single(obj->dev, pt_dma, IOPTE_TABLE_SIZE,
477					 DMA_TO_DEVICE);
478		}
479
480		kmem_cache_free(iopte_cachep, iopte);
481	}
482}
483
484static u32 *iopte_alloc(struct omap_iommu *obj, u32 *iopgd,
485			dma_addr_t *pt_dma, u32 da)
486{
487	u32 *iopte;
488	unsigned long offset = iopgd_index(da) * sizeof(da);
489
490	/* a table has already existed */
491	if (*iopgd)
492		goto pte_ready;
493
494	/*
495	 * do the allocation outside the page table lock
496	 */
497	spin_unlock(&obj->page_table_lock);
498	iopte = kmem_cache_zalloc(iopte_cachep, GFP_KERNEL);
499	spin_lock(&obj->page_table_lock);
500
501	if (!*iopgd) {
502		if (!iopte)
503			return ERR_PTR(-ENOMEM);
504
505		*pt_dma = dma_map_single(obj->dev, iopte, IOPTE_TABLE_SIZE,
506					 DMA_TO_DEVICE);
507		if (dma_mapping_error(obj->dev, *pt_dma)) {
508			dev_err(obj->dev, "DMA map error for L2 table\n");
509			iopte_free(obj, iopte, false);
510			return ERR_PTR(-ENOMEM);
511		}
512
513		/*
514		 * we rely on dma address and the physical address to be
515		 * the same for mapping the L2 table
516		 */
517		if (WARN_ON(*pt_dma != virt_to_phys(iopte))) {
518			dev_err(obj->dev, "DMA translation error for L2 table\n");
519			dma_unmap_single(obj->dev, *pt_dma, IOPTE_TABLE_SIZE,
520					 DMA_TO_DEVICE);
521			iopte_free(obj, iopte, false);
522			return ERR_PTR(-ENOMEM);
523		}
524
525		*iopgd = virt_to_phys(iopte) | IOPGD_TABLE;
526
527		flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
528		dev_vdbg(obj->dev, "%s: a new pte:%p\n", __func__, iopte);
529	} else {
530		/* We raced, free the reduniovant table */
531		iopte_free(obj, iopte, false);
532	}
533
534pte_ready:
535	iopte = iopte_offset(iopgd, da);
536	*pt_dma = iopgd_page_paddr(iopgd);
537	dev_vdbg(obj->dev,
538		 "%s: da:%08x pgd:%p *pgd:%08x pte:%p *pte:%08x\n",
539		 __func__, da, iopgd, *iopgd, iopte, *iopte);
540
541	return iopte;
542}
543
544static int iopgd_alloc_section(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
545{
546	u32 *iopgd = iopgd_offset(obj, da);
547	unsigned long offset = iopgd_index(da) * sizeof(da);
548
549	if ((da | pa) & ~IOSECTION_MASK) {
550		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
551			__func__, da, pa, IOSECTION_SIZE);
552		return -EINVAL;
553	}
554
555	*iopgd = (pa & IOSECTION_MASK) | prot | IOPGD_SECTION;
556	flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
557	return 0;
558}
559
560static int iopgd_alloc_super(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
561{
562	u32 *iopgd = iopgd_offset(obj, da);
563	unsigned long offset = iopgd_index(da) * sizeof(da);
564	int i;
565
566	if ((da | pa) & ~IOSUPER_MASK) {
567		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
568			__func__, da, pa, IOSUPER_SIZE);
569		return -EINVAL;
570	}
571
572	for (i = 0; i < 16; i++)
573		*(iopgd + i) = (pa & IOSUPER_MASK) | prot | IOPGD_SUPER;
574	flush_iopte_range(obj->dev, obj->pd_dma, offset, 16);
575	return 0;
576}
577
578static int iopte_alloc_page(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
579{
580	u32 *iopgd = iopgd_offset(obj, da);
581	dma_addr_t pt_dma;
582	u32 *iopte = iopte_alloc(obj, iopgd, &pt_dma, da);
583	unsigned long offset = iopte_index(da) * sizeof(da);
584
585	if (IS_ERR(iopte))
586		return PTR_ERR(iopte);
587
588	*iopte = (pa & IOPAGE_MASK) | prot | IOPTE_SMALL;
589	flush_iopte_range(obj->dev, pt_dma, offset, 1);
590
591	dev_vdbg(obj->dev, "%s: da:%08x pa:%08x pte:%p *pte:%08x\n",
592		 __func__, da, pa, iopte, *iopte);
593
594	return 0;
595}
596
597static int iopte_alloc_large(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
598{
599	u32 *iopgd = iopgd_offset(obj, da);
600	dma_addr_t pt_dma;
601	u32 *iopte = iopte_alloc(obj, iopgd, &pt_dma, da);
602	unsigned long offset = iopte_index(da) * sizeof(da);
603	int i;
604
605	if ((da | pa) & ~IOLARGE_MASK) {
606		dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
607			__func__, da, pa, IOLARGE_SIZE);
608		return -EINVAL;
609	}
610
611	if (IS_ERR(iopte))
612		return PTR_ERR(iopte);
613
614	for (i = 0; i < 16; i++)
615		*(iopte + i) = (pa & IOLARGE_MASK) | prot | IOPTE_LARGE;
616	flush_iopte_range(obj->dev, pt_dma, offset, 16);
617	return 0;
618}
619
620static int
621iopgtable_store_entry_core(struct omap_iommu *obj, struct iotlb_entry *e)
622{
623	int (*fn)(struct omap_iommu *, u32, u32, u32);
624	u32 prot;
625	int err;
