Commit | Line | Data |
---|---|---|
d53e54b4 HD |
1 | /* |
2 | * IOMMU API for GART in Tegra20 | |
3 | * | |
4 | * Copyright (c) 2010-2012, NVIDIA CORPORATION. All rights reserved. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms and conditions of the GNU General Public License, | |
8 | * version 2, as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | */ | |
19 | ||
20 | #define pr_fmt(fmt) "%s(): " fmt, __func__ | |
21 | ||
22 | #include <linux/module.h> | |
23 | #include <linux/platform_device.h> | |
24 | #include <linux/spinlock.h> | |
25 | #include <linux/slab.h> | |
26 | #include <linux/vmalloc.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/list.h> | |
29 | #include <linux/device.h> | |
30 | #include <linux/io.h> | |
31 | #include <linux/iommu.h> | |
7cffae42 | 32 | #include <linux/of.h> |
d53e54b4 HD |
33 | |
34 | #include <asm/cacheflush.h> | |
35 | ||
36 | /* bitmap of the page sizes currently supported */ | |
37 | #define GART_IOMMU_PGSIZES (SZ_4K) | |
38 | ||
774dfc9b HD |
39 | #define GART_REG_BASE 0x24 |
40 | #define GART_CONFIG (0x24 - GART_REG_BASE) | |
41 | #define GART_ENTRY_ADDR (0x28 - GART_REG_BASE) | |
42 | #define GART_ENTRY_DATA (0x2c - GART_REG_BASE) | |
d53e54b4 HD |
43 | #define GART_ENTRY_PHYS_ADDR_VALID (1 << 31) |
44 | ||
45 | #define GART_PAGE_SHIFT 12 | |
46 | #define GART_PAGE_SIZE (1 << GART_PAGE_SHIFT) | |
47 | #define GART_PAGE_MASK \ | |
48 | (~(GART_PAGE_SIZE - 1) & ~GART_ENTRY_PHYS_ADDR_VALID) | |
49 | ||
50 | struct gart_client { | |
51 | struct device *dev; | |
52 | struct list_head list; | |
53 | }; | |
54 | ||
55 | struct gart_device { | |
56 | void __iomem *regs; | |
57 | u32 *savedata; | |
58 | u32 page_count; /* total remappable size */ | |
59 | dma_addr_t iovmm_base; /* offset to vmm_area */ | |
60 | spinlock_t pte_lock; /* for pagetable */ | |
61 | struct list_head client; | |
62 | spinlock_t client_lock; /* for client list */ | |
63 | struct device *dev; | |
64 | }; | |
65 | ||
66 | static struct gart_device *gart_handle; /* unique for a system */ | |
67 | ||
68 | #define GART_PTE(_pfn) \ | |
69 | (GART_ENTRY_PHYS_ADDR_VALID | ((_pfn) << PAGE_SHIFT)) | |
70 | ||
71 | /* | |
72 | * Any interaction between any block on PPSB and a block on APB or AHB | |
73 | * must have these read-back to ensure the APB/AHB bus transaction is | |
74 | * complete before initiating activity on the PPSB block. | |
75 | */ | |
76 | #define FLUSH_GART_REGS(gart) ((void)readl((gart)->regs + GART_CONFIG)) | |
77 | ||
78 | #define for_each_gart_pte(gart, iova) \ | |
79 | for (iova = gart->iovmm_base; \ | |
80 | iova < gart->iovmm_base + GART_PAGE_SIZE * gart->page_count; \ | |
81 | iova += GART_PAGE_SIZE) | |
82 | ||
83 | static inline void gart_set_pte(struct gart_device *gart, | |
84 | unsigned long offs, u32 pte) | |
85 | { | |
86 | writel(offs, gart->regs + GART_ENTRY_ADDR); | |
87 | writel(pte, gart->regs + GART_ENTRY_DATA); | |
88 | ||
89 | dev_dbg(gart->dev, "%s %08lx:%08x\n", | |
90 | pte ? "map" : "unmap", offs, pte & GART_PAGE_MASK); | |
