Steve Muckle | f132c6c | 2012-06-06 18:30:57 -0700 | [diff] [blame] | 1 | /* Copyright (c) 2012 Code Aurora Forum. All rights reserved. |
| 2 | * |
| 3 | * This program is free software; you can redistribute it and/or modify |
| 4 | * it under the terms of the GNU General Public License version 2 and |
| 5 | * only version 2 as published by the Free Software Foundation. |
| 6 | * |
| 7 | * This program is distributed in the hope that it will be useful, |
| 8 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 9 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 10 | * GNU General Public License for more details. |
| 11 | */ |
| 12 | |
| 13 | #include <linux/kernel.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <linux/errno.h> |
| 16 | #include <linux/io.h> |
| 17 | #include <linux/iommu.h> |
| 18 | #include <linux/scatterlist.h> |
| 19 | |
| 20 | #include <asm/cacheflush.h> |
| 21 | |
| 22 | #include <mach/iommu.h> |
| 23 | #include "msm_iommu_pagetable.h" |
| 24 | |
| 25 | /* Sharability attributes of MSM IOMMU mappings */ |
| 26 | #define MSM_IOMMU_ATTR_NON_SH 0x0 |
| 27 | #define MSM_IOMMU_ATTR_SH 0x4 |
| 28 | |
| 29 | /* Cacheability attributes of MSM IOMMU mappings */ |
| 30 | #define MSM_IOMMU_ATTR_NONCACHED 0x0 |
| 31 | #define MSM_IOMMU_ATTR_CACHED_WB_WA 0x1 |
| 32 | #define MSM_IOMMU_ATTR_CACHED_WB_NWA 0x2 |
| 33 | #define MSM_IOMMU_ATTR_CACHED_WT 0x3 |
| 34 | |
| 35 | static int msm_iommu_tex_class[4]; |
| 36 | |
| 37 | static inline void clean_pte(unsigned long *start, unsigned long *end, |
| 38 | int redirect) |
| 39 | { |
| 40 | if (!redirect) |
| 41 | dmac_flush_range(start, end); |
| 42 | } |
| 43 | |
| 44 | int msm_iommu_pagetable_alloc(struct iommu_pt *pt) |
| 45 | { |
| 46 | pt->fl_table = (unsigned long *)__get_free_pages(GFP_KERNEL, |
| 47 | get_order(SZ_16K)); |
| 48 | if (!pt->fl_table) |
| 49 | return -ENOMEM; |
| 50 | |
| 51 | memset(pt->fl_table, 0, SZ_16K); |
| 52 | clean_pte(pt->fl_table, pt->fl_table + NUM_FL_PTE, pt->redirect); |
| 53 | |
| 54 | return 0; |
| 55 | } |
| 56 | |
| 57 | void msm_iommu_pagetable_free(struct iommu_pt *pt) |
| 58 | { |
| 59 | unsigned long *fl_table; |
| 60 | int i; |
| 61 | |
| 62 | fl_table = pt->fl_table; |
| 63 | for (i = 0; i < NUM_FL_PTE; i++) |
| 64 | if ((fl_table[i] & 0x03) == FL_TYPE_TABLE) |
| 65 | free_page((unsigned long) __va(((fl_table[i]) & |
| 66 | FL_BASE_MASK))); |
| 67 | free_pages((unsigned long)fl_table, get_order(SZ_16K)); |
| 68 | pt->fl_table = 0; |
| 69 | } |
| 70 | |
| 71 | static int __get_pgprot(int prot, int len) |
| 72 | { |
| 73 | unsigned int pgprot; |
| 74 | int tex; |
| 75 | |
| 76 | if (!(prot & (IOMMU_READ | IOMMU_WRITE))) { |
