1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * IOMMU API for Rockchip
4  *
5  * Module Authors:	Simon Xue <xxm@rock-chips.com>
6  *			Daniel Kurtz <djkurtz@chromium.org>
7  */
8 
9 #include <linux/clk.h>
10 #include <linux/compiler.h>
11 #include <linux/delay.h>
12 #include <linux/device.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/errno.h>
15 #include <linux/interrupt.h>
16 #include <linux/io.h>
17 #include <linux/iommu.h>
18 #include <linux/iopoll.h>
19 #include <linux/list.h>
20 #include <linux/mm.h>
21 #include <linux/init.h>
22 #include <linux/of.h>
23 #include <linux/of_platform.h>
24 #include <linux/platform_device.h>
25 #include <linux/pm_runtime.h>
26 #include <linux/slab.h>
27 #include <linux/spinlock.h>
28 
29 /** MMU register offsets */
30 #define RK_MMU_DTE_ADDR		0x00	/* Directory table address */
31 #define RK_MMU_STATUS		0x04
32 #define RK_MMU_COMMAND		0x08
33 #define RK_MMU_PAGE_FAULT_ADDR	0x0C	/* IOVA of last page fault */
34 #define RK_MMU_ZAP_ONE_LINE	0x10	/* Shootdown one IOTLB entry */
35 #define RK_MMU_INT_RAWSTAT	0x14	/* IRQ status ignoring mask */
36 #define RK_MMU_INT_CLEAR	0x18	/* Acknowledge and re-arm irq */
37 #define RK_MMU_INT_MASK		0x1C	/* IRQ enable */
38 #define RK_MMU_INT_STATUS	0x20	/* IRQ status after masking */
39 #define RK_MMU_AUTO_GATING	0x24
40 
41 #define DTE_ADDR_DUMMY		0xCAFEBABE
42 
43 #define RK_MMU_POLL_PERIOD_US		100
44 #define RK_MMU_FORCE_RESET_TIMEOUT_US	100000
45 #define RK_MMU_POLL_TIMEOUT_US		1000
46 
47 /* RK_MMU_STATUS fields */
48 #define RK_MMU_STATUS_PAGING_ENABLED       BIT(0)
49 #define RK_MMU_STATUS_PAGE_FAULT_ACTIVE    BIT(1)
50 #define RK_MMU_STATUS_STALL_ACTIVE         BIT(2)
51 #define RK_MMU_STATUS_IDLE                 BIT(3)
52 #define RK_MMU_STATUS_REPLAY_BUFFER_EMPTY  BIT(4)
53 #define RK_MMU_STATUS_PAGE_FAULT_IS_WRITE  BIT(5)
54 #define RK_MMU_STATUS_STALL_NOT_ACTIVE     BIT(31)
55 
56 /* RK_MMU_COMMAND command values */
57 #define RK_MMU_CMD_ENABLE_PAGING    0  /* Enable memory translation */
58 #define RK_MMU_CMD_DISABLE_PAGING   1  /* Disable memory translation */
59 #define RK_MMU_CMD_ENABLE_STALL     2  /* Stall paging to allow other cmds */
60 #define RK_MMU_CMD_DISABLE_STALL    3  /* Stop stall re-enables paging */
61 #define RK_MMU_CMD_ZAP_CACHE        4  /* Shoot down entire IOTLB */
62 #define RK_MMU_CMD_PAGE_FAULT_DONE  5  /* Clear page fault */
63 #define RK_MMU_CMD_FORCE_RESET      6  /* Reset all registers */
64 
65 /* RK_MMU_INT_* register fields */
66 #define RK_MMU_IRQ_PAGE_FAULT    0x01  /* page fault */
67 #define RK_MMU_IRQ_BUS_ERROR     0x02  /* bus read error */
68 #define RK_MMU_IRQ_MASK          (RK_MMU_IRQ_PAGE_FAULT | RK_MMU_IRQ_BUS_ERROR)
69 
70 #define NUM_DT_ENTRIES 1024
71 #define NUM_PT_ENTRIES 1024
72 
73 #define SPAGE_ORDER 12
74 #define SPAGE_SIZE (1 << SPAGE_ORDER)
75 
76  /*
77   * Support mapping any size that fits in one page table:
78   *   4 KiB to 4 MiB
79   */
80 #define RK_IOMMU_PGSIZE_BITMAP 0x007ff000
81 
82 struct rk_iommu_domain {
83 	struct list_head iommus;
84 	u32 *dt; /* page directory table */
85 	dma_addr_t dt_dma;
86 	spinlock_t iommus_lock; /* lock for iommus list */
87 	spinlock_t dt_lock; /* lock for modifying page directory table */
88 
89 	struct iommu_domain domain;
90 };
91 
92 /* list of clocks required by IOMMU */
93 static const char * const rk_iommu_clocks[] = {
94 	"aclk", "iface",
95 };
96 
97 struct rk_iommu_ops {
98 	phys_addr_t (*pt_address)(u32 dte);
99 	u32 (*mk_dtentries)(dma_addr_t pt_dma);
100 	u32 (*mk_ptentries)(phys_addr_t page, int prot);
101 	phys_addr_t (*dte_addr_phys)(u32 addr);
102 	u32 (*dma_addr_dte)(dma_addr_t dt_dma);
103 	u64 dma_bit_mask;
104 };
105 
106 struct rk_iommu {
107 	struct device *dev;
108 	void __iomem **bases;
109 	int num_mmu;
110 	int num_irq;
111 	struct clk_bulk_data *clocks;
112 	int num_clocks;
113 	bool reset_disabled;
114 	struct iommu_device iommu;
115 	struct list_head node; /* entry in rk_iommu_domain.iommus */
116 	struct iommu_domain *domain; /* domain to which iommu is attached */
117 	struct iommu_group *group;
118 };
119 
120 struct rk_iommudata {
121 	struct device_link *link; /* runtime PM link from IOMMU to master */
122 	struct rk_iommu *iommu;
123 };
124 
125 static struct device *dma_dev;
126 static const struct rk_iommu_ops *rk_ops;
127 
128 static inline void rk_table_flush(struct rk_iommu_domain *dom, dma_addr_t dma,
129 				  unsigned int count)
130 {
131 	size_t size = count * sizeof(u32); /* count of u32 entry */
132 
133 	dma_sync_single_for_device(dma_dev, dma, size, DMA_TO_DEVICE);
134 }
135 
136 static struct rk_iommu_domain *to_rk_domain(struct iommu_domain *dom)
137 {
138 	return container_of(dom, struct rk_iommu_domain, domain);
139 }
140 
141 /*
142  * The Rockchip rk3288 iommu uses a 2-level page table.
143  * The first level is the "Directory Table" (DT).
144  * The DT consists of 1024 4-byte Directory Table Entries (DTEs), each pointing
145  * to a "Page Table".
146  * The second level is the 1024 Page Tables (PT).
147  * Each PT consists of 1024 4-byte Page Table Entries (PTEs), each pointing to
148  * a 4 KB page of physical memory.
149  *
150  * The DT and each PT fits in a single 4 KB page (4-bytes * 1024 entries).
151  * Each iommu device has a MMU_DTE_ADDR register that contains the physical
152  * address of the start of the DT page.