626
627	if (!obj || !e)
628		return -EINVAL;
629
630	switch (e->pgsz) {
631	case MMU_CAM_PGSZ_16M:
632		fn = iopgd_alloc_super;
633		break;
634	case MMU_CAM_PGSZ_1M:
635		fn = iopgd_alloc_section;
636		break;
637	case MMU_CAM_PGSZ_64K:
638		fn = iopte_alloc_large;
639		break;
640	case MMU_CAM_PGSZ_4K:
641		fn = iopte_alloc_page;
642		break;
643	default:
644		fn = NULL;
645		break;
646	}
647
648	if (WARN_ON(!fn))
649		return -EINVAL;
650
651	prot = get_iopte_attr(e);
652
653	spin_lock(&obj->page_table_lock);
654	err = fn(obj, e->da, e->pa, prot);
655	spin_unlock(&obj->page_table_lock);
656
657	return err;
658}
659
660/**
661 * omap_iopgtable_store_entry - Make an iommu pte entry
662 * @obj:	target iommu
663 * @e:		an iommu tlb entry info
664 **/
665static int
666omap_iopgtable_store_entry(struct omap_iommu *obj, struct iotlb_entry *e)
667{
668	int err;
669
670	flush_iotlb_page(obj, e->da);
671	err = iopgtable_store_entry_core(obj, e);
672	if (!err)
673		prefetch_iotlb_entry(obj, e);
674	return err;
675}
676
677/**
678 * iopgtable_lookup_entry - Lookup an iommu pte entry
679 * @obj:	target iommu
680 * @da:		iommu device virtual address
681 * @ppgd:	iommu pgd entry pointer to be returned
682 * @ppte:	iommu pte entry pointer to be returned
683 **/
684static void
685iopgtable_lookup_entry(struct omap_iommu *obj, u32 da, u32 **ppgd, u32 **ppte)
686{
687	u32 *iopgd, *iopte = NULL;
688
689	iopgd = iopgd_offset(obj, da);
690	if (!*iopgd)
691		goto out;
692
693	if (iopgd_is_table(*iopgd))
694		iopte = iopte_offset(iopgd, da);
695out:
696	*ppgd = iopgd;
697	*ppte = iopte;
698}
699
700static size_t iopgtable_clear_entry_core(struct omap_iommu *obj, u32 da)
701{
702	size_t bytes;
703	u32 *iopgd = iopgd_offset(obj, da);
704	int nent = 1;
705	dma_addr_t pt_dma;
706	unsigned long pd_offset = iopgd_index(da) * sizeof(da);
707	unsigned long pt_offset = iopte_index(da) * sizeof(da);
708
709	if (!*iopgd)
710		return 0;
711
712	if (iopgd_is_table(*iopgd)) {
713		int i;
714		u32 *iopte = iopte_offset(iopgd, da);
715
716		bytes = IOPTE_SIZE;
717		if (*iopte & IOPTE_LARGE) {
718			nent *= 16;
719			/* rewind to the 1st entry */
720			iopte = iopte_offset(iopgd, (da & IOLARGE_MASK));
721		}
722		bytes *= nent;
723		memset(iopte, 0, nent * sizeof(*iopte));
724		pt_dma = iopgd_page_paddr(iopgd);
725		flush_iopte_range(obj->dev, pt_dma, pt_offset, nent);
726
727		/*
728		 * do table walk to check if this table is necessary or not
729		 */
730		iopte = iopte_offset(iopgd, 0);
731		for (i = 0; i < PTRS_PER_IOPTE; i++)
732			if (iopte[i])
733				goto out;
734
735		iopte_free(obj, iopte, true);
736		nent = 1; /* for the next L1 entry */
737	} else {
738		bytes = IOPGD_SIZE;
739		if ((*iopgd & IOPGD_SUPER) == IOPGD_SUPER) {
740			nent *= 16;
741			/* rewind to the 1st entry */
742			iopgd = iopgd_offset(obj, (da & IOSUPER_MASK));
743		}
744		bytes *= nent;
745	}
746	memset(iopgd, 0, nent * sizeof(*iopgd));
747	flush_iopte_range(obj->dev, obj->pd_dma, pd_offset, nent);
748out:
749	return bytes;
750}
751
752/**
753 * iopgtable_clear_entry - Remove an iommu pte entry
754 * @obj:	target iommu
755 * @da:		iommu device virtual address
756 **/
757static size_t iopgtable_clear_entry(struct omap_iommu *obj, u32 da)
758{
759	size_t bytes;
760
761	spin_lock(&obj->page_table_lock);
762
763	bytes = iopgtable_clear_entry_core(obj, da);
764	flush_iotlb_page(obj, da);
765
766	spin_unlock(&obj->page_table_lock);
767
768	return bytes;
769}
770
771static void iopgtable_clear_entry_all(struct omap_iommu *obj)
772{
773	unsigned long offset;
774	int i;
775
776	spin_lock(&obj->page_table_lock);
777
778	for (i = 0; i < PTRS_PER_IOPGD; i++) {
779		u32 da;
780		u32 *iopgd;
781
782		da = i << IOPGD_SHIFT;
783		iopgd = iopgd_offset(obj, da);
784		offset = iopgd_index(da) * sizeof(da);
785
786		if (!*iopgd)
787			continue;
788
789		if (iopgd_is_table(*iopgd))
790			iopte_free(obj, iopte_offset(iopgd, 0), true);
791
792		*iopgd = 0;
793		flush_iopte_range(obj->dev, obj->pd_dma, offset, 1);
794	}
795
796	flush_iotlb_all(obj);
797
798	spin_unlock(&obj->page_table_lock);
799}
800
801/*
802 *	Device IOMMU generic operations
803 */
804static irqreturn_t iommu_fault_handler(int irq, void *data)
805{
806	u32 da, errs;
807	u32 *iopgd, *iopte;
808	struct omap_iommu *obj = data;
809	struct iommu_domain *domain = obj->domain;