| 77 | prot |= IOMMU_READ | IOMMU_WRITE; |
| 78 | WARN_ONCE(1, "No attributes in iommu mapping; assuming RW\n"); |
| 79 | } |
| 80 | |
| 81 | if ((prot & IOMMU_WRITE) && !(prot & IOMMU_READ)) { |
| 82 | prot |= IOMMU_READ; |
| 83 | WARN_ONCE(1, "Write-only unsupported; falling back to RW\n"); |
| 84 | } |
| 85 | |
| 86 | if (prot & IOMMU_CACHE) |
| 87 | tex = (pgprot_kernel >> 2) & 0x07; |
| 88 | else |
| 89 | tex = msm_iommu_tex_class[MSM_IOMMU_ATTR_NONCACHED]; |
| 90 | |
| 91 | if (tex < 0 || tex > NUM_TEX_CLASS - 1) |
| 92 | return 0; |
| 93 | |
| 94 | if (len == SZ_16M || len == SZ_1M) { |
| 95 | pgprot = FL_SHARED; |
| 96 | pgprot |= tex & 0x01 ? FL_BUFFERABLE : 0; |
| 97 | pgprot |= tex & 0x02 ? FL_CACHEABLE : 0; |
| 98 | pgprot |= tex & 0x04 ? FL_TEX0 : 0; |
| 99 | pgprot |= FL_AP0 | FL_AP1; |
| 100 | pgprot |= prot & IOMMU_WRITE ? 0 : FL_AP2; |
| 101 | } else { |
| 102 | pgprot = SL_SHARED; |
| 103 | pgprot |= tex & 0x01 ? SL_BUFFERABLE : 0; |
| 104 | pgprot |= tex & 0x02 ? SL_CACHEABLE : 0; |
| 105 | pgprot |= tex & 0x04 ? SL_TEX0 : 0; |
| 106 | pgprot |= SL_AP0 | SL_AP1; |
| 107 | pgprot |= prot & IOMMU_WRITE ? 0 : SL_AP2; |
| 108 | } |
| 109 | |
| 110 | return pgprot; |
| 111 | } |
| 112 | |
| 113 | int msm_iommu_pagetable_map(struct iommu_pt *pt, unsigned long va, |
| 114 | phys_addr_t pa, size_t len, int prot) |
| 115 | { |
| 116 | unsigned long *fl_pte; |
| 117 | unsigned long fl_offset; |
| 118 | unsigned long *sl_table; |
| 119 | unsigned long *sl_pte; |
| 120 | unsigned long sl_offset; |
| 121 | unsigned int pgprot; |
| 122 | int ret = 0; |
| 123 | |
| 124 | if (len != SZ_16M && len != SZ_1M && |
| 125 | len != SZ_64K && len != SZ_4K) { |
| 126 | pr_debug("Bad size: %d\n", len); |
| 127 | ret = -EINVAL; |
| 128 | goto fail; |
| 129 | } |
| 130 | |
| 131 | if (!pt->fl_table) { |
| 132 | pr_debug("Null page table\n"); |
| 133 | ret = -EINVAL; |
| 134 | goto fail; |
| 135 | } |
| 136 | |
| 137 | pgprot = __get_pgprot(prot, len); |
| 138 | if (!pgprot) { |
| 139 | ret = -EINVAL; |
| 140 | goto fail; |
| 141 | } |
| 142 | |
| 143 | fl_offset = FL_OFFSET(va); /* Upper 12 bits */ |
| 144 | fl_pte = pt->fl_table + fl_offset; /* int pointers, 4 bytes */ |
| 145 | |
| 146 | if (len == SZ_16M) { |
| 147 | int i = 0; |
| 148 | |
| 149 | for (i = 0; i < 16; i++) |
| 150 | if (*(fl_pte+i)) { |
| 151 | ret = -EBUSY; |
| 152 | goto fail; |
| 153 | } |
| 154 | |
| 155 | for (i = 0; i < 16; i++) |
| 156 | *(fl_pte+i) = (pa & 0xFF000000) | FL_SUPERSECTION | |
| 157 | FL_TYPE_SECT | FL_SHARED | FL_NG | pgprot; |
| 158 | clean_pte(fl_pte, fl_pte + 16, pt->redirect); |
| 159 | } |
| 160 | |
| 161 | if (len == SZ_1M) { |
| 162 | if (*fl_pte) { |
