xref: /linux/drivers/iommu/rockchip-iommu.c (revision 0cc6f45cecb46cefe89c17ec816dc8cd58a2229a)
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 #include "iommu-pages.h"
30 
31 /** MMU register offsets */
32 #define RK_MMU_DTE_ADDR		0x00	/* Directory table address */
33 #define RK_MMU_STATUS		0x04
34 #define RK_MMU_COMMAND		0x08
35 #define RK_MMU_PAGE_FAULT_ADDR	0x0C	/* IOVA of last page fault */
36 #define RK_MMU_ZAP_ONE_LINE	0x10	/* Shootdown one IOTLB entry */
37 #define RK_MMU_INT_RAWSTAT	0x14	/* IRQ status ignoring mask */
38 #define RK_MMU_INT_CLEAR	0x18	/* Acknowledge and re-arm irq */
39 #define RK_MMU_INT_MASK		0x1C	/* IRQ enable */
40 #define RK_MMU_INT_STATUS	0x20	/* IRQ status after masking */
41 #define RK_MMU_AUTO_GATING	0x24
42 
43 #define DTE_ADDR_DUMMY		0xCAFEBABE
44 
45 #define RK_MMU_POLL_PERIOD_US		100
46 #define RK_MMU_FORCE_RESET_TIMEOUT_US	100000
47 #define RK_MMU_POLL_TIMEOUT_US		1000
48 
49 /* RK_MMU_STATUS fields */
50 #define RK_MMU_STATUS_PAGING_ENABLED       BIT(0)
51 #define RK_MMU_STATUS_PAGE_FAULT_ACTIVE    BIT(1)
52 #define RK_MMU_STATUS_STALL_ACTIVE         BIT(2)
53 #define RK_MMU_STATUS_IDLE                 BIT(3)
54 #define RK_MMU_STATUS_REPLAY_BUFFER_EMPTY  BIT(4)
55 #define RK_MMU_STATUS_PAGE_FAULT_IS_WRITE  BIT(5)
56 #define RK_MMU_STATUS_STALL_NOT_ACTIVE     BIT(31)
57 
58 /* RK_MMU_COMMAND command values */
59 #define RK_MMU_CMD_ENABLE_PAGING    0  /* Enable memory translation */
60 #define RK_MMU_CMD_DISABLE_PAGING   1  /* Disable memory translation */
61 #define RK_MMU_CMD_ENABLE_STALL     2  /* Stall paging to allow other cmds */
62 #define RK_MMU_CMD_DISABLE_STALL    3  /* Stop stall re-enables paging */
63 #define RK_MMU_CMD_ZAP_CACHE        4  /* Shoot down entire IOTLB */
64 #define RK_MMU_CMD_PAGE_FAULT_DONE  5  /* Clear page fault */
65 #define RK_MMU_CMD_FORCE_RESET      6  /* Reset all registers */
66 
67 /* RK_MMU_INT_* register fields */
68 #define RK_MMU_IRQ_PAGE_FAULT    0x01  /* page fault */
69 #define RK_MMU_IRQ_BUS_ERROR     0x02  /* bus read error */
70 #define RK_MMU_IRQ_MASK          (RK_MMU_IRQ_PAGE_FAULT | RK_MMU_IRQ_BUS_ERROR)
71 
72 #define NUM_DT_ENTRIES 1024
73 #define NUM_PT_ENTRIES 1024
74 
75 #define SPAGE_ORDER 12
76 #define SPAGE_SIZE (1 << SPAGE_ORDER)
77 
78  /*
79   * Support mapping any size that fits in one page table:
80   *   4 KiB to 4 MiB
81   */
82 #define RK_IOMMU_PGSIZE_BITMAP 0x007ff000
83 
84 struct rk_iommu_domain {
85 	struct list_head iommus;
86 	u32 *dt; /* page directory table */
87 	dma_addr_t dt_dma;
88 	spinlock_t iommus_lock; /* lock for iommus list */
89 	spinlock_t dt_lock; /* lock for modifying page directory table */
90 
91 	struct iommu_domain domain;
92 };
93 
94 /* list of clocks required by IOMMU */
95 static const char * const rk_iommu_clocks[] = {
96 	"aclk", "iface",
97 };
98 
99 struct rk_iommu_ops {
100 	phys_addr_t (*pt_address)(u32 dte);
101 	u32 (*mk_dtentries)(dma_addr_t pt_dma);
102 	u32 (*mk_ptentries)(phys_addr_t page, int prot);
103 	u64 dma_bit_mask;
104 	gfp_t gfp_flags;
105 };
106 
107 struct rk_iommu {
108 	struct device *dev;
109 	void __iomem **bases;
110 	int num_mmu;
111 	int num_irq;
112 	struct clk_bulk_data *clocks;
113 	int num_clocks;
114 	bool reset_disabled;
115 	struct iommu_device iommu;
116 	struct list_head node; /* entry in rk_iommu_domain.iommus */
117 	struct iommu_domain *domain; /* domain to which iommu is attached */
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 static struct iommu_domain rk_identity_domain;
128 
rk_table_flush(struct rk_iommu_domain * dom,dma_addr_t dma,unsigned int count)129 static inline void rk_table_flush(struct rk_iommu_domain *dom, dma_addr_t dma,
130 				  unsigned int count)
131 {
132 	size_t size = count * sizeof(u32); /* count of u32 entry */
133 
134 	dma_sync_single_for_device(dma_dev, dma, size, DMA_TO_DEVICE);
135 }
136 
to_rk_domain(struct iommu_domain * dom)137 static struct rk_iommu_domain *to_rk_domain(struct iommu_domain *dom)
138 {
139 	return container_of(dom, struct rk_iommu_domain, domain);
140 }
141 
142 /*
143  * The Rockchip rk3288 iommu uses a 2-level page table.
144  * The first level is the "Directory Table" (DT).
145  * The DT consists of 1024 4-byte Directory Table Entries (DTEs), each pointing
146  * to a "Page Table".
147  * The second level is the 1024 Page Tables (PT).
148  * Each PT consists of 1024 4-byte Page Table Entries (PTEs), each pointing to
149  * a 4 KB page of physical memory.
150  *
151  * The DT and each PT fits in a single 4 KB page (4-bytes * 1024 entries).
152  * Each iommu device has a MMU_DTE_ADDR register that contains the physical
153  * address of the start of the DT page.
