xref: /linux/drivers/iommu/sun50i-iommu.c (revision c5dbf04160005e07e8ca7232a7faa77ab1547ae0)
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 kmem_cache *pt_pool;
111 };
112 
113 struct sun50i_iommu_domain {
114 	struct iommu_domain domain;
115 
116 	/* Number of devices attached to the domain */
117 	refcount_t refcnt;
118 
119 	/* L1 Page Table */
120 	u32 *dt;
121 	dma_addr_t dt_dma;
122 
123 	struct sun50i_iommu *iommu;
124 };
125 
126 static struct sun50i_iommu_domain *to_sun50i_domain(struct iommu_domain *domain)
127 {
128 	return container_of(domain, struct sun50i_iommu_domain, domain);
129 }
130 
131 static struct sun50i_iommu *sun50i_iommu_from_dev(struct device *dev)
132 {
133 	return dev_iommu_priv_get(dev);
134 }
135 
136 static u32 iommu_read(struct sun50i_iommu *iommu, u32 offset)
137 {
138 	return readl(iommu->base + offset);
139 }
140 
141 static void iommu_write(struct sun50i_iommu *iommu, u32 offset, u32 value)
142 {
143 	writel(value, iommu->base + offset);
144 }
145 
146 /*
147  * The Allwinner H6 IOMMU uses a 2-level page table.
148  *
149  * The first level is the usual Directory Table (DT), that consists of
150  * 4096 4-bytes Directory Table Entries (DTE), each pointing to a Page
151  * Table (PT).
152  *
153  * Each PT consits of 256 4-bytes Page Table Entries (PTE), each
154  * pointing to a 4kB page of physical memory.
155  *
156  * The IOMMU supports a single DT, pointed by the IOMMU_TTB_REG
157  * register that contains its physical address.
158  */
159 
160 #define SUN50I_IOVA_DTE_MASK	GENMASK(31, 20)
161 #define SUN50I_IOVA_PTE_MASK	GENMASK(19, 12)
162 #define SUN50I_IOVA_PAGE_MASK	GENMASK(11, 0)
163 
164 static u32 sun50i_iova_get_dte_index(dma_addr_t iova)
165 {
166 	return FIELD_GET(SUN50I_IOVA_DTE_MASK, iova);
167 }
168 
169 static u32 sun50i_iova_get_pte_index(dma_addr_t iova)
170 {
171 	return FIELD_GET(SUN50I_IOVA_PTE_MASK, iova);
172 }
173 
174 static u32 sun50i_iova_get_page_offset(dma_addr_t iova)
175 {
176 	return FIELD_GET(SUN50I_IOVA_PAGE_MASK, iova);
177 }
178 
179 /*
180  * Each Directory Table Entry has a Page Table address and a valid
181  * bit:
182 
183  * +---------------------+-----------+-+
184  * | PT address          | Reserved  |V|
185  * +---------------------+-----------+-+
186  *  31:10 - Page Table address
187  *   9:2  - Reserved
188  *   1:0  - 1 if the entry is valid
189  */
190 
191 #define SUN50I_DTE_PT_ADDRESS_MASK	GENMASK(31, 10)
192 #define SUN50I_DTE_PT_ATTRS		GENMASK(1, 0)
193 #define SUN50I_DTE_PT_VALID		1
194 
195 static phys_addr_t sun50i_dte_get_pt_address(u32 dte)
196 {
197 	return (phys_addr_t)dte & SUN50I_DTE_PT_ADDRESS_MASK;
198 }
199 
200 static bool sun50i_dte_is_pt_valid(u32 dte)
201 {
202 	return (dte & SUN50I_DTE_PT_ATTRS) == SUN50I_DTE_PT_VALID;
203 }
204 
205 static u32 sun50i_mk_dte(dma_addr_t pt_dma)
206 {
207 	return (pt_dma & SUN50I_DTE_PT_ADDRESS_MASK) | SUN50I_DTE_PT_VALID;
208 }
209 
210 /*
211  * Each PTE has a Page address, an authority index and a valid bit:
212  *
213  * +----------------+-----+-----+-----+---+-----+
214  * | Page address   | Rsv | ACI | Rsv | V | Rsv |
215  * +----------------+-----+-----+-----+---+-----+
216  *  31:12 - Page address
217  *  11:8  - Reserved
218  *   7:4  - Authority Control Index
219  *   3:2  - Reserved
220  *     1  - 1 if the entry is valid
221  *     0  - Reserved
222  *
223  * The way permissions work is that the IOMMU has 16 "domains" that
224  * can be configured to give each masters either read or write
225  * permissions through the IOMMU_DM_AUT_CTRL_REG registers. The domain
226  * 0 seems like the default domain, and its permissions in the
227  * IOMMU_DM_AUT_CTRL_REG are only read-only, so it's not really
228  * useful to enforce any particular permission.