153  *
154  * The structure of the page table is as follows:
155  *
156  *                   DT
157  * MMU_DTE_ADDR -> +-----+
158  *                 |     |
159  *                 +-----+     PT
160  *                 | DTE | -> +-----+
161  *                 +-----+    |     |     Memory
162  *                 |     |    +-----+     Page
163  *                 |     |    | PTE | -> +-----+
164  *                 +-----+    +-----+    |     |
165  *                            |     |    |     |
166  *                            |     |    |     |
167  *                            +-----+    |     |
168  *                                       |     |
169  *                                       |     |
170  *                                       +-----+
171  */
172 
173 /*
174  * Each DTE has a PT address and a valid bit:
175  * +---------------------+-----------+-+
176  * | PT address          | Reserved  |V|
177  * +---------------------+-----------+-+
178  *  31:12 - PT address (PTs always starts on a 4 KB boundary)
179  *  11: 1 - Reserved
180  *      0 - 1 if PT @ PT address is valid
181  */
182 #define RK_DTE_PT_ADDRESS_MASK    0xfffff000
183 #define RK_DTE_PT_VALID           BIT(0)
184 
185 static inline phys_addr_t rk_dte_pt_address(u32 dte)
186 {
187 	return (phys_addr_t)dte & RK_DTE_PT_ADDRESS_MASK;
188 }
189 
190 /*
191  * In v2:
192  * 31:12 - PT address bit 31:0
193  * 11: 8 - PT address bit 35:32
194  *  7: 4 - PT address bit 39:36
195  *  3: 1 - Reserved
196  *     0 - 1 if PT @ PT address is valid
197  */
198 #define RK_DTE_PT_ADDRESS_MASK_V2 GENMASK_ULL(31, 4)
199 #define DTE_HI_MASK1	GENMASK(11, 8)
200 #define DTE_HI_MASK2	GENMASK(7, 4)
201 #define DTE_HI_SHIFT1	24 /* shift bit 8 to bit 32 */
202 #define DTE_HI_SHIFT2	32 /* shift bit 4 to bit 36 */
203 #define PAGE_DESC_HI_MASK1	GENMASK_ULL(35, 32)
204 #define PAGE_DESC_HI_MASK2	GENMASK_ULL(39, 36)
205 
206 static inline phys_addr_t rk_dte_pt_address_v2(u32 dte)
207 {
208 	u64 dte_v2 = dte;
209 
210 	dte_v2 = ((dte_v2 & DTE_HI_MASK2) << DTE_HI_SHIFT2) |
211 		 ((dte_v2 & DTE_HI_MASK1) << DTE_HI_SHIFT1) |
212 		 (dte_v2 & RK_DTE_PT_ADDRESS_MASK);
213 
214 	return (phys_addr_t)dte_v2;
215 }
216 
217 static inline bool rk_dte_is_pt_valid(u32 dte)
218 {
219 	return dte & RK_DTE_PT_VALID;
220 }
221 
222 static inline u32 rk_mk_dte(dma_addr_t pt_dma)
223 {
224 	return (pt_dma & RK_DTE_PT_ADDRESS_MASK) | RK_DTE_PT_VALID;
225 }
226 
227 static inline u32 rk_mk_dte_v2(dma_addr_t pt_dma)
228 {
229 	pt_dma = (pt_dma & RK_DTE_PT_ADDRESS_MASK) |
230 		 ((pt_dma & PAGE_DESC_HI_MASK1) >> DTE_HI_SHIFT1) |
231 		 (pt_dma & PAGE_DESC_HI_MASK2) >> DTE_HI_SHIFT2;
232 
233 	return (pt_dma & RK_DTE_PT_ADDRESS_MASK_V2) | RK_DTE_PT_VALID;
234 }
235 
236 /*
237  * Each PTE has a Page address, some flags and a valid bit:
238  * +---------------------+---+-------+-+
239  * | Page address        |Rsv| Flags |V|
240  * +---------------------+---+-------+-+
241  *  31:12 - Page address (Pages always start on a 4 KB boundary)
242  *  11: 9 - Reserved
243  *   8: 1 - Flags
244  *      8 - Read allocate - allocate cache space on read misses
245  *      7 - Read cache - enable cache & prefetch of data
246  *      6 - Write buffer - enable delaying writes on their way to memory
247  *      5 - Write allocate - allocate cache space on write misses
248  *      4 - Write cache - different writes can be merged together
249  *      3 - Override cache attributes
250  *          if 1, bits 4-8 control cache attributes
251  *          if 0, the system bus defaults are used
252  *      2 - Writable
253  *      1 - Readable
254  *      0 - 1 if Page @ Page address is valid
255  */
256 #define RK_PTE_PAGE_ADDRESS_MASK  0xfffff000
257 #define RK_PTE_PAGE_FLAGS_MASK    0x000001fe
258 #define RK_PTE_PAGE_WRITABLE      BIT(2)
259 #define RK_PTE_PAGE_READABLE      BIT(1)
260 #define RK_PTE_PAGE_VALID         BIT(0)
261 
262 static inline bool rk_pte_is_page_valid(u32 pte)
263 {
264 	return pte & RK_PTE_PAGE_VALID;
265 }
266 
267 /* TODO: set cache flags per prot IOMMU_CACHE */
268 static u32 rk_mk_pte(phys_addr_t page, int prot)
269 {
270 	u32 flags = 0;
271 	flags |= (prot & IOMMU_READ) ? RK_PTE_PAGE_READABLE : 0;
272 	flags |= (prot & IOMMU_WRITE) ? RK_PTE_PAGE_WRITABLE : 0;
273 	page &= RK_PTE_PAGE_ADDRESS_MASK;
274 	return page | flags | RK_PTE_PAGE_VALID;
275 }
276 
277 /*
278  * In v2:
279  * 31:12 - Page address bit 31:0
280  *  11:9 - Page address bit 34:32
281  *   8:4 - Page address bit 39:35
282  *     3 - Security
283  *     2 - Writable
284  *     1 - Readable
285  *     0 - 1 if Page @ Page address is valid
286  */
287 
288 static u32 rk_mk_pte_v2(phys_addr_t page, int prot)
289 {
290 	u32 flags = 0;
291 
292 	flags |= (prot & IOMMU_READ) ? RK_PTE_PAGE_READABLE : 0;
293 	flags |= (prot & IOMMU_WRITE) ? RK_PTE_PAGE_WRITABLE : 0;
294 
295 	return rk_mk_dte_v2(page) | flags;
296 }
297 
298 static u32 rk_mk_pte_invalid(u32 pte)
299 {
300 	return pte & ~RK_PTE_PAGE_VALID;
301 }
302 
303 /*
304  * rk3288 iova (IOMMU Virtual Address) format
305  *  31       22.21       12.11          0
306  * +-----------+-----------+-------------+
307  * | DTE index | PTE index | Page offset |
308  * +-----------+-----------+-------------+
309  *  31:22 - DTE index   - index of DTE in DT
310  *  21:12 - PTE index   - index of PTE in PT @ DTE.pt_address
311  *  11: 0 - Page offset - offset into page @ PTE.page_address
312  */
313 #define RK_IOVA_DTE_MASK    0xffc00000
314 #define RK_IOVA_DTE_SHIFT   22
315 #define RK_IOVA_PTE_MASK    0x003ff000
316 #define RK_IOVA_PTE_SHIFT   12
317 #define RK_IOVA_PAGE_MASK   0x00000fff
318 #define RK_IOVA_PAGE_SHIFT  0
319 
320 static u32 rk_iova_dte_index(dma_addr_t iova)
321 {
322 	return (u32)(iova & RK_IOVA_DTE_MASK) >> RK_IOVA_DTE_SHIFT;
323 }
324 
325 static u32 rk_iova_pte_index(dma_addr_t iova)
326 {
327 	return (u32)(iova & RK_IOVA_PTE_MASK) >> RK_IOVA_PTE_SHIFT;
328 }
329 
330 static u32 rk_iova_page_offset(dma_addr_t iova)
331 {
332 	return (u32)(iova & RK_IOVA_PAGE_MASK) >> RK_IOVA_PAGE_SHIFT;
333 }
334 
335 static u32 rk_iommu_read(void __iomem *base, u32 offset)
336 {
337 	return readl(base + offset);
338 }
339 
340 static void rk_iommu_write(void __iomem *base, u32 offset, u32 value)
341 {
342 	writel(value, base + offset);
343 }
344 
345 static void rk_iommu_command(struct rk_iommu *iommu, u32 command)
346 {
347 	int i;
348 
349 	for (i = 0; i < iommu->num_mmu; i++)
350 		writel(command, iommu->bases[i] + RK_MMU_COMMAND);
351 }
352 
353 static void rk_iommu_base_command(void __iomem *base, u32 command)
354 {
355 	writel(command, base + RK_MMU_COMMAND);
356 }
357 static void rk_iommu_zap_lines(struct rk_iommu *iommu, dma_addr_t iova_start,
358 			       size_t size)
359 {
360 	int i;
361 	dma_addr_t iova_end = iova_start + size;
362 	/*
363 	 * TODO(djkurtz): Figure out when it is more efficient to shootdown the
364 	 * entire iotlb rather than iterate over individual iovas.
