xref: /linux/drivers/iommu/sun50i-iommu.c (revision 60684c2bd35064043360e6f716d1b7c20e967b7d)
1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 // Copyright (C) 2016-2018, Allwinner Technology CO., LTD.
3 // Copyright (C) 2019-2020, Cerno
4 
5 #include <linux/bitfield.h>
6 #include <linux/bug.h>
7 #include <linux/clk.h>
8 #include <linux/device.h>
9 #include <linux/dma-direction.h>
10 #include <linux/dma-mapping.h>
11 #include <linux/err.h>
12 #include <linux/errno.h>
13 #include <linux/interrupt.h>
14 #include <linux/iommu.h>
15 #include <linux/iopoll.h>
16 #include <linux/ioport.h>
17 #include <linux/log2.h>
18 #include <linux/module.h>
19 #include <linux/of_platform.h>
20 #include <linux/platform_device.h>
21 #include <linux/pm.h>
22 #include <linux/pm_runtime.h>
23 #include <linux/reset.h>
24 #include <linux/sizes.h>
25 #include <linux/slab.h>
26 #include <linux/spinlock.h>
27 #include <linux/types.h>
28 
29 #define IOMMU_RESET_REG			0x010
30 #define IOMMU_RESET_RELEASE_ALL			0xffffffff
31 #define IOMMU_ENABLE_REG		0x020
32 #define IOMMU_ENABLE_ENABLE			BIT(0)
33 
34 #define IOMMU_BYPASS_REG		0x030
35 #define IOMMU_AUTO_GATING_REG		0x040
36 #define IOMMU_AUTO_GATING_ENABLE		BIT(0)
37 
38 #define IOMMU_WBUF_CTRL_REG		0x044
39 #define IOMMU_OOO_CTRL_REG		0x048
40 #define IOMMU_4KB_BDY_PRT_CTRL_REG	0x04c
41 #define IOMMU_TTB_REG			0x050
42 #define IOMMU_TLB_ENABLE_REG		0x060
43 #define IOMMU_TLB_PREFETCH_REG		0x070
44 #define IOMMU_TLB_PREFETCH_MASTER_ENABLE(m)	BIT(m)
45 
46 #define IOMMU_TLB_FLUSH_REG		0x080
47 #define IOMMU_TLB_FLUSH_PTW_CACHE		BIT(17)
48 #define IOMMU_TLB_FLUSH_MACRO_TLB		BIT(16)
49 #define IOMMU_TLB_FLUSH_MICRO_TLB(i)		(BIT(i) & GENMASK(5, 0))
50 
51 #define IOMMU_TLB_IVLD_ADDR_REG		0x090
52 #define IOMMU_TLB_IVLD_ADDR_MASK_REG	0x094
53 #define IOMMU_TLB_IVLD_ENABLE_REG	0x098
54 #define IOMMU_TLB_IVLD_ENABLE_ENABLE		BIT(0)
55 
56 #define IOMMU_PC_IVLD_ADDR_REG		0x0a0
57 #define IOMMU_PC_IVLD_ENABLE_REG	0x0a8
58 #define IOMMU_PC_IVLD_ENABLE_ENABLE		BIT(0)
59 
60 #define IOMMU_DM_AUT_CTRL_REG(d)	(0x0b0 + ((d) / 2) * 4)
61 #define IOMMU_DM_AUT_CTRL_RD_UNAVAIL(d, m)	(1 << (((d & 1) * 16) + ((m) * 2)))
62 #define IOMMU_DM_AUT_CTRL_WR_UNAVAIL(d, m)	(1 << (((d & 1) * 16) + ((m) * 2) + 1))
63 
64 #define IOMMU_DM_AUT_OVWT_REG		0x0d0
65 #define IOMMU_INT_ENABLE_REG		0x100
66 #define IOMMU_INT_CLR_REG		0x104
67 #define IOMMU_INT_STA_REG		0x108
68 #define IOMMU_INT_ERR_ADDR_REG(i)	(0x110 + (i) * 4)
69 #define IOMMU_INT_ERR_ADDR_L1_REG	0x130
70 #define IOMMU_INT_ERR_ADDR_L2_REG	0x134
71 #define IOMMU_INT_ERR_DATA_REG(i)	(0x150 + (i) * 4)
72 #define IOMMU_L1PG_INT_REG		0x0180
73 #define IOMMU_L2PG_INT_REG		0x0184
74 
75 #define IOMMU_INT_INVALID_L2PG			BIT(17)
76 #define IOMMU_INT_INVALID_L1PG			BIT(16)
77 #define IOMMU_INT_MASTER_PERMISSION(m)		BIT(m)
78 #define IOMMU_INT_MASTER_MASK			(IOMMU_INT_MASTER_PERMISSION(0) | \
79 						 IOMMU_INT_MASTER_PERMISSION(1) | \
80 						 IOMMU_INT_MASTER_PERMISSION(2) | \
81 						 IOMMU_INT_MASTER_PERMISSION(3) | \
82 						 IOMMU_INT_MASTER_PERMISSION(4) | \
83 						 IOMMU_INT_MASTER_PERMISSION(5))
84 #define IOMMU_INT_MASK				(IOMMU_INT_INVALID_L1PG | \
85 						 IOMMU_INT_INVALID_L2PG | \
86 						 IOMMU_INT_MASTER_MASK)
87 
88 #define PT_ENTRY_SIZE			sizeof(u32)
89 
90 #define NUM_DT_ENTRIES			4096
91 #define DT_SIZE				(NUM_DT_ENTRIES * PT_ENTRY_SIZE)
92 
93 #define NUM_PT_ENTRIES			256
94 #define PT_SIZE				(NUM_PT_ENTRIES * PT_ENTRY_SIZE)
95 
96 #define SPAGE_SIZE			4096
97 
98 struct sun50i_iommu {
99 	struct iommu_device iommu;
100 
101 	/* Lock to modify the IOMMU registers */
102 	spinlock_t iommu_lock;
103 
104 	struct device *dev;
105 	void __iomem *base;
106 	struct reset_control *reset;
107 	struct clk *clk;
108 
109 	struct iommu_domain *domain;
110 	struct iommu_group *group;
111 	struct kmem_cache *pt_pool;
112 };
113 
114 struct sun50i_iommu_domain {
115 	struct iommu_domain domain;
116 
117 	/* Number of devices attached to the domain */
118 	refcount_t refcnt;
119 
120 	/* L1 Page Table */
121 	u32 *dt;
