Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame^] | 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> |
| 32 | #include <linux/of.h> |
| 33 | |
| 34 | #include <asm/cacheflush.h> |
| 35 | |
| 36 | /* bitmap of the page sizes currently supported */ |
| 37 | #define GART_IOMMU_PGSIZES (SZ_4K) |
| 38 | |
| 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) |
| 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 | struct gart_domain { |
| 67 | struct iommu_domain domain; /* generic domain handle */ |
| 68 | struct gart_device *gart; /* link to gart device */ |
| 69 | }; |
| 70 | |
| 71 | static struct gart_device *gart_handle; /* unique for a system */ |
| 72 | |
| 73 | #define GART_PTE(_pfn) \ |
| 74 | (GART_ENTRY_PHYS_ADDR_VALID | ((_pfn) << PAGE_SHIFT)) |
| 75 | |
| 76 | static struct gart_domain *to_gart_domain(struct iommu_domain *dom) |
| 77 | { |
| 78 | return container_of(dom, struct gart_domain, domain); |
| 79 | } |
| 80 | |
| 81 | /* |
| 82 | * Any interaction between any block on PPSB and a block on APB or AHB |
| 83 | * must have these read-back to ensure the APB/AHB bus transaction is |
| 84 | * complete before initiating activity on the PPSB block. |
| 85 | */ |
| 86 | #define FLUSH_GART_REGS(gart) ((void)readl((gart)->regs + GART_CONFIG)) |
| 87 | |
| 88 | #define for_each_gart_pte(gart, iova) \ |
| 89 | for (iova = gart->iovmm_base; \ |
| 90 | iova < gart->iovmm_base + GART_PAGE_SIZE * gart->page_count; \ |
| 91 | iova += GART_PAGE_SIZE) |
| 92 | |
| 93 | static inline void gart_set_pte(struct gart_device *gart, |
| 94 | unsigned long offs, u32 pte) |
| 95 | { |
| 96 | writel(offs, gart->regs + GART_ENTRY_ADDR); |
| 97 | writel(pte, gart->regs + GART_ENTRY_DATA); |
| 98 | |
| 99 | dev_dbg(gart->dev, "%s %08lx:%08x\n", |
| 100 | pte ? "map" : "unmap", offs, pte & GART_PAGE_MASK); |
| 101 | } |
| 102 | |
| 103 | static inline unsigned long gart_read_pte(struct gart_device *gart, |
| 104 | unsigned long offs) |
| 105 | { |
| 106 | unsigned long pte; |
| 107 | |
| 108 | writel(offs, gart->regs + GART_ENTRY_ADDR); |
| 109 | pte = readl(gart->regs + GART_ENTRY_DATA); |
| 110 | |
| 111 | return pte; |
| 112 | } |
| 113 | |
| 114 | static void do_gart_setup(struct gart_device *gart, const u32 *data) |
| 115 | { |
| 116 | unsigned long iova; |
| 117 | |
| 118 | for_each_gart_pte(gart, iova) |
| 119 | gart_set_pte(gart, iova, data ? *(data++) : 0); |
| 120 | |
| 121 | writel(1, gart->regs + GART_CONFIG); |
| 122 | FLUSH_GART_REGS(gart); |
| 123 | } |
| 124 | |
| 125 | #ifdef DEBUG |
| 126 | static void gart_dump_table(struct gart_device *gart) |
| 127 | { |
| 128 | unsigned long iova; |
| 129 | unsigned long flags; |
| 130 | |
| 131 | spin_lock_irqsave(&gart->pte_lock, flags); |
| 132 | for_each_gart_pte(gart, iova) { |
| 133 | unsigned long pte; |
| 134 | |
| 135 | pte = gart_read_pte(gart, iova); |
| 136 | |