365 	 */
366 	for (i = 0; i < iommu->num_mmu; i++) {
367 		dma_addr_t iova;
368 
369 		for (iova = iova_start; iova < iova_end; iova += SPAGE_SIZE)
370 			rk_iommu_write(iommu->bases[i], RK_MMU_ZAP_ONE_LINE, iova);
371 	}
372 }
373 
374 static bool rk_iommu_is_stall_active(struct rk_iommu *iommu)
375 {
376 	bool active = true;
377 	int i;
378 
379 	for (i = 0; i < iommu->num_mmu; i++)
380 		active &= !!(rk_iommu_read(iommu->bases[i], RK_MMU_STATUS) &
381 					   RK_MMU_STATUS_STALL_ACTIVE);
382 
383 	return active;
384 }
385 
386 static bool rk_iommu_is_paging_enabled(struct rk_iommu *iommu)
387 {
388 	bool enable = true;
389 	int i;
390 
391 	for (i = 0; i < iommu->num_mmu; i++)
392 		enable &= !!(rk_iommu_read(iommu->bases[i], RK_MMU_STATUS) &
393 					   RK_MMU_STATUS_PAGING_ENABLED);
394 
395 	return enable;
396 }
397 
398 static bool rk_iommu_is_reset_done(struct rk_iommu *iommu)
399 {
400 	bool done = true;
401 	int i;
402 
403 	for (i = 0; i < iommu->num_mmu; i++)
404 		done &= rk_iommu_read(iommu->bases[i], RK_MMU_DTE_ADDR) == 0;
405 
406 	return done;
407 }
408 
409 static int rk_iommu_enable_stall(struct rk_iommu *iommu)
410 {
411 	int ret, i;
412 	bool val;
413 
414 	if (rk_iommu_is_stall_active(iommu))
415 		return 0;
416 
417 	/* Stall can only be enabled if paging is enabled */
418 	if (!rk_iommu_is_paging_enabled(iommu))
419 		return 0;
420 
421 	rk_iommu_command(iommu, RK_MMU_CMD_ENABLE_STALL);
422 
423 	ret = readx_poll_timeout(rk_iommu_is_stall_active, iommu, val,
424 				 val, RK_MMU_POLL_PERIOD_US,
425 				 RK_MMU_POLL_TIMEOUT_US);
426 	if (ret)
427 		for (i = 0; i < iommu->num_mmu; i++)
428 			dev_err(iommu->dev, "Enable stall request timed out, status: %#08x\n",
429 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
430 
431 	return ret;
432 }
433 
434 static int rk_iommu_disable_stall(struct rk_iommu *iommu)
435 {
436 	int ret, i;
437 	bool val;
438 
439 	if (!rk_iommu_is_stall_active(iommu))
440 		return 0;
441 
442 	rk_iommu_command(iommu, RK_MMU_CMD_DISABLE_STALL);
443 
444 	ret = readx_poll_timeout(rk_iommu_is_stall_active, iommu, val,
445 				 !val, RK_MMU_POLL_PERIOD_US,
446 				 RK_MMU_POLL_TIMEOUT_US);
447 	if (ret)
448 		for (i = 0; i < iommu->num_mmu; i++)
449 			dev_err(iommu->dev, "Disable stall request timed out, status: %#08x\n",
450 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
451 
452 	return ret;
453 }
454 
455 static int rk_iommu_enable_paging(struct rk_iommu *iommu)
456 {
457 	int ret, i;
458 	bool val;
459 
460 	if (rk_iommu_is_paging_enabled(iommu))
461 		return 0;
462 
463 	rk_iommu_command(iommu, RK_MMU_CMD_ENABLE_PAGING);
464 
465 	ret = readx_poll_timeout(rk_iommu_is_paging_enabled, iommu, val,
466 				 val, RK_MMU_POLL_PERIOD_US,
467 				 RK_MMU_POLL_TIMEOUT_US);
468 	if (ret)
469 		for (i = 0; i < iommu->num_mmu; i++)
470 			dev_err(iommu->dev, "Enable paging request timed out, status: %#08x\n",
471 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
472 
473 	return ret;
474 }
475 
476 static int rk_iommu_disable_paging(struct rk_iommu *iommu)
477 {
478 	int ret, i;
479 	bool val;
480 
481 	if (!rk_iommu_is_paging_enabled(iommu))
482 		return 0;
483 
484 	rk_iommu_command(iommu, RK_MMU_CMD_DISABLE_PAGING);
485 
486 	ret = readx_poll_timeout(rk_iommu_is_paging_enabled, iommu, val,
487 				 !val, RK_MMU_POLL_PERIOD_US,
488 				 RK_MMU_POLL_TIMEOUT_US);
489 	if (ret)
490 		for (i = 0; i < iommu->num_mmu; i++)
491 			dev_err(iommu->dev, "Disable paging request timed out, status: %#08x\n",
492 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
493 
494 	return ret;
495 }
496 
497 static int rk_iommu_force_reset(struct rk_iommu *iommu)
498 {
499 	int ret, i;
500 	u32 dte_addr;
501 	bool val;
502 
503 	if (iommu->reset_disabled)
504 		return 0;
505 
506 	/*
507 	 * Check if register DTE_ADDR is working by writing DTE_ADDR_DUMMY
508 	 * and verifying that upper 5 nybbles are read back.