229  *
230  * Each page entry will then have a reference to the domain they are
231  * affected to, so that we can actually enforce them on a per-page
232  * basis.
233  *
234  * In order to make it work with the IOMMU framework, we will be using
235  * 4 different domains, starting at 1: RD_WR, RD, WR and NONE
236  * depending on the permission we want to enforce. Each domain will
237  * have each master setup in the same way, since the IOMMU framework
238  * doesn't seem to restrict page access on a per-device basis. And
239  * then we will use the relevant domain index when generating the page
240  * table entry depending on the permissions we want to be enforced.
241  */
242 
243 enum sun50i_iommu_aci {
244 	SUN50I_IOMMU_ACI_DO_NOT_USE = 0,
245 	SUN50I_IOMMU_ACI_NONE,
246 	SUN50I_IOMMU_ACI_RD,
247 	SUN50I_IOMMU_ACI_WR,
248 	SUN50I_IOMMU_ACI_RD_WR,
249 };
250 
251 #define SUN50I_PTE_PAGE_ADDRESS_MASK	GENMASK(31, 12)
252 #define SUN50I_PTE_ACI_MASK		GENMASK(7, 4)
253 #define SUN50I_PTE_PAGE_VALID		BIT(1)
254 
255 static phys_addr_t sun50i_pte_get_page_address(u32 pte)
256 {
257 	return (phys_addr_t)pte & SUN50I_PTE_PAGE_ADDRESS_MASK;
258 }
259 
260 static enum sun50i_iommu_aci sun50i_get_pte_aci(u32 pte)
261 {
262 	return FIELD_GET(SUN50I_PTE_ACI_MASK, pte);
263 }
264 
265 static bool sun50i_pte_is_page_valid(u32 pte)
266 {
267 	return pte & SUN50I_PTE_PAGE_VALID;
268 }
269 
270 static u32 sun50i_mk_pte(phys_addr_t page, int prot)
271 {
272 	enum sun50i_iommu_aci aci;
273 	u32 flags = 0;
274 
275 	if ((prot & (IOMMU_READ | IOMMU_WRITE)) == (IOMMU_READ | IOMMU_WRITE))
276 		aci = SUN50I_IOMMU_ACI_RD_WR;
277 	else if (prot & IOMMU_READ)
278 		aci = SUN50I_IOMMU_ACI_RD;
279 	else if (prot & IOMMU_WRITE)
280 		aci = SUN50I_IOMMU_ACI_WR;
281 	else
282 		aci = SUN50I_IOMMU_ACI_NONE;
283 
284 	flags |= FIELD_PREP(SUN50I_PTE_ACI_MASK, aci);
285 	page &= SUN50I_PTE_PAGE_ADDRESS_MASK;
286 	return page | flags | SUN50I_PTE_PAGE_VALID;
287 }
288 
289 static void sun50i_table_flush(struct sun50i_iommu_domain *sun50i_domain,
290 			       void *vaddr, unsigned int count)
291 {
292 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
293 	dma_addr_t dma = virt_to_phys(vaddr);
294 	size_t size = count * PT_ENTRY_SIZE;
295 
296 	dma_sync_single_for_device(iommu->dev, dma, size, DMA_TO_DEVICE);
297 }
298 
299 static void sun50i_iommu_zap_iova(struct sun50i_iommu *iommu,
300 				  unsigned long iova)
301 {
302 	u32 reg;
303 	int ret;
304 
305 	iommu_write(iommu, IOMMU_TLB_IVLD_ADDR_REG, iova);
306 	iommu_write(iommu, IOMMU_TLB_IVLD_ADDR_MASK_REG, GENMASK(31, 12));
307 	iommu_write(iommu, IOMMU_TLB_IVLD_ENABLE_REG,
308 		    IOMMU_TLB_IVLD_ENABLE_ENABLE);
309 
310 	ret = readl_poll_timeout_atomic(iommu->base + IOMMU_TLB_IVLD_ENABLE_REG,