154  *
155  * The structure of the page table is as follows:
156  *
157  *                   DT
158  * MMU_DTE_ADDR -> +-----+
159  *                 |     |
160  *                 +-----+     PT
161  *                 | DTE | -> +-----+
162  *                 +-----+    |     |     Memory
163  *                 |     |    +-----+     Page
164  *                 |     |    | PTE | -> +-----+
165  *                 +-----+    +-----+    |     |
166  *                            |     |    |     |
167  *                            |     |    |     |
168  *                            +-----+    |     |
169  *                                       |     |
170  *                                       |     |
171  *                                       +-----+
172  */
173 
174 /*
175  * Each DTE has a PT address and a valid bit:
176  * +---------------------+-----------+-+
177  * | PT address          | Reserved  |V|
178  * +---------------------+-----------+-+
179  *  31:12 - PT address (PTs always starts on a 4 KB boundary)
180  *  11: 1 - Reserved
181  *      0 - 1 if PT @ PT address is valid
182  */
183 #define RK_DTE_PT_ADDRESS_MASK    0xfffff000
184 #define RK_DTE_PT_VALID           BIT(0)
185 
rk_dte_pt_address(u32 dte)186 static inline phys_addr_t rk_dte_pt_address(u32 dte)
187 {
188 	return (phys_addr_t)dte & RK_DTE_PT_ADDRESS_MASK;
189 }
190 
191 /*
192  * In v2:
193  * 31:12 - PT address bit 31:0
194  * 11: 8 - PT address bit 35:32
195  *  7: 4 - PT address bit 39:36
196  *  3: 1 - Reserved
197  *     0 - 1 if PT @ PT address is valid
198  */
199 #define RK_DTE_PT_ADDRESS_MASK_V2 GENMASK_ULL(31, 4)
200 #define DTE_HI_MASK1	GENMASK(11, 8)
201 #define DTE_HI_MASK2	GENMASK(7, 4)
202 #define DTE_HI_SHIFT1	24 /* shift bit 8 to bit 32 */
203 #define DTE_HI_SHIFT2	32 /* shift bit 4 to bit 36 */
204 #define PAGE_DESC_HI_MASK1	GENMASK_ULL(35, 32)
205 #define PAGE_DESC_HI_MASK2	GENMASK_ULL(39, 36)
206 
rk_dte_pt_address_v2(u32 dte)207 static inline phys_addr_t rk_dte_pt_address_v2(u32 dte)
208 {
209 	u64 dte_v2 = dte;
210 
211 	dte_v2 = ((dte_v2 & DTE_HI_MASK2) << DTE_HI_SHIFT2) |
212 		 ((dte_v2 & DTE_HI_MASK1) << DTE_HI_SHIFT1) |
213 		 (dte_v2 & RK_DTE_PT_ADDRESS_MASK);
214 
215 	return (phys_addr_t)dte_v2;
216 }
217 
rk_dte_is_pt_valid(u32 dte)218 static inline bool rk_dte_is_pt_valid(u32 dte)
219 {
220 	return dte & RK_DTE_PT_VALID;
221 }
222 
rk_mk_dte(dma_addr_t pt_dma)223 static inline u32 rk_mk_dte(dma_addr_t pt_dma)
224 {
225 	return (pt_dma & RK_DTE_PT_ADDRESS_MASK) | RK_DTE_PT_VALID;
226 }
227 
rk_mk_dte_v2(dma_addr_t pt_dma)228 static inline u32 rk_mk_dte_v2(dma_addr_t pt_dma)
229 {
230 	pt_dma = (pt_dma & RK_DTE_PT_ADDRESS_MASK) |
231 		 ((pt_dma & PAGE_DESC_HI_MASK1) >> DTE_HI_SHIFT1) |
232 		 (pt_dma & PAGE_DESC_HI_MASK2) >> DTE_HI_SHIFT2;
233 
234 	return (pt_dma & RK_DTE_PT_ADDRESS_MASK_V2) | RK_DTE_PT_VALID;
235 }
236 
237 /*
238  * Each PTE has a Page address, some flags and a valid bit:
239  * +---------------------+---+-------+-+
240  * | Page address        |Rsv| Flags |V|
241  * +---------------------+---+-------+-+
242  *  31:12 - Page address (Pages always start on a 4 KB boundary)
243  *  11: 9 - Reserved
244  *   8: 1 - Flags
245  *      8 - Read allocate - allocate cache space on read misses
246  *      7 - Read cache - enable cache & prefetch of data
247  *      6 - Write buffer - enable delaying writes on their way to memory
248  *      5 - Write allocate - allocate cache space on write misses
249  *      4 - Write cache - different writes can be merged together
250  *      3 - Override cache attributes
251  *          if 1, bits 4-8 control cache attributes
252  *          if 0, the system bus defaults are used
253  *      2 - Writable
254  *      1 - Readable
255  *      0 - 1 if Page @ Page address is valid
256  */
257 #define RK_PTE_PAGE_ADDRESS_MASK  0xfffff000
258 #define RK_PTE_PAGE_FLAGS_MASK    0x000001fe
259 #define RK_PTE_PAGE_WRITABLE      BIT(2)
260 #define RK_PTE_PAGE_READABLE      BIT(1)
261 #define RK_PTE_PAGE_VALID         BIT(0)
262 
rk_pte_is_page_valid(u32 pte)263 static inline bool rk_pte_is_page_valid(u32 pte)
264 {
265 	return pte & RK_PTE_PAGE_VALID;
266 }
267 
268 /* TODO: set cache flags per prot IOMMU_CACHE */
rk_mk_pte(phys_addr_t page,int prot)269 static u32 rk_mk_pte(phys_addr_t page, int prot)
270 {
271 	u32 flags = 0;
272 	flags |= (prot & IOMMU_READ) ? RK_PTE_PAGE_READABLE : 0;
273 	flags |= (prot & IOMMU_WRITE) ? RK_PTE_PAGE_WRITABLE : 0;
274 	page &= RK_PTE_PAGE_ADDRESS_MASK;
275 	return page | flags | RK_PTE_PAGE_VALID;
276 }
277 
278 /*
279  * In v2:
280  * 31:12 - Page address bit 31:0
281  * 11: 8 - Page address bit 35:32
282  *  7: 4 - Page address bit 39:36
283  *     3 - Security
284  *     2 - Writable
285  *     1 - Readable
286  *     0 - 1 if Page @ Page address is valid
287  */
288 
rk_mk_pte_v2(phys_addr_t page,int prot)289 static u32 rk_mk_pte_v2(phys_addr_t page, int prot)
290 {
291 	u32 flags = 0;
292 
293 	flags |= (prot & IOMMU_READ) ? RK_PTE_PAGE_READABLE : 0;
294 	flags |= (prot & IOMMU_WRITE) ? RK_PTE_PAGE_WRITABLE : 0;
295 
296 	return rk_mk_dte_v2(page) | flags;
297 }
298 
rk_mk_pte_invalid(u32 pte)299 static u32 rk_mk_pte_invalid(u32 pte)
300 {
301 	return pte & ~RK_PTE_PAGE_VALID;
302 }
303 
304 /*
305  * rk3288 iova (IOMMU Virtual Address) format
306  *  31       22.21       12.11          0
307  * +-----------+-----------+-------------+
308  * | DTE index | PTE index | Page offset |
309  * +-----------+-----------+-------------+
310  *  31:22 - DTE index   - index of DTE in DT
311  *  21:12 - PTE index   - index of PTE in PT @ DTE.pt_address
312  *  11: 0 - Page offset - offset into page @ PTE.page_address
313  */
314 #define RK_IOVA_DTE_MASK    0xffc00000
315 #define RK_IOVA_DTE_SHIFT   22
316 #define RK_IOVA_PTE_MASK    0x003ff000
317 #define RK_IOVA_PTE_SHIFT   12
318 #define RK_IOVA_PAGE_MASK   0x00000fff
319 #define RK_IOVA_PAGE_SHIFT  0
320 
rk_iova_dte_index(dma_addr_t iova)321 static u32 rk_iova_dte_index(dma_addr_t iova)
322 {
323 	return (u32)(iova & RK_IOVA_DTE_MASK) >> RK_IOVA_DTE_SHIFT;
324 }
325 
rk_iova_pte_index(dma_addr_t iova)326 static u32 rk_iova_pte_index(dma_addr_t iova)
327 {
328 	return (u32)(iova & RK_IOVA_PTE_MASK) >> RK_IOVA_PTE_SHIFT;
329 }
330 
rk_iova_page_offset(dma_addr_t iova)331 static u32 rk_iova_page_offset(dma_addr_t iova)
332 {
333 	return (u32)(iova & RK_IOVA_PAGE_MASK) >> RK_IOVA_PAGE_SHIFT;
334 }
335 
rk_iommu_read(void __iomem * base,u32 offset)336 static u32 rk_iommu_read(void __iomem *base, u32 offset)
337 {
338 	return readl(base + offset);
339 }
340 
rk_iommu_write(void __iomem * base,u32 offset,u32 value)341 static void rk_iommu_write(void __iomem *base, u32 offset, u32 value)
342 {
343 	writel(value, base + offset);
344 }
345 
rk_iommu_command(struct rk_iommu * iommu,u32 command)346 static void rk_iommu_command(struct rk_iommu *iommu, u32 command)
347 {
348 	int i;
349 
350 	for (i = 0; i < iommu->num_mmu; i++)
351 		writel(command, iommu->bases[i] + RK_MMU_COMMAND);
352 }
353 
rk_iommu_base_command(void __iomem * base,u32 command)354 static void rk_iommu_base_command(void __iomem *base, u32 command)
355 {
356 	writel(command, base + RK_MMU_COMMAND);
357 }
rk_iommu_zap_lines(struct rk_iommu * iommu,dma_addr_t iova_start,size_t size)358 static void rk_iommu_zap_lines(struct rk_iommu *iommu, dma_addr_t iova_start,
359 			       size_t size)
360 {
361 	int i;
362 	dma_addr_t iova_end = iova_start + size;
363 	/*
364 	 * TODO(djkurtz): Figure out when it is more efficient to shootdown the
365 	 * entire iotlb rather than iterate over individual iovas.