509 	 */
510 	for (i = 0; i < iommu->num_mmu; i++) {
511 		dte_addr = rk_ops->pt_address(DTE_ADDR_DUMMY);
512 		rk_iommu_write(iommu->bases[i], RK_MMU_DTE_ADDR, dte_addr);
513 
514 		if (dte_addr != rk_iommu_read(iommu->bases[i], RK_MMU_DTE_ADDR)) {
515 			dev_err(iommu->dev, "Error during raw reset. MMU_DTE_ADDR is not functioning\n");
516 			return -EFAULT;
517 		}
518 	}
519 
520 	rk_iommu_command(iommu, RK_MMU_CMD_FORCE_RESET);
521 
522 	ret = readx_poll_timeout(rk_iommu_is_reset_done, iommu, val,
523 				 val, RK_MMU_FORCE_RESET_TIMEOUT_US,
524 				 RK_MMU_POLL_TIMEOUT_US);
525 	if (ret) {
526 		dev_err(iommu->dev, "FORCE_RESET command timed out\n");
527 		return ret;
528 	}
529 
530 	return 0;
531 }
532 
533 static inline phys_addr_t rk_dte_addr_phys(u32 addr)
534 {
535 	return (phys_addr_t)addr;
536 }
537 
538 static inline u32 rk_dma_addr_dte(dma_addr_t dt_dma)
539 {
540 	return dt_dma;
541 }
542 
543 #define DT_HI_MASK GENMASK_ULL(39, 32)
544 #define DTE_BASE_HI_MASK GENMASK(11, 4)
545 #define DT_SHIFT   28
546 
547 static inline phys_addr_t rk_dte_addr_phys_v2(u32 addr)
548 {
549 	u64 addr64 = addr;
550 	return (phys_addr_t)(addr64 & RK_DTE_PT_ADDRESS_MASK) |
551 	       ((addr64 & DTE_BASE_HI_MASK) << DT_SHIFT);
552 }
553 
554 static inline u32 rk_dma_addr_dte_v2(dma_addr_t dt_dma)
555 {
556 	return (dt_dma & RK_DTE_PT_ADDRESS_MASK) |
557 	       ((dt_dma & DT_HI_MASK) >> DT_SHIFT);
558 }
559 
560 static void log_iova(struct rk_iommu *iommu, int index, dma_addr_t iova)
561 {
562 	void __iomem *base = iommu->bases[index];
563 	u32 dte_index, pte_index, page_offset;
564 	u32 mmu_dte_addr;
565 	phys_addr_t mmu_dte_addr_phys, dte_addr_phys;
566 	u32 *dte_addr;
567 	u32 dte;
568 	phys_addr_t pte_addr_phys = 0;
569 	u32 *pte_addr = NULL;
570 	u32 pte = 0;
571 	phys_addr_t page_addr_phys = 0;
572 	u32 page_flags = 0;
573 
574 	dte_index = rk_iova_dte_index(iova);
575 	pte_index = rk_iova_pte_index(iova);
576 	page_offset = rk_iova_page_offset(iova);
577 
578 	mmu_dte_addr = rk_iommu_read(base, RK_MMU_DTE_ADDR);
579 	mmu_dte_addr_phys = rk_ops->dte_addr_phys(mmu_dte_addr);
580 
581 	dte_addr_phys = mmu_dte_addr_phys + (4 * dte_index);
582 	dte_addr = phys_to_virt(dte_addr_phys);
583 	dte = *dte_addr;
584 
585 	if (!rk_dte_is_pt_valid(dte))
586 		goto print_it;
587 
588 	pte_addr_phys = rk_ops->pt_address(dte) + (pte_index * 4);
589 	pte_addr = phys_to_virt(pte_addr_phys);
590 	pte = *pte_addr;
591 
592 	if (!rk_pte_is_page_valid(pte))
593 		goto print_it;
594 
595 	page_addr_phys = rk_ops->pt_address(pte) + page_offset;
596 	page_flags = pte & RK_PTE_PAGE_FLAGS_MASK;
597 
598 print_it:
599 	dev_err(iommu->dev, "iova = %pad: dte_index: %#03x pte_index: %#03x page_offset: %#03x\n",
600 		&iova, dte_index, pte_index, page_offset);
601 	dev_err(iommu->dev, "mmu_dte_addr: %pa dte@%pa: %#08x valid: %u pte@%pa: %#08x valid: %u page@%pa flags: %#03x\n",
602 		&mmu_dte_addr_phys, &dte_addr_phys, dte,
603 		rk_dte_is_pt_valid(dte), &pte_addr_phys, pte,
604 		rk_pte_is_page_valid(pte), &page_addr_phys, page_flags);
605 }
606 
607 static irqreturn_t rk_iommu_irq(int irq, void *dev_id)
608 {
609 	struct rk_iommu *iommu = dev_id;
610 	u32 status;
611 	u32 int_status;
612 	dma_addr_t iova;
613 	irqreturn_t ret = IRQ_NONE;
614 	int i, err;
615 
616 	err = pm_runtime_get_if_in_use(iommu->dev);
617 	if (!err || WARN_ON_ONCE(err < 0))
618 		return ret;
619 
620 	if (WARN_ON(clk_bulk_enable(iommu->num_clocks, iommu->clocks)))
621 		goto out;
622 
623 	for (i = 0; i < iommu->num_mmu; i++) {
624 		int_status = rk_iommu_read(iommu->bases[i], RK_MMU_INT_STATUS);
625 		if (int_status == 0)
626 			continue;
627 
628 		ret = IRQ_HANDLED;
629 		iova = rk_iommu_read(iommu->bases[i], RK_MMU_PAGE_FAULT_ADDR);
630 
631 		if (int_status & RK_MMU_IRQ_PAGE_FAULT) {
632 			int flags;
633 
634 			status = rk_iommu_read(iommu->bases[i], RK_MMU_STATUS);
635 			flags = (status & RK_MMU_STATUS_PAGE_FAULT_IS_WRITE) ?
636 					IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
637 
638 			dev_err(iommu->dev, "Page fault at %pad of type %s\n",
639 				&iova,
640 				(flags == IOMMU_FAULT_WRITE) ? "write" : "read");
641 
642 			log_iova(iommu, i, iova);
643 
644 			/*
645 			 * Report page fault to any installed handlers.
646 			 * Ignore the return code, though, since we always zap cache
647 			 * and clear the page fault anyway.