| 137 | dev_dbg(gart->dev, "%s %08lx:%08lx\n", |
| 138 | (GART_ENTRY_PHYS_ADDR_VALID & pte) ? "v" : " ", |
| 139 | iova, pte & GART_PAGE_MASK); |
| 140 | } |
| 141 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
| 142 | } |
| 143 | #else |
| 144 | static inline void gart_dump_table(struct gart_device *gart) |
| 145 | { |
| 146 | } |
| 147 | #endif |
| 148 | |
| 149 | static inline bool gart_iova_range_valid(struct gart_device *gart, |
| 150 | unsigned long iova, size_t bytes) |
| 151 | { |
| 152 | unsigned long iova_start, iova_end, gart_start, gart_end; |
| 153 | |
| 154 | iova_start = iova; |
| 155 | iova_end = iova_start + bytes - 1; |
| 156 | gart_start = gart->iovmm_base; |
| 157 | gart_end = gart_start + gart->page_count * GART_PAGE_SIZE - 1; |
| 158 | |
| 159 | if (iova_start < gart_start) |
| 160 | return false; |
| 161 | if (iova_end > gart_end) |
| 162 | return false; |
| 163 | return true; |
| 164 | } |
| 165 | |
| 166 | static int gart_iommu_attach_dev(struct iommu_domain *domain, |
| 167 | struct device *dev) |
| 168 | { |
| 169 | struct gart_domain *gart_domain = to_gart_domain(domain); |
| 170 | struct gart_device *gart = gart_domain->gart; |
| 171 | struct gart_client *client, *c; |
| 172 | int err = 0; |
| 173 | |
| 174 | client = devm_kzalloc(gart->dev, sizeof(*c), GFP_KERNEL); |
| 175 | if (!client) |
| 176 | return -ENOMEM; |
| 177 | client->dev = dev; |
| 178 | |
| 179 | spin_lock(&gart->client_lock); |
| 180 | list_for_each_entry(c, &gart->client, list) { |
| 181 | if (c->dev == dev) { |
| 182 | dev_err(gart->dev, |
| 183 | "%s is already attached\n", dev_name(dev)); |
| 184 | err = -EINVAL; |
| 185 | goto fail; |
| 186 | } |
| 187 | } |
| 188 | list_add(&client->list, &gart->client); |
| 189 | spin_unlock(&gart->client_lock); |
| 190 | dev_dbg(gart->dev, "Attached %s\n", dev_name(dev)); |
| 191 | return 0; |
| 192 | |
| 193 | fail: |
| 194 | devm_kfree(gart->dev, client); |
| 195 | spin_unlock(&gart->client_lock); |
| 196 | return err; |
| 197 | } |
| 198 | |
| 199 | static void gart_iommu_detach_dev(struct iommu_domain *domain, |
| 200 | struct device *dev) |
| 201 | { |
| 202 | struct gart_domain *gart_domain = to_gart_domain(domain); |
| 203 | struct gart_device *gart = gart_domain->gart; |
| 204 | struct gart_client *c; |
| 205 | |
| 206 | spin_lock(&gart->client_lock); |
| 207 | |
| 208 | list_for_each_entry(c, &gart->client, list) { |
| 209 | if (c->dev == dev) { |
| 210 | list_del(&c->list); |
| 211 | devm_kfree(gart->dev, c); |
| 212 | dev_dbg(gart->dev, "Detached %s\n", dev_name(dev)); |
| 213 | goto out; |
| 214 | } |
| 215 | } |
| 216 | dev_err(gart->dev, "Couldn't find\n"); |
| 217 | out: |
| 218 | spin_unlock(&gart->client_lock); |
| 219 | } |
| 220 | |
| 221 | static struct iommu_domain *gart_iommu_domain_alloc(unsigned type) |
| 222 | { |
| 223 | struct gart_domain *gart_domain; |