311 					reg, !reg, 1, 2000);
312 	if (ret)
313 		dev_warn(iommu->dev, "TLB invalidation timed out!\n");
314 }
315 
316 static void sun50i_iommu_zap_ptw_cache(struct sun50i_iommu *iommu,
317 				       unsigned long iova)
318 {
319 	u32 reg;
320 	int ret;
321 
322 	iommu_write(iommu, IOMMU_PC_IVLD_ADDR_REG, iova);
323 	iommu_write(iommu, IOMMU_PC_IVLD_ENABLE_REG,
324 		    IOMMU_PC_IVLD_ENABLE_ENABLE);
325 
326 	ret = readl_poll_timeout_atomic(iommu->base + IOMMU_PC_IVLD_ENABLE_REG,
327 					reg, !reg, 1, 2000);
328 	if (ret)
329 		dev_warn(iommu->dev, "PTW cache invalidation timed out!\n");
330 }
331 
332 static void sun50i_iommu_zap_range(struct sun50i_iommu *iommu,
333 				   unsigned long iova, size_t size)
334 {
335 	assert_spin_locked(&iommu->iommu_lock);
336 
337 	iommu_write(iommu, IOMMU_AUTO_GATING_REG, 0);
338 
339 	sun50i_iommu_zap_iova(iommu, iova);
340 	sun50i_iommu_zap_iova(iommu, iova + SPAGE_SIZE);
341 	if (size > SPAGE_SIZE) {
342 		sun50i_iommu_zap_iova(iommu, iova + size);
343 		sun50i_iommu_zap_iova(iommu, iova + size + SPAGE_SIZE);
344 	}
345 	sun50i_iommu_zap_ptw_cache(iommu, iova);
346 	sun50i_iommu_zap_ptw_cache(iommu, iova + SZ_1M);
347 	if (size > SZ_1M) {
348 		sun50i_iommu_zap_ptw_cache(iommu, iova + size);
349 		sun50i_iommu_zap_ptw_cache(iommu, iova + size + SZ_1M);
350 	}
351 
352 	iommu_write(iommu, IOMMU_AUTO_GATING_REG, IOMMU_AUTO_GATING_ENABLE);
353 }
354 
355 static int sun50i_iommu_flush_all_tlb(struct sun50i_iommu *iommu)
356 {
357 	u32 reg;
358 	int ret;
359 
360 	assert_spin_locked(&iommu->iommu_lock);
361 
362 	iommu_write(iommu,
363 		    IOMMU_TLB_FLUSH_REG,
364 		    IOMMU_TLB_FLUSH_PTW_CACHE |
365 		    IOMMU_TLB_FLUSH_MACRO_TLB |
366 		    IOMMU_TLB_FLUSH_MICRO_TLB(5) |
367 		    IOMMU_TLB_FLUSH_MICRO_TLB(4) |
368 		    IOMMU_TLB_FLUSH_MICRO_TLB(3) |
369 		    IOMMU_TLB_FLUSH_MICRO_TLB(2) |
370 		    IOMMU_TLB_FLUSH_MICRO_TLB(1) |
371 		    IOMMU_TLB_FLUSH_MICRO_TLB(0));
372 
373 	ret = readl_poll_timeout_atomic(iommu->base + IOMMU_TLB_FLUSH_REG,
374 					reg, !reg,
375 					1, 2000);
376 	if (ret)
377 		dev_warn(iommu->dev, "TLB Flush timed out!\n");
378 
379 	return ret;
380 }
381 
382 static void sun50i_iommu_flush_iotlb_all(struct iommu_domain *domain)
383 {
384 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
385 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
386 	unsigned long flags;
387 
388 	/*
389 	 * At boot, we'll have a first call into .flush_iotlb_all right after
390 	 * .probe_device, and since we link our (single) domain to our iommu in
391 	 * the .attach_device callback, we don't have that pointer set.
392 	 *
393 	 * It shouldn't really be any trouble to ignore it though since we flush
394 	 * all caches as part of the device powerup.