648 			 */
649 			if (iommu->domain)
650 				report_iommu_fault(iommu->domain, iommu->dev, iova,
651 						   flags);
652 			else
653 				dev_err(iommu->dev, "Page fault while iommu not attached to domain?\n");
654 
655 			rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_ZAP_CACHE);
656 			rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_PAGE_FAULT_DONE);
657 		}
658 
659 		if (int_status & RK_MMU_IRQ_BUS_ERROR)
660 			dev_err(iommu->dev, "BUS_ERROR occurred at %pad\n", &iova);
661 
662 		if (int_status & ~RK_MMU_IRQ_MASK)
663 			dev_err(iommu->dev, "unexpected int_status: %#08x\n",
664 				int_status);
665 
666 		rk_iommu_write(iommu->bases[i], RK_MMU_INT_CLEAR, int_status);
667 	}
668 
669 	clk_bulk_disable(iommu->num_clocks, iommu->clocks);
670 
671 out:
672 	pm_runtime_put(iommu->dev);
673 	return ret;
674 }
675 
676 static phys_addr_t rk_iommu_iova_to_phys(struct iommu_domain *domain,
677 					 dma_addr_t iova)
678 {
679 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
680 	unsigned long flags;
681 	phys_addr_t pt_phys, phys = 0;
682 	u32 dte, pte;
683 	u32 *page_table;
684 
685 	spin_lock_irqsave(&rk_domain->dt_lock, flags);
686 
687 	dte = rk_domain->dt[rk_iova_dte_index(iova)];
688 	if (!rk_dte_is_pt_valid(dte))
689 		goto out;
690 
691 	pt_phys = rk_ops->pt_address(dte);
692 	page_table = (u32 *)phys_to_virt(pt_phys);
693 	pte = page_table[rk_iova_pte_index(iova)];
694 	if (!rk_pte_is_page_valid(pte))
695 		goto out;
696 
697 	phys = rk_ops->pt_address(pte) + rk_iova_page_offset(iova);
698 out:
699 	spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
700 
701 	return phys;
702 }
703 
704 static void rk_iommu_zap_iova(struct rk_iommu_domain *rk_domain,
705 			      dma_addr_t iova, size_t size)
706 {
707 	struct list_head *pos;
708 	unsigned long flags;
709 
710 	/* shootdown these iova from all iommus using this domain */
711 	spin_lock_irqsave(&rk_domain->iommus_lock, flags);
712 	list_for_each(pos, &rk_domain->iommus) {
713 		struct rk_iommu *iommu;
714 		int ret;
715 
716 		iommu = list_entry(pos, struct rk_iommu, node);
717 
718 		/* Only zap TLBs of IOMMUs that are powered on. */
719 		ret = pm_runtime_get_if_in_use(iommu->dev);
720 		if (WARN_ON_ONCE(ret < 0))
721 			continue;
722 		if (ret) {
723 			WARN_ON(clk_bulk_enable(iommu->num_clocks,
724 						iommu->clocks));
725 			rk_iommu_zap_lines(iommu, iova, size);
726 			clk_bulk_disable(iommu->num_clocks, iommu->clocks);
727 			pm_runtime_put(iommu->dev);
728 		}
729 	}
730 	spin_unlock_irqrestore(&rk_domain->iommus_lock, flags);
731 }
732 
733 static void rk_iommu_zap_iova_first_last(struct rk_iommu_domain *rk_domain,
734 					 dma_addr_t iova, size_t size)
735 {
736 	rk_iommu_zap_iova(rk_domain, iova, SPAGE_SIZE);
737 	if (size > SPAGE_SIZE)
738 		rk_iommu_zap_iova(rk_domain, iova + size - SPAGE_SIZE,
739 					SPAGE_SIZE);
740 }
741 
742 static u32 *rk_dte_get_page_table(struct rk_iommu_domain *rk_domain,
743 				  dma_addr_t iova)
744 {
745 	u32 *page_table, *dte_addr;
746 	u32 dte_index, dte;
747 	phys_addr_t pt_phys;
748 	dma_addr_t pt_dma;
749 
750 	assert_spin_locked(&rk_domain->dt_lock);
751 
752 	dte_index = rk_iova_dte_index(iova);
753 	dte_addr = &rk_domain->dt[dte_index];
754 	dte = *dte_addr;
755 	if (rk_dte_is_pt_valid(dte))
756 		goto done;
757 
758 	page_table = (u32 *)get_zeroed_page(GFP_ATOMIC | GFP_DMA32);
759 	if (!page_table)
760 		return ERR_PTR(-ENOMEM);
761 
762 	pt_dma = dma_map_single(dma_dev, page_table, SPAGE_SIZE, DMA_TO_DEVICE);
763 	if (dma_mapping_error(dma_dev, pt_dma)) {
764 		dev_err(dma_dev, "DMA mapping error while allocating page table\n");
765 		free_page((unsigned long)page_table);
766 		return ERR_PTR(-ENOMEM);
767 	}
768 
769 	dte = rk_ops->mk_dtentries(pt_dma);
770 	*dte_addr = dte;
771 
772 	rk_table_flush(rk_domain,
773 		       rk_domain->dt_dma + dte_index * sizeof(u32), 1);
774 done:
775 	pt_phys = rk_ops->pt_address(dte);
776 	return (u32 *)phys_to_virt(pt_phys);
777 }
778 
779 static size_t rk_iommu_unmap_iova(struct rk_iommu_domain *rk_domain,
780 				  u32 *pte_addr, dma_addr_t pte_dma,
781 				  size_t size)
782 {
783 	unsigned int pte_count;
784 	unsigned int pte_total = size / SPAGE_SIZE;
785 
786 	assert_spin_locked(&rk_domain->dt_lock);
787 
788 	for (pte_count = 0; pte_count < pte_total; pte_count++) {
789 		u32 pte = pte_addr[pte_count];
790 		if (!rk_pte_is_page_valid(pte))
791 			break;
792 
793 		pte_addr[pte_count] = rk_mk_pte_invalid(pte);
794 	}
795 
796 	rk_table_flush(rk_domain, pte_dma, pte_count);
797 
798 	return pte_count * SPAGE_SIZE;
799 }
800 
801 static int rk_iommu_map_iova(struct rk_iommu_domain *rk_domain, u32 *pte_addr,
802 			     dma_addr_t pte_dma, dma_addr_t iova,
803 			     phys_addr_t paddr, size_t size, int prot)
804 {
805 	unsigned int pte_count;
806 	unsigned int pte_total = size / SPAGE_SIZE;
807 	phys_addr_t page_phys;
808 
809 	assert_spin_locked(&rk_domain->dt_lock);
810 
811 	for (pte_count = 0; pte_count < pte_total; pte_count++) {
812 		u32 pte = pte_addr[pte_count];
813 
814 		if (rk_pte_is_page_valid(pte))
815 			goto unwind;
816 
817 		pte_addr[pte_count] = rk_ops->mk_ptentries(paddr, prot);
818 
819 		paddr += SPAGE_SIZE;
820 	}
821 
822 	rk_table_flush(rk_domain, pte_dma, pte_total);
823 
824 	/*
825 	 * Zap the first and last iova to evict from iotlb any previously
826 	 * mapped cachelines holding stale values for its dte and pte.
827 	 * We only zap the first and last iova, since only they could have
828 	 * dte or pte shared with an existing mapping.
829 	 */
830 	rk_iommu_zap_iova_first_last(rk_domain, iova, size);
831 
832 	return 0;
833 unwind:
834 	/* Unmap the range of iovas that we just mapped */
835 	rk_iommu_unmap_iova(rk_domain, pte_addr, pte_dma,
836 			    pte_count * SPAGE_SIZE);
837 
838 	iova += pte_count * SPAGE_SIZE;
839 	page_phys = rk_ops->pt_address(pte_addr[pte_count]);
840 	pr_err("iova: %pad already mapped to %pa cannot remap to phys: %pa prot: %#x\n",
841 	       &iova, &page_phys, &paddr, prot);
842 
843 	return -EADDRINUSE;
844 }
845 
846 static int rk_iommu_map(struct iommu_domain *domain, unsigned long _iova,
847 			phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
848 {
849 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
850 	unsigned long flags;
851 	dma_addr_t pte_dma, iova = (dma_addr_t)_iova;
852 	u32 *page_table, *pte_addr;
853 	u32 dte_index, pte_index;
854 	int ret;
855 
856 	spin_lock_irqsave(&rk_domain->dt_lock, flags);
857 
858 	/*
859 	 * pgsize_bitmap specifies iova sizes that fit in one page table
860 	 * (1024 4-KiB pages = 4 MiB).