810	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
811
812	if (!omap_domain->dev)
813		return IRQ_NONE;
814
815	errs = iommu_report_fault(obj, &da);
816	if (errs == 0)
817		return IRQ_HANDLED;
818
819	/* Fault callback or TLB/PTE Dynamic loading */
820	if (!report_iommu_fault(domain, obj->dev, da, 0))
821		return IRQ_HANDLED;
822
823	iommu_write_reg(obj, 0, MMU_IRQENABLE);
824
825	iopgd = iopgd_offset(obj, da);
826
827	if (!iopgd_is_table(*iopgd)) {
828		dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:px%08x\n",
829			obj->name, errs, da, iopgd, *iopgd);
830		return IRQ_NONE;
831	}
832
833	iopte = iopte_offset(iopgd, da);
834
835	dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:0x%08x pte:0x%p *pte:0x%08x\n",
836		obj->name, errs, da, iopgd, *iopgd, iopte, *iopte);
837
838	return IRQ_NONE;
839}
840
841/**
842 * omap_iommu_attach() - attach iommu device to an iommu domain
843 * @obj:	target omap iommu device
844 * @iopgd:	page table
845 **/
846static int omap_iommu_attach(struct omap_iommu *obj, u32 *iopgd)
847{
848	int err;
849
850	spin_lock(&obj->iommu_lock);
851
852	obj->pd_dma = dma_map_single(obj->dev, iopgd, IOPGD_TABLE_SIZE,
853				     DMA_TO_DEVICE);
854	if (dma_mapping_error(obj->dev, obj->pd_dma)) {
855		dev_err(obj->dev, "DMA map error for L1 table\n");
856		err = -ENOMEM;
857		goto out_err;
858	}
859
860	obj->iopgd = iopgd;
861	err = iommu_enable(obj);
862	if (err)
863		goto out_err;
864	flush_iotlb_all(obj);
865
866	spin_unlock(&obj->iommu_lock);
867
868	dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
869
870	return 0;
871
872out_err:
873	spin_unlock(&obj->iommu_lock);
874
875	return err;
876}
877
878/**
879 * omap_iommu_detach - release iommu device
880 * @obj:	target iommu
881 **/
882static void omap_iommu_detach(struct omap_iommu *obj)
883{
884	if (!obj || IS_ERR(obj))
885		return;
886
887	spin_lock(&obj->iommu_lock);
888
889	dma_unmap_single(obj->dev, obj->pd_dma, IOPGD_TABLE_SIZE,
890			 DMA_TO_DEVICE);
891	obj->pd_dma = 0;
892	obj->iopgd = NULL;
893	iommu_disable(obj);
894
895	spin_unlock(&obj->iommu_lock);
896
897	dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
898}
899
900static void omap_iommu_save_tlb_entries(struct omap_iommu *obj)
901{
902	struct iotlb_lock lock;
903	struct cr_regs cr;
904	struct cr_regs *tmp;
905	int i;
906
907	/* check if there are any locked tlbs to save */
908	iotlb_lock_get(obj, &lock);
909	obj->num_cr_ctx = lock.base;
910	if (!obj->num_cr_ctx)
911		return;
912
913	tmp = obj->cr_ctx;
914	for_each_iotlb_cr(obj, obj->num_cr_ctx, i, cr)
915		* tmp++ = cr;
916}
917
918static void omap_iommu_restore_tlb_entries(struct omap_iommu *obj)
919{
920	struct iotlb_lock l;
921	struct cr_regs *tmp;
922	int i;
923
924	/* no locked tlbs to restore */
925	if (!obj->num_cr_ctx)
926		return;
927
928	l.base = 0;
929	tmp = obj->cr_ctx;
930	for (i = 0; i < obj->num_cr_ctx; i++, tmp++) {
931		l.vict = i;
932		iotlb_lock_set(obj, &l);
933		iotlb_load_cr(obj, tmp);
934	}
935	l.base = obj->num_cr_ctx;
936	l.vict = i;
937	iotlb_lock_set(obj, &l);
938}
939
940/**
941 * omap_iommu_domain_deactivate - deactivate attached iommu devices
942 * @domain: iommu domain attached to the target iommu device
943 *
944 * This API allows the client devices of IOMMU devices to suspend
945 * the IOMMUs they control at runtime, after they are idled and
946 * suspended all activity. System Suspend will leverage the PM
947 * driver late callbacks.
948 **/
949int omap_iommu_domain_deactivate(struct iommu_domain *domain)
950{
951	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
952	struct omap_iommu_device *iommu;
953	struct omap_iommu *oiommu;
954	int i;
955
956	if (!omap_domain->dev)
957		return 0;
958
959	iommu = omap_domain->iommus;
960	iommu += (omap_domain->num_iommus - 1);
961	for (i = 0; i < omap_domain->num_iommus; i++, iommu--) {
962		oiommu = iommu->iommu_dev;
963		pm_runtime_put_sync(oiommu->dev);
964	}
965
966	return 0;
967}
968EXPORT_SYMBOL_GPL(omap_iommu_domain_deactivate);
969
970/**
971 * omap_iommu_domain_activate - activate attached iommu devices
972 * @domain: iommu domain attached to the target iommu device
973 *
974 * This API allows the client devices of IOMMU devices to resume the
975 * IOMMUs they control at runtime, before they can resume operations.
976 * System Resume will leverage the PM driver late callbacks.