122 	dma_addr_t dt_dma;
123 
124 	struct sun50i_iommu *iommu;
125 };
126 
127 static struct sun50i_iommu_domain *to_sun50i_domain(struct iommu_domain *domain)
128 {
129 	return container_of(domain, struct sun50i_iommu_domain, domain);
130 }
131 
132 static struct sun50i_iommu *sun50i_iommu_from_dev(struct device *dev)
133 {
134 	return dev_iommu_priv_get(dev);
135 }
136 
137 static u32 iommu_read(struct sun50i_iommu *iommu, u32 offset)
138 {
139 	return readl(iommu->base + offset);
140 }
141 
142 static void iommu_write(struct sun50i_iommu *iommu, u32 offset, u32 value)
143 {
144 	writel(value, iommu->base + offset);
145 }
146 
147 /*
148  * The Allwinner H6 IOMMU uses a 2-level page table.
149  *
150  * The first level is the usual Directory Table (DT), that consists of
151  * 4096 4-bytes Directory Table Entries (DTE), each pointing to a Page
152  * Table (PT).
153  *
154  * Each PT consits of 256 4-bytes Page Table Entries (PTE), each
155  * pointing to a 4kB page of physical memory.
156  *
157  * The IOMMU supports a single DT, pointed by the IOMMU_TTB_REG
158  * register that contains its physical address.
159  */
160 
161 #define SUN50I_IOVA_DTE_MASK	GENMASK(31, 20)
162 #define SUN50I_IOVA_PTE_MASK	GENMASK(19, 12)
163 #define SUN50I_IOVA_PAGE_MASK	GENMASK(11, 0)
164 
165 static u32 sun50i_iova_get_dte_index(dma_addr_t iova)
166 {
167 	return FIELD_GET(SUN50I_IOVA_DTE_MASK, iova);
168 }
169 
170 static u32 sun50i_iova_get_pte_index(dma_addr_t iova)
171 {
172 	return FIELD_GET(SUN50I_IOVA_PTE_MASK, iova);
173 }
174 
175 static u32 sun50i_iova_get_page_offset(dma_addr_t iova)
176 {
177 	return FIELD_GET(SUN50I_IOVA_PAGE_MASK, iova);
178 }
179 
180 /*
181  * Each Directory Table Entry has a Page Table address and a valid
182  * bit:
183 
184  * +---------------------+-----------+-+
185  * | PT address          | Reserved  |V|
186  * +---------------------+-----------+-+
187  *  31:10 - Page Table address
188  *   9:2  - Reserved
189  *   1:0  - 1 if the entry is valid
190  */
191 
192 #define SUN50I_DTE_PT_ADDRESS_MASK	GENMASK(31, 10)
193 #define SUN50I_DTE_PT_ATTRS		GENMASK(1, 0)
194 #define SUN50I_DTE_PT_VALID		1
195 
196 static phys_addr_t sun50i_dte_get_pt_address(u32 dte)
197 {
198 	return (phys_addr_t)dte & SUN50I_DTE_PT_ADDRESS_MASK;
199 }
200 
201 static bool sun50i_dte_is_pt_valid(u32 dte)
202 {
203 	return (dte & SUN50I_DTE_PT_ATTRS) == SUN50I_DTE_PT_VALID;
204 }
205 
206 static u32 sun50i_mk_dte(dma_addr_t pt_dma)
207 {
208 	return (pt_dma & SUN50I_DTE_PT_ADDRESS_MASK) | SUN50I_DTE_PT_VALID;
209 }
210 
211 /*
212  * Each PTE has a Page address, an authority index and a valid bit:
213  *
214  * +----------------+-----+-----+-----+---+-----+
215  * | Page address   | Rsv | ACI | Rsv | V | Rsv |
216  * +----------------+-----+-----+-----+---+-----+
217  *  31:12 - Page address
218  *  11:8  - Reserved
219  *   7:4  - Authority Control Index
220  *   3:2  - Reserved
221  *     1  - 1 if the entry is valid
222  *     0  - Reserved
223  *
224  * The way permissions work is that the IOMMU has 16 "domains" that
225  * can be configured to give each masters either read or write
226  * permissions through the IOMMU_DM_AUT_CTRL_REG registers. The domain
227  * 0 seems like the default domain, and its permissions in the
228  * IOMMU_DM_AUT_CTRL_REG are only read-only, so it's not really
229  * useful to enforce any particular permission.
230  *
231  * Each page entry will then have a reference to the domain they are
232  * affected to, so that we can actually enforce them on a per-page
233  * basis.
234  *
235  * In order to make it work with the IOMMU framework, we will be using
236  * 4 different domains, starting at 1: RD_WR, RD, WR and NONE
237  * depending on the permission we want to enforce. Each domain will
238  * have each master setup in the same way, since the IOMMU framework
239  * doesn't seem to restrict page access on a per-device basis. And
240  * then we will use the relevant domain index when generating the page
241  * table entry depending on the permissions we want to be enforced.