861 	 * So, size will always be 4096 <= size <= 4194304.
862 	 * Since iommu_map() guarantees that both iova and size will be
863 	 * aligned, we will always only be mapping from a single dte here.
864 	 */
865 	page_table = rk_dte_get_page_table(rk_domain, iova);
866 	if (IS_ERR(page_table)) {
867 		spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
868 		return PTR_ERR(page_table);
869 	}
870 
871 	dte_index = rk_domain->dt[rk_iova_dte_index(iova)];
872 	pte_index = rk_iova_pte_index(iova);
873 	pte_addr = &page_table[pte_index];
874 
875 	pte_dma = rk_ops->pt_address(dte_index) + pte_index * sizeof(u32);
876 	ret = rk_iommu_map_iova(rk_domain, pte_addr, pte_dma, iova,
877 				paddr, size, prot);
878 
879 	spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
880 
881 	return ret;
882 }
883 
884 static size_t rk_iommu_unmap(struct iommu_domain *domain, unsigned long _iova,
885 			     size_t size, struct iommu_iotlb_gather *gather)
886 {
887 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
888 	unsigned long flags;
889 	dma_addr_t pte_dma, iova = (dma_addr_t)_iova;
890 	phys_addr_t pt_phys;
891 	u32 dte;
892 	u32 *pte_addr;
893 	size_t unmap_size;
894 
895 	spin_lock_irqsave(&rk_domain->dt_lock, flags);
896 
897 	/*
898 	 * pgsize_bitmap specifies iova sizes that fit in one page table
899 	 * (1024 4-KiB pages = 4 MiB).
900 	 * So, size will always be 4096 <= size <= 4194304.
901 	 * Since iommu_unmap() guarantees that both iova and size will be
902 	 * aligned, we will always only be unmapping from a single dte here.
903 	 */
904 	dte = rk_domain->dt[rk_iova_dte_index(iova)];
905 	/* Just return 0 if iova is unmapped */
906 	if (!rk_dte_is_pt_valid(dte)) {
907 		spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
908 		return 0;
909 	}
910 
911 	pt_phys = rk_ops->pt_address(dte);
912 	pte_addr = (u32 *)phys_to_virt(pt_phys) + rk_iova_pte_index(iova);
913 	pte_dma = pt_phys + rk_iova_pte_index(iova) * sizeof(u32);
914 	unmap_size = rk_iommu_unmap_iova(rk_domain, pte_addr, pte_dma, size);
915 
916 	spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
917 
918 	/* Shootdown iotlb entries for iova range that was just unmapped */
919 	rk_iommu_zap_iova(rk_domain, iova, unmap_size);
920 
921 	return unmap_size;
922 }
923 
924 static struct rk_iommu *rk_iommu_from_dev(struct device *dev)
925 {
926 	struct rk_iommudata *data = dev_iommu_priv_get(dev);
927 
928 	return data ? data->iommu : NULL;
929 }
930 
931 /* Must be called with iommu powered on and attached */
932 static void rk_iommu_disable(struct rk_iommu *iommu)
933 {
934 	int i;
935 
936 	/* Ignore error while disabling, just keep going */
937 	WARN_ON(clk_bulk_enable(iommu->num_clocks, iommu->clocks));
938 	rk_iommu_enable_stall(iommu);
939 	rk_iommu_disable_paging(iommu);
940 	for (i = 0; i < iommu->num_mmu; i++) {
941 		rk_iommu_write(iommu->bases[i], RK_MMU_INT_MASK, 0);
942 		rk_iommu_write(iommu->bases[i], RK_MMU_DTE_ADDR, 0);
943 	}
944 	rk_iommu_disable_stall(iommu);
945 	clk_bulk_disable(iommu->num_clocks, iommu->clocks);
946 }
947 
948 /* Must be called with iommu powered on and attached */
949 static int rk_iommu_enable(struct rk_iommu *iommu)
950 {
951 	struct iommu_domain *domain = iommu->domain;
952 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
953 	int ret, i;
954 
955 	ret = clk_bulk_enable(iommu->num_clocks, iommu->clocks);
956 	if (ret)
957 		return ret;
958 
959 	ret = rk_iommu_enable_stall(iommu);
960 	if (ret)
961 		goto out_disable_clocks;
962 
963 	ret = rk_iommu_force_reset(iommu);
964 	if (ret)
965 		goto out_disable_stall;
966 
967 	for (i = 0; i < iommu->num_mmu; i++) {
968 		rk_iommu_write(iommu->bases[i], RK_MMU_DTE_ADDR,
969 			       rk_ops->dma_addr_dte(rk_domain->dt_dma));
970 		rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_ZAP_CACHE);
971 		rk_iommu_write(iommu->bases[i], RK_MMU_INT_MASK, RK_MMU_IRQ_MASK);
972 	}
973 
974 	ret = rk_iommu_enable_paging(iommu);
975 
976 out_disable_stall:
977 	rk_iommu_disable_stall(iommu);
978 out_disable_clocks:
979 	clk_bulk_disable(iommu->num_clocks, iommu->clocks);
980 	return ret;
981 }
982 
983 static void rk_iommu_detach_device(struct iommu_domain *domain,
984 				   struct device *dev)
985 {
986 	struct rk_iommu *iommu;
987 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
988 	unsigned long flags;
989 	int ret;
990 
991 	/* Allow 'virtual devices' (eg drm) to detach from domain */
992 	iommu = rk_iommu_from_dev(dev);
993 	if (!iommu)
994 		return;
995 
996 	dev_dbg(dev, "Detaching from iommu domain\n");
997 
998 	/* iommu already detached */
999 	if (iommu->domain != domain)
1000 		return;
1001 
1002 	iommu->domain = NULL;
1003 
1004 	spin_lock_irqsave(&rk_domain->iommus_lock, flags);
1005 	list_del_init(&iommu->node);
1006 	spin_unlock_irqrestore(&rk_domain->iommus_lock, flags);
1007 
1008 	ret = pm_runtime_get_if_in_use(iommu->dev);
1009 	WARN_ON_ONCE(ret < 0);
1010 	if (ret > 0) {
1011 		rk_iommu_disable(iommu);
1012 		pm_runtime_put(iommu->dev);
1013 	}
1014 }
1015 
1016 static int rk_iommu_attach_device(struct iommu_domain *domain,
1017 		struct device *dev)
1018 {
1019 	struct rk_iommu *iommu;
1020 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
1021 	unsigned long flags;
1022 	int ret;
1023 
1024 	/*
1025 	 * Allow 'virtual devices' (e.g., drm) to attach to domain.
1026 	 * Such a device does not belong to an iommu group.
