drivers/iommu/tegra-gart.c

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