| 163 | ret = -EBUSY; |
| 164 | goto fail; |
| 165 | } |
| 166 | |
| 167 | *fl_pte = (pa & 0xFFF00000) | FL_NG | FL_TYPE_SECT |
| 168 | | FL_SHARED | pgprot; |
| 169 | clean_pte(fl_pte, fl_pte + 1, pt->redirect); |
| 170 | } |
| 171 | |
| 172 | /* Need a 2nd level table */ |
| 173 | if (len == SZ_4K || len == SZ_64K) { |
| 174 | |
| 175 | if (*fl_pte == 0) { |
| 176 | unsigned long *sl; |
| 177 | sl = (unsigned long *) __get_free_pages(GFP_KERNEL, |
| 178 | get_order(SZ_4K)); |
| 179 | |
| 180 | if (!sl) { |
| 181 | pr_debug("Could not allocate second level table\n"); |
| 182 | ret = -ENOMEM; |
| 183 | goto fail; |
| 184 | } |
| 185 | memset(sl, 0, SZ_4K); |
| 186 | clean_pte(sl, sl + NUM_SL_PTE, pt->redirect); |
| 187 | |
| 188 | *fl_pte = ((((int)__pa(sl)) & FL_BASE_MASK) | \ |
| 189 | FL_TYPE_TABLE); |
| 190 | clean_pte(fl_pte, fl_pte + 1, pt->redirect); |
| 191 | } |
| 192 | |
| 193 | if (!(*fl_pte & FL_TYPE_TABLE)) { |
| 194 | ret = -EBUSY; |
| 195 | goto fail; |
| 196 | } |
| 197 | } |
| 198 | |
| 199 | sl_table = (unsigned long *) __va(((*fl_pte) & FL_BASE_MASK)); |
| 200 | sl_offset = SL_OFFSET(va); |
| 201 | sl_pte = sl_table + sl_offset; |
| 202 | |
| 203 | if (len == SZ_4K) { |
| 204 | if (*sl_pte) { |
| 205 | ret = -EBUSY; |
| 206 | goto fail; |
| 207 | } |
| 208 | |
| 209 | *sl_pte = (pa & SL_BASE_MASK_SMALL) | SL_NG | SL_SHARED |
| 210 | | SL_TYPE_SMALL | pgprot; |
| 211 | clean_pte(sl_pte, sl_pte + 1, pt->redirect); |
| 212 | } |
| 213 | |
| 214 | if (len == SZ_64K) { |
| 215 | int i; |
| 216 | |
| 217 | for (i = 0; i < 16; i++) |
| 218 | if (*(sl_pte+i)) { |
| 219 | ret = -EBUSY; |
| 220 | goto fail; |
| 221 | } |
| 222 | |
| 223 | for (i = 0; i < 16; i++) |
| 224 | *(sl_pte+i) = (pa & SL_BASE_MASK_LARGE) | SL_NG |
| 225 | | SL_SHARED | SL_TYPE_LARGE | pgprot; |
| 226 | |
| 227 | clean_pte(sl_pte, sl_pte + 16, pt->redirect); |
| 228 | } |
| 229 | |
| 230 | fail: |
| 231 | return ret; |
| 232 | } |
| 233 | |
| 234 | size_t msm_iommu_pagetable_unmap(struct iommu_pt *pt, unsigned long va, |
| 235 | size_t len) |
| 236 | { |
| 237 | unsigned long *fl_pte; |
| 238 | unsigned long fl_offset; |
| 239 | unsigned long *sl_table; |
| 240 | unsigned long *sl_pte; |
| 241 | unsigned long sl_offset; |
| 242 | int i, ret = 0; |
| 243 | |
| 244 | if (len != SZ_16M && len != SZ_1M && |
| 245 | len != SZ_64K && len != SZ_4K) { |
| 246 | pr_debug("Bad length: %d\n", len); |
| 247 | ret = -EINVAL; |
| 248 | goto fail; |
| 249 | } |
| 250 | |
| 251 | if (!pt->fl_table) { |
| 252 | pr_debug("Null page table\n"); |
| 253 | ret = -EINVAL; |
| 254 | goto fail; |
| 255 | } |
| 256 | |
| 257 | fl_offset = FL_OFFSET(va); /* Upper 12 bits */ |