395 	 */
396 	if (!iommu)
397 		return;
398 
399 	spin_lock_irqsave(&iommu->iommu_lock, flags);
400 	sun50i_iommu_flush_all_tlb(iommu);
401 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
402 }
403 
404 static int sun50i_iommu_iotlb_sync_map(struct iommu_domain *domain,
405 				       unsigned long iova, size_t size)
406 {
407 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
408 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
409 	unsigned long flags;
410 
411 	spin_lock_irqsave(&iommu->iommu_lock, flags);
412 	sun50i_iommu_zap_range(iommu, iova, size);
413 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
414 
415 	return 0;
416 }
417 
418 static void sun50i_iommu_iotlb_sync(struct iommu_domain *domain,
419 				    struct iommu_iotlb_gather *gather)
420 {
421 	sun50i_iommu_flush_iotlb_all(domain);
422 }
423 
424 static int sun50i_iommu_enable(struct sun50i_iommu *iommu)
425 {
426 	struct sun50i_iommu_domain *sun50i_domain;
427 	unsigned long flags;
428 	int ret;
429 
430 	if (!iommu->domain)
431 		return 0;
432 
433 	sun50i_domain = to_sun50i_domain(iommu->domain);
434 
435 	ret = reset_control_deassert(iommu->reset);
436 	if (ret)
437 		return ret;
438 
439 	ret = clk_prepare_enable(iommu->clk);
440 	if (ret)
441 		goto err_reset_assert;
442 
443 	spin_lock_irqsave(&iommu->iommu_lock, flags);
444 
445 	iommu_write(iommu, IOMMU_TTB_REG, sun50i_domain->dt_dma);
446 	iommu_write(iommu, IOMMU_TLB_PREFETCH_REG,
447 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(0) |
448 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(1) |
449 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(2) |
450 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(3) |
451 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(4) |
452 		    IOMMU_TLB_PREFETCH_MASTER_ENABLE(5));
453 	iommu_write(iommu, IOMMU_INT_ENABLE_REG, IOMMU_INT_MASK);
454 	iommu_write(iommu, IOMMU_DM_AUT_CTRL_REG(SUN50I_IOMMU_ACI_NONE),
455 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 0) |
456 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 0) |
457 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 1) |
458 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 1) |
459 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 2) |
460 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 2) |
461 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 3) |
462 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 3) |
463 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 4) |
464 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 4) |
465 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 5) |
466 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_NONE, 5));
467 
468 	iommu_write(iommu, IOMMU_DM_AUT_CTRL_REG(SUN50I_IOMMU_ACI_RD),
469 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 0) |
470 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 1) |
471 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 2) |
472 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 3) |
473 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 4) |
474 		    IOMMU_DM_AUT_CTRL_WR_UNAVAIL(SUN50I_IOMMU_ACI_RD, 5));
475 
476 	iommu_write(iommu, IOMMU_DM_AUT_CTRL_REG(SUN50I_IOMMU_ACI_WR),
477 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 0) |
478 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 1) |
479 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 2) |
480 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 3) |
481 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 4) |
482 		    IOMMU_DM_AUT_CTRL_RD_UNAVAIL(SUN50I_IOMMU_ACI_WR, 5));
483 
484 	ret = sun50i_iommu_flush_all_tlb(iommu);
485 	if (ret) {
486 		spin_unlock_irqrestore(&iommu->iommu_lock, flags);
487 		goto err_clk_disable;
488 	}
489 
490 	iommu_write(iommu, IOMMU_AUTO_GATING_REG, IOMMU_AUTO_GATING_ENABLE);
491 	iommu_write(iommu, IOMMU_ENABLE_REG, IOMMU_ENABLE_ENABLE);
492 
493 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
494 
495 	return 0;
496 
497 err_clk_disable:
498 	clk_disable_unprepare(iommu->clk);
499 
500 err_reset_assert:
501 	reset_control_assert(iommu->reset);
502 
503 	return ret;
504 }
505 
506 static void sun50i_iommu_disable(struct sun50i_iommu *iommu)
507 {
508 	unsigned long flags;
509 
510 	spin_lock_irqsave(&iommu->iommu_lock, flags);
511 
512 	iommu_write(iommu, IOMMU_ENABLE_REG, 0);
513 	iommu_write(iommu, IOMMU_TTB_REG, 0);
514 
515 	spin_unlock_irqrestore(&iommu->iommu_lock, flags);
516 