91 | } | |
92 | ||
93 | static inline unsigned long gart_read_pte(struct gart_device *gart, | |
94 | unsigned long offs) | |
95 | { | |
96 | unsigned long pte; | |
97 | ||
98 | writel(offs, gart->regs + GART_ENTRY_ADDR); | |
99 | pte = readl(gart->regs + GART_ENTRY_DATA); | |
100 | ||
101 | return pte; | |
102 | } | |
103 | ||
104 | static void do_gart_setup(struct gart_device *gart, const u32 *data) | |
105 | { | |
106 | unsigned long iova; | |
107 | ||
108 | for_each_gart_pte(gart, iova) | |
109 | gart_set_pte(gart, iova, data ? *(data++) : 0); | |
110 | ||
111 | writel(1, gart->regs + GART_CONFIG); | |
112 | FLUSH_GART_REGS(gart); | |
113 | } | |
114 | ||
115 | #ifdef DEBUG | |
116 | static void gart_dump_table(struct gart_device *gart) | |
117 | { | |
118 | unsigned long iova; | |
119 | unsigned long flags; | |
120 | ||
121 | spin_lock_irqsave(&gart->pte_lock, flags); | |
122 | for_each_gart_pte(gart, iova) { | |
123 | unsigned long pte; | |
124 | ||
125 | pte = gart_read_pte(gart, iova); | |
126 | ||
127 | dev_dbg(gart->dev, "%s %08lx:%08lx\n", | |
128 | (GART_ENTRY_PHYS_ADDR_VALID & pte) ? "v" : " ", | |
129 | iova, pte & GART_PAGE_MASK); | |
130 | } | |
131 | spin_unlock_irqrestore(&gart->pte_lock, flags); | |
132 | } | |
133 | #else | |
134 | static inline void gart_dump_table(struct gart_device *gart) | |
135 | { | |
136 | } | |
137 | #endif | |
138 | ||
139 | static inline bool gart_iova_range_valid(struct gart_device *gart, | |
140 | unsigned long iova, size_t bytes) | |
141 | { | |
142 | unsigned long iova_start, iova_end, gart_start, gart_end; | |
143 | ||
144 | iova_start = iova; | |
145 | iova_end = iova_start + bytes - 1; | |
146 | gart_start = gart->iovmm_base; | |
147 | gart_end = gart_start + gart->page_count * GART_PAGE_SIZE - 1; | |
148 | ||
149 | if (iova_start < gart_start) | |
150 | return false; | |
151 | if (iova_end > gart_end) | |
152 | return false; | |
153 | return true; | |
154 | } | |
155 | ||
156 | static int gart_iommu_attach_dev(struct iommu_domain *domain, | |
157 | struct device *dev) | |
158 | { | |
159 | struct gart_device *gart; | |
160 | struct gart_client *client, *c; | |
161 | int err = 0; | |
162 | ||
543f3f33 | 163 | gart = gart_handle; |
d53e54b4 HD |
164 | if (!gart) |
165 | return -EINVAL; | |
166 | domain->priv = gart; | |
167 | ||
168 | client = devm_kzalloc(gart->dev, sizeof(*c), GFP_KERNEL); | |
169 | if (!client) | |
170 | return -ENOMEM; | |
171 | client->dev = dev; | |
172 | ||
173 | spin_lock(&gart->client_lock); | |
174 | list_for_each_entry(c, &gart->client, list) { | |
175 | if (c->dev == dev) { | |
176 | dev_err(gart->dev, | |
177 | "%s is already attached\n", dev_name(dev)); | |
178 | err = -EINVAL; | |
179 | goto fail; | |
180 | } | |
181 | } | |
182 | list_add(&client->list, &gart->client); | |
183 | spin_unlock(&gart->client_lock); | |
184 | dev_dbg(gart->dev, "Attached %s\n", dev_name(dev)); | |
185 | return 0; | |
186 | ||
187 | fail: | |
188 | devm_kfree(gart->dev, client); | |
189 | spin_unlock(&gart->client_lock); | |