977 **/
978int omap_iommu_domain_activate(struct iommu_domain *domain)
979{
980	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
981	struct omap_iommu_device *iommu;
982	struct omap_iommu *oiommu;
983	int i;
984
985	if (!omap_domain->dev)
986		return 0;
987
988	iommu = omap_domain->iommus;
989	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
990		oiommu = iommu->iommu_dev;
991		pm_runtime_get_sync(oiommu->dev);
992	}
993
994	return 0;
995}
996EXPORT_SYMBOL_GPL(omap_iommu_domain_activate);
997
998/**
999 * omap_iommu_runtime_suspend - disable an iommu device
1000 * @dev:	iommu device
1001 *
1002 * This function performs all that is necessary to disable an
1003 * IOMMU device, either during final detachment from a client
1004 * device, or during system/runtime suspend of the device. This
1005 * includes programming all the appropriate IOMMU registers, and
1006 * managing the associated omap_hwmod's state and the device's
1007 * reset line. This function also saves the context of any
1008 * locked TLBs if suspending.
1009 **/
1010static __maybe_unused int omap_iommu_runtime_suspend(struct device *dev)
1011{
1012	struct platform_device *pdev = to_platform_device(dev);
1013	struct iommu_platform_data *pdata = dev_get_platdata(dev);
1014	struct omap_iommu *obj = to_iommu(dev);
1015	int ret;
1016
1017	/* save the TLBs only during suspend, and not for power down */
1018	if (obj->domain && obj->iopgd)
1019		omap_iommu_save_tlb_entries(obj);
1020
1021	omap2_iommu_disable(obj);
1022
1023	if (pdata && pdata->device_idle)
1024		pdata->device_idle(pdev);
1025
1026	if (pdata && pdata->assert_reset)
1027		pdata->assert_reset(pdev, pdata->reset_name);
1028
1029	if (pdata && pdata->set_pwrdm_constraint) {
1030		ret = pdata->set_pwrdm_constraint(pdev, false, &obj->pwrst);
1031		if (ret) {
1032			dev_warn(obj->dev, "pwrdm_constraint failed to be reset, status = %d\n",
1033				 ret);
1034		}
1035	}
1036
1037	return 0;
1038}
1039
1040/**
1041 * omap_iommu_runtime_resume - enable an iommu device
1042 * @dev:	iommu device
1043 *
1044 * This function performs all that is necessary to enable an
1045 * IOMMU device, either during initial attachment to a client
1046 * device, or during system/runtime resume of the device. This
1047 * includes programming all the appropriate IOMMU registers, and
1048 * managing the associated omap_hwmod's state and the device's
1049 * reset line. The function also restores any locked TLBs if
1050 * resuming after a suspend.
1051 **/
1052static __maybe_unused int omap_iommu_runtime_resume(struct device *dev)
1053{
1054	struct platform_device *pdev = to_platform_device(dev);
1055	struct iommu_platform_data *pdata = dev_get_platdata(dev);
1056	struct omap_iommu *obj = to_iommu(dev);
1057	int ret = 0;
1058
1059	if (pdata && pdata->set_pwrdm_constraint) {
1060		ret = pdata->set_pwrdm_constraint(pdev, true, &obj->pwrst);
1061		if (ret) {
1062			dev_warn(obj->dev, "pwrdm_constraint failed to be set, status = %d\n",
1063				 ret);
1064		}
1065	}
1066
1067	if (pdata && pdata->deassert_reset) {
1068		ret = pdata->deassert_reset(pdev, pdata->reset_name);
1069		if (ret) {
1070			dev_err(dev, "deassert_reset failed: %d\n", ret);
1071			return ret;
1072		}
1073	}
1074
1075	if (pdata && pdata->device_enable)
1076		pdata->device_enable(pdev);
1077
1078	/* restore the TLBs only during resume, and not for power up */
1079	if (obj->domain)
1080		omap_iommu_restore_tlb_entries(obj);
1081
1082	ret = omap2_iommu_enable(obj);
1083
1084	return ret;
1085}
1086
1087/**
1088 * omap_iommu_prepare - prepare() dev_pm_ops implementation
1089 * @dev:	iommu device
1090 *
1091 * This function performs the necessary checks to determine if the IOMMU
1092 * device needs suspending or not. The function checks if the runtime_pm
1093 * status of the device is suspended, and returns 1 in that case. This
1094 * results in the PM core to skip invoking any of the Sleep PM callbacks
1095 * (suspend, suspend_late, resume, resume_early etc).
1096 */
1097static int omap_iommu_prepare(struct device *dev)
1098{
1099	if (pm_runtime_status_suspended(dev))
1100		return 1;
1101	return 0;
1102}
1103
1104static bool omap_iommu_can_register(struct platform_device *pdev)
1105{
1106	struct device_node *np = pdev->dev.of_node;
1107
1108	if (!of_device_is_compatible(np, "ti,dra7-dsp-iommu"))
1109		return true;
1110
1111	/*
1112	 * restrict IOMMU core registration only for processor-port MDMA MMUs
1113	 * on DRA7 DSPs
1114	 */
1115	if ((!strcmp(dev_name(&pdev->dev), "40d01000.mmu")) ||
1116	    (!strcmp(dev_name(&pdev->dev), "41501000.mmu")))