242  */
243 
244 enum sun50i_iommu_aci {
245 	SUN50I_IOMMU_ACI_DO_NOT_USE = 0,
246 	SUN50I_IOMMU_ACI_NONE,
247 	SUN50I_IOMMU_ACI_RD,
248 	SUN50I_IOMMU_ACI_WR,
249 	SUN50I_IOMMU_ACI_RD_WR,
250 };
251 
252 #define SUN50I_PTE_PAGE_ADDRESS_MASK	GENMASK(31, 12)
253 #define SUN50I_PTE_ACI_MASK		GENMASK(7, 4)
254 #define SUN50I_PTE_PAGE_VALID		BIT(1)
255 
256 static phys_addr_t sun50i_pte_get_page_address(u32 pte)
257 {
258 	return (phys_addr_t)pte & SUN50I_PTE_PAGE_ADDRESS_MASK;
259 }
260 
261 static enum sun50i_iommu_aci sun50i_get_pte_aci(u32 pte)
262 {
263 	return FIELD_GET(SUN50I_PTE_ACI_MASK, pte);
264 }
265 
266 static bool sun50i_pte_is_page_valid(u32 pte)
267 {
268 	return pte & SUN50I_PTE_PAGE_VALID;
269 }
270 
271 static u32 sun50i_mk_pte(phys_addr_t page, int prot)
272 {
273 	enum sun50i_iommu_aci aci;
274 	u32 flags = 0;
275 
276 	if ((prot & (IOMMU_READ | IOMMU_WRITE)) == (IOMMU_READ | IOMMU_WRITE))
277 		aci = SUN50I_IOMMU_ACI_RD_WR;
278 	else if (prot & IOMMU_READ)
279 		aci = SUN50I_IOMMU_ACI_RD;
280 	else if (prot & IOMMU_WRITE)
281 		aci = SUN50I_IOMMU_ACI_WR;
282 	else
283 		aci = SUN50I_IOMMU_ACI_NONE;
284 
285 	flags |= FIELD_PREP(SUN50I_PTE_ACI_MASK, aci);
286 	page &= SUN50I_PTE_PAGE_ADDRESS_MASK;
287 	return page | flags | SUN50I_PTE_PAGE_VALID;
288 }
289 
290 static void sun50i_table_flush(struct sun50i_iommu_domain *sun50i_domain,
291 			       void *vaddr, unsigned int count)
292 {
293 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
294 	dma_addr_t dma = virt_to_phys(vaddr);
295 	size_t size = count * PT_ENTRY_SIZE;
296 
297 	dma_sync_single_for_device(iommu->dev, dma, size, DMA_TO_DEVICE);
298 }
299 
300 static void sun50i_iommu_zap_iova(struct sun50i_iommu *iommu,
301 				  unsigned long iova)
302 {
303 	u32 reg;
304 	int ret;
305 
306 	iommu_write(iommu, IOMMU_TLB_IVLD_ADDR_REG, iova);
307 	iommu_write(iommu, IOMMU_TLB_IVLD_ADDR_MASK_REG, GENMASK(31, 12));
308 	iommu_write(iommu, IOMMU_TLB_IVLD_ENABLE_REG,
309 		    IOMMU_TLB_IVLD_ENABLE_ENABLE);
310 
311 	ret = readl_poll_timeout_atomic(iommu->base + IOMMU_TLB_IVLD_ENABLE_REG,
312 					reg, !reg, 1, 2000);
313 	if (ret)
314 		dev_warn(iommu->dev, "TLB invalidation timed out!\n");
315 }
316 
317 static void sun50i_iommu_zap_ptw_cache(struct sun50i_iommu *iommu,
318 				       unsigned long iova)
319 {
320 	u32 reg;
321 	int ret;
322 
323 	iommu_write(iommu, IOMMU_PC_IVLD_ADDR_REG, iova);
324 	iommu_write(iommu, IOMMU_PC_IVLD_ENABLE_REG,
325 		    IOMMU_PC_IVLD_ENABLE_ENABLE);
326 
327 	ret = readl_poll_timeout_atomic(iommu->base + IOMMU_PC_IVLD_ENABLE_REG,
328 					reg, !reg, 1, 2000);
329 	if (ret)
330 		dev_warn(iommu->dev, "PTW cache invalidation timed out!\n");
331 }
332 
333 static void sun50i_iommu_zap_range(struct sun50i_iommu *iommu,
334 				   unsigned long iova, size_t size)
335 {
336 	assert_spin_locked(&iommu->iommu_lock);
337 
338 	iommu_write(iommu, IOMMU_AUTO_GATING_REG, 0);
339 
340 	sun50i_iommu_zap_iova(iommu, iova);
341 	sun50i_iommu_zap_iova(iommu, iova + SPAGE_SIZE);
342 	if (size > SPAGE_SIZE) {
343 		sun50i_iommu_zap_iova(iommu, iova + size);
344 		sun50i_iommu_zap_iova(iommu, iova + size + SPAGE_SIZE);
345 	}
346 	sun50i_iommu_zap_ptw_cache(iommu, iova);
347 	sun50i_iommu_zap_ptw_cache(iommu, iova + SZ_1M);
348 	if (size > SZ_1M) {
349 		sun50i_iommu_zap_ptw_cache(iommu, iova + size);
350 		sun50i_iommu_zap_ptw_cache(iommu, iova + size + SZ_1M);
351 	}
352 
353 	iommu_write(iommu, IOMMU_AUTO_GATING_REG, IOMMU_AUTO_GATING_ENABLE);
354 }
355 
356 static int sun50i_iommu_flush_all_tlb(struct sun50i_iommu *iommu)
357 {
358 	u32 reg;
359 	int ret;
360 
361 	assert_spin_locked(&iommu->iommu_lock);
362 
363 	iommu_write(iommu,
364 		    IOMMU_TLB_FLUSH_REG,
365 		    IOMMU_TLB_FLUSH_PTW_CACHE |
366 		    IOMMU_TLB_FLUSH_MACRO_TLB |
367 		    IOMMU_TLB_FLUSH_MICRO_TLB(5) |
368 		    IOMMU_TLB_FLUSH_MICRO_TLB(4) |
369 		    IOMMU_TLB_FLUSH_MICRO_TLB(3) |
370 		    IOMMU_TLB_FLUSH_MICRO_TLB(2) |
371 		    IOMMU_TLB_FLUSH_MICRO_TLB(1) |
372 		    IOMMU_TLB_FLUSH_MICRO_TLB(0));
373 
374 	ret = readl_poll_timeout_atomic(iommu->base + IOMMU_TLB_FLUSH_REG,
375 					reg, !reg,
376 					1, 2000);
377 	if (ret)
378 		dev_warn(iommu->dev, "TLB Flush timed out!\n");
379 
380 	return ret;
381 }
382 
383 static void sun50i_iommu_flush_iotlb_all(struct iommu_domain *domain)
384 {
385 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
386 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
387 	unsigned long flags;
388 
389 	/*
390 	 * At boot, we'll have a first call into .flush_iotlb_all right after
391 	 * .probe_device, and since we link our (single) domain to our iommu in
392 	 * the .attach_device callback, we don't have that pointer set.