190 | return err; | |
191 | } | |
192 | ||
193 | static void gart_iommu_detach_dev(struct iommu_domain *domain, | |
194 | struct device *dev) | |
195 | { | |
196 | struct gart_device *gart = domain->priv; | |
197 | struct gart_client *c; | |
198 | ||
199 | spin_lock(&gart->client_lock); | |
200 | ||
201 | list_for_each_entry(c, &gart->client, list) { | |
202 | if (c->dev == dev) { | |
203 | list_del(&c->list); | |
204 | devm_kfree(gart->dev, c); | |
205 | dev_dbg(gart->dev, "Detached %s\n", dev_name(dev)); | |
206 | goto out; | |
207 | } | |
208 | } | |
209 | dev_err(gart->dev, "Couldn't find\n"); | |
210 | out: | |
211 | spin_unlock(&gart->client_lock); | |
212 | } | |
213 | ||
214 | static int gart_iommu_domain_init(struct iommu_domain *domain) | |
215 | { | |
216 | return 0; | |
217 | } | |
218 | ||
219 | static void gart_iommu_domain_destroy(struct iommu_domain *domain) | |
220 | { | |
221 | struct gart_device *gart = domain->priv; | |
222 | ||
223 | if (!gart) | |
224 | return; | |
225 | ||
226 | spin_lock(&gart->client_lock); | |
227 | if (!list_empty(&gart->client)) { | |
228 | struct gart_client *c; | |
229 | ||
230 | list_for_each_entry(c, &gart->client, list) | |
231 | gart_iommu_detach_dev(domain, c->dev); | |
232 | } | |
233 | spin_unlock(&gart->client_lock); | |
234 | domain->priv = NULL; | |
235 | } | |
236 | ||
237 | static int gart_iommu_map(struct iommu_domain *domain, unsigned long iova, | |
238 | phys_addr_t pa, size_t bytes, int prot) | |
239 | { | |
240 | struct gart_device *gart = domain->priv; | |
241 | unsigned long flags; | |
242 | unsigned long pfn; | |
243 | ||
244 | if (!gart_iova_range_valid(gart, iova, bytes)) | |
245 | return -EINVAL; | |
246 | ||
247 | spin_lock_irqsave(&gart->pte_lock, flags); | |
248 | pfn = __phys_to_pfn(pa); | |
249 | if (!pfn_valid(pfn)) { | |
250 | dev_err(gart->dev, "Invalid page: %08x\n", pa); | |
09c32533 | 251 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
d53e54b4 HD |
252 | return -EINVAL; |
253 | } | |
254 | gart_set_pte(gart, iova, GART_PTE(pfn)); | |
255 | FLUSH_GART_REGS(gart); | |
256 | spin_unlock_irqrestore(&gart->pte_lock, flags); | |
257 | return 0; | |
258 | } | |
259 | ||
260 | static size_t gart_iommu_unmap(struct iommu_domain *domain, unsigned long iova, | |
261 | size_t bytes) | |
262 | { | |
263 | struct gart_device *gart = domain->priv; | |
264 | unsigned long flags; | |
265 | ||
266 | if (!gart_iova_range_valid(gart, iova, bytes)) | |
267 | return 0; | |
268 | ||
269 | spin_lock_irqsave(&gart->pte_lock, flags); | |
270 | gart_set_pte(gart, iova, 0); | |
271 | FLUSH_GART_REGS(gart); | |
272 | spin_unlock_irqrestore(&gart->pte_lock, flags); | |
273 | return 0; | |
274 | } | |
275 | ||
276 | static phys_addr_t gart_iommu_iova_to_phys(struct iommu_domain *domain, | |
277 | unsigned long iova) | |
278 | { | |
279 | struct gart_device *gart = domain->priv; | |
280 | unsigned long pte; | |
281 | phys_addr_t pa; | |
282 | unsigned long flags; | |
283 | ||