366 	 */
367 	for (i = 0; i < iommu->num_mmu; i++) {
368 		dma_addr_t iova;
369 
370 		for (iova = iova_start; iova < iova_end; iova += SPAGE_SIZE)
371 			rk_iommu_write(iommu->bases[i], RK_MMU_ZAP_ONE_LINE, iova);
372 	}
373 }
374 
rk_iommu_is_stall_active(struct rk_iommu * iommu)375 static bool rk_iommu_is_stall_active(struct rk_iommu *iommu)
376 {
377 	bool active = true;
378 	int i;
379 
380 	for (i = 0; i < iommu->num_mmu; i++)
381 		active &= !!(rk_iommu_read(iommu->bases[i], RK_MMU_STATUS) &
382 					   RK_MMU_STATUS_STALL_ACTIVE);
383 
384 	return active;
385 }
386 
rk_iommu_is_paging_enabled(struct rk_iommu * iommu)387 static bool rk_iommu_is_paging_enabled(struct rk_iommu *iommu)
388 {
389 	bool enable = true;
390 	int i;
391 
392 	for (i = 0; i < iommu->num_mmu; i++)
393 		enable &= !!(rk_iommu_read(iommu->bases[i], RK_MMU_STATUS) &
394 					   RK_MMU_STATUS_PAGING_ENABLED);
395 
396 	return enable;
397 }
398 
rk_iommu_is_reset_done(struct rk_iommu * iommu)399 static bool rk_iommu_is_reset_done(struct rk_iommu *iommu)
400 {
401 	bool done = true;
402 	int i;
403 
404 	for (i = 0; i < iommu->num_mmu; i++)
405 		done &= rk_iommu_read(iommu->bases[i], RK_MMU_DTE_ADDR) == 0;
406 
407 	return done;
408 }
409 
rk_iommu_enable_stall(struct rk_iommu * iommu)410 static int rk_iommu_enable_stall(struct rk_iommu *iommu)
411 {
412 	int ret, i;
413 	bool val;
414 
415 	if (rk_iommu_is_stall_active(iommu))
416 		return 0;
417 
418 	/* Stall can only be enabled if paging is enabled */
419 	if (!rk_iommu_is_paging_enabled(iommu))
420 		return 0;
421 
422 	rk_iommu_command(iommu, RK_MMU_CMD_ENABLE_STALL);
423 
424 	ret = readx_poll_timeout(rk_iommu_is_stall_active, iommu, val,
425 				 val, RK_MMU_POLL_PERIOD_US,
426 				 RK_MMU_POLL_TIMEOUT_US);
427 	if (ret)
428 		for (i = 0; i < iommu->num_mmu; i++)
429 			dev_err(iommu->dev, "Enable stall request timed out, status: %#08x\n",
430 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
431 
432 	return ret;
433 }
434 
rk_iommu_disable_stall(struct rk_iommu * iommu)435 static int rk_iommu_disable_stall(struct rk_iommu *iommu)
436 {
437 	int ret, i;
438 	bool val;
439 
440 	if (!rk_iommu_is_stall_active(iommu))
441 		return 0;
442 
443 	rk_iommu_command(iommu, RK_MMU_CMD_DISABLE_STALL);
444 
445 	ret = readx_poll_timeout(rk_iommu_is_stall_active, iommu, val,
446 				 !val, RK_MMU_POLL_PERIOD_US,
447 				 RK_MMU_POLL_TIMEOUT_US);
448 	if (ret)
449 		for (i = 0; i < iommu->num_mmu; i++)
450 			dev_err(iommu->dev, "Disable stall request timed out, status: %#08x\n",
451 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
452 
453 	return ret;
454 }
455 
rk_iommu_enable_paging(struct rk_iommu * iommu)456 static int rk_iommu_enable_paging(struct rk_iommu *iommu)
457 {
458 	int ret, i;
459 	bool val;
460 
461 	if (rk_iommu_is_paging_enabled(iommu))
462 		return 0;
463 
464 	rk_iommu_command(iommu, RK_MMU_CMD_ENABLE_PAGING);
465 
466 	ret = readx_poll_timeout(rk_iommu_is_paging_enabled, iommu, val,
467 				 val, RK_MMU_POLL_PERIOD_US,
468 				 RK_MMU_POLL_TIMEOUT_US);
469 	if (ret)
470 		for (i = 0; i < iommu->num_mmu; i++)
471 			dev_err(iommu->dev, "Enable paging request timed out, status: %#08x\n",
472 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
473 
474 	return ret;
475 }
476 
rk_iommu_disable_paging(struct rk_iommu * iommu)477 static int rk_iommu_disable_paging(struct rk_iommu *iommu)
478 {
479 	int ret, i;
480 	bool val;
481 
482 	if (!rk_iommu_is_paging_enabled(iommu))
483 		return 0;
484 
485 	rk_iommu_command(iommu, RK_MMU_CMD_DISABLE_PAGING);
486 
487 	ret = readx_poll_timeout(rk_iommu_is_paging_enabled, iommu, val,
488 				 !val, RK_MMU_POLL_PERIOD_US,
489 				 RK_MMU_POLL_TIMEOUT_US);
490 	if (ret)
491 		for (i = 0; i < iommu->num_mmu; i++)
492 			dev_err(iommu->dev, "Disable paging request timed out, status: %#08x\n",
493 				rk_iommu_read(iommu->bases[i], RK_MMU_STATUS));
494 
495 	return ret;
496 }
497 
rk_iommu_force_reset(struct rk_iommu * iommu)498 static int rk_iommu_force_reset(struct rk_iommu *iommu)
499 {
500 	int ret, i;
501 	u32 dte_addr;
502 	bool val;
503 
504 	if (iommu->reset_disabled)
505 		return 0;
506 
507 	/*
508 	 * Check if register DTE_ADDR is working by writing DTE_ADDR_DUMMY
509 	 * and verifying that upper 5 (v1) or 7 (v2) nybbles are read back.