393 	 *
394 	 * It shouldn't really be any trouble to ignore it though since we flush
395 	 * all caches as part of the device powerup.
396 	 */
397 	if (!iommu)
398 		return;
399 
400 	spin_lock_irqsave(&iommu->iommu_lock, flags);
401 	sun50i_iommu_flush_all_tlb(iommu);
402 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
403 }
404 
405 static void sun50i_iommu_iotlb_sync_map(struct iommu_domain *domain,
406 					unsigned long iova, size_t size)
407 {
408 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
409 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
410 	unsigned long flags;
411 
412 	spin_lock_irqsave(&iommu->iommu_lock, flags);
413 	sun50i_iommu_zap_range(iommu, iova, size);
414 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
415 }
416 
417 static void sun50i_iommu_iotlb_sync(struct iommu_domain *domain,
418 				    struct iommu_iotlb_gather *gather)
419 {
420 	sun50i_iommu_flush_iotlb_all(domain);
421 }
422 
423 static int sun50i_iommu_enable(struct sun50i_iommu *iommu)
424 {
425 	struct sun50i_iommu_domain *sun50i_domain;
426 	unsigned long flags;
427 	int ret;
428 
429 	if (!iommu->domain)
430 		return 0;
431 
432 	sun50i_domain = to_sun50i_domain(iommu->domain);
433 
434 	ret = reset_control_deassert(iommu->reset);
435 	if (ret)
436 		return ret;
437 
438 	ret = clk_prepare_enable(iommu->clk);
439 	if (ret)
440 		goto err_reset_assert;
441 
442 	spin_lock_irqsave(&iommu->iommu_lock, flags);
443 
444 	iommu_write(iommu, IOMMU_TTB_REG, sun50i_domain->dt_dma);
445 	iommu_write(iommu, IOMMU_TLB_PREFETCH_REG,
446 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(0) |
447 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(1) |
448 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(2) |
449 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(3) |
450 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(4) |
451 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(5));
452 	iommu_write(iommu, IOMMU_INT_ENABLE_REG, IOMMU_INT_MASK);
453 	iommu_write(iommu, IOMMU_DM_AUT_CTRL_REG(SUN50I_IOMMU_ACI_NONE),
454 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 0) |
455 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 0) |
456 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 1) |
457 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 1) |
458 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 2) |
459 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 2) |
460 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 3) |
461 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 3) |
462 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 4) |
463 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 4) |
464 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 5) |
465 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 5));
466 
467 	iommu_write(iommu, IOMMU_DM_AUT_CTRL_REG(SUN50I_IOMMU_ACI_RD),
468 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 0) |
469 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 1) |
470 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 2) |
471 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 3) |
472 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 4) |
473 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 5));
474 
475 	iommu_write(iommu, IOMMU_DM_AUT_CTRL_REG(SUN50I_IOMMU_ACI_WR),
476 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 0) |
477 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 1) |
478 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 2) |
479 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 3) |
480 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 4) |
481 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 5));
482 
483 	ret = sun50i_iommu_flush_all_tlb(iommu);
484 	if (ret) {
485 		spin_unlock_irqrestore(&iommu->iommu_lock, flags);
486 		goto err_clk_disable;
487 	}
488 
489 	iommu_write(iommu, IOMMU_AUTO_GATING_REG, IOMMU_AUTO_GATING_ENABLE);
490 	iommu_write(iommu, IOMMU_ENABLE_REG, IOMMU_ENABLE_ENABLE);
491 
492 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
493 
494 	return 0;
495 
496 err_clk_disable:
497 	clk_disable_unprepare(iommu->clk);
498 
499 err_reset_assert:
500 	reset_control_assert(iommu->reset);
501 
502 	return ret;
503 }