| 224 | struct gart_device *gart; |
| 225 | |
| 226 | if (type != IOMMU_DOMAIN_UNMANAGED) |
| 227 | return NULL; |
| 228 | |
| 229 | gart = gart_handle; |
| 230 | if (!gart) |
| 231 | return NULL; |
| 232 | |
| 233 | gart_domain = kzalloc(sizeof(*gart_domain), GFP_KERNEL); |
| 234 | if (!gart_domain) |
| 235 | return NULL; |
| 236 | |
| 237 | gart_domain->gart = gart; |
| 238 | gart_domain->domain.geometry.aperture_start = gart->iovmm_base; |
| 239 | gart_domain->domain.geometry.aperture_end = gart->iovmm_base + |
| 240 | gart->page_count * GART_PAGE_SIZE - 1; |
| 241 | gart_domain->domain.geometry.force_aperture = true; |
| 242 | |
| 243 | return &gart_domain->domain; |
| 244 | } |
| 245 | |
| 246 | static void gart_iommu_domain_free(struct iommu_domain *domain) |
| 247 | { |
| 248 | struct gart_domain *gart_domain = to_gart_domain(domain); |
| 249 | struct gart_device *gart = gart_domain->gart; |
| 250 | |
| 251 | if (gart) { |
| 252 | spin_lock(&gart->client_lock); |
| 253 | if (!list_empty(&gart->client)) { |
| 254 | struct gart_client *c; |
| 255 | |
| 256 | list_for_each_entry(c, &gart->client, list) |
| 257 | gart_iommu_detach_dev(domain, c->dev); |
| 258 | } |
| 259 | spin_unlock(&gart->client_lock); |
| 260 | } |
| 261 | |
| 262 | kfree(gart_domain); |
| 263 | } |
| 264 | |
| 265 | static int gart_iommu_map(struct iommu_domain *domain, unsigned long iova, |
| 266 | phys_addr_t pa, size_t bytes, int prot) |
| 267 | { |
| 268 | struct gart_domain *gart_domain = to_gart_domain(domain); |
| 269 | struct gart_device *gart = gart_domain->gart; |
| 270 | unsigned long flags; |
| 271 | unsigned long pfn; |
| 272 | |
| 273 | if (!gart_iova_range_valid(gart, iova, bytes)) |
| 274 | return -EINVAL; |
| 275 | |
| 276 | spin_lock_irqsave(&gart->pte_lock, flags); |
| 277 | pfn = __phys_to_pfn(pa); |
| 278 | if (!pfn_valid(pfn)) { |
| 279 | dev_err(gart->dev, "Invalid page: %pa\n", &pa); |
| 280 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
| 281 | return -EINVAL; |
| 282 | } |
| 283 | gart_set_pte(gart, iova, GART_PTE(pfn)); |
| 284 | FLUSH_GART_REGS(gart); |
| 285 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
| 286 | return 0; |
| 287 | } |
| 288 | |
| 289 | static size_t gart_iommu_unmap(struct iommu_domain *domain, unsigned long iova, |
| 290 | size_t bytes) |
| 291 | { |
| 292 | struct gart_domain *gart_domain = to_gart_domain(domain); |
| 293 | struct gart_device *gart = gart_domain->gart; |
| 294 | unsigned long flags; |
| 295 | |
| 296 | if (!gart_iova_range_valid(gart, iova, bytes)) |
| 297 | return 0; |
| 298 | |
| 299 | spin_lock_irqsave(&gart->pte_lock, flags); |
| 300 | gart_set_pte(gart, iova, 0); |
| 301 | FLUSH_GART_REGS(gart); |
| 302 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
| 303 | return 0; |
| 304 | } |
| 305 | |
| 306 | static phys_addr_t gart_iommu_iova_to_phys(struct iommu_domain *domain, |