510 	 */
511 	for (i = 0; i < iommu->num_mmu; i++) {
512 		dte_addr = rk_ops->pt_address(DTE_ADDR_DUMMY);
513 		rk_iommu_write(iommu->bases[i], RK_MMU_DTE_ADDR, dte_addr);
514 
515 		if (dte_addr != rk_iommu_read(iommu->bases[i], RK_MMU_DTE_ADDR)) {
516 			dev_err(iommu->dev, "Error during raw reset. MMU_DTE_ADDR is not functioning\n");
517 			return -EFAULT;
518 		}
519 	}
520 
521 	rk_iommu_command(iommu, RK_MMU_CMD_FORCE_RESET);
522 
523 	ret = readx_poll_timeout(rk_iommu_is_reset_done, iommu, val,
524 				 val, RK_MMU_FORCE_RESET_TIMEOUT_US,
525 				 RK_MMU_POLL_TIMEOUT_US);
526 	if (ret) {
527 		dev_err(iommu->dev, "FORCE_RESET command timed out\n");
528 		return ret;
529 	}
530 
531 	return 0;
532 }
533 
log_iova(struct rk_iommu * iommu,int index,dma_addr_t iova)534 static void log_iova(struct rk_iommu *iommu, int index, dma_addr_t iova)
535 {
536 	void __iomem *base = iommu->bases[index];
537 	u32 dte_index, pte_index, page_offset;
538 	u32 mmu_dte_addr;
539 	phys_addr_t mmu_dte_addr_phys, dte_addr_phys;
540 	u32 *dte_addr;
541 	u32 dte;
542 	phys_addr_t pte_addr_phys = 0;
543 	u32 *pte_addr = NULL;
544 	u32 pte = 0;
545 	phys_addr_t page_addr_phys = 0;
546 	u32 page_flags = 0;
547 
548 	dte_index = rk_iova_dte_index(iova);
549 	pte_index = rk_iova_pte_index(iova);
550 	page_offset = rk_iova_page_offset(iova);
551 
552 	mmu_dte_addr = rk_iommu_read(base, RK_MMU_DTE_ADDR);
553 	mmu_dte_addr_phys = rk_ops->pt_address(mmu_dte_addr);
554 
555 	dte_addr_phys = mmu_dte_addr_phys + (4 * dte_index);
556 	dte_addr = phys_to_virt(dte_addr_phys);
557 	dte = *dte_addr;
558 
559 	if (!rk_dte_is_pt_valid(dte))
560 		goto print_it;
561 
562 	pte_addr_phys = rk_ops->pt_address(dte) + (pte_index * 4);
563 	pte_addr = phys_to_virt(pte_addr_phys);
564 	pte = *pte_addr;
565 
566 	if (!rk_pte_is_page_valid(pte))
567 		goto print_it;
568 
569 	page_addr_phys = rk_ops->pt_address(pte) + page_offset;
570 	page_flags = pte & RK_PTE_PAGE_FLAGS_MASK;
571 
572 print_it:
573 	dev_err(iommu->dev, "iova = %pad: dte_index: %#03x pte_index: %#03x page_offset: %#03x\n",
574 		&iova, dte_index, pte_index, page_offset);
575 	dev_err(iommu->dev, "mmu_dte_addr: %pa dte@%pa: %#08x valid: %u pte@%pa: %#08x valid: %u page@%pa flags: %#03x\n",
576 		&mmu_dte_addr_phys, &dte_addr_phys, dte,
577 		rk_dte_is_pt_valid(dte), &pte_addr_phys, pte,
578 		rk_pte_is_page_valid(pte), &page_addr_phys, page_flags);
579 }
580 
rk_iommu_irq(int irq,void * dev_id)581 static irqreturn_t rk_iommu_irq(int irq, void *dev_id)
582 {
583 	struct rk_iommu *iommu = dev_id;
584 	u32 status;
585 	u32 int_status;
586 	dma_addr_t iova;
587 	irqreturn_t ret = IRQ_NONE;
588 	int i, err;
589 
590 	err = pm_runtime_get_if_in_use(iommu->dev);
591 	if (!err || WARN_ON_ONCE(err < 0))
592 		return ret;
593 
594 	if (WARN_ON(clk_bulk_enable(iommu->num_clocks, iommu->clocks)))
595 		goto out;
596 
597 	for (i = 0; i < iommu->num_mmu; i++) {
598 		int_status = rk_iommu_read(iommu->bases[i], RK_MMU_INT_STATUS);
599 		if (int_status == 0)
600 			continue;
601 
602 		ret = IRQ_HANDLED;
603 		iova = rk_iommu_read(iommu->bases[i], RK_MMU_PAGE_FAULT_ADDR);
604 
605 		if (int_status & RK_MMU_IRQ_PAGE_FAULT) {
606 			int flags;
607 
608 			status = rk_iommu_read(iommu->bases[i], RK_MMU_STATUS);
609 			flags = (status & RK_MMU_STATUS_PAGE_FAULT_IS_WRITE) ?
610 					IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
611 
612 			dev_err(iommu->dev, "Page fault at %pad of type %s\n",
613 				&iova,
614 				(flags == IOMMU_FAULT_WRITE) ? "write" : "read");
615 
616 			log_iova(iommu, i, iova);
617 
618 			/*
619 			 * Report page fault to any installed handlers.
620 			 * Ignore the return code, though, since we always zap cache
621 			 * and clear the page fault anyway.
622 			 */
623 			if (iommu->domain != &rk_identity_domain)
624 				report_iommu_fault(iommu->domain, iommu->dev, iova,
625 						   flags);
626 			else
627 				dev_err(iommu->dev, "Page fault while iommu not attached to domain?\n");
628 
629 			rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_ZAP_CACHE);
630 			rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_PAGE_FAULT_DONE);
631 		}
632 
633 		if (int_status & RK_MMU_IRQ_BUS_ERROR)
634 			dev_err(iommu->dev, "BUS_ERROR occurred at %pad\n", &iova);
635 
636 		if (int_status & ~RK_MMU_IRQ_MASK)
637 			dev_err(iommu->dev, "unexpected int_status: %#08x\n",
638 				int_status);
639 
640 		rk_iommu_write(iommu->bases[i], RK_MMU_INT_CLEAR, int_status);
641 	}
642 
643 	clk_bulk_disable(iommu->num_clocks, iommu->clocks);
644 
645 out:
646 	pm_runtime_put(iommu->dev);
647 	return ret;
648 }
649 
rk_iommu_iova_to_phys(struct iommu_domain * domain,dma_addr_t iova)650 static phys_addr_t rk_iommu_iova_to_phys(struct iommu_domain *domain,
651 					 dma_addr_t iova)
652 {
653 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
654 	unsigned long flags;
655 	phys_addr_t pt_phys, phys = 0;
656 	u32 dte, pte;
657 	u32 *page_table;
658 
659 	spin_lock_irqsave(&rk_domain->dt_lock, flags);
660 
661 	dte = rk_domain->dt[rk_iova_dte_index(iova)];
662 	if (!rk_dte_is_pt_valid(dte))
663 		goto out;
664 
665 	pt_phys = rk_ops->pt_address(dte);
666 	page_table = (u32 *)phys_to_virt(pt_phys);
667 	pte = page_table[rk_iova_pte_index(iova)];
668 	if (!rk_pte_is_page_valid(pte))
669 		goto out;
670 
671 	phys = rk_ops->pt_address(pte) + rk_iova_page_offset(iova);
672 out:
673 	spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
674 
675 	return phys;
676 }
677 
rk_iommu_zap_iova(struct rk_iommu_domain * rk_domain,dma_addr_t iova,size_t size)678 static void rk_iommu_zap_iova(struct rk_iommu_domain *rk_domain,
