hypervisor/paging.c

   1 /*
   2  * Jailhouse, a Linux-based partitioning hypervisor
   3  *
   4  * Copyright (c) Siemens AG, 2013
   5  *
   6  * Authors:
   7  *  Jan Kiszka <jan.kiszka@siemens.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  */
  12
  13 #include <jailhouse/paging.h>
  14 #include <jailhouse/printk.h>
  15 #include <jailhouse/string.h>
  16 #include <jailhouse/control.h>
  17 #include <asm/bitops.h>
  18
  19 #define BITS_PER_PAGE           (PAGE_SIZE * 8)
  20
  21 #define INVALID_PAGE_NR         (~0UL)
  22
  23 #define PAGE_SCRUB_ON_FREE      0x1
  24
  25 extern u8 __page_pool[];
  26
  27 struct page_pool mem_pool;
  28 struct page_pool remap_pool = {
  29         .base_address = (void *)REMAP_BASE,
  30         .pages = BITS_PER_PAGE * NUM_REMAP_BITMAP_PAGES,
  31 };
  32
  33 struct paging_structures hv_paging_structs;
  34
  35 unsigned long page_map_get_phys_invalid(pt_entry_t pte, unsigned long virt)
  36 {
  37         return INVALID_PHYS_ADDR;
  38 }
  39
  40 static unsigned long find_next_free_page(struct page_pool *pool,
  41                                          unsigned long start)
  42 {
  43         unsigned long start_mask =
  44                 ~0UL >> (BITS_PER_LONG - (start % BITS_PER_LONG));
  45         unsigned long bmp_pos, bmp_val, page_nr;
  46
  47         if (start >= pool->pages)
  48                 return INVALID_PAGE_NR;
  49
  50         for (bmp_pos = start / BITS_PER_LONG;
  51              bmp_pos < pool->pages / BITS_PER_LONG; bmp_pos++) {
  52                 bmp_val = pool->used_bitmap[bmp_pos] | start_mask;
  53                 start_mask = 0;
  54                 if (bmp_val != ~0UL) {
  55                         page_nr = ffzl(bmp_val) + bmp_pos * BITS_PER_LONG;
  56                         if (page_nr >= pool->pages)
  57                                 break;
  58                         return page_nr;
  59                 }
  60         }
  61
  62         return INVALID_PAGE_NR;
  63 }
  64
  65 void *page_alloc(struct page_pool *pool, unsigned int num)
  66 {
  67         unsigned long start, last, next;
  68         unsigned int allocated;
  69
  70         start = find_next_free_page(pool, 0);
  71         if (start == INVALID_PAGE_NR)
  72                 return NULL;
  73
  74 restart:
  75         for (allocated = 1, last = start; allocated < num;
  76              allocated++, last = next) {
  77                 next = find_next_free_page(pool, last + 1);
  78                 if (next == INVALID_PAGE_NR)
  79                         return NULL;
  80                 if (next != last + 1) {
  81                         start = next;
  82                         goto restart;
  83                 }
  84         }
  85
  86         for (allocated = 0; allocated < num; allocated++)
  87                 set_bit(start + allocated, pool->used_bitmap);
  88
  89         pool->used_pages += num;
  90
  91         return pool->base_address + start * PAGE_SIZE;
  92 }
  93
  94 void page_free(struct page_pool *pool, void *page, unsigned int num)
  95 {
  96         unsigned long page_nr;
  97
  98         if (!page)
  99                 return;
 100
 101         while (num-- > 0) {
 102                 if (pool->flags & PAGE_SCRUB_ON_FREE)
 103                         memset(page, 0, PAGE_SIZE);
 104                 page_nr = (page - pool->base_address) / PAGE_SIZE;