| 258 | fl_pte = pt->fl_table + fl_offset; /* int pointers, 4 bytes */ |
| 259 | |
| 260 | if (*fl_pte == 0) { |
| 261 | pr_debug("First level PTE is 0\n"); |
| 262 | ret = -ENODEV; |
| 263 | goto fail; |
| 264 | } |
| 265 | |
| 266 | /* Unmap supersection */ |
| 267 | if (len == SZ_16M) { |
| 268 | for (i = 0; i < 16; i++) |
| 269 | *(fl_pte+i) = 0; |
| 270 | |
| 271 | clean_pte(fl_pte, fl_pte + 16, pt->redirect); |
| 272 | } |
| 273 | |
| 274 | if (len == SZ_1M) { |
| 275 | *fl_pte = 0; |
| 276 | clean_pte(fl_pte, fl_pte + 1, pt->redirect); |
| 277 | } |
| 278 | |
| 279 | sl_table = (unsigned long *) __va(((*fl_pte) & FL_BASE_MASK)); |
| 280 | sl_offset = SL_OFFSET(va); |
| 281 | sl_pte = sl_table + sl_offset; |
| 282 | |
| 283 | if (len == SZ_64K) { |
| 284 | for (i = 0; i < 16; i++) |
| 285 | *(sl_pte+i) = 0; |
| 286 | |
| 287 | clean_pte(sl_pte, sl_pte + 16, pt->redirect); |
| 288 | } |
| 289 | |
| 290 | if (len == SZ_4K) { |
| 291 | *sl_pte = 0; |
| 292 | clean_pte(sl_pte, sl_pte + 1, pt->redirect); |
| 293 | } |
| 294 | |
| 295 | if (len == SZ_4K || len == SZ_64K) { |
| 296 | int used = 0; |
| 297 | |
| 298 | for (i = 0; i < NUM_SL_PTE; i++) |
| 299 | if (sl_table[i]) |
| 300 | used = 1; |
| 301 | if (!used) { |
| 302 | free_page((unsigned long)sl_table); |
| 303 | *fl_pte = 0; |
| 304 | clean_pte(fl_pte, fl_pte + 1, pt->redirect); |
| 305 | } |
| 306 | } |
| 307 | |
| 308 | fail: |
| 309 | return ret; |
| 310 | } |
| 311 | |
| 312 | static unsigned int get_phys_addr(struct scatterlist *sg) |
| 313 | { |
| 314 | /* |
| 315 | * Try sg_dma_address first so that we can |
| 316 | * map carveout regions that do not have a |
| 317 | * struct page associated with them. |
| 318 | */ |
| 319 | unsigned int pa = sg_dma_address(sg); |
| 320 | if (pa == 0) |
| 321 | pa = sg_phys(sg); |
| 322 | return pa; |
| 323 | } |
| 324 | |
| 325 | int msm_iommu_pagetable_map_range(struct iommu_pt *pt, unsigned int va, |
| 326 | struct scatterlist *sg, unsigned int len, int prot) |
| 327 | { |
| 328 | unsigned int pa; |
| 329 | unsigned int offset = 0; |
| 330 | unsigned int pgprot; |
| 331 | unsigned long *fl_pte; |
| 332 | unsigned long fl_offset; |
| 333 | unsigned long *sl_table; |
| 334 | unsigned long sl_offset, sl_start; |
| 335 | unsigned int chunk_offset = 0; |
| 336 | unsigned int chunk_pa; |
| 337 | int ret = 0; |
| 338 | |
| 339 | BUG_ON(len & (SZ_4K - 1)); |
| 340 | |
| 341 | pgprot = __get_pgprot(prot, SZ_4K); |
| 342 | if (!pgprot) { |
| 343 | ret = -EINVAL; |
| 344 | goto fail; |
| 345 | } |
| 346 | |
| 347 | fl_offset = FL_OFFSET(va); /* Upper 12 bits */ |
| 348 | fl_pte = pt->fl_table + fl_offset; /* int pointers, 4 bytes */ |
| 349 | |