504 
505 static void sun50i_iommu_disable(struct sun50i_iommu *iommu)
506 {
507 	unsigned long flags;
508 
509 	spin_lock_irqsave(&iommu->iommu_lock, flags);
510 
511 	iommu_write(iommu, IOMMU_ENABLE_REG, 0);
512 	iommu_write(iommu, IOMMU_TTB_REG, 0);
513 
514 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
515 
516 	clk_disable_unprepare(iommu->clk);
517 	reset_control_assert(iommu->reset);
518 }
519 
520 static void *sun50i_iommu_alloc_page_table(struct sun50i_iommu *iommu,
521 					   gfp_t gfp)
522 {
523 	dma_addr_t pt_dma;
524 	u32 *page_table;
525 
526 	page_table = kmem_cache_zalloc(iommu->pt_pool, gfp);
527 	if (!page_table)
528 		return ERR_PTR(-ENOMEM);
529 
530 	pt_dma = dma_map_single(iommu->dev, page_table, PT_SIZE, DMA_TO_DEVICE);
531 	if (dma_mapping_error(iommu->dev, pt_dma)) {
532 		dev_err(iommu->dev, "Couldn't map L2 Page Table\n");
533 		kmem_cache_free(iommu->pt_pool, page_table);
534 		return ERR_PTR(-ENOMEM);
535 	}
536 
537 	/* We rely on the physical address and DMA address being the same */
538 	WARN_ON(pt_dma != virt_to_phys(page_table));
539 
540 	return page_table;
541 }
542 
543 static void sun50i_iommu_free_page_table(struct sun50i_iommu *iommu,
544 					 u32 *page_table)
545 {
546 	phys_addr_t pt_phys = virt_to_phys(page_table);
547 
548 	dma_unmap_single(iommu->dev, pt_phys, PT_SIZE, DMA_TO_DEVICE);
549 	kmem_cache_free(iommu->pt_pool, page_table);
550 }
551 
552 static u32 *sun50i_dte_get_page_table(struct sun50i_iommu_domain *sun50i_domain,
553 				      dma_addr_t iova, gfp_t gfp)
554 {
555 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
556 	u32 *page_table;
557 	u32 *dte_addr;
558 	u32 old_dte;
559 	u32 dte;
560 
561 	dte_addr = &sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
562 	dte = *dte_addr;
563 	if (sun50i_dte_is_pt_valid(dte)) {
564 		phys_addr_t pt_phys = sun50i_dte_get_pt_address(dte);
565 		return (u32 *)phys_to_virt(pt_phys);
566 	}
567 
568 	page_table = sun50i_iommu_alloc_page_table(iommu, gfp);
569 	if (IS_ERR(page_table))
570 		return page_table;
571 
572 	dte = sun50i_mk_dte(virt_to_phys(page_table));
573 	old_dte = cmpxchg(dte_addr, 0, dte);
574 	if (old_dte) {
575 		phys_addr_t installed_pt_phys =
576 			sun50i_dte_get_pt_address(old_dte);
577 		u32 *installed_pt = phys_to_virt(installed_pt_phys);
578 		u32 *drop_pt = page_table;
579 
580 		page_table = installed_pt;
581 		dte = old_dte;
582 		sun50i_iommu_free_page_table(iommu, drop_pt);
583 	}
584 
585 	sun50i_table_flush(sun50i_domain, page_table, NUM_PT_ENTRIES);
586 	sun50i_table_flush(sun50i_domain, dte_addr, 1);
587 
588 	return page_table;
589 }
590 
591 static int sun50i_iommu_map(struct iommu_domain *domain, unsigned long iova,
592 			    phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
593 {
594 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
595 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
596 	u32 pte_index;
597 	u32 *page_table, *pte_addr;
598 	int ret = 0;
599 
600 	page_table = sun50i_dte_get_page_table(sun50i_domain, iova, gfp);
601 	if (IS_ERR(page_table)) {
602 		ret = PTR_ERR(page_table);
603 		goto out;
604 	}
605 
606 	pte_index = sun50i_iova_get_pte_index(iova);
607 	pte_addr = &page_table[pte_index];
608 	if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
609 		phys_addr_t page_phys = sun50i_pte_get_page_address(*pte_addr);
610 		dev_err(iommu->dev,
611 			"iova %pad already mapped to %pa cannot remap to %pa prot: %#x\n",
612 			&iova, &page_phys, &paddr, prot);
613 		ret = -EBUSY;
614 		goto out;
615 	}
616 
617 	*pte_addr = sun50i_mk_pte(paddr, prot);
618 	sun50i_table_flush(sun50i_domain, pte_addr, 1);
619 
620 out:
621 	return ret;
622 }
623 
624 static size_t sun50i_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
625 				 size_t size, struct iommu_iotlb_gather *gather)
626 {
627 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
628 	phys_addr_t pt_phys;
629 	u32 *pte_addr;
630 	u32 dte;
631 
632 	dte = sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
633 	if (!sun50i_dte_is_pt_valid(dte))
634 		return 0;
635 
636 	pt_phys = sun50i_dte_get_pt_address(dte);
637 	pte_addr = (u32 *)phys_to_virt(pt_phys) + sun50i_iova_get_pte_index(iova);
638 
639 	if (!sun50i_pte_is_page_valid(*pte_addr))
640 		return 0;