679 			      dma_addr_t iova, size_t size)
680 {
681 	struct list_head *pos;
682 	unsigned long flags;
683 
684 	/* shootdown these iova from all iommus using this domain */
685 	spin_lock_irqsave(&rk_domain->iommus_lock, flags);
686 	list_for_each(pos, &rk_domain->iommus) {
687 		struct rk_iommu *iommu;
688 		int ret;
689 
690 		iommu = list_entry(pos, struct rk_iommu, node);
691 
692 		/* Only zap TLBs of IOMMUs that are powered on. */
693 		ret = pm_runtime_get_if_in_use(iommu->dev);
694 		if (WARN_ON_ONCE(ret < 0))
695 			continue;
696 		if (ret) {
697 			WARN_ON(clk_bulk_enable(iommu->num_clocks,
698 						iommu->clocks));
699 			rk_iommu_zap_lines(iommu, iova, size);
700 			clk_bulk_disable(iommu->num_clocks, iommu->clocks);
701 			pm_runtime_put(iommu->dev);
702 		}
703 	}
704 	spin_unlock_irqrestore(&rk_domain->iommus_lock, flags);
705 }
706 
rk_iommu_zap_iova_first_last(struct rk_iommu_domain * rk_domain,dma_addr_t iova,size_t size)707 static void rk_iommu_zap_iova_first_last(struct rk_iommu_domain *rk_domain,
708 					 dma_addr_t iova, size_t size)
709 {
710 	rk_iommu_zap_iova(rk_domain, iova, SPAGE_SIZE);
711 	if (size > SPAGE_SIZE)
712 		rk_iommu_zap_iova(rk_domain, iova + size - SPAGE_SIZE,
713 					SPAGE_SIZE);
714 }
715 
rk_dte_get_page_table(struct rk_iommu_domain * rk_domain,dma_addr_t iova)716 static u32 *rk_dte_get_page_table(struct rk_iommu_domain *rk_domain,
717 				  dma_addr_t iova)
718 {
719 	u32 *page_table, *dte_addr;
720 	u32 dte_index, dte;
721 	phys_addr_t pt_phys;
722 	dma_addr_t pt_dma;
723 
724 	assert_spin_locked(&rk_domain->dt_lock);
725 
726 	dte_index = rk_iova_dte_index(iova);
727 	dte_addr = &rk_domain->dt[dte_index];
728 	dte = *dte_addr;
729 	if (rk_dte_is_pt_valid(dte))
730 		goto done;
731 
732 	page_table = iommu_alloc_page(GFP_ATOMIC | rk_ops->gfp_flags);
733 	if (!page_table)
734 		return ERR_PTR(-ENOMEM);
735 
736 	pt_dma = dma_map_single(dma_dev, page_table, SPAGE_SIZE, DMA_TO_DEVICE);
737 	if (dma_mapping_error(dma_dev, pt_dma)) {
738 		dev_err(dma_dev, "DMA mapping error while allocating page table\n");
739 		iommu_free_page(page_table);
740 		return ERR_PTR(-ENOMEM);
741 	}
742 
743 	dte = rk_ops->mk_dtentries(pt_dma);
744 	*dte_addr = dte;
745 
746 	rk_table_flush(rk_domain,
747 		       rk_domain->dt_dma + dte_index * sizeof(u32), 1);
748 done:
749 	pt_phys = rk_ops->pt_address(dte);
750 	return (u32 *)phys_to_virt(pt_phys);
751 }
752 
rk_iommu_unmap_iova(struct rk_iommu_domain * rk_domain,u32 * pte_addr,dma_addr_t pte_dma,size_t size)753 static size_t rk_iommu_unmap_iova(struct rk_iommu_domain *rk_domain,
754 				  u32 *pte_addr, dma_addr_t pte_dma,
755 				  size_t size)
756 {
757 	unsigned int pte_count;
758 	unsigned int pte_total = size / SPAGE_SIZE;
759 
760 	assert_spin_locked(&rk_domain->dt_lock);
761 
762 	for (pte_count = 0; pte_count < pte_total; pte_count++) {
763 		u32 pte = pte_addr[pte_count];
764 		if (!rk_pte_is_page_valid(pte))
765 			break;
766 
767 		pte_addr[pte_count] = rk_mk_pte_invalid(pte);
768 	}
769 
770 	rk_table_flush(rk_domain, pte_dma, pte_count);
771 
772 	return pte_count * SPAGE_SIZE;
773 }
774 
rk_iommu_map_iova(struct rk_iommu_domain * rk_domain,u32 * pte_addr,dma_addr_t pte_dma,dma_addr_t iova,phys_addr_t paddr,size_t size,int prot)775 static int rk_iommu_map_iova(struct rk_iommu_domain *rk_domain, u32 *pte_addr,
776 			     dma_addr_t pte_dma, dma_addr_t iova,
777 			     phys_addr_t paddr, size_t size, int prot)
778 {
779 	unsigned int pte_count;
780 	unsigned int pte_total = size / SPAGE_SIZE;
781 	phys_addr_t page_phys;
782 
783 	assert_spin_locked(&rk_domain->dt_lock);
784 
785 	for (pte_count = 0; pte_count < pte_total; pte_count++) {
786 		u32 pte = pte_addr[pte_count];
787 
788 		if (rk_pte_is_page_valid(pte))
789 			goto unwind;
790 
791 		pte_addr[pte_count] = rk_ops->mk_ptentries(paddr, prot);
792 
793 		paddr += SPAGE_SIZE;
794 	}
795 
796 	rk_table_flush(rk_domain, pte_dma, pte_total);
797 
798 	/*
799 	 * Zap the first and last iova to evict from iotlb any previously
800 	 * mapped cachelines holding stale values for its dte and pte.
801 	 * We only zap the first and last iova, since only they could have
802 	 * dte or pte shared with an existing mapping.
803 	 */
804 	rk_iommu_zap_iova_first_last(rk_domain, iova, size);
805 
806 	return 0;
807 unwind:
808 	/* Unmap the range of iovas that we just mapped */
809 	rk_iommu_unmap_iova(rk_domain, pte_addr, pte_dma,
810 			    pte_count * SPAGE_SIZE);
811 
812 	iova += pte_count * SPAGE_SIZE;
813 	page_phys = rk_ops->pt_address(pte_addr[pte_count]);
814 	pr_err("iova: %pad already mapped to %pa cannot remap to phys: %pa prot: %#x\n",
815 	       &iova, &page_phys, &paddr, prot);
816 
817 	return -EADDRINUSE;
818 }
819 
rk_iommu_map(struct iommu_domain * domain,unsigned long _iova,phys_addr_t paddr,size_t size,size_t count,int prot,gfp_t gfp,size_t * mapped)820 static int rk_iommu_map(struct iommu_domain *domain, unsigned long _iova,
821 			phys_addr_t paddr, size_t size, size_t count,
822 			int prot, gfp_t gfp, size_t *mapped)
823 {
824 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
825 	unsigned long flags;
826 	dma_addr_t pte_dma, iova = (dma_addr_t)_iova;
827 	u32 *page_table, *pte_addr;
828 	u32 dte_index, pte_index;
829 	int ret;
830 
831 	spin_lock_irqsave(&rk_domain->dt_lock, flags);
832 
833 	/*
834 	 * pgsize_bitmap specifies iova sizes that fit in one page table
835 	 * (1024 4-KiB pages = 4 MiB).
836 	 * So, size will always be 4096 <= size <= 4194304.
837 	 * Since iommu_map() guarantees that both iova and size will be
838 	 * aligned, we will always only be mapping from a single dte here.