1117		return true;
1118
1119	return false;
1120}
1121
1122static int omap_iommu_dra7_get_dsp_system_cfg(struct platform_device *pdev,
1123					      struct omap_iommu *obj)
1124{
1125	struct device_node *np = pdev->dev.of_node;
1126	int ret;
1127
1128	if (!of_device_is_compatible(np, "ti,dra7-dsp-iommu"))
1129		return 0;
1130
1131	if (!of_property_read_bool(np, "ti,syscon-mmuconfig")) {
1132		dev_err(&pdev->dev, "ti,syscon-mmuconfig property is missing\n");
1133		return -EINVAL;
1134	}
1135
1136	obj->syscfg =
1137		syscon_regmap_lookup_by_phandle(np, "ti,syscon-mmuconfig");
1138	if (IS_ERR(obj->syscfg)) {
1139		/* can fail with -EPROBE_DEFER */
1140		ret = PTR_ERR(obj->syscfg);
1141		return ret;
1142	}
1143
1144	if (of_property_read_u32_index(np, "ti,syscon-mmuconfig", 1,
1145				       &obj->id)) {
1146		dev_err(&pdev->dev, "couldn't get the IOMMU instance id within subsystem\n");
1147		return -EINVAL;
1148	}
1149
1150	if (obj->id != 0 && obj->id != 1) {
1151		dev_err(&pdev->dev, "invalid IOMMU instance id\n");
1152		return -EINVAL;
1153	}
1154
1155	return 0;
1156}
1157
1158/*
1159 *	OMAP Device MMU(IOMMU) detection
1160 */
1161static int omap_iommu_probe(struct platform_device *pdev)
1162{
1163	int err = -ENODEV;
1164	int irq;
1165	struct omap_iommu *obj;
1166	struct resource *res;
1167	struct device_node *of = pdev->dev.of_node;
1168
1169	if (!of) {
1170		pr_err("%s: only DT-based devices are supported\n", __func__);
1171		return -ENODEV;
1172	}
1173
1174	obj = devm_kzalloc(&pdev->dev, sizeof(*obj) + MMU_REG_SIZE, GFP_KERNEL);
1175	if (!obj)
1176		return -ENOMEM;
1177
1178	/*
1179	 * self-manage the ordering dependencies between omap_device_enable/idle
1180	 * and omap_device_assert/deassert_hardreset API
1181	 */
1182	if (pdev->dev.pm_domain) {
1183		dev_dbg(&pdev->dev, "device pm_domain is being reset\n");
1184		pdev->dev.pm_domain = NULL;
1185	}
1186
1187	obj->name = dev_name(&pdev->dev);
1188	obj->nr_tlb_entries = 32;
1189	err = of_property_read_u32(of, "ti,#tlb-entries", &obj->nr_tlb_entries);
1190	if (err && err != -EINVAL)
1191		return err;
1192	if (obj->nr_tlb_entries != 32 && obj->nr_tlb_entries != 8)
1193		return -EINVAL;
1194	if (of_property_read_bool(of, "ti,iommu-bus-err-back"))
1195		obj->has_bus_err_back = MMU_GP_REG_BUS_ERR_BACK_EN;
1196
1197	obj->dev = &pdev->dev;
1198	obj->ctx = (void *)obj + sizeof(*obj);
1199	obj->cr_ctx = devm_kzalloc(&pdev->dev,
1200				   sizeof(*obj->cr_ctx) * obj->nr_tlb_entries,
1201				   GFP_KERNEL);
1202	if (!obj->cr_ctx)
1203		return -ENOMEM;
1204
1205	spin_lock_init(&obj->iommu_lock);
1206	spin_lock_init(&obj->page_table_lock);
1207
1208	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1209	obj->regbase = devm_ioremap_resource(obj->dev, res);
1210	if (IS_ERR(obj->regbase))
1211		return PTR_ERR(obj->regbase);
1212
1213	err = omap_iommu_dra7_get_dsp_system_cfg(pdev, obj);
1214	if (err)
1215		return err;
1216
1217	irq = platform_get_irq(pdev, 0);
1218	if (irq < 0)
1219		return -ENODEV;
1220
1221	err = devm_request_irq(obj->dev, irq, iommu_fault_handler, IRQF_SHARED,
1222			       dev_name(obj->dev), obj);
1223	if (err < 0)
1224		return err;
1225	platform_set_drvdata(pdev, obj);
1226
1227	if (omap_iommu_can_register(pdev)) {
1228		err = iommu_device_sysfs_add(&obj->iommu, obj->dev, NULL,
1229					     obj->name);
1230		if (err)
1231			return err;
1232
1233		err = iommu_device_register(&obj->iommu, &omap_iommu_ops, &pdev->dev);
1234		if (err)
1235			goto out_sysfs;
1236		obj->has_iommu_driver = true;
1237	}
1238
1239	pm_runtime_enable(obj->dev);
1240
1241	omap_iommu_debugfs_add(obj);
1242
1243	dev_info(&pdev->dev, "%s registered\n", obj->name);
1244
1245	/* Re-probe bus to probe device attached to this IOMMU */
1246	bus_iommu_probe(&platform_bus_type);
1247
1248	return 0;
1249
1250out_sysfs:
1251	iommu_device_sysfs_remove(&obj->iommu);
1252	return err;
1253}
1254
1255static void omap_iommu_remove(struct platform_device *pdev)
1256{
1257	struct omap_iommu *obj = platform_get_drvdata(pdev);
1258
1259	if (obj->has_iommu_driver) {
1260		iommu_device_sysfs_remove(&obj->iommu);
1261		iommu_device_unregister(&obj->iommu);
1262	}
1263
1264	omap_iommu_debugfs_remove(obj);
1265
1266	pm_runtime_disable(obj->dev);
1267
1268	dev_info(&pdev->dev, "%s removed\n", obj->name);
1269}
1270
1271static const struct dev_pm_ops omap_iommu_pm_ops = {
1272	.prepare = omap_iommu_prepare,
1273	SET_LATE_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1274				     pm_runtime_force_resume)
1275	SET_RUNTIME_PM_OPS(omap_iommu_runtime_suspend,