641 
642 	memset(pte_addr, 0, sizeof(*pte_addr));
643 	sun50i_table_flush(sun50i_domain, pte_addr, 1);
644 
645 	return SZ_4K;
646 }
647 
648 static phys_addr_t sun50i_iommu_iova_to_phys(struct iommu_domain *domain,
649 					     dma_addr_t iova)
650 {
651 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
652 	phys_addr_t pt_phys;
653 	u32 *page_table;
654 	u32 dte, pte;
655 
656 	dte = sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
657 	if (!sun50i_dte_is_pt_valid(dte))
658 		return 0;
659 
660 	pt_phys = sun50i_dte_get_pt_address(dte);
661 	page_table = (u32 *)phys_to_virt(pt_phys);
662 	pte = page_table[sun50i_iova_get_pte_index(iova)];
663 	if (!sun50i_pte_is_page_valid(pte))
664 		return 0;
665 
666 	return sun50i_pte_get_page_address(pte) +
667 		sun50i_iova_get_page_offset(iova);
668 }
669 
670 static struct iommu_domain *sun50i_iommu_domain_alloc(unsigned type)
671 {
672 	struct sun50i_iommu_domain *sun50i_domain;
673 
674 	if (type != IOMMU_DOMAIN_DMA &&
675 	    type != IOMMU_DOMAIN_UNMANAGED)
676 		return NULL;
677 
678 	sun50i_domain = kzalloc(sizeof(*sun50i_domain), GFP_KERNEL);
679 	if (!sun50i_domain)
680 		return NULL;
681 
682 	sun50i_domain->dt = (u32 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
683 						    get_order(DT_SIZE));
684 	if (!sun50i_domain->dt)
685 		goto err_free_domain;
686 
687 	refcount_set(&sun50i_domain->refcnt, 1);
688 
689 	sun50i_domain->domain.geometry.aperture_start = 0;
690 	sun50i_domain->domain.geometry.aperture_end = DMA_BIT_MASK(32);
691 	sun50i_domain->domain.geometry.force_aperture = true;
692 
693 	return &sun50i_domain->domain;
694 
695 err_free_domain:
696 	kfree(sun50i_domain);
697 
698 	return NULL;
699 }
700 
701 static void sun50i_iommu_domain_free(struct iommu_domain *domain)
702 {
703 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
704 
705 	free_pages((unsigned long)sun50i_domain->dt, get_order(DT_SIZE));
706 	sun50i_domain->dt = NULL;
707 
708 	kfree(sun50i_domain);
709 }
710 
711 static int sun50i_iommu_attach_domain(struct sun50i_iommu *iommu,
712 				      struct sun50i_iommu_domain *sun50i_domain)
713 {
714 	iommu->domain = &sun50i_domain->domain;
715 	sun50i_domain->iommu = iommu;
716 
717 	sun50i_domain->dt_dma = dma_map_single(iommu->dev, sun50i_domain->dt,
718 					       DT_SIZE, DMA_TO_DEVICE);
719 	if (dma_mapping_error(iommu->dev, sun50i_domain->dt_dma)) {
720 		dev_err(iommu->dev, "Couldn't map L1 Page Table\n");
721 		return -ENOMEM;
722 	}
723 
724 	return sun50i_iommu_enable(iommu);
725 }
726 
727 static void sun50i_iommu_detach_domain(struct sun50i_iommu *iommu,
728 				       struct sun50i_iommu_domain *sun50i_domain)
729 {
730 	unsigned int i;
731 
732 	for (i = 0; i < NUM_DT_ENTRIES; i++) {
733 		phys_addr_t pt_phys;
734 		u32 *page_table;
735 		u32 *dte_addr;
736 		u32 dte;
737 
738 		dte_addr = &sun50i_domain->dt[i];
739 		dte = *dte_addr;
740 		if (!sun50i_dte_is_pt_valid(dte))
741 			continue;
742 
743 		memset(dte_addr, 0, sizeof(*dte_addr));
744 		sun50i_table_flush(sun50i_domain, dte_addr, 1);
745 
746 		pt_phys = sun50i_dte_get_pt_address(dte);
747 		page_table = phys_to_virt(pt_phys);
748 		sun50i_iommu_free_page_table(iommu, page_table);
749 	}
750 
751 
752 	sun50i_iommu_disable(iommu);
753 
754 	dma_unmap_single(iommu->dev, virt_to_phys(sun50i_domain->dt),
755 			 DT_SIZE, DMA_TO_DEVICE);
756 
757 	iommu->domain = NULL;
758 }
759 
760 static void sun50i_iommu_detach_device(struct iommu_domain *domain,
761 				       struct device *dev)
762 {
763 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
764 	struct sun50i_iommu *iommu = dev_iommu_priv_get(dev);
765 
766 	dev_dbg(dev, "Detaching from IOMMU domain\n");
767 
768 	if (iommu->domain != domain)
769 		return;
770 
771 	if (refcount_dec_and_test(&sun50i_domain->refcnt))
772 		sun50i_iommu_detach_domain(iommu, sun50i_domain);
773 }
774 
775 static int sun50i_iommu_attach_device(struct iommu_domain *domain,
776 				      struct device *dev)
777 {
778 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
779 	struct sun50i_iommu *iommu;
780 
781 	iommu = sun50i_iommu_from_dev(dev);
782 	if (!iommu)
783 		return -ENODEV;
784 
785 	dev_dbg(dev, "Attaching to IOMMU domain\n");
786 
787 	refcount_inc(&sun50i_domain->refcnt);
788 