517 	clk_disable_unprepare(iommu->clk);
518 	reset_control_assert(iommu->reset);
519 }
520 
521 static void *sun50i_iommu_alloc_page_table(struct sun50i_iommu *iommu,
522 					   gfp_t gfp)
523 {
524 	dma_addr_t pt_dma;
525 	u32 *page_table;
526 
527 	page_table = kmem_cache_zalloc(iommu->pt_pool, gfp);
528 	if (!page_table)
529 		return ERR_PTR(-ENOMEM);
530 
531 	pt_dma = dma_map_single(iommu->dev, page_table, PT_SIZE, DMA_TO_DEVICE);
532 	if (dma_mapping_error(iommu->dev, pt_dma)) {
533 		dev_err(iommu->dev, "Couldn't map L2 Page Table\n");
534 		kmem_cache_free(iommu->pt_pool, page_table);
535 		return ERR_PTR(-ENOMEM);
536 	}
537 
538 	/* We rely on the physical address and DMA address being the same */
539 	WARN_ON(pt_dma != virt_to_phys(page_table));
540 
541 	return page_table;
542 }
543 
544 static void sun50i_iommu_free_page_table(struct sun50i_iommu *iommu,
545 					 u32 *page_table)
546 {
547 	phys_addr_t pt_phys = virt_to_phys(page_table);
548 
549 	dma_unmap_single(iommu->dev, pt_phys, PT_SIZE, DMA_TO_DEVICE);
550 	kmem_cache_free(iommu->pt_pool, page_table);
551 }
552 
553 static u32 *sun50i_dte_get_page_table(struct sun50i_iommu_domain *sun50i_domain,
554 				      dma_addr_t iova, gfp_t gfp)
555 {
556 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
557 	u32 *page_table;
558 	u32 *dte_addr;
559 	u32 old_dte;
560 	u32 dte;
561 
562 	dte_addr = &sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
563 	dte = *dte_addr;
564 	if (sun50i_dte_is_pt_valid(dte)) {
565 		phys_addr_t pt_phys = sun50i_dte_get_pt_address(dte);
566 		return (u32 *)phys_to_virt(pt_phys);
567 	}
568 
569 	page_table = sun50i_iommu_alloc_page_table(iommu, gfp);
570 	if (IS_ERR(page_table))
571 		return page_table;
572 
573 	dte = sun50i_mk_dte(virt_to_phys(page_table));
574 	old_dte = cmpxchg(dte_addr, 0, dte);
575 	if (old_dte) {
576 		phys_addr_t installed_pt_phys =
577 			sun50i_dte_get_pt_address(old_dte);
578 		u32 *installed_pt = phys_to_virt(installed_pt_phys);
579 		u32 *drop_pt = page_table;
580 
581 		page_table = installed_pt;
582 		dte = old_dte;
583 		sun50i_iommu_free_page_table(iommu, drop_pt);
584 	}
585 
586 	sun50i_table_flush(sun50i_domain, page_table, NUM_PT_ENTRIES);
587 	sun50i_table_flush(sun50i_domain, dte_addr, 1);
588 
589 	return page_table;
590 }
591 
592 static int sun50i_iommu_map(struct iommu_domain *domain, unsigned long iova,
593 			    phys_addr_t paddr, size_t size, size_t count,
594 			    int prot, gfp_t gfp, size_t *mapped)
595 {
596 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
597 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
598 	u32 pte_index;
599 	u32 *page_table, *pte_addr;
600 	int ret = 0;
601 
602 	page_table = sun50i_dte_get_page_table(sun50i_domain, iova, gfp);
603 	if (IS_ERR(page_table)) {
604 		ret = PTR_ERR(page_table);
605 		goto out;
606 	}
607 
608 	pte_index = sun50i_iova_get_pte_index(iova);
609 	pte_addr = &page_table[pte_index];
610 	if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
611 		phys_addr_t page_phys = sun50i_pte_get_page_address(*pte_addr);
612 		dev_err(iommu->dev,
613 			"iova %pad already mapped to %pa cannot remap to %pa prot: %#x\n",
614 			&iova, &page_phys, &paddr, prot);
615 		ret = -EBUSY;
616 		goto out;
617 	}
618 
619 	*pte_addr = sun50i_mk_pte(paddr, prot);
620 	sun50i_table_flush(sun50i_domain, pte_addr, 1);
621 	*mapped = size;
622 
623 out:
624 	return ret;
625 }
626 
627 static size_t sun50i_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
628 				 size_t size, size_t count, struct iommu_iotlb_gather *gather)
629 {
630 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
631 	phys_addr_t pt_phys;
632 	u32 *pte_addr;
633 	u32 dte;
634 
635 	dte = sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
636 	if (!sun50i_dte_is_pt_valid(dte))
637 		return 0;
638 
639 	pt_phys = sun50i_dte_get_pt_address(dte);
640 	pte_addr = (u32 *)phys_to_virt(pt_phys) + sun50i_iova_get_pte_index(iova);
641 
642 	if (!sun50i_pte_is_page_valid(*pte_addr))
643 		return 0;
644 
645 	memset(pte_addr, 0, sizeof(*pte_addr));
646 	sun50i_table_flush(sun50i_domain, pte_addr, 1);
647 
648 	return SZ_4K;
649 }
650 
651 static phys_addr_t sun50i_iommu_iova_to_phys(struct iommu_domain *domain,
652 					     dma_addr_t iova)
653 {
654 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
655 	phys_addr_t pt_phys;
656 	u32 *page_table;
657 	u32 dte, pte;
658 
659 	dte = sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