1276			   omap_iommu_runtime_resume, NULL)
1277};
1278
1279static const struct of_device_id omap_iommu_of_match[] = {
1280	{ .compatible = "ti,omap2-iommu" },
1281	{ .compatible = "ti,omap4-iommu" },
1282	{ .compatible = "ti,dra7-iommu"	},
1283	{ .compatible = "ti,dra7-dsp-iommu" },
1284	{},
1285};
1286
1287static struct platform_driver omap_iommu_driver = {
1288	.probe	= omap_iommu_probe,
1289	.remove_new = omap_iommu_remove,
1290	.driver	= {
1291		.name	= "omap-iommu",
1292		.pm	= &omap_iommu_pm_ops,
1293		.of_match_table = of_match_ptr(omap_iommu_of_match),
1294	},
1295};
1296
1297static u32 iotlb_init_entry(struct iotlb_entry *e, u32 da, u32 pa, int pgsz)
1298{
1299	memset(e, 0, sizeof(*e));
1300
1301	e->da		= da;
1302	e->pa		= pa;
1303	e->valid	= MMU_CAM_V;
1304	e->pgsz		= pgsz;
1305	e->endian	= MMU_RAM_ENDIAN_LITTLE;
1306	e->elsz		= MMU_RAM_ELSZ_8;
1307	e->mixed	= 0;
1308
1309	return iopgsz_to_bytes(e->pgsz);
1310}
1311
1312static int omap_iommu_map(struct iommu_domain *domain, unsigned long da,
1313			  phys_addr_t pa, size_t bytes, size_t count,
1314			  int prot, gfp_t gfp, size_t *mapped)
1315{
1316	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1317	struct device *dev = omap_domain->dev;
1318	struct omap_iommu_device *iommu;
1319	struct omap_iommu *oiommu;
1320	struct iotlb_entry e;
1321	int omap_pgsz;
1322	u32 ret = -EINVAL;
1323	int i;
1324
1325	omap_pgsz = bytes_to_iopgsz(bytes);
1326	if (omap_pgsz < 0) {
1327		dev_err(dev, "invalid size to map: %zu\n", bytes);
1328		return -EINVAL;
1329	}
1330
1331	dev_dbg(dev, "mapping da 0x%lx to pa %pa size 0x%zx\n", da, &pa, bytes);
1332
1333	iotlb_init_entry(&e, da, pa, omap_pgsz);
1334
1335	iommu = omap_domain->iommus;
1336	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
1337		oiommu = iommu->iommu_dev;
1338		ret = omap_iopgtable_store_entry(oiommu, &e);
1339		if (ret) {
1340			dev_err(dev, "omap_iopgtable_store_entry failed: %d\n",
1341				ret);
1342			break;
1343		}
1344	}
1345
1346	if (ret) {
1347		while (i--) {
1348			iommu--;
1349			oiommu = iommu->iommu_dev;
1350			iopgtable_clear_entry(oiommu, da);
1351		}
1352	} else {
1353		*mapped = bytes;
1354	}
1355
1356	return ret;
1357}
1358
1359static size_t omap_iommu_unmap(struct iommu_domain *domain, unsigned long da,
1360			       size_t size, size_t count, struct iommu_iotlb_gather *gather)
1361{
1362	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1363	struct device *dev = omap_domain->dev;
1364	struct omap_iommu_device *iommu;
1365	struct omap_iommu *oiommu;
1366	bool error = false;
1367	size_t bytes = 0;
1368	int i;
1369
1370	dev_dbg(dev, "unmapping da 0x%lx size %zu\n", da, size);
1371
1372	iommu = omap_domain->iommus;
1373	for (i = 0; i < omap_domain->num_iommus; i++, iommu++) {
1374		oiommu = iommu->iommu_dev;
1375		bytes = iopgtable_clear_entry(oiommu, da);
1376		if (!bytes)
1377			error = true;
1378	}
1379
1380	/*
1381	 * simplify return - we are only checking if any of the iommus
1382	 * reported an error, but not if all of them are unmapping the
1383	 * same number of entries. This should not occur due to the
1384	 * mirror programming.
1385	 */
1386	return error ? 0 : bytes;
1387}
1388
1389static int omap_iommu_count(struct device *dev)
1390{
1391	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1392	int count = 0;
1393
1394	while (arch_data->iommu_dev) {
1395		count++;
1396		arch_data++;
1397	}
1398
1399	return count;
1400}
1401
1402/* caller should call cleanup if this function fails */
1403static int omap_iommu_attach_init(struct device *dev,
1404				  struct omap_iommu_domain *odomain)
1405{
1406	struct omap_iommu_device *iommu;
1407	int i;
1408
1409	odomain->num_iommus = omap_iommu_count(dev);
1410	if (!odomain->num_iommus)
1411		return -ENODEV;
1412
1413	odomain->iommus = kcalloc(odomain->num_iommus, sizeof(*iommu),
1414				  GFP_ATOMIC);
1415	if (!odomain->iommus)
1416		return -ENOMEM;
1417
1418	iommu = odomain->iommus;
1419	for (i = 0; i < odomain->num_iommus; i++, iommu++) {
1420		iommu->pgtable = kzalloc(IOPGD_TABLE_SIZE, GFP_ATOMIC);
1421		if (!iommu->pgtable)
1422			return -ENOMEM;
1423
1424		/*
1425		 * should never fail, but please keep this around to ensure
1426		 * we keep the hardware happy
1427		 */
1428		if (WARN_ON(!IS_ALIGNED((long)iommu->pgtable,
1429					IOPGD_TABLE_SIZE)))
1430			return -EINVAL;
1431	}
1432
1433	return 0;
1434}
1435
1436static void omap_iommu_detach_fini(struct omap_iommu_domain *odomain)
1437{
1438	int i;
1439	struct omap_iommu_device *iommu = odomain->iommus;
1440