 105                 clear_bit(page_nr, pool->used_bitmap);
 106                 pool->used_pages--;
 107                 page += PAGE_SIZE;
 108         }
 109 }
 110
 111 unsigned long page_map_virt2phys(const struct paging_structures *pg_structs,
 112                                  unsigned long virt)
 113 {
 114         const struct paging *paging = pg_structs->root_paging;
 115         page_table_t pt = pg_structs->root_table;
 116         unsigned long phys;
 117         pt_entry_t pte;
 118
 119         while (1) {
 120                 pte = paging->get_entry(pt, virt);
 121                 if (!paging->entry_valid(pte))
 122                         return INVALID_PHYS_ADDR;
 123                 phys = paging->get_phys(pte, virt);
 124                 if (phys != INVALID_PHYS_ADDR)
 125                         return phys;
 126                 pt = page_map_phys2hvirt(paging->get_next_pt(pte));
 127                 paging++;
 128         }
 129 }
 130
 131 static void flush_pt_entry(pt_entry_t pte, enum page_map_coherent coherent)
 132 {
 133         if (coherent == PAGE_MAP_COHERENT)
 134                 flush_cache(pte, sizeof(*pte));
 135 }
 136
 137 static int split_hugepage(const struct paging *paging, pt_entry_t pte,
 138                           unsigned long virt, enum page_map_coherent coherent)
 139 {
 140         unsigned long phys = paging->get_phys(pte, virt);
 141         struct paging_structures sub_structs;
 142         unsigned long page_mask, flags;
 143
 144         if (phys == INVALID_PHYS_ADDR)
 145                 return 0;
 146
 147         page_mask = ~(paging->page_size - 1);
 148         phys &= page_mask;
 149         virt &= page_mask;
 150
 151         flags = paging->get_flags(pte);
 152
 153         sub_structs.root_paging = paging + 1;
 154         sub_structs.root_table = page_alloc(&mem_pool, 1);
 155         if (!sub_structs.root_table)
 156                 return -ENOMEM;
 157         paging->set_next_pt(pte, page_map_hvirt2phys(sub_structs.root_table));
 158         flush_pt_entry(pte, coherent);
 159
 160         return page_map_create(&sub_structs, phys, paging->page_size, virt,
 161                                flags, coherent);
 162 }
 163
 164 int page_map_create(const struct paging_structures *pg_structs,
 165                     unsigned long phys, unsigned long size, unsigned long virt,
 166                     unsigned long flags, enum page_map_coherent coherent)
 167 {
 168         phys &= PAGE_MASK;
 169         virt &= PAGE_MASK;
 170         size = PAGE_ALIGN(size);
 171
 172         while (size > 0) {
 173                 const struct paging *paging = pg_structs->root_paging;
 174                 page_table_t pt = pg_structs->root_table;
 175                 pt_entry_t pte;
 176                 int err;
 177
 178                 while (1) {
 179                         pte = paging->get_entry(pt, virt);
 180                         if (paging->page_size > 0 &&
 181                             paging->page_size <= size &&
 182                             ((phys | virt) & (paging->page_size - 1)) == 0) {
 183                                 /*
 184                                  * We might be overwriting a more fine-grained
 185                                  * mapping, so release it first. This cannot
 186                                  * fail as we are working along hugepage
 187                                  * boundaries.