660 	if (!sun50i_dte_is_pt_valid(dte))
661 		return 0;
662 
663 	pt_phys = sun50i_dte_get_pt_address(dte);
664 	page_table = (u32 *)phys_to_virt(pt_phys);
665 	pte = page_table[sun50i_iova_get_pte_index(iova)];
666 	if (!sun50i_pte_is_page_valid(pte))
667 		return 0;
668 
669 	return sun50i_pte_get_page_address(pte) +
670 		sun50i_iova_get_page_offset(iova);
671 }
672 
673 static struct iommu_domain *
674 sun50i_iommu_domain_alloc_paging(struct device *dev)
675 {
676 	struct sun50i_iommu_domain *sun50i_domain;
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 int sun50i_iommu_identity_attach(struct iommu_domain *identity_domain,
761 					struct device *dev)
762 {
763 	struct sun50i_iommu *iommu = dev_iommu_priv_get(dev);
764 	struct sun50i_iommu_domain *sun50i_domain;
765 
766 	dev_dbg(dev, "Detaching from IOMMU domain\n");
767 
768 	if (iommu->domain == identity_domain)
769 		return 0;
770 
771 	sun50i_domain = to_sun50i_domain(iommu->domain);
772 	if (refcount_dec_and_test(&sun50i_domain->refcnt))
773 		sun50i_iommu_detach_domain(iommu, sun50i_domain);
774 	return 0;
775 }
776 
777 static struct iommu_domain_ops sun50i_iommu_identity_ops = {
778 	.attach_dev = sun50i_iommu_identity_attach,
779 };
780 
781 static struct iommu_domain sun50i_iommu_identity_domain = {
782 	.type = IOMMU_DOMAIN_IDENTITY,
783 	.ops = &sun50i_iommu_identity_ops,
784 };
785 
786 static int sun50i_iommu_attach_device(struct iommu_domain *domain,
787 				      struct device *dev)
788 {
789 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
790 	struct sun50i_iommu *iommu;
791 
792 	iommu = sun50i_iommu_from_dev(dev);
793 	if (!iommu)
794 		return -ENODEV;
795 
796 	dev_dbg(dev, "Attaching to IOMMU domain\n");
797 
798 	refcount_inc(&sun50i_domain->refcnt);
799 
800 	if (iommu->domain == domain)
801 		return 0;
802 
803 	sun50i_iommu_identity_attach(&sun50i_iommu_identity_domain, dev);
804 
805 	sun50i_iommu_attach_domain(iommu, sun50i_domain);
806 
807 	return 0;
808 }
809 
810 static struct iommu_device *sun50i_iommu_probe_device(struct device *dev)
811 {
812 	struct sun50i_iommu *iommu;
813 
814 	iommu = sun50i_iommu_from_dev(dev);
815 	if (!iommu)
816 		return ERR_PTR(-ENODEV);
817 
818 	return &iommu->iommu;
819 }
820 
821 static int sun50i_iommu_of_xlate(struct device *dev,
822 				 struct of_phandle_args *args)
823 {
824 	struct platform_device *iommu_pdev = of_find_device_by_node(args->np);
825 	unsigned id = args->args[0];
826 
827 	dev_iommu_priv_set(dev, platform_get_drvdata(iommu_pdev));
828 
829 	return iommu_fwspec_add_ids(dev, &id, 1);
830 }
831 
832 static const struct iommu_ops sun50i_iommu_ops = {
833 	.identity_domain = &sun50i_iommu_identity_domain,
834 	.pgsize_bitmap	= SZ_4K,
835 	.device_group	= generic_single_device_group,
836 	.domain_alloc_paging = sun50i_iommu_domain_alloc_paging,
837 	.of_xlate	= sun50i_iommu_of_xlate,
838 	.probe_device	= sun50i_iommu_probe_device,
839 	.default_domain_ops = &(const struct iommu_domain_ops) {
840 		.attach_dev	= sun50i_iommu_attach_device,
841 		.flush_iotlb_all = sun50i_iommu_flush_iotlb_all,
842 		.iotlb_sync_map = sun50i_iommu_iotlb_sync_map,
843 		.iotlb_sync	= sun50i_iommu_iotlb_sync,
844 		.iova_to_phys	= sun50i_iommu_iova_to_phys,
845 		.map_pages	= sun50i_iommu_map,
846 		.unmap_pages	= sun50i_iommu_unmap,
847 		.free		= sun50i_iommu_domain_free,
848 	}
849 };
850 
851 static void sun50i_iommu_report_fault(struct sun50i_iommu *iommu,
852 				      unsigned master, phys_addr_t iova,
853 				      unsigned prot)
854 {
855 	dev_err(iommu->dev, "Page fault for %pad (master %d, dir %s)\n",
856 		&iova, master, (prot == IOMMU_FAULT_WRITE) ? "wr" : "rd");
857 
858 	if (iommu->domain)
859 		report_iommu_fault(iommu->domain, iommu->dev, iova, prot);
860 	else
861 		dev_err(iommu->dev, "Page fault while iommu not attached to any domain?\n");
862 
863 	sun50i_iommu_zap_range(iommu, iova, SPAGE_SIZE);
864 }
865 
866 static phys_addr_t sun50i_iommu_handle_pt_irq(struct sun50i_iommu *iommu,
867 					      unsigned addr_reg,
868 					      unsigned blame_reg)
869 {
870 	phys_addr_t iova;
871 	unsigned master;
872 	u32 blame;
873 
874 	assert_spin_locked(&iommu->iommu_lock);
875 
876 	iova = iommu_read(iommu, addr_reg);
877 	blame = iommu_read(iommu, blame_reg);
878 	master = ilog2(blame & IOMMU_INT_MASTER_MASK);
879 
880 	/*
881 	 * If the address is not in the page table, we can't get what
882 	 * operation triggered the fault. Assume it's a read
883 	 * operation.