1441	for (i = 0; iommu && i < odomain->num_iommus; i++, iommu++)
1442		kfree(iommu->pgtable);
1443
1444	kfree(odomain->iommus);
1445	odomain->num_iommus = 0;
1446	odomain->iommus = NULL;
1447}
1448
1449static int
1450omap_iommu_attach_dev(struct iommu_domain *domain, struct device *dev)
1451{
1452	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1453	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1454	struct omap_iommu_device *iommu;
1455	struct omap_iommu *oiommu;
1456	int ret = 0;
1457	int i;
1458
1459	if (!arch_data || !arch_data->iommu_dev) {
1460		dev_err(dev, "device doesn't have an associated iommu\n");
1461		return -ENODEV;
1462	}
1463
1464	spin_lock(&omap_domain->lock);
1465
1466	/* only a single client device can be attached to a domain */
1467	if (omap_domain->dev) {
1468		dev_err(dev, "iommu domain is already attached\n");
1469		ret = -EINVAL;
1470		goto out;
1471	}
1472
1473	ret = omap_iommu_attach_init(dev, omap_domain);
1474	if (ret) {
1475		dev_err(dev, "failed to allocate required iommu data %d\n",
1476			ret);
1477		goto init_fail;
1478	}
1479
1480	iommu = omap_domain->iommus;
1481	for (i = 0; i < omap_domain->num_iommus; i++, iommu++, arch_data++) {
1482		/* configure and enable the omap iommu */
1483		oiommu = arch_data->iommu_dev;
1484		ret = omap_iommu_attach(oiommu, iommu->pgtable);
1485		if (ret) {
1486			dev_err(dev, "can't get omap iommu: %d\n", ret);
1487			goto attach_fail;
1488		}
1489
1490		oiommu->domain = domain;
1491		iommu->iommu_dev = oiommu;
1492	}
1493
1494	omap_domain->dev = dev;
1495
1496	goto out;
1497
1498attach_fail:
1499	while (i--) {
1500		iommu--;
1501		arch_data--;
1502		oiommu = iommu->iommu_dev;
1503		omap_iommu_detach(oiommu);
1504		iommu->iommu_dev = NULL;
1505		oiommu->domain = NULL;
1506	}
1507init_fail:
1508	omap_iommu_detach_fini(omap_domain);
1509out:
1510	spin_unlock(&omap_domain->lock);
1511	return ret;
1512}
1513
1514static void _omap_iommu_detach_dev(struct omap_iommu_domain *omap_domain,
1515				   struct device *dev)
1516{
1517	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1518	struct omap_iommu_device *iommu = omap_domain->iommus;
1519	struct omap_iommu *oiommu;
1520	int i;
1521
1522	if (!omap_domain->dev) {
1523		dev_err(dev, "domain has no attached device\n");
1524		return;
1525	}
1526
1527	/* only a single device is supported per domain for now */
1528	if (omap_domain->dev != dev) {
1529		dev_err(dev, "invalid attached device\n");
1530		return;
1531	}
1532
1533	/*
1534	 * cleanup in the reverse order of attachment - this addresses
1535	 * any h/w dependencies between multiple instances, if any
1536	 */
1537	iommu += (omap_domain->num_iommus - 1);
1538	arch_data += (omap_domain->num_iommus - 1);
1539	for (i = 0; i < omap_domain->num_iommus; i++, iommu--, arch_data--) {
1540		oiommu = iommu->iommu_dev;
1541		iopgtable_clear_entry_all(oiommu);
1542
1543		omap_iommu_detach(oiommu);
1544		iommu->iommu_dev = NULL;
1545		oiommu->domain = NULL;
1546	}
1547
1548	omap_iommu_detach_fini(omap_domain);
1549
1550	omap_domain->dev = NULL;
1551}
1552
1553static int omap_iommu_identity_attach(struct iommu_domain *identity_domain,
1554				      struct device *dev)
1555{
1556	struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1557	struct omap_iommu_domain *omap_domain;
1558
1559	if (domain == identity_domain || !domain)
1560		return 0;
1561
1562	omap_domain = to_omap_domain(domain);
1563	spin_lock(&omap_domain->lock);
1564	_omap_iommu_detach_dev(omap_domain, dev);
1565	spin_unlock(&omap_domain->lock);
1566	return 0;
1567}
1568
1569static struct iommu_domain_ops omap_iommu_identity_ops = {
1570	.attach_dev = omap_iommu_identity_attach,
1571};
1572
1573static struct iommu_domain omap_iommu_identity_domain = {
1574	.type = IOMMU_DOMAIN_IDENTITY,
1575	.ops = &omap_iommu_identity_ops,
1576};
1577
1578static struct iommu_domain *omap_iommu_domain_alloc_paging(struct device *dev)
1579{
1580	struct omap_iommu_domain *omap_domain;
1581
1582	omap_domain = kzalloc(sizeof(*omap_domain), GFP_KERNEL);
1583	if (!omap_domain)
1584		return NULL;
1585
1586	spin_lock_init(&omap_domain->lock);
1587
1588	omap_domain->domain.geometry.aperture_start = 0;
1589	omap_domain->domain.geometry.aperture_end   = (1ULL << 32) - 1;
1590	omap_domain->domain.geometry.force_aperture = true;
1591
1592	return &omap_domain->domain;
1593}
1594
1595static void omap_iommu_domain_free(struct iommu_domain *domain)
1596{
1597	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1598
1599	/*
1600	 * An iommu device is still attached
1601	 * (currently, only one device can be attached) ?