839 	 */
840 	page_table = rk_dte_get_page_table(rk_domain, iova);
841 	if (IS_ERR(page_table)) {
842 		spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
843 		return PTR_ERR(page_table);
844 	}
845 
846 	dte_index = rk_domain->dt[rk_iova_dte_index(iova)];
847 	pte_index = rk_iova_pte_index(iova);
848 	pte_addr = &page_table[pte_index];
849 
850 	pte_dma = rk_ops->pt_address(dte_index) + pte_index * sizeof(u32);
851 	ret = rk_iommu_map_iova(rk_domain, pte_addr, pte_dma, iova,
852 				paddr, size, prot);
853 
854 	spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
855 	if (!ret)
856 		*mapped = size;
857 
858 	return ret;
859 }
860 
rk_iommu_unmap(struct iommu_domain * domain,unsigned long _iova,size_t size,size_t count,struct iommu_iotlb_gather * gather)861 static size_t rk_iommu_unmap(struct iommu_domain *domain, unsigned long _iova,
862 			     size_t size, size_t count, struct iommu_iotlb_gather *gather)
863 {
864 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
865 	unsigned long flags;
866 	dma_addr_t pte_dma, iova = (dma_addr_t)_iova;
867 	phys_addr_t pt_phys;
868 	u32 dte;
869 	u32 *pte_addr;
870 	size_t unmap_size;
871 
872 	spin_lock_irqsave(&rk_domain->dt_lock, flags);
873 
874 	/*
875 	 * pgsize_bitmap specifies iova sizes that fit in one page table
876 	 * (1024 4-KiB pages = 4 MiB).
877 	 * So, size will always be 4096 <= size <= 4194304.
878 	 * Since iommu_unmap() guarantees that both iova and size will be
879 	 * aligned, we will always only be unmapping from a single dte here.
880 	 */
881 	dte = rk_domain->dt[rk_iova_dte_index(iova)];
882 	/* Just return 0 if iova is unmapped */
883 	if (!rk_dte_is_pt_valid(dte)) {
884 		spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
885 		return 0;
886 	}
887 
888 	pt_phys = rk_ops->pt_address(dte);
889 	pte_addr = (u32 *)phys_to_virt(pt_phys) + rk_iova_pte_index(iova);
890 	pte_dma = pt_phys + rk_iova_pte_index(iova) * sizeof(u32);
891 	unmap_size = rk_iommu_unmap_iova(rk_domain, pte_addr, pte_dma, size);
892 
893 	spin_unlock_irqrestore(&rk_domain->dt_lock, flags);
894 
895 	/* Shootdown iotlb entries for iova range that was just unmapped */
896 	rk_iommu_zap_iova(rk_domain, iova, unmap_size);
897 
898 	return unmap_size;
899 }
900 
rk_iommu_from_dev(struct device * dev)901 static struct rk_iommu *rk_iommu_from_dev(struct device *dev)
902 {
903 	struct rk_iommudata *data = dev_iommu_priv_get(dev);
904 
905 	return data ? data->iommu : NULL;
906 }
907 
908 /* Must be called with iommu powered on and attached */
rk_iommu_disable(struct rk_iommu * iommu)909 static void rk_iommu_disable(struct rk_iommu *iommu)
910 {
911 	int i;
912 
913 	/* Ignore error while disabling, just keep going */
914 	WARN_ON(clk_bulk_enable(iommu->num_clocks, iommu->clocks));
915 	rk_iommu_enable_stall(iommu);
916 	rk_iommu_disable_paging(iommu);
917 	for (i = 0; i < iommu->num_mmu; i++) {
918 		rk_iommu_write(iommu->bases[i], RK_MMU_INT_MASK, 0);
919 		rk_iommu_write(iommu->bases[i], RK_MMU_DTE_ADDR, 0);
920 	}
921 	rk_iommu_disable_stall(iommu);
922 	clk_bulk_disable(iommu->num_clocks, iommu->clocks);
923 }
924 
925 /* Must be called with iommu powered on and attached */
rk_iommu_enable(struct rk_iommu * iommu)926 static int rk_iommu_enable(struct rk_iommu *iommu)
927 {
928 	struct iommu_domain *domain = iommu->domain;
929 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
930 	int ret, i;
931 
932 	ret = clk_bulk_enable(iommu->num_clocks, iommu->clocks);
933 	if (ret)
934 		return ret;
935 
936 	ret = rk_iommu_enable_stall(iommu);
937 	if (ret)
938 		goto out_disable_clocks;
939 
940 	ret = rk_iommu_force_reset(iommu);
941 	if (ret)
942 		goto out_disable_stall;
943 
944 	for (i = 0; i < iommu->num_mmu; i++) {
945 		rk_iommu_write(iommu->bases[i], RK_MMU_DTE_ADDR,
946 			       rk_ops->mk_dtentries(rk_domain->dt_dma));
947 		rk_iommu_base_command(iommu->bases[i], RK_MMU_CMD_ZAP_CACHE);
948 		rk_iommu_write(iommu->bases[i], RK_MMU_INT_MASK, RK_MMU_IRQ_MASK);
949 	}
950 
951 	ret = rk_iommu_enable_paging(iommu);
952 
953 out_disable_stall:
954 	rk_iommu_disable_stall(iommu);
955 out_disable_clocks:
956 	clk_bulk_disable(iommu->num_clocks, iommu->clocks);
957 	return ret;
958 }
959 
rk_iommu_identity_attach(struct iommu_domain * identity_domain,struct device * dev)960 static int rk_iommu_identity_attach(struct iommu_domain *identity_domain,
961 				    struct device *dev)
962 {
963 	struct rk_iommu *iommu;
964 	struct rk_iommu_domain *rk_domain;
965 	unsigned long flags;
966 	int ret;
967 
968 	/* Allow 'virtual devices' (eg drm) to detach from domain */
969 	iommu = rk_iommu_from_dev(dev);
970 	if (!iommu)
971 		return -ENODEV;
972 
973 	rk_domain = to_rk_domain(iommu->domain);
974 
975 	dev_dbg(dev, "Detaching from iommu domain\n");
976 
977 	if (iommu->domain == identity_domain)
978 		return 0;
979 
980 	iommu->domain = identity_domain;
981 
982 	spin_lock_irqsave(&rk_domain->iommus_lock, flags);
983 	list_del_init(&iommu->node);
984 	spin_unlock_irqrestore(&rk_domain->iommus_lock, flags);
985 
986 	ret = pm_runtime_get_if_in_use(iommu->dev);
987 	WARN_ON_ONCE(ret < 0);
988 	if (ret > 0) {
989 		rk_iommu_disable(iommu);
990 		pm_runtime_put(iommu->dev);
991 	}
992 
993 	return 0;
994 }
995 
996 static struct iommu_domain_ops rk_identity_ops = {
997 	.attach_dev = rk_iommu_identity_attach,
998 };
999 
1000 static struct iommu_domain rk_identity_domain = {
1001 	.type = IOMMU_DOMAIN_IDENTITY,
1002 	.ops = &rk_identity_ops,
1003 };
1004 
rk_iommu_attach_device(struct iommu_domain * domain,struct device * dev)1005 static int rk_iommu_attach_device(struct iommu_domain *domain,
1006 		struct device *dev)
1007 {
1008 	struct rk_iommu *iommu;
1009 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
1010 	unsigned long flags;
1011 	int ret;
1012 
1013 	/*
1014 	 * Allow 'virtual devices' (e.g., drm) to attach to domain.
1015 	 * Such a device does not belong to an iommu group.