884 	 */
885 	sun50i_iommu_report_fault(iommu, master, iova, IOMMU_FAULT_READ);
886 
887 	return iova;
888 }
889 
890 static phys_addr_t sun50i_iommu_handle_perm_irq(struct sun50i_iommu *iommu)
891 {
892 	enum sun50i_iommu_aci aci;
893 	phys_addr_t iova;
894 	unsigned master;
895 	unsigned dir;
896 	u32 blame;
897 
898 	assert_spin_locked(&iommu->iommu_lock);
899 
900 	blame = iommu_read(iommu, IOMMU_INT_STA_REG);
901 	master = ilog2(blame & IOMMU_INT_MASTER_MASK);
902 	iova = iommu_read(iommu, IOMMU_INT_ERR_ADDR_REG(master));
903 	aci = sun50i_get_pte_aci(iommu_read(iommu,
904 					    IOMMU_INT_ERR_DATA_REG(master)));
905 
906 	switch (aci) {
907 		/*
908 		 * If we are in the read-only domain, then it means we
909 		 * tried to write.
910 		 */
911 	case SUN50I_IOMMU_ACI_RD:
912 		dir = IOMMU_FAULT_WRITE;
913 		break;
914 
915 		/*
916 		 * If we are in the write-only domain, then it means
917 		 * we tried to read.
918 		 */
919 	case SUN50I_IOMMU_ACI_WR:
920 
921 		/*
922 		 * If we are in the domain without any permission, we
923 		 * can't really tell. Let's default to a read
924 		 * operation.
925 		 */
926 	case SUN50I_IOMMU_ACI_NONE:
927 
928 		/* WTF? */
929 	case SUN50I_IOMMU_ACI_RD_WR:
930 	default:
931 		dir = IOMMU_FAULT_READ;
932 		break;
933 	}
934 
935 	/*
936 	 * If the address is not in the page table, we can't get what
937 	 * operation triggered the fault. Assume it's a read
938 	 * operation.
939 	 */
940 	sun50i_iommu_report_fault(iommu, master, iova, dir);
941 
942 	return iova;
943 }
944 
945 static irqreturn_t sun50i_iommu_irq(int irq, void *dev_id)
946 {
947 	u32 status, l1_status, l2_status, resets;
948 	struct sun50i_iommu *iommu = dev_id;
949 
950 	spin_lock(&iommu->iommu_lock);
951 
952 	status = iommu_read(iommu, IOMMU_INT_STA_REG);
953 	if (!(status & IOMMU_INT_MASK)) {
954 		spin_unlock(&iommu->iommu_lock);
955 		return IRQ_NONE;
956 	}
957 
958 	l1_status = iommu_read(iommu, IOMMU_L1PG_INT_REG);
959 	l2_status = iommu_read(iommu, IOMMU_L2PG_INT_REG);
960 
961 	if (status & IOMMU_INT_INVALID_L2PG)
962 		sun50i_iommu_handle_pt_irq(iommu,
963 					    IOMMU_INT_ERR_ADDR_L2_REG,
964 					    IOMMU_L2PG_INT_REG);
965 	else if (status & IOMMU_INT_INVALID_L1PG)
966 		sun50i_iommu_handle_pt_irq(iommu,
967 					   IOMMU_INT_ERR_ADDR_L1_REG,
968 					   IOMMU_L1PG_INT_REG);
969 	else
970 		sun50i_iommu_handle_perm_irq(iommu);
971 
972 	iommu_write(iommu, IOMMU_INT_CLR_REG, status);
973 
974 	resets = (status | l1_status | l2_status) & IOMMU_INT_MASTER_MASK;
975 	iommu_write(iommu, IOMMU_RESET_REG, ~resets);
976 	iommu_write(iommu, IOMMU_RESET_REG, IOMMU_RESET_RELEASE_ALL);
977 
978 	spin_unlock(&iommu->iommu_lock);
979 
980 	return IRQ_HANDLED;
981 }
982 