1602	 */
1603	if (omap_domain->dev)
1604		_omap_iommu_detach_dev(omap_domain, omap_domain->dev);
1605
1606	kfree(omap_domain);
1607}
1608
1609static phys_addr_t omap_iommu_iova_to_phys(struct iommu_domain *domain,
1610					   dma_addr_t da)
1611{
1612	struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
1613	struct omap_iommu_device *iommu = omap_domain->iommus;
1614	struct omap_iommu *oiommu = iommu->iommu_dev;
1615	struct device *dev = oiommu->dev;
1616	u32 *pgd, *pte;
1617	phys_addr_t ret = 0;
1618
1619	/*
1620	 * all the iommus within the domain will have identical programming,
1621	 * so perform the lookup using just the first iommu
1622	 */
1623	iopgtable_lookup_entry(oiommu, da, &pgd, &pte);
1624
1625	if (pte) {
1626		if (iopte_is_small(*pte))
1627			ret = omap_iommu_translate(*pte, da, IOPTE_MASK);
1628		else if (iopte_is_large(*pte))
1629			ret = omap_iommu_translate(*pte, da, IOLARGE_MASK);
1630		else
1631			dev_err(dev, "bogus pte 0x%x, da 0x%llx", *pte,
1632				(unsigned long long)da);
1633	} else {
1634		if (iopgd_is_section(*pgd))
1635			ret = omap_iommu_translate(*pgd, da, IOSECTION_MASK);
1636		else if (iopgd_is_super(*pgd))
1637			ret = omap_iommu_translate(*pgd, da, IOSUPER_MASK);
1638		else
1639			dev_err(dev, "bogus pgd 0x%x, da 0x%llx", *pgd,
1640				(unsigned long long)da);
1641	}
1642
1643	return ret;
1644}
1645
1646static struct iommu_device *omap_iommu_probe_device(struct device *dev)
1647{
1648	struct omap_iommu_arch_data *arch_data, *tmp;
1649	struct platform_device *pdev;
1650	struct omap_iommu *oiommu;
1651	struct device_node *np;
1652	int num_iommus, i;
1653
1654	/*
1655	 * Allocate the per-device iommu structure for DT-based devices.
1656	 *
1657	 * TODO: Simplify this when removing non-DT support completely from the
1658	 * IOMMU users.
1659	 */
1660	if (!dev->of_node)
1661		return ERR_PTR(-ENODEV);
1662
1663	/*
1664	 * retrieve the count of IOMMU nodes using phandle size as element size
1665	 * since #iommu-cells = 0 for OMAP
1666	 */
1667	num_iommus = of_property_count_elems_of_size(dev->of_node, "iommus",
1668						     sizeof(phandle));
1669	if (num_iommus < 0)
1670		return ERR_PTR(-ENODEV);
1671
1672	arch_data = kcalloc(num_iommus + 1, sizeof(*arch_data), GFP_KERNEL);
1673	if (!arch_data)
1674		return ERR_PTR(-ENOMEM);
1675
1676	for (i = 0, tmp = arch_data; i < num_iommus; i++, tmp++) {
1677		np = of_parse_phandle(dev->of_node, "iommus", i);
1678		if (!np) {
1679			kfree(arch_data);
1680			return ERR_PTR(-EINVAL);
1681		}
1682
1683		pdev = of_find_device_by_node(np);
1684		if (!pdev) {
1685			of_node_put(np);
1686			kfree(arch_data);
1687			return ERR_PTR(-ENODEV);
1688		}
1689
1690		oiommu = platform_get_drvdata(pdev);
1691		if (!oiommu) {
1692			of_node_put(np);
1693			kfree(arch_data);
1694			return ERR_PTR(-EINVAL);
1695		}
1696
1697		tmp->iommu_dev = oiommu;
1698		tmp->dev = &pdev->dev;
1699
1700		of_node_put(np);
1701	}
1702
1703	dev_iommu_priv_set(dev, arch_data);
1704
1705	/*
1706	 * use the first IOMMU alone for the sysfs device linking.
1707	 * TODO: Evaluate if a single iommu_group needs to be
1708	 * maintained for both IOMMUs
1709	 */
1710	oiommu = arch_data->iommu_dev;
1711
1712	return &oiommu->iommu;
1713}
1714
1715static void omap_iommu_release_device(struct device *dev)
1716{
1717	struct omap_iommu_arch_data *arch_data = dev_iommu_priv_get(dev);
1718
1719	if (!dev->of_node || !arch_data)
1720		return;
1721
1722	kfree(arch_data);
1723
1724}
1725
1726static const struct iommu_ops omap_iommu_ops = {
1727	.identity_domain = &omap_iommu_identity_domain,
1728	.domain_alloc_paging = omap_iommu_domain_alloc_paging,
1729	.probe_device	= omap_iommu_probe_device,
1730	.release_device	= omap_iommu_release_device,
1731	.device_group	= generic_single_device_group,
1732	.pgsize_bitmap	= OMAP_IOMMU_PGSIZES,
1733	.default_domain_ops = &(const struct iommu_domain_ops) {
1734		.attach_dev	= omap_iommu_attach_dev,
1735		.map_pages	= omap_iommu_map,
1736		.unmap_pages	= omap_iommu_unmap,
1737		.iova_to_phys	= omap_iommu_iova_to_phys,
1738		.free		= omap_iommu_domain_free,
1739	}
1740};
1741
1742static int __init omap_iommu_init(void)
1743{
1744	struct kmem_cache *p;
1745	const slab_flags_t flags = SLAB_HWCACHE_ALIGN;
1746	size_t align = 1 << 10; /* L2 pagetable alignement */
1747	struct device_node *np;
1748	int ret;
1749
1750	np = of_find_matching_node(NULL, omap_iommu_of_match);
1751	if (!np)
1752		return 0;
1753
1754	of_node_put(np);
1755
1756	p = kmem_cache_create("iopte_cache", IOPTE_TABLE_SIZE, align, flags,
1757			      NULL);
1758	if (!p)
1759		return -ENOMEM;
1760	iopte_cachep = p;
1761
1762	omap_iommu_debugfs_init();
1763
1764	ret = platform_driver_register(&omap_iommu_driver);
1765	if (ret) {
1766		pr_err("%s: failed to register driver\n", __func__);
1767		goto fail_driver;
1768	}
1769
1770	return 0;
1771
1772fail_driver:
1773	kmem_cache_destroy(iopte_cachep);
1774	return ret;
1775}
1776subsys_initcall(omap_iommu_init);
1777/* must be ready before omap3isp is probed */
1778