789 	if (iommu->domain == domain)
790 		return 0;
791 
792 	if (iommu->domain)
793 		sun50i_iommu_detach_device(iommu->domain, dev);
794 
795 	sun50i_iommu_attach_domain(iommu, sun50i_domain);
796 
797 	return 0;
798 }
799 
800 static struct iommu_device *sun50i_iommu_probe_device(struct device *dev)
801 {
802 	struct sun50i_iommu *iommu;
803 
804 	iommu = sun50i_iommu_from_dev(dev);
805 	if (!iommu)
806 		return ERR_PTR(-ENODEV);
807 
808 	return &iommu->iommu;
809 }
810 
811 static struct iommu_group *sun50i_iommu_device_group(struct device *dev)
812 {
813 	struct sun50i_iommu *iommu = sun50i_iommu_from_dev(dev);
814 
815 	return iommu_group_ref_get(iommu->group);
816 }
817 
818 static int sun50i_iommu_of_xlate(struct device *dev,
819 				 struct of_phandle_args *args)
820 {
821 	struct platform_device *iommu_pdev = of_find_device_by_node(args->np);
822 	unsigned id = args->args[0];
823 
824 	dev_iommu_priv_set(dev, platform_get_drvdata(iommu_pdev));
825 
826 	return iommu_fwspec_add_ids(dev, &id, 1);
827 }
828 
829 static const struct iommu_ops sun50i_iommu_ops = {
830 	.pgsize_bitmap	= SZ_4K,
831 	.device_group	= sun50i_iommu_device_group,
832 	.domain_alloc	= sun50i_iommu_domain_alloc,
833 	.of_xlate	= sun50i_iommu_of_xlate,
834 	.probe_device	= sun50i_iommu_probe_device,
835 	.default_domain_ops = &(const struct iommu_domain_ops) {
836 		.attach_dev	= sun50i_iommu_attach_device,
837 		.flush_iotlb_all = sun50i_iommu_flush_iotlb_all,
838 		.iotlb_sync_map = sun50i_iommu_iotlb_sync_map,
839 		.iotlb_sync	= sun50i_iommu_iotlb_sync,
840 		.iova_to_phys	= sun50i_iommu_iova_to_phys,
841 		.map		= sun50i_iommu_map,
842 		.unmap		= sun50i_iommu_unmap,
843 		.free		= sun50i_iommu_domain_free,
844 	}
845 };
846 
847 static void sun50i_iommu_report_fault(struct sun50i_iommu *iommu,
848 				      unsigned master, phys_addr_t iova,
849 				      unsigned prot)
850 {
851 	dev_err(iommu->dev, "Page fault for %pad (master %d, dir %s)\n",
852 		&iova, master, (prot == IOMMU_FAULT_WRITE) ? "wr" : "rd");
853 
854 	if (iommu->domain)
855 		report_iommu_fault(iommu->domain, iommu->dev, iova, prot);
856 	else
857 		dev_err(iommu->dev, "Page fault while iommu not attached to any domain?\n");
858 
859 	sun50i_iommu_zap_range(iommu, iova, SPAGE_SIZE);
860 }
861 
862 static phys_addr_t sun50i_iommu_handle_pt_irq(struct sun50i_iommu *iommu,
863 					      unsigned addr_reg,
864 					      unsigned blame_reg)
865 {
866 	phys_addr_t iova;
867 	unsigned master;
868 	u32 blame;
869 
870 	assert_spin_locked(&iommu->iommu_lock);
871 
872 	iova = iommu_read(iommu, addr_reg);
873 	blame = iommu_read(iommu, blame_reg);
874 	master = ilog2(blame & IOMMU_INT_MASTER_MASK);
875 
876 	/*
877 	 * If the address is not in the page table, we can't get what
878 	 * operation triggered the fault. Assume it's a read
879 	 * operation.
880 	 */
881 	sun50i_iommu_report_fault(iommu, master, iova, IOMMU_FAULT_READ);
882 
883 	return iova;
884 }
885 
886 static phys_addr_t sun50i_iommu_handle_perm_irq(struct sun50i_iommu *iommu)
887 {
888 	enum sun50i_iommu_aci aci;
889 	phys_addr_t iova;
890 	unsigned master;
891 	unsigned dir;
892 	u32 blame;
893 
894 	assert_spin_locked(&iommu->iommu_lock);
895 
896 	blame = iommu_read(iommu, IOMMU_INT_STA_REG);
897 	master = ilog2(blame & IOMMU_INT_MASTER_MASK);
898 	iova = iommu_read(iommu, IOMMU_INT_ERR_ADDR_REG(master));
899 	aci = sun50i_get_pte_aci(iommu_read(iommu,
900 					    IOMMU_INT_ERR_DATA_REG(master)));
901 
902 	switch (aci) {
903 		/*
904 		 * If we are in the read-only domain, then it means we
905 		 * tried to write.
906 		 */
907 	case SUN50I_IOMMU_ACI_RD:
908 		dir = IOMMU_FAULT_WRITE;
909 		break;
910 
911 		/*
912 		 * If we are in the write-only domain, then it means
913 		 * we tried to read.
914 		 */
915 	case SUN50I_IOMMU_ACI_WR:
916 
917 		/*
918 		 * If we are in the domain without any permission, we
919 		 * can't really tell. Let's default to a read
920 		 * operation.
921 		 */
922 	case SUN50I_IOMMU_ACI_NONE:
923 
924 		/* WTF? */
925 	case SUN50I_IOMMU_ACI_RD_WR:
926 	default:
927 		dir = IOMMU_FAULT_READ;
928 		break;
929 	}
930 
931 	/*
932 	 * If the address is not in the page table, we can't get what
933 	 * operation triggered the fault. Assume it's a read
934 	 * operation.