1027 	 */
1028 	iommu = rk_iommu_from_dev(dev);
1029 	if (!iommu)
1030 		return 0;
1031 
1032 	dev_dbg(dev, "Attaching to iommu domain\n");
1033 
1034 	/* iommu already attached */
1035 	if (iommu->domain == domain)
1036 		return 0;
1037 
1038 	if (iommu->domain)
1039 		rk_iommu_detach_device(iommu->domain, dev);
1040 
1041 	iommu->domain = domain;
1042 
1043 	spin_lock_irqsave(&rk_domain->iommus_lock, flags);
1044 	list_add_tail(&iommu->node, &rk_domain->iommus);
1045 	spin_unlock_irqrestore(&rk_domain->iommus_lock, flags);
1046 
1047 	ret = pm_runtime_get_if_in_use(iommu->dev);
1048 	if (!ret || WARN_ON_ONCE(ret < 0))
1049 		return 0;
1050 
1051 	ret = rk_iommu_enable(iommu);
1052 	if (ret)
1053 		rk_iommu_detach_device(iommu->domain, dev);
1054 
1055 	pm_runtime_put(iommu->dev);
1056 
1057 	return ret;
1058 }
1059 
1060 static struct iommu_domain *rk_iommu_domain_alloc(unsigned type)
1061 {
1062 	struct rk_iommu_domain *rk_domain;
1063 
1064 	if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
1065 		return NULL;
1066 
1067 	if (!dma_dev)
1068 		return NULL;
1069 
1070 	rk_domain = kzalloc(sizeof(*rk_domain), GFP_KERNEL);
1071 	if (!rk_domain)
1072 		return NULL;
1073 
1074 	/*
1075 	 * rk32xx iommus use a 2 level pagetable.
1076 	 * Each level1 (dt) and level2 (pt) table has 1024 4-byte entries.
1077 	 * Allocate one 4 KiB page for each table.
1078 	 */
1079 	rk_domain->dt = (u32 *)get_zeroed_page(GFP_KERNEL | GFP_DMA32);
1080 	if (!rk_domain->dt)
1081 		goto err_free_domain;
1082 
1083 	rk_domain->dt_dma = dma_map_single(dma_dev, rk_domain->dt,
1084 					   SPAGE_SIZE, DMA_TO_DEVICE);
1085 	if (dma_mapping_error(dma_dev, rk_domain->dt_dma)) {
1086 		dev_err(dma_dev, "DMA map error for DT\n");
1087 		goto err_free_dt;
1088 	}
1089 
1090 	spin_lock_init(&rk_domain->iommus_lock);
1091 	spin_lock_init(&rk_domain->dt_lock);
1092 	INIT_LIST_HEAD(&rk_domain->iommus);
1093 
1094 	rk_domain->domain.geometry.aperture_start = 0;
1095 	rk_domain->domain.geometry.aperture_end   = DMA_BIT_MASK(32);
1096 	rk_domain->domain.geometry.force_aperture = true;
1097 
1098 	return &rk_domain->domain;
1099 
1100 err_free_dt:
1101 	free_page((unsigned long)rk_domain->dt);
1102 err_free_domain:
1103 	kfree(rk_domain);
1104 
1105 	return NULL;
1106 }
1107 
1108 static void rk_iommu_domain_free(struct iommu_domain *domain)
1109 {
1110 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
1111 	int i;
1112 
1113 	WARN_ON(!list_empty(&rk_domain->iommus));
1114 
1115 	for (i = 0; i < NUM_DT_ENTRIES; i++) {
1116 		u32 dte = rk_domain->dt[i];
1117 		if (rk_dte_is_pt_valid(dte)) {
1118 			phys_addr_t pt_phys = rk_ops->pt_address(dte);
1119 			u32 *page_table = phys_to_virt(pt_phys);
1120 			dma_unmap_single(dma_dev, pt_phys,
1121 					 SPAGE_SIZE, DMA_TO_DEVICE);
1122 			free_page((unsigned long)page_table);
1123 		}
1124 	}
1125 
1126 	dma_unmap_single(dma_dev, rk_domain->dt_dma,
1127 			 SPAGE_SIZE, DMA_TO_DEVICE);
1128 	free_page((unsigned long)rk_domain->dt);
1129 
1130 	kfree(rk_domain);
1131 }
1132 
1133 static struct iommu_device *rk_iommu_probe_device(struct device *dev)
1134 {
1135 	struct rk_iommudata *data;
1136 	struct rk_iommu *iommu;
1137 
1138 	data = dev_iommu_priv_get(dev);
1139 	if (!data)
1140 		return ERR_PTR(-ENODEV);
1141 
1142 	iommu = rk_iommu_from_dev(dev);
1143 
1144 	data->link = device_link_add(dev, iommu->dev,
1145 				     DL_FLAG_STATELESS | DL_FLAG_PM_RUNTIME);
1146 
1147 	return &iommu->iommu;
1148 }
1149 
1150 static void rk_iommu_release_device(struct device *dev)
1151 {
1152 	struct rk_iommudata *data = dev_iommu_priv_get(dev);
1153 
1154 	device_link_del(data->link);
1155 }
1156 
1157 static struct iommu_group *rk_iommu_device_group(struct device *dev)
1158 {
1159 	struct rk_iommu *iommu;
1160 
1161 	iommu = rk_iommu_from_dev(dev);
1162 
1163 	return iommu_group_ref_get(iommu->group);
1164 }
1165 
1166 static int rk_iommu_of_xlate(struct device *dev,
1167 			     struct of_phandle_args *args)
1168 {
1169 	struct platform_device *iommu_dev;
1170 	struct rk_iommudata *data;
1171 
1172 	data = devm_kzalloc(dma_dev, sizeof(*data), GFP_KERNEL);
1173 	if (!data)
1174 		return -ENOMEM;
1175 
1176 	iommu_dev = of_find_device_by_node(args->np);
1177 
1178 	data->iommu = platform_get_drvdata(iommu_dev);
1179 	dev_iommu_priv_set(dev, data);
1180 
1181 	platform_device_put(iommu_dev);
1182 
1183 	return 0;
1184 }
1185 
1186 static const struct iommu_ops rk_iommu_ops = {
1187 	.domain_alloc = rk_iommu_domain_alloc,
1188 	.probe_device = rk_iommu_probe_device,
1189 	.release_device = rk_iommu_release_device,
1190 	.device_group = rk_iommu_device_group,
1191 	.pgsize_bitmap = RK_IOMMU_PGSIZE_BITMAP,
1192 	.of_xlate = rk_iommu_of_xlate,
1193 	.default_domain_ops = &(const struct iommu_domain_ops) {
1194 		.attach_dev	= rk_iommu_attach_device,
1195 		.map		= rk_iommu_map,
1196 		.unmap		= rk_iommu_unmap,
1197 		.iova_to_phys	= rk_iommu_iova_to_phys,
1198 		.free		= rk_iommu_domain_free,
1199 	}
1200 };
1201 
1202 static int rk_iommu_probe(struct platform_device *pdev)
1203 {
1204 	struct device *dev = &pdev->dev;
1205 	struct rk_iommu *iommu;
1206 	struct resource *res;
1207 	const struct rk_iommu_ops *ops;
1208 	int num_res = pdev->num_resources;
1209 	int err, i;
1210 
1211 	iommu = devm_kzalloc(dev, sizeof(*iommu), GFP_KERNEL);
1212 	if (!iommu)
1213 		return -ENOMEM;
1214 
1215 	platform_set_drvdata(pdev, iommu);
1216 	iommu->dev = dev;
1217 	iommu->num_mmu = 0;
1218 
1219 	ops = of_device_get_match_data(dev);
1220 	if (!rk_ops)
1221 		rk_ops = ops;
1222 
1223 	/*
1224 	 * That should not happen unless different versions of the
1225 	 * hardware block are embedded the same SoC
1226 	 */
1227 	if (WARN_ON(rk_ops != ops))
1228 		return -EINVAL;
1229 
1230 	iommu->bases = devm_kcalloc(dev, num_res, sizeof(*iommu->bases),
1231 				    GFP_KERNEL);
1232 	if (!iommu->bases)
1233 		return -ENOMEM;
1234 
1235 	for (i = 0; i < num_res; i++) {
1236 		res = platform_get_resource(pdev, IORESOURCE_MEM, i);
1237 		if (!res)
1238 			continue;
1239 		iommu->bases[i] = devm_ioremap_resource(&pdev->dev, res);
1240 		if (IS_ERR(iommu->bases[i]))
1241 			continue;
1242 		iommu->num_mmu++;
1243 	}
1244 	if (iommu->num_mmu == 0)
1245 		return PTR_ERR(iommu->bases[0]);
1246 
1247 	iommu->num_irq = platform_irq_count(pdev);
1248 	if (iommu->num_irq < 0)
1249 		return iommu->num_irq;
1250 
1251 	iommu->reset_disabled = device_property_read_bool(dev,
1252 					"rockchip,disable-mmu-reset");
1253 
1254 	iommu->num_clocks = ARRAY_SIZE(rk_iommu_clocks);
1255 	iommu->clocks = devm_kcalloc(iommu->dev, iommu->num_clocks,
1256 				     sizeof(*iommu->clocks), GFP_KERNEL);
1257 	if (!iommu->clocks)
1258 		return -ENOMEM;
1259 
1260 	for (i = 0; i < iommu->num_clocks; ++i)
1261 		iommu->clocks[i].id = rk_iommu_clocks[i];
1262 
1263 	/*
1264 	 * iommu clocks should be present for all new devices and devicetrees
1265 	 * but there are older devicetrees without clocks out in the wild.