| 350 | sl_table = (unsigned long *) __va(((*fl_pte) & FL_BASE_MASK)); |
| 351 | sl_offset = SL_OFFSET(va); |
| 352 | |
| 353 | chunk_pa = get_phys_addr(sg); |
| 354 | if (chunk_pa == 0) { |
| 355 | pr_debug("No dma address for sg %p\n", sg); |
| 356 | ret = -EINVAL; |
| 357 | goto fail; |
| 358 | } |
| 359 | |
| 360 | while (offset < len) { |
| 361 | /* Set up a 2nd level page table if one doesn't exist */ |
| 362 | if (*fl_pte == 0) { |
| 363 | sl_table = (unsigned long *) |
| 364 | __get_free_pages(GFP_KERNEL, get_order(SZ_4K)); |
| 365 | |
| 366 | if (!sl_table) { |
| 367 | pr_debug("Could not allocate second level table\n"); |
| 368 | ret = -ENOMEM; |
| 369 | goto fail; |
| 370 | } |
| 371 | |
| 372 | memset(sl_table, 0, SZ_4K); |
| 373 | clean_pte(sl_table, sl_table + NUM_SL_PTE, |
| 374 | pt->redirect); |
| 375 | |
| 376 | *fl_pte = ((((int)__pa(sl_table)) & FL_BASE_MASK) | |
| 377 | FL_TYPE_TABLE); |
| 378 | clean_pte(fl_pte, fl_pte + 1, pt->redirect); |
| 379 | } else |
| 380 | sl_table = (unsigned long *) |
| 381 | __va(((*fl_pte) & FL_BASE_MASK)); |
| 382 | |
| 383 | /* Keep track of initial position so we |
| 384 | * don't clean more than we have to |
| 385 | */ |
| 386 | sl_start = sl_offset; |
| 387 | |
| 388 | /* Build the 2nd level page table */ |
| 389 | while (offset < len && sl_offset < NUM_SL_PTE) { |
| 390 | pa = chunk_pa + chunk_offset; |
| 391 | sl_table[sl_offset] = (pa & SL_BASE_MASK_SMALL) | |
| 392 | pgprot | SL_NG | SL_SHARED | SL_TYPE_SMALL; |
| 393 | sl_offset++; |
| 394 | offset += SZ_4K; |
| 395 | |
| 396 | chunk_offset += SZ_4K; |
| 397 | |
| 398 | if (chunk_offset >= sg->length && offset < len) { |
| 399 | chunk_offset = 0; |
| 400 | sg = sg_next(sg); |
| 401 | chunk_pa = get_phys_addr(sg); |
| 402 | if (chunk_pa == 0) { |
| 403 | pr_debug("No dma address for sg %p\n", |
| 404 | sg); |
| 405 | ret = -EINVAL; |
| 406 | goto fail; |
| 407 | } |
| 408 | } |
| 409 | } |
| 410 | |
| 411 | clean_pte(sl_table + sl_start, sl_table + sl_offset, |
| 412 | pt->redirect); |
| 413 | fl_pte++; |
| 414 | sl_offset = 0; |
| 415 | } |
| 416 | |
| 417 | fail: |
| 418 | return ret; |
| 419 | } |
| 420 | |
| 421 | void msm_iommu_pagetable_unmap_range(struct iommu_pt *pt, unsigned int va, |
| 422 | unsigned int len) |
| 423 | { |
| 424 | unsigned int offset = 0; |
| 425 | unsigned long *fl_pte; |
| 426 | unsigned long fl_offset; |
| 427 | unsigned long *sl_table; |
| 428 | unsigned long sl_start, sl_end; |
| 429 | int used, i; |
| 430 | |
| 431 | BUG_ON(len & (SZ_4K - 1)); |
| 432 | |
| 433 | fl_offset = FL_OFFSET(va); /* Upper 12 bits */ |
| 434 | fl_pte = pt->fl_table + fl_offset; /* int pointers, 4 bytes */ |
| 435 | |
| 436 | sl_start = SL_OFFSET(va); |