 188                                  */
 189                                 if (paging->page_size > PAGE_SIZE)
 190                                         page_map_destroy(pg_structs, virt,
 191                                                          paging->page_size,
 192                                                          coherent);
 193                                 paging->set_terminal(pte, phys, flags);
 194                                 flush_pt_entry(pte, coherent);
 195                                 break;
 196                         }
 197                         if (paging->entry_valid(pte)) {
 198                                 err = split_hugepage(paging, pte, virt,
 199                                                      coherent);
 200                                 if (err)
 201                                         return err;
 202                                 pt = page_map_phys2hvirt(
 203                                                 paging->get_next_pt(pte));
 204                         } else {
 205                                 pt = page_alloc(&mem_pool, 1);
 206                                 if (!pt)
 207                                         return -ENOMEM;
 208                                 paging->set_next_pt(pte,
 209                                                     page_map_hvirt2phys(pt));
 210                                 flush_pt_entry(pte, coherent);
 211                         }
 212                         paging++;
 213                 }
 214                 arch_tlb_flush_page(virt);
 215
 216                 phys += paging->page_size;
 217                 virt += paging->page_size;
 218                 size -= paging->page_size;
 219         }
 220         return 0;
 221 }
 222
 223 int page_map_destroy(const struct paging_structures *pg_structs,
 224                      unsigned long virt, unsigned long size,
 225                      enum page_map_coherent coherent)
 226 {
 227         size = PAGE_ALIGN(size);
 228
 229         while (size > 0) {
 230                 const struct paging *paging = pg_structs->root_paging;
 231                 page_table_t pt[MAX_PAGE_DIR_LEVELS];
 232                 unsigned long page_size;
 233                 pt_entry_t pte;
 234                 int n = 0;
 235                 int err;
 236
 237                 /* walk down the page table, saving intermediate tables */
 238                 pt[0] = pg_structs->root_table;
 239                 while (1) {
 240                         pte = paging->get_entry(pt[n], virt);
 241                         if (!paging->entry_valid(pte))
 242                                 break;
 243                         if (paging->get_phys(pte, virt) != INVALID_PHYS_ADDR) {
 244                                 if (paging->page_size > size) {
 245                                         err = split_hugepage(paging, pte, virt,
 246                                                              coherent);
 247                                         if (err)
 248                                                 return err;
 249                                 } else
 250                                         break;
 251                         }
 252                         pt[++n] = page_map_phys2hvirt(
 253                                         paging->get_next_pt(pte));
 254                         paging++;
 255                 }
 256                 /* advance by page size of current level paging */
 257                 page_size = paging->page_size ? paging->page_size : PAGE_SIZE;
 258
 259                 /* walk up again, clearing entries, releasing empty tables */
 260                 while (1) {
 261                         paging->clear_entry(pte);
 262                         flush_pt_entry(pte, coherent);
 263                         if (n == 0 || !paging->page_table_empty(pt[n]))
 264                                 break;
 265                         page_free(&mem_pool, pt[n], 1);
 266                         paging--;
 267                         pte = paging->get_entry(pt[--n], virt);
 268                 }
 269                 arch_tlb_flush_page(virt);
 270
 271                 if (page_size > size)
 272                         break;
 273                 virt += page_size;
 274                 size -= page_size;
 275         }
 276         return 0;
 277 }
 278
 279 void *page_map_get_guest_page(struct per_cpu *cpu_data,
 280                               const struct guest_paging_structures *pg_structs,
 281                               unsigned long virt, unsigned long flags)
 282 {
 283         unsigned long page_table_gphys = pg_structs->root_table_gphys;
 284         const struct paging *paging = pg_structs->root_paging;
 285         unsigned long page_virt, phys, gphys;
 286         pt_entry_t pte;
 287         int err;
 288
 289         page_virt = TEMPORARY_MAPPING_BASE +