1016 	 */
1017 	iommu = rk_iommu_from_dev(dev);
1018 	if (!iommu)
1019 		return 0;
1020 
1021 	dev_dbg(dev, "Attaching to iommu domain\n");
1022 
1023 	/* iommu already attached */
1024 	if (iommu->domain == domain)
1025 		return 0;
1026 
1027 	ret = rk_iommu_identity_attach(&rk_identity_domain, dev);
1028 	if (ret)
1029 		return ret;
1030 
1031 	iommu->domain = domain;
1032 
1033 	spin_lock_irqsave(&rk_domain->iommus_lock, flags);
1034 	list_add_tail(&iommu->node, &rk_domain->iommus);
1035 	spin_unlock_irqrestore(&rk_domain->iommus_lock, flags);
1036 
1037 	ret = pm_runtime_get_if_in_use(iommu->dev);
1038 	if (!ret || WARN_ON_ONCE(ret < 0))
1039 		return 0;
1040 
1041 	ret = rk_iommu_enable(iommu);
1042 	if (ret)
1043 		WARN_ON(rk_iommu_identity_attach(&rk_identity_domain, dev));
1044 
1045 	pm_runtime_put(iommu->dev);
1046 
1047 	return ret;
1048 }
1049 
rk_iommu_domain_alloc_paging(struct device * dev)1050 static struct iommu_domain *rk_iommu_domain_alloc_paging(struct device *dev)
1051 {
1052 	struct rk_iommu_domain *rk_domain;
1053 
1054 	if (!dma_dev)
1055 		return NULL;
1056 
1057 	rk_domain = kzalloc(sizeof(*rk_domain), GFP_KERNEL);
1058 	if (!rk_domain)
1059 		return NULL;
1060 
1061 	/*
1062 	 * rk32xx iommus use a 2 level pagetable.
1063 	 * Each level1 (dt) and level2 (pt) table has 1024 4-byte entries.
1064 	 * Allocate one 4 KiB page for each table.
1065 	 */
1066 	rk_domain->dt = iommu_alloc_page(GFP_KERNEL | rk_ops->gfp_flags);
1067 	if (!rk_domain->dt)
1068 		goto err_free_domain;
1069 
1070 	rk_domain->dt_dma = dma_map_single(dma_dev, rk_domain->dt,
1071 					   SPAGE_SIZE, DMA_TO_DEVICE);
1072 	if (dma_mapping_error(dma_dev, rk_domain->dt_dma)) {
1073 		dev_err(dma_dev, "DMA map error for DT\n");
1074 		goto err_free_dt;
1075 	}
1076 
1077 	spin_lock_init(&rk_domain->iommus_lock);
1078 	spin_lock_init(&rk_domain->dt_lock);
1079 	INIT_LIST_HEAD(&rk_domain->iommus);
1080 
1081 	rk_domain->domain.geometry.aperture_start = 0;
1082 	rk_domain->domain.geometry.aperture_end   = DMA_BIT_MASK(32);
1083 	rk_domain->domain.geometry.force_aperture = true;
1084 
1085 	return &rk_domain->domain;
1086 
1087 err_free_dt:
1088 	iommu_free_page(rk_domain->dt);
1089 err_free_domain:
1090 	kfree(rk_domain);
1091 
1092 	return NULL;
1093 }
1094 
rk_iommu_domain_free(struct iommu_domain * domain)1095 static void rk_iommu_domain_free(struct iommu_domain *domain)
1096 {
1097 	struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
1098 	int i;
1099 
1100 	WARN_ON(!list_empty(&rk_domain->iommus));
1101 
1102 	for (i = 0; i < NUM_DT_ENTRIES; i++) {
1103 		u32 dte = rk_domain->dt[i];
1104 		if (rk_dte_is_pt_valid(dte)) {
1105 			phys_addr_t pt_phys = rk_ops->pt_address(dte);
1106 			u32 *page_table = phys_to_virt(pt_phys);
1107 			dma_unmap_single(dma_dev, pt_phys,
1108 					 SPAGE_SIZE, DMA_TO_DEVICE);
1109 			iommu_free_page(page_table);
1110 		}
1111 	}
1112 
1113 	dma_unmap_single(dma_dev, rk_domain->dt_dma,
1114 			 SPAGE_SIZE, DMA_TO_DEVICE);
1115 	iommu_free_page(rk_domain->dt);
1116 
1117 	kfree(rk_domain);
1118 }
1119 
rk_iommu_probe_device(struct device * dev)1120 static struct iommu_device *rk_iommu_probe_device(struct device *dev)
1121 {
1122 	struct rk_iommudata *data;
1123 	struct rk_iommu *iommu;
1124 
1125 	data = dev_iommu_priv_get(dev);
1126 	if (!data)
1127 		return ERR_PTR(-ENODEV);
1128 
1129 	iommu = rk_iommu_from_dev(dev);
1130 
1131 	data->link = device_link_add(dev, iommu->dev,
1132 				     DL_FLAG_STATELESS | DL_FLAG_PM_RUNTIME);
1133 
1134 	return &iommu->iommu;
1135 }
1136 
rk_iommu_release_device(struct device * dev)1137 static void rk_iommu_release_device(struct device *dev)
1138 {
1139 	struct rk_iommudata *data = dev_iommu_priv_get(dev);
1140 
1141 	device_link_del(data->link);
1142 }
1143 
rk_iommu_of_xlate(struct device * dev,const struct of_phandle_args * args)1144 static int rk_iommu_of_xlate(struct device *dev,
1145 			     const struct of_phandle_args *args)
1146 {
1147 	struct platform_device *iommu_dev;
1148 	struct rk_iommudata *data;
1149 
1150 	data = devm_kzalloc(dma_dev, sizeof(*data), GFP_KERNEL);
1151 	if (!data)
1152 		return -ENOMEM;
1153 
1154 	iommu_dev = of_find_device_by_node(args->np);
1155 
1156 	data->iommu = platform_get_drvdata(iommu_dev);
1157 	data->iommu->domain = &rk_identity_domain;
1158 	dev_iommu_priv_set(dev, data);
1159 
1160 	platform_device_put(iommu_dev);
1161 
1162 	return 0;
1163 }
1164 
1165 static const struct iommu_ops rk_iommu_ops = {
1166 	.identity_domain = &rk_identity_domain,
1167 	.domain_alloc_paging = rk_iommu_domain_alloc_paging,
1168 	.probe_device = rk_iommu_probe_device,
1169 	.release_device = rk_iommu_release_device,
1170 	.device_group = generic_single_device_group,
1171 	.pgsize_bitmap = RK_IOMMU_PGSIZE_BITMAP,
1172 	.of_xlate = rk_iommu_of_xlate,
1173 	.default_domain_ops = &(const struct iommu_domain_ops) {
1174 		.attach_dev	= rk_iommu_attach_device,
1175 		.map_pages	= rk_iommu_map,
1176 		.unmap_pages	= rk_iommu_unmap,
1177 		.iova_to_phys	= rk_iommu_iova_to_phys,
1178 		.free		= rk_iommu_domain_free,
1179 	}
1180 };
1181 
rk_iommu_probe(struct platform_device * pdev)1182 static int rk_iommu_probe(struct platform_device *pdev)
1183 {
1184 	struct device *dev = &pdev->dev;
1185 	struct rk_iommu *iommu;
1186 	struct resource *res;
1187 	const struct rk_iommu_ops *ops;
1188 	int num_res = pdev->num_resources;
1189 	int err, i;
1190 
1191 	iommu = devm_kzalloc(dev, sizeof(*iommu), GFP_KERNEL);
1192 	if (!iommu)
1193 		return -ENOMEM;
1194 
1195 	platform_set_drvdata(pdev, iommu);
1196 	iommu->dev = dev;
1197 	iommu->num_mmu = 0;
1198 
1199 	ops = of_device_get_match_data(dev);
1200 	if (!rk_ops)
1201 		rk_ops = ops;
1202 
1203 	/*
1204 	 * That should not happen unless different versions of the
1205 	 * hardware block are embedded the same SoC
1206 	 */
1207 	if (WARN_ON(rk_ops != ops))
1208 		return -EINVAL;
1209 
1210 	iommu->bases = devm_kcalloc(dev, num_res, sizeof(*iommu->bases),
1211 				    GFP_KERNEL);
1212 	if (!iommu->bases)
1213 		return -ENOMEM;
1214 
1215 	for (i = 0; i < num_res; i++) {
1216 		res = platform_get_resource(pdev, IORESOURCE_MEM, i);
1217 		if (!res)
1218 			continue;
1219 		iommu->bases[i] = devm_ioremap_resource(&pdev->dev, res);
1220 		if (IS_ERR(iommu->bases[i]))
1221 			continue;
1222 		iommu->num_mmu++;
1223 	}
1224 	if (iommu->num_mmu == 0)
1225 		return PTR_ERR(iommu->bases[0]);
1226 
1227 	iommu->num_irq = platform_irq_count(pdev);
1228 	if (iommu->num_irq < 0)
1229 		return iommu->num_irq;
1230 
1231 	iommu->reset_disabled = device_property_read_bool(dev,
1232 					"rockchip,disable-mmu-reset");
1233 
1234 	iommu->num_clocks = ARRAY_SIZE(rk_iommu_clocks);
1235 	iommu->clocks = devm_kcalloc(iommu->dev, iommu->num_clocks,
1236 				     sizeof(*iommu->clocks), GFP_KERNEL);
1237 	if (!iommu->clocks)
1238 		return -ENOMEM;
1239 
1240 	for (i = 0; i < iommu->num_clocks; ++i)
1241 		iommu->clocks[i].id = rk_iommu_clocks[i];
1242 
1243 	/*
1244 	 * iommu clocks should be present for all new devices and devicetrees
1245 	 * but there are older devicetrees without clocks out in the wild.