| 307 | dma_addr_t iova) |
| 308 | { |
| 309 | struct gart_domain *gart_domain = to_gart_domain(domain); |
| 310 | struct gart_device *gart = gart_domain->gart; |
| 311 | unsigned long pte; |
| 312 | phys_addr_t pa; |
| 313 | unsigned long flags; |
| 314 | |
| 315 | if (!gart_iova_range_valid(gart, iova, 0)) |
| 316 | return -EINVAL; |
| 317 | |
| 318 | spin_lock_irqsave(&gart->pte_lock, flags); |
| 319 | pte = gart_read_pte(gart, iova); |
| 320 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
| 321 | |
| 322 | pa = (pte & GART_PAGE_MASK); |
| 323 | if (!pfn_valid(__phys_to_pfn(pa))) { |
| 324 | dev_err(gart->dev, "No entry for %08llx:%pa\n", |
| 325 | (unsigned long long)iova, &pa); |
| 326 | gart_dump_table(gart); |
| 327 | return -EINVAL; |
| 328 | } |
| 329 | return pa; |
| 330 | } |
| 331 | |
| 332 | static bool gart_iommu_capable(enum iommu_cap cap) |
| 333 | { |
| 334 | return false; |
| 335 | } |
| 336 | |
| 337 | static const struct iommu_ops gart_iommu_ops = { |
| 338 | .capable = gart_iommu_capable, |
| 339 | .domain_alloc = gart_iommu_domain_alloc, |
| 340 | .domain_free = gart_iommu_domain_free, |
| 341 | .attach_dev = gart_iommu_attach_dev, |
| 342 | .detach_dev = gart_iommu_detach_dev, |
| 343 | .map = gart_iommu_map, |
| 344 | .map_sg = default_iommu_map_sg, |
| 345 | .unmap = gart_iommu_unmap, |
| 346 | .iova_to_phys = gart_iommu_iova_to_phys, |
| 347 | .pgsize_bitmap = GART_IOMMU_PGSIZES, |
| 348 | }; |
| 349 | |
| 350 | static int tegra_gart_suspend(struct device *dev) |
| 351 | { |
| 352 | struct gart_device *gart = dev_get_drvdata(dev); |
| 353 | unsigned long iova; |
| 354 | u32 *data = gart->savedata; |
| 355 | unsigned long flags; |
| 356 | |
| 357 | spin_lock_irqsave(&gart->pte_lock, flags); |
| 358 | for_each_gart_pte(gart, iova) |
| 359 | *(data++) = gart_read_pte(gart, iova); |
| 360 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
| 361 | return 0; |
| 362 | } |
| 363 | |
| 364 | static int tegra_gart_resume(struct device *dev) |
| 365 | { |
| 366 | struct gart_device *gart = dev_get_drvdata(dev); |
| 367 | unsigned long flags; |
| 368 | |
| 369 | spin_lock_irqsave(&gart->pte_lock, flags); |
| 370 | do_gart_setup(gart, gart->savedata); |
| 371 | spin_unlock_irqrestore(&gart->pte_lock, flags); |
| 372 | return 0; |
| 373 | } |
| 374 | |
| 375 | static int tegra_gart_probe(struct platform_device *pdev) |
| 376 | { |
| 377 | struct gart_device *gart; |
| 378 | struct resource *res, *res_remap; |
| 379 | void __iomem *gart_regs; |
| 380 | struct device *dev = &pdev->dev; |
| 381 | |
| 382 | if (gart_handle) |
| 383 | return -EIO; |
| 384 | |
| 385 | BUILD_BUG_ON(PAGE_SHIFT != GART_PAGE_SHIFT); |
| 386 | |
| 387 | /* the GART memory aperture is required */ |
| 388 | res = platform_get_resource(pdev, IORESOURCE_MEM, 0); |
| 389 | res_remap = platform_get_resource(pdev, IORESOURCE_MEM, 1); |