1266 	 * So clocks as optional for the time being.
1267 	 */
1268 	err = devm_clk_bulk_get(iommu->dev, iommu->num_clocks, iommu->clocks);
1269 	if (err == -ENOENT)
1270 		iommu->num_clocks = 0;
1271 	else if (err)
1272 		return err;
1273 
1274 	err = clk_bulk_prepare(iommu->num_clocks, iommu->clocks);
1275 	if (err)
1276 		return err;
1277 
1278 	iommu->group = iommu_group_alloc();
1279 	if (IS_ERR(iommu->group)) {
1280 		err = PTR_ERR(iommu->group);
1281 		goto err_unprepare_clocks;
1282 	}
1283 
1284 	err = iommu_device_sysfs_add(&iommu->iommu, dev, NULL, dev_name(dev));
1285 	if (err)
1286 		goto err_put_group;
1287 
1288 	err = iommu_device_register(&iommu->iommu, &rk_iommu_ops, dev);
1289 	if (err)
1290 		goto err_remove_sysfs;
1291 
1292 	/*
1293 	 * Use the first registered IOMMU device for domain to use with DMA
1294 	 * API, since a domain might not physically correspond to a single
1295 	 * IOMMU device..
1296 	 */
1297 	if (!dma_dev)
1298 		dma_dev = &pdev->dev;
1299 
1300 	pm_runtime_enable(dev);
1301 
1302 	for (i = 0; i < iommu->num_irq; i++) {
1303 		int irq = platform_get_irq(pdev, i);
1304 
1305 		if (irq < 0)
1306 			return irq;
1307 
1308 		err = devm_request_irq(iommu->dev, irq, rk_iommu_irq,
1309 				       IRQF_SHARED, dev_name(dev), iommu);
1310 		if (err) {
1311 			pm_runtime_disable(dev);
1312 			goto err_remove_sysfs;
1313 		}
1314 	}
1315 
1316 	dma_set_mask_and_coherent(dev, rk_ops->dma_bit_mask);
1317 
1318 	return 0;
1319 err_remove_sysfs:
1320 	iommu_device_sysfs_remove(&iommu->iommu);
1321 err_put_group:
1322 	iommu_group_put(iommu->group);
1323 err_unprepare_clocks:
1324 	clk_bulk_unprepare(iommu->num_clocks, iommu->clocks);
1325 	return err;
1326 }
1327 
1328 static void rk_iommu_shutdown(struct platform_device *pdev)
1329 {
1330 	struct rk_iommu *iommu = platform_get_drvdata(pdev);
1331 	int i;
1332 
1333 	for (i = 0; i < iommu->num_irq; i++) {
1334 		int irq = platform_get_irq(pdev, i);
1335 
1336 		devm_free_irq(iommu->dev, irq, iommu);
1337 	}
1338 
1339 	pm_runtime_force_suspend(&pdev->dev);
1340 }
1341 
1342 static int __maybe_unused rk_iommu_suspend(struct device *dev)
1343 {
1344 	struct rk_iommu *iommu = dev_get_drvdata(dev);
1345 
1346 	if (!iommu->domain)
1347 		return 0;
1348 
1349 	rk_iommu_disable(iommu);
1350 	return 0;
1351 }
1352 
1353 static int __maybe_unused rk_iommu_resume(struct device *dev)
1354 {
1355 	struct rk_iommu *iommu = dev_get_drvdata(dev);
1356 
1357 	if (!iommu->domain)
1358 		return 0;
1359 
1360 	return rk_iommu_enable(iommu);
1361 }
1362 
1363 static const struct dev_pm_ops rk_iommu_pm_ops = {
1364 	SET_RUNTIME_PM_OPS(rk_iommu_suspend, rk_iommu_resume, NULL)
1365 	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1366 				pm_runtime_force_resume)
1367 };
1368 
1369 static struct rk_iommu_ops iommu_data_ops_v1 = {
1370 	.pt_address = &rk_dte_pt_address,
1371 	.mk_dtentries = &rk_mk_dte,
1372 	.mk_ptentries = &rk_mk_pte,
1373 	.dte_addr_phys = &rk_dte_addr_phys,
1374 	.dma_addr_dte = &rk_dma_addr_dte,
1375 	.dma_bit_mask = DMA_BIT_MASK(32),
1376 };
1377 
1378 static struct rk_iommu_ops iommu_data_ops_v2 = {
1379 	.pt_address = &rk_dte_pt_address_v2,
1380 	.mk_dtentries = &rk_mk_dte_v2,
1381 	.mk_ptentries = &rk_mk_pte_v2,
1382 	.dte_addr_phys = &rk_dte_addr_phys_v2,
1383 	.dma_addr_dte = &rk_dma_addr_dte_v2,
1384 	.dma_bit_mask = DMA_BIT_MASK(40),
1385 };
1386 
1387 static const struct of_device_id rk_iommu_dt_ids[] = {
1388 	{	.compatible = "rockchip,iommu",
1389 		.data = &iommu_data_ops_v1,
1390 	},
1391 	{	.compatible = "rockchip,rk3568-iommu",
1392 		.data = &iommu_data_ops_v2,
1393 	},
1394 	{ /* sentinel */ }
1395 };
1396 
1397 static struct platform_driver rk_iommu_driver = {
1398 	.probe = rk_iommu_probe,
1399 	.shutdown = rk_iommu_shutdown,
1400 	.driver = {
1401 		   .name = "rk_iommu",
1402 		   .of_match_table = rk_iommu_dt_ids,
1403 		   .pm = &rk_iommu_pm_ops,
1404 		   .suppress_bind_attrs = true,
1405 	},
1406 };
1407 builtin_platform_driver(rk_iommu_driver);
1408