 290                 cpu_data->cpu_id * PAGE_SIZE * NUM_TEMPORARY_PAGES;
 291
 292         while (1) {
 293                 /* map guest page table */
 294                 phys = arch_page_map_gphys2phys(cpu_data, page_table_gphys);
 295                 if (phys == INVALID_PHYS_ADDR)
 296                         return NULL;
 297                 err = page_map_create(&hv_paging_structs, phys,
 298                                       PAGE_SIZE, page_virt,
 299                                       PAGE_READONLY_FLAGS,
 300                                       PAGE_MAP_NON_COHERENT);
 301                 if (err)
 302                         return NULL;
 303
 304                 /* evaluate page table entry */
 305                 pte = paging->get_entry((page_table_t)page_virt, virt);
 306                 if (!paging->entry_valid(pte))
 307                         return NULL;
 308                 gphys = paging->get_phys(pte, virt);
 309                 if (gphys != INVALID_PHYS_ADDR)
 310                         break;
 311                 page_table_gphys = paging->get_next_pt(pte);
 312                 paging++;
 313         }
 314
 315         phys = arch_page_map_gphys2phys(cpu_data, gphys);
 316         if (phys == INVALID_PHYS_ADDR)
 317                 return NULL;
 318         /* map guest page */
 319         err = page_map_create(&hv_paging_structs, phys, PAGE_SIZE, page_virt,
 320                               flags, PAGE_MAP_NON_COHERENT);
 321         if (err)
 322                 return NULL;
 323
 324         return (void *)page_virt;
 325 }
 326
 327 int paging_init(void)
 328 {
 329         unsigned long per_cpu_pages, config_pages, bitmap_pages;
 330         unsigned long n;
 331         int err;
 332
 333         per_cpu_pages = hypervisor_header.possible_cpus *
 334                 sizeof(struct per_cpu) / PAGE_SIZE;
 335
 336         system_config = (struct jailhouse_system *)
 337                 (__page_pool + per_cpu_pages * PAGE_SIZE);
 338         config_pages = (jailhouse_system_config_size(system_config) +
 339                         PAGE_SIZE - 1) / PAGE_SIZE;
 340
 341         mem_pool.pages = (system_config->hypervisor_memory.size -
 342                 (__page_pool - (u8 *)&hypervisor_header)) / PAGE_SIZE;
 343         bitmap_pages = (mem_pool.pages + BITS_PER_PAGE - 1) / BITS_PER_PAGE;
 344
 345         if (mem_pool.pages <= per_cpu_pages + config_pages + bitmap_pages)
 346                 goto error_nomem;
 347
 348         mem_pool.base_address = __page_pool;
 349         mem_pool.used_bitmap =
 350                 (unsigned long *)(__page_pool + per_cpu_pages * PAGE_SIZE +
 351                                   config_pages * PAGE_SIZE);
 352         mem_pool.used_pages = per_cpu_pages + config_pages + bitmap_pages;
 353         for (n = 0; n < mem_pool.used_pages; n++)
 354                 set_bit(n, mem_pool.used_bitmap);
 355         mem_pool.flags = PAGE_SCRUB_ON_FREE;
 356
 357         remap_pool.used_bitmap = page_alloc(&mem_pool, NUM_REMAP_BITMAP_PAGES);
 358         remap_pool.used_pages =
 359                 hypervisor_header.possible_cpus * NUM_TEMPORARY_PAGES;
 360         for (n = 0; n < remap_pool.used_pages; n++)
 361                 set_bit(n, remap_pool.used_bitmap);
 362
 363         arch_paging_init();
 364
 365         hv_paging_structs.root_paging = hv_paging;
 366         hv_paging_structs.root_table = page_alloc(&mem_pool, 1);
 367         if (!hv_paging_structs.root_table)
 368                 goto error_nomem;
 369
 370         /* Replicate hypervisor mapping of Linux */
 371         err = page_map_create(&hv_paging_structs,
 372                               page_map_hvirt2phys(&hypervisor_header),
 373                               system_config->hypervisor_memory.size,
 374                               (unsigned long)&hypervisor_header,
 375                               PAGE_DEFAULT_FLAGS, PAGE_MAP_NON_COHERENT);
 376         if (err)
 377                 goto error_nomem;
 378
 379         /* Make sure any remappings to the temporary regions can be performed
 380          * without allocations of page table pages. */
 381         err = page_map_create(&hv_paging_structs, 0,
 382                               remap_pool.used_pages * PAGE_SIZE,
 383                               TEMPORARY_MAPPING_BASE, PAGE_NONPRESENT_FLAGS,
 384                               PAGE_MAP_NON_COHERENT);
 385         if (err)
 386                 goto error_nomem;
 387
 388         return 0;
 389
 390 error_nomem:
 391         printk("FATAL: page pool much too small\n");
 392         return -ENOMEM;
 393 }
 394
 395 void page_map_dump_stats(const char *when)
 396 {
 397         printk("Page pool usage %s: mem %d/%d, remap %d/%d\n", when,
 398                mem_pool.used_pages, mem_pool.pages,
 399                remap_pool.used_pages, remap_pool.pages);
 400 }