| 437 | |
| 438 | while (offset < len) { |
| 439 | sl_table = (unsigned long *) __va(((*fl_pte) & FL_BASE_MASK)); |
| 440 | sl_end = ((len - offset) / SZ_4K) + sl_start; |
| 441 | |
| 442 | if (sl_end > NUM_SL_PTE) |
| 443 | sl_end = NUM_SL_PTE; |
| 444 | |
| 445 | memset(sl_table + sl_start, 0, (sl_end - sl_start) * 4); |
| 446 | clean_pte(sl_table + sl_start, sl_table + sl_end, |
| 447 | pt->redirect); |
| 448 | |
| 449 | offset += (sl_end - sl_start) * SZ_4K; |
| 450 | |
| 451 | /* Unmap and free the 2nd level table if all mappings in it |
| 452 | * were removed. This saves memory, but the table will need |
| 453 | * to be re-allocated the next time someone tries to map these |
| 454 | * VAs. |
| 455 | */ |
| 456 | used = 0; |
| 457 | |
| 458 | /* If we just unmapped the whole table, don't bother |
| 459 | * seeing if there are still used entries left. |
| 460 | */ |
| 461 | if (sl_end - sl_start != NUM_SL_PTE) |
| 462 | for (i = 0; i < NUM_SL_PTE; i++) |
| 463 | if (sl_table[i]) { |
| 464 | used = 1; |
| 465 | break; |
| 466 | } |
| 467 | if (!used) { |
| 468 | free_page((unsigned long)sl_table); |
| 469 | *fl_pte = 0; |
| 470 | clean_pte(fl_pte, fl_pte + 1, pt->redirect); |
| 471 | } |
| 472 | |
| 473 | sl_start = 0; |
| 474 | fl_pte++; |
| 475 | } |
| 476 | } |
| 477 | |
| 478 | static int __init get_tex_class(int icp, int ocp, int mt, int nos) |
| 479 | { |
| 480 | int i = 0; |
| 481 | unsigned int prrr = 0; |
| 482 | unsigned int nmrr = 0; |
| 483 | int c_icp, c_ocp, c_mt, c_nos; |
| 484 | |
| 485 | RCP15_PRRR(prrr); |
| 486 | RCP15_NMRR(nmrr); |
| 487 | |
| 488 | for (i = 0; i < NUM_TEX_CLASS; i++) { |
| 489 | c_nos = PRRR_NOS(prrr, i); |
| 490 | c_mt = PRRR_MT(prrr, i); |
| 491 | c_icp = NMRR_ICP(nmrr, i); |
| 492 | c_ocp = NMRR_OCP(nmrr, i); |
| 493 | |
| 494 | if (icp == c_icp && ocp == c_ocp && c_mt == mt && c_nos == nos) |
| 495 | return i; |
| 496 | } |
| 497 | |
| 498 | return -ENODEV; |
| 499 | } |
| 500 | |
| 501 | static void __init setup_iommu_tex_classes(void) |
| 502 | { |
| 503 | msm_iommu_tex_class[MSM_IOMMU_ATTR_NONCACHED] = |
| 504 | get_tex_class(CP_NONCACHED, CP_NONCACHED, MT_NORMAL, 1); |
| 505 | |
| 506 | msm_iommu_tex_class[MSM_IOMMU_ATTR_CACHED_WB_WA] = |
| 507 | get_tex_class(CP_WB_WA, CP_WB_WA, MT_NORMAL, 1); |
| 508 | |
| 509 | msm_iommu_tex_class[MSM_IOMMU_ATTR_CACHED_WB_NWA] = |
| 510 | get_tex_class(CP_WB_NWA, CP_WB_NWA, MT_NORMAL, 1); |
| 511 | |
| 512 | msm_iommu_tex_class[MSM_IOMMU_ATTR_CACHED_WT] = |
| 513 | get_tex_class(CP_WT, CP_WT, MT_NORMAL, 1); |
| 514 | } |
| 515 | |
| 516 | void __init msm_iommu_pagetable_init(void) |
| 517 | { |
| 518 | setup_iommu_tex_classes(); |
| 519 | } |