983 static int sun50i_iommu_probe(struct platform_device *pdev)
984 {
985 	struct sun50i_iommu *iommu;
986 	int ret, irq;
987 
988 	iommu = devm_kzalloc(&pdev->dev, sizeof(*iommu), GFP_KERNEL);
989 	if (!iommu)
990 		return -ENOMEM;
991 	spin_lock_init(&iommu->iommu_lock);
992 	iommu->domain = &sun50i_iommu_identity_domain;
993 	platform_set_drvdata(pdev, iommu);
994 	iommu->dev = &pdev->dev;
995 
996 	iommu->pt_pool = kmem_cache_create(dev_name(&pdev->dev),
997 					   PT_SIZE, PT_SIZE,
998 					   SLAB_HWCACHE_ALIGN,
999 					   NULL);
1000 	if (!iommu->pt_pool)
1001 		return -ENOMEM;
1002 
1003 	iommu->base = devm_platform_ioremap_resource(pdev, 0);
1004 	if (IS_ERR(iommu->base)) {
1005 		ret = PTR_ERR(iommu->base);
1006 		goto err_free_cache;
1007 	}
1008 
1009 	irq = platform_get_irq(pdev, 0);
1010 	if (irq < 0) {
1011 		ret = irq;
1012 		goto err_free_cache;
1013 	}
1014 
1015 	iommu->clk = devm_clk_get(&pdev->dev, NULL);
1016 	if (IS_ERR(iommu->clk)) {
1017 		dev_err(&pdev->dev, "Couldn't get our clock.\n");
1018 		ret = PTR_ERR(iommu->clk);
1019 		goto err_free_cache;
1020 	}
1021 
1022 	iommu->reset = devm_reset_control_get(&pdev->dev, NULL);
1023 	if (IS_ERR(iommu->reset)) {
1024 		dev_err(&pdev->dev, "Couldn't get our reset line.\n");
1025 		ret = PTR_ERR(iommu->reset);
1026 		goto err_free_cache;
1027 	}
1028 
1029 	ret = iommu_device_sysfs_add(&iommu->iommu, &pdev->dev,
1030 				     NULL, dev_name(&pdev->dev));
1031 	if (ret)
1032 		goto err_free_cache;
1033 
1034 	ret = iommu_device_register(&iommu->iommu, &sun50i_iommu_ops, &pdev->dev);
1035 	if (ret)
1036 		goto err_remove_sysfs;
1037 
1038 	ret = devm_request_irq(&pdev->dev, irq, sun50i_iommu_irq, 0,
1039 			       dev_name(&pdev->dev), iommu);
1040 	if (ret < 0)
1041 		goto err_unregister;
1042 
1043 	return 0;
1044 
1045 err_unregister:
1046 	iommu_device_unregister(&iommu->iommu);
1047 
1048 err_remove_sysfs:
1049 	iommu_device_sysfs_remove(&iommu->iommu);
1050 
1051 err_free_cache:
1052 	kmem_cache_destroy(iommu->pt_pool);
1053 
1054 	return ret;
1055 }
1056 
1057 static const struct of_device_id sun50i_iommu_dt[] = {
1058 	{ .compatible = "allwinner,sun50i-h6-iommu", },
1059 	{ /* sentinel */ },
1060 };
1061 MODULE_DEVICE_TABLE(of, sun50i_iommu_dt);
1062 
1063 static struct platform_driver sun50i_iommu_driver = {
1064 	.driver		= {
1065 		.name			= "sun50i-iommu",
1066 		.of_match_table 	= sun50i_iommu_dt,
1067 		.suppress_bind_attrs	= true,
1068 	}
1069 };
1070 builtin_platform_driver_probe(sun50i_iommu_driver, sun50i_iommu_probe);
1071 
1072 MODULE_DESCRIPTION("Allwinner H6 IOMMU driver");
1073 MODULE_AUTHOR("Maxime Ripard <maxime@cerno.tech>");
1074 MODULE_AUTHOR("zhuxianbin <zhuxianbin@allwinnertech.com>");
1075