284 | if (!gart_iova_range_valid(gart, iova, 0)) | |
285 | return -EINVAL; | |
286 | ||
287 | spin_lock_irqsave(&gart->pte_lock, flags); | |
288 | pte = gart_read_pte(gart, iova); | |
289 | spin_unlock_irqrestore(&gart->pte_lock, flags); | |
290 | ||
291 | pa = (pte & GART_PAGE_MASK); | |
292 | if (!pfn_valid(__phys_to_pfn(pa))) { | |
293 | dev_err(gart->dev, "No entry for %08lx:%08x\n", iova, pa); | |
294 | gart_dump_table(gart); | |
295 | return -EINVAL; | |
296 | } | |
297 | return pa; | |
298 | } | |
299 | ||
300 | static int gart_iommu_domain_has_cap(struct iommu_domain *domain, | |
301 | unsigned long cap) | |
302 | { | |
303 | return 0; | |
304 | } | |
305 | ||
306 | static struct iommu_ops gart_iommu_ops = { | |
307 | .domain_init = gart_iommu_domain_init, | |
308 | .domain_destroy = gart_iommu_domain_destroy, | |
309 | .attach_dev = gart_iommu_attach_dev, | |
310 | .detach_dev = gart_iommu_detach_dev, | |
311 | .map = gart_iommu_map, | |
312 | .unmap = gart_iommu_unmap, | |
313 | .iova_to_phys = gart_iommu_iova_to_phys, | |
314 | .domain_has_cap = gart_iommu_domain_has_cap, | |
315 | .pgsize_bitmap = GART_IOMMU_PGSIZES, | |
316 | }; | |
317 | ||
318 | static int tegra_gart_suspend(struct device *dev) | |
319 | { | |
320 | struct gart_device *gart = dev_get_drvdata(dev); | |
321 | unsigned long iova; | |
322 | u32 *data = gart->savedata; | |
323 | unsigned long flags; | |
324 | ||
325 | spin_lock_irqsave(&gart->pte_lock, flags); | |
326 | for_each_gart_pte(gart, iova) | |
327 | *(data++) = gart_read_pte(gart, iova); | |
328 | spin_unlock_irqrestore(&gart->pte_lock, flags); | |
329 | return 0; | |
330 | } | |
331 | ||
332 | static int tegra_gart_resume(struct device *dev) | |
333 | { | |
334 | struct gart_device *gart = dev_get_drvdata(dev); | |
335 | unsigned long flags; | |
336 | ||
337 | spin_lock_irqsave(&gart->pte_lock, flags); | |
338 | do_gart_setup(gart, gart->savedata); | |
339 | spin_unlock_irqrestore(&gart->pte_lock, flags); | |
340 | return 0; | |
341 | } | |
342 | ||
343 | static int tegra_gart_probe(struct platform_device *pdev) | |
344 | { | |
345 | struct gart_device *gart; | |
346 | struct resource *res, *res_remap; | |
347 | void __iomem *gart_regs; | |
348 | int err; | |
349 | struct device *dev = &pdev->dev; | |
350 | ||
351 | if (gart_handle) | |
352 | return -EIO; | |
353 | ||
354 | BUILD_BUG_ON(PAGE_SHIFT != GART_PAGE_SHIFT); | |
355 | ||
356 | /* the GART memory aperture is required */ | |
357 | res = platform_get_resource(pdev, IORESOURCE_MEM, 0); | |
358 | res_remap = platform_get_resource(pdev, IORESOURCE_MEM, 1); | |
359 | if (!res || !res_remap) { | |
360 | dev_err(dev, "GART memory aperture expected\n"); | |
361 | return -ENXIO; | |
362 | } | |
363 | ||
364 | gart = devm_kzalloc(dev, sizeof(*gart), GFP_KERNEL); | |
365 | if (!gart) { | |
366 | dev_err(dev, "failed to allocate gart_device\n"); | |
367 | return -ENOMEM; | |
368 | } | |
369 | ||
370 | gart_regs = devm_ioremap(dev, res->start, resource_size(res)); | |
371 | if (!gart_regs) { | |