1246 	 * So clocks as optional for the time being.
1247 	 */
1248 	err = devm_clk_bulk_get(iommu->dev, iommu->num_clocks, iommu->clocks);
1249 	if (err == -ENOENT)
1250 		iommu->num_clocks = 0;
1251 	else if (err)
1252 		return err;
1253 
1254 	err = clk_bulk_prepare(iommu->num_clocks, iommu->clocks);
1255 	if (err)
1256 		return err;
1257 
1258 	err = iommu_device_sysfs_add(&iommu->iommu, dev, NULL, dev_name(dev));
1259 	if (err)
1260 		goto err_unprepare_clocks;
1261 
1262 	err = iommu_device_register(&iommu->iommu, &rk_iommu_ops, dev);
1263 	if (err)
1264 		goto err_remove_sysfs;
1265 
1266 	/*
1267 	 * Use the first registered IOMMU device for domain to use with DMA
1268 	 * API, since a domain might not physically correspond to a single
1269 	 * IOMMU device..
1270 	 */
1271 	if (!dma_dev)
1272 		dma_dev = &pdev->dev;
1273 
1274 	pm_runtime_enable(dev);
1275 
1276 	for (i = 0; i < iommu->num_irq; i++) {
1277 		int irq = platform_get_irq(pdev, i);
1278 
1279 		if (irq < 0) {
1280 			err = irq;
1281 			goto err_pm_disable;
1282 		}
1283 
1284 		err = devm_request_irq(iommu->dev, irq, rk_iommu_irq,
1285 				       IRQF_SHARED, dev_name(dev), iommu);
1286 		if (err)
1287 			goto err_pm_disable;
1288 	}
1289 
1290 	dma_set_mask_and_coherent(dev, rk_ops->dma_bit_mask);
1291 
1292 	return 0;
1293 err_pm_disable:
1294 	pm_runtime_disable(dev);
1295 err_remove_sysfs:
1296 	iommu_device_sysfs_remove(&iommu->iommu);
1297 err_unprepare_clocks:
1298 	clk_bulk_unprepare(iommu->num_clocks, iommu->clocks);
1299 	return err;
1300 }
1301 
rk_iommu_shutdown(struct platform_device * pdev)1302 static void rk_iommu_shutdown(struct platform_device *pdev)
1303 {
1304 	struct rk_iommu *iommu = platform_get_drvdata(pdev);
1305 	int i;
1306 
1307 	for (i = 0; i < iommu->num_irq; i++) {
1308 		int irq = platform_get_irq(pdev, i);
1309 
1310 		devm_free_irq(iommu->dev, irq, iommu);
1311 	}
1312 
1313 	pm_runtime_force_suspend(&pdev->dev);
1314 }
1315 
rk_iommu_suspend(struct device * dev)1316 static int __maybe_unused rk_iommu_suspend(struct device *dev)
1317 {
1318 	struct rk_iommu *iommu = dev_get_drvdata(dev);
1319 
1320 	if (iommu->domain == &rk_identity_domain)
1321 		return 0;
1322 
1323 	rk_iommu_disable(iommu);
1324 	return 0;
1325 }
1326 
rk_iommu_resume(struct device * dev)1327 static int __maybe_unused rk_iommu_resume(struct device *dev)
1328 {
1329 	struct rk_iommu *iommu = dev_get_drvdata(dev);
1330 
1331 	if (iommu->domain == &rk_identity_domain)
1332 		return 0;
1333 
1334 	return rk_iommu_enable(iommu);
1335 }
1336 
1337 static const struct dev_pm_ops rk_iommu_pm_ops = {
1338 	SET_RUNTIME_PM_OPS(rk_iommu_suspend, rk_iommu_resume, NULL)
1339 	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1340 				pm_runtime_force_resume)
1341 };
1342 
1343 static struct rk_iommu_ops iommu_data_ops_v1 = {
1344 	.pt_address = &rk_dte_pt_address,
1345 	.mk_dtentries = &rk_mk_dte,
1346 	.mk_ptentries = &rk_mk_pte,
1347 	.dma_bit_mask = DMA_BIT_MASK(32),
1348 	.gfp_flags = GFP_DMA32,
1349 };
1350 
1351 static struct rk_iommu_ops iommu_data_ops_v2 = {
1352 	.pt_address = &rk_dte_pt_address_v2,
1353 	.mk_dtentries = &rk_mk_dte_v2,
1354 	.mk_ptentries = &rk_mk_pte_v2,
1355 	.dma_bit_mask = DMA_BIT_MASK(40),
1356 	.gfp_flags = 0,
1357 };
1358 
1359 static const struct of_device_id rk_iommu_dt_ids[] = {
1360 	{	.compatible = "rockchip,iommu",
1361 		.data = &iommu_data_ops_v1,
1362 	},
1363 	{	.compatible = "rockchip,rk3568-iommu",
1364 		.data = &iommu_data_ops_v2,
1365 	},
1366 	{ /* sentinel */ }
1367 };
1368 
1369 static struct platform_driver rk_iommu_driver = {
1370 	.probe = rk_iommu_probe,
1371 	.shutdown = rk_iommu_shutdown,
1372 	.driver = {
1373 		   .name = "rk_iommu",
1374 		   .of_match_table = rk_iommu_dt_ids,
1375 		   .pm = &rk_iommu_pm_ops,
1376 		   .suppress_bind_attrs = true,
1377 	},
1378 };
1379 builtin_platform_driver(rk_iommu_driver);
1380