| 390 | if (!res || !res_remap) { |
| 391 | dev_err(dev, "GART memory aperture expected\n"); |
| 392 | return -ENXIO; |
| 393 | } |
| 394 | |
| 395 | gart = devm_kzalloc(dev, sizeof(*gart), GFP_KERNEL); |
| 396 | if (!gart) { |
| 397 | dev_err(dev, "failed to allocate gart_device\n"); |
| 398 | return -ENOMEM; |
| 399 | } |
| 400 | |
| 401 | gart_regs = devm_ioremap(dev, res->start, resource_size(res)); |
| 402 | if (!gart_regs) { |
| 403 | dev_err(dev, "failed to remap GART registers\n"); |
| 404 | return -ENXIO; |
| 405 | } |
| 406 | |
| 407 | gart->dev = &pdev->dev; |
| 408 | spin_lock_init(&gart->pte_lock); |
| 409 | spin_lock_init(&gart->client_lock); |
| 410 | INIT_LIST_HEAD(&gart->client); |
| 411 | gart->regs = gart_regs; |
| 412 | gart->iovmm_base = (dma_addr_t)res_remap->start; |
| 413 | gart->page_count = (resource_size(res_remap) >> GART_PAGE_SHIFT); |
| 414 | |
| 415 | gart->savedata = vmalloc(sizeof(u32) * gart->page_count); |
| 416 | if (!gart->savedata) { |
| 417 | dev_err(dev, "failed to allocate context save area\n"); |
| 418 | return -ENOMEM; |
| 419 | } |
| 420 | |
| 421 | platform_set_drvdata(pdev, gart); |
| 422 | do_gart_setup(gart, NULL); |
| 423 | |
| 424 | gart_handle = gart; |
| 425 | |
| 426 | return 0; |
| 427 | } |
| 428 | |
| 429 | static int tegra_gart_remove(struct platform_device *pdev) |
| 430 | { |
| 431 | struct gart_device *gart = platform_get_drvdata(pdev); |
| 432 | |
| 433 | writel(0, gart->regs + GART_CONFIG); |
| 434 | if (gart->savedata) |
| 435 | vfree(gart->savedata); |
| 436 | gart_handle = NULL; |
| 437 | return 0; |
| 438 | } |
| 439 | |
| 440 | static const struct dev_pm_ops tegra_gart_pm_ops = { |
| 441 | .suspend = tegra_gart_suspend, |
| 442 | .resume = tegra_gart_resume, |
| 443 | }; |
| 444 | |
| 445 | static const struct of_device_id tegra_gart_of_match[] = { |
| 446 | { .compatible = "nvidia,tegra20-gart", }, |
| 447 | { }, |
| 448 | }; |
| 449 | MODULE_DEVICE_TABLE(of, tegra_gart_of_match); |
| 450 | |
| 451 | static struct platform_driver tegra_gart_driver = { |
| 452 | .probe = tegra_gart_probe, |
| 453 | .remove = tegra_gart_remove, |
| 454 | .driver = { |
| 455 | .name = "tegra-gart", |
| 456 | .pm = &tegra_gart_pm_ops, |
| 457 | .of_match_table = tegra_gart_of_match, |
| 458 | }, |
| 459 | }; |
| 460 | |
| 461 | static int tegra_gart_init(void) |
| 462 | { |
| 463 | return platform_driver_register(&tegra_gart_driver); |
| 464 | } |
| 465 | |
| 466 | static void __exit tegra_gart_exit(void) |
| 467 | { |
| 468 | platform_driver_unregister(&tegra_gart_driver); |
| 469 | } |
| 470 | |
| 471 | subsys_initcall(tegra_gart_init); |
| 472 | module_exit(tegra_gart_exit); |
| 473 | |
| 474 | MODULE_DESCRIPTION("IOMMU API for GART in Tegra20"); |
| 475 | MODULE_AUTHOR("Hiroshi DOYU <hdoyu@nvidia.com>"); |
| 476 | MODULE_ALIAS("platform:tegra-gart"); |
| 477 | MODULE_LICENSE("GPL v2"); |