935 	 */
936 	sun50i_iommu_report_fault(iommu, master, iova, dir);
937 
938 	return iova;
939 }
940 
941 static irqreturn_t sun50i_iommu_irq(int irq, void *dev_id)
942 {
943 	u32 status, l1_status, l2_status, resets;
944 	struct sun50i_iommu *iommu = dev_id;
945 
946 	spin_lock(&iommu->iommu_lock);
947 
948 	status = iommu_read(iommu, IOMMU_INT_STA_REG);
949 	if (!(status & IOMMU_INT_MASK)) {
950 		spin_unlock(&iommu->iommu_lock);
951 		return IRQ_NONE;
952 	}
953 
954 	l1_status = iommu_read(iommu, IOMMU_L1PG_INT_REG);
955 	l2_status = iommu_read(iommu, IOMMU_L2PG_INT_REG);
956 
957 	if (status & IOMMU_INT_INVALID_L2PG)
958 		sun50i_iommu_handle_pt_irq(iommu,
959 					    IOMMU_INT_ERR_ADDR_L2_REG,
960 					    IOMMU_L2PG_INT_REG);
961 	else if (status & IOMMU_INT_INVALID_L1PG)
962 		sun50i_iommu_handle_pt_irq(iommu,
963 					   IOMMU_INT_ERR_ADDR_L1_REG,
964 					   IOMMU_L1PG_INT_REG);
965 	else
966 		sun50i_iommu_handle_perm_irq(iommu);
967 
968 	iommu_write(iommu, IOMMU_INT_CLR_REG, status);
969 
970 	resets = (status | l1_status | l2_status) & IOMMU_INT_MASTER_MASK;
971 	iommu_write(iommu, IOMMU_RESET_REG, ~resets);
972 	iommu_write(iommu, IOMMU_RESET_REG, IOMMU_RESET_RELEASE_ALL);
973 
974 	spin_unlock(&iommu->iommu_lock);
975 
976 	return IRQ_HANDLED;
977 }
978 
979 static int sun50i_iommu_probe(struct platform_device *pdev)
980 {
981 	struct sun50i_iommu *iommu;
982 	int ret, irq;
983 
984 	iommu = devm_kzalloc(&pdev->dev, sizeof(*iommu), GFP_KERNEL);
985 	if (!iommu)
986 		return -ENOMEM;
987 	spin_lock_init(&iommu->iommu_lock);
988 	platform_set_drvdata(pdev, iommu);
989 	iommu->dev = &pdev->dev;
990 
991 	iommu->pt_pool = kmem_cache_create(dev_name(&pdev->dev),
992 					   PT_SIZE, PT_SIZE,
993 					   SLAB_HWCACHE_ALIGN,
994 					   NULL);
995 	if (!iommu->pt_pool)
996 		return -ENOMEM;
997 
998 	iommu->group = iommu_group_alloc();
999 	if (IS_ERR(iommu->group)) {
1000 		ret = PTR_ERR(iommu->group);
1001 		goto err_free_cache;
1002 	}
1003 
1004 	iommu->base = devm_platform_ioremap_resource(pdev, 0);
1005 	if (IS_ERR(iommu->base)) {
1006 		ret = PTR_ERR(iommu->base);
1007 		goto err_free_group;
1008 	}
1009 
1010 	irq = platform_get_irq(pdev, 0);
1011 	if (irq < 0) {
1012 		ret = irq;
1013 		goto err_free_group;
1014 	}
1015 
1016 	iommu->clk = devm_clk_get(&pdev->dev, NULL);
1017 	if (IS_ERR(iommu->clk)) {
1018 		dev_err(&pdev->dev, "Couldn't get our clock.\n");
1019 		ret = PTR_ERR(iommu->clk);
1020 		goto err_free_group;
1021 	}
1022 
1023 	iommu->reset = devm_reset_control_get(&pdev->dev, NULL);
1024 	if (IS_ERR(iommu->reset)) {
1025 		dev_err(&pdev->dev, "Couldn't get our reset line.\n");
1026 		ret = PTR_ERR(iommu->reset);
1027 		goto err_free_group;
1028 	}
1029 
1030 	ret = iommu_device_sysfs_add(&iommu->iommu, &pdev->dev,
1031 				     NULL, dev_name(&pdev->dev));
1032 	if (ret)
1033 		goto err_free_group;
1034 
1035 	ret = iommu_device_register(&iommu->iommu, &sun50i_iommu_ops, &pdev->dev);
1036 	if (ret)
1037 		goto err_remove_sysfs;
1038 
1039 	ret = devm_request_irq(&pdev->dev, irq, sun50i_iommu_irq, 0,
1040 			       dev_name(&pdev->dev), iommu);
1041 	if (ret < 0)
1042 		goto err_unregister;
1043 
1044 	return 0;
1045 
1046 err_unregister:
1047 	iommu_device_unregister(&iommu->iommu);
1048 
1049 err_remove_sysfs:
1050 	iommu_device_sysfs_remove(&iommu->iommu);
1051 
1052 err_free_group:
1053 	iommu_group_put(iommu->group);
1054 
1055 err_free_cache:
1056 	kmem_cache_destroy(iommu->pt_pool);
1057 
1058 	return ret;
1059 }
1060 
1061 static const struct of_device_id sun50i_iommu_dt[] = {
1062 	{ .compatible = "allwinner,sun50i-h6-iommu", },
1063 	{ /* sentinel */ },
1064 };
1065 MODULE_DEVICE_TABLE(of, sun50i_iommu_dt);
1066 
1067 static struct platform_driver sun50i_iommu_driver = {
1068 	.driver		= {
1069 		.name			= "sun50i-iommu",
1070 		.of_match_table 	= sun50i_iommu_dt,
1071 		.suppress_bind_attrs	= true,
1072 	}
1073 };
1074 builtin_platform_driver_probe(sun50i_iommu_driver, sun50i_iommu_probe);
1075 
1076 MODULE_DESCRIPTION("Allwinner H6 IOMMU driver");
1077 MODULE_AUTHOR("Maxime Ripard <maxime@cerno.tech>");
1078 MODULE_AUTHOR("zhuxianbin <zhuxianbin@allwinnertech.com>");
1079 MODULE_LICENSE("Dual BSD/GPL");
1080