372 | dev_err(dev, "failed to remap GART registers\n"); | |
373 | err = -ENXIO; | |
374 | goto fail; | |
375 | } | |
376 | ||
377 | gart->dev = &pdev->dev; | |
378 | spin_lock_init(&gart->pte_lock); | |
379 | spin_lock_init(&gart->client_lock); | |
380 | INIT_LIST_HEAD(&gart->client); | |
381 | gart->regs = gart_regs; | |
382 | gart->iovmm_base = (dma_addr_t)res_remap->start; | |
383 | gart->page_count = (resource_size(res_remap) >> GART_PAGE_SHIFT); | |
384 | ||
385 | gart->savedata = vmalloc(sizeof(u32) * gart->page_count); | |
386 | if (!gart->savedata) { | |
387 | dev_err(dev, "failed to allocate context save area\n"); | |
388 | err = -ENOMEM; | |
389 | goto fail; | |
390 | } | |
391 | ||
392 | platform_set_drvdata(pdev, gart); | |
393 | do_gart_setup(gart, NULL); | |
394 | ||
395 | gart_handle = gart; | |
396 | return 0; | |
397 | ||
398 | fail: | |
399 | if (gart_regs) | |
400 | devm_iounmap(dev, gart_regs); | |
401 | if (gart && gart->savedata) | |
402 | vfree(gart->savedata); | |
403 | devm_kfree(dev, gart); | |
404 | return err; | |
405 | } | |
406 | ||
407 | static int tegra_gart_remove(struct platform_device *pdev) | |
408 | { | |
409 | struct gart_device *gart = platform_get_drvdata(pdev); | |
410 | struct device *dev = gart->dev; | |
411 | ||
412 | writel(0, gart->regs + GART_CONFIG); | |
413 | if (gart->savedata) | |
414 | vfree(gart->savedata); | |
415 | if (gart->regs) | |
416 | devm_iounmap(dev, gart->regs); | |
417 | devm_kfree(dev, gart); | |
418 | gart_handle = NULL; | |
419 | return 0; | |
420 | } | |
421 | ||
422 | const struct dev_pm_ops tegra_gart_pm_ops = { | |
423 | .suspend = tegra_gart_suspend, | |
424 | .resume = tegra_gart_resume, | |
425 | }; | |
426 | ||
7cffae42 TR |
427 | #ifdef CONFIG_OF |
428 | static struct of_device_id tegra_gart_of_match[] __devinitdata = { | |
429 | { .compatible = "nvidia,tegra20-gart", }, | |
430 | { }, | |
431 | }; | |
432 | MODULE_DEVICE_TABLE(of, tegra_gart_of_match); | |
433 | #endif | |
434 | ||
d53e54b4 HD |
435 | static struct platform_driver tegra_gart_driver = { |
436 | .probe = tegra_gart_probe, | |
437 | .remove = tegra_gart_remove, | |
438 | .driver = { | |
439 | .owner = THIS_MODULE, | |
440 | .name = "tegra-gart", | |
441 | .pm = &tegra_gart_pm_ops, | |
7cffae42 | 442 | .of_match_table = of_match_ptr(tegra_gart_of_match), |
d53e54b4 HD |
443 | }, |
444 | }; | |
445 | ||
446 | static int __devinit tegra_gart_init(void) | |
447 | { | |
448 | bus_set_iommu(&platform_bus_type, &gart_iommu_ops); | |
449 | return platform_driver_register(&tegra_gart_driver); | |
450 | } | |
451 | ||
452 | static void __exit tegra_gart_exit(void) | |
453 | { | |
454 | platform_driver_unregister(&tegra_gart_driver); | |
455 | } | |
456 | ||
457 | subsys_initcall(tegra_gart_init); | |
458 | module_exit(tegra_gart_exit); | |
459 | ||
460 | MODULE_DESCRIPTION("IOMMU API for GART in Tegra20"); | |
461 | MODULE_AUTHOR("Hiroshi DOYU <hdoyu@nvidia.com>"); | |
7cffae42 | 462 | MODULE_ALIAS("platform:tegra-gart"); |
d53e54b4 | 463 | MODULE_LICENSE("GPL v2"); |