| 1 | |
| 2 | #include <linux/alloc_tag.h>
|
| 3 | #include <linux/execmem.h>
|
| 4 | #include <linux/fs.h>
|
| 5 | #include <linux/gfp.h>
|
| 6 | #include <linux/kallsyms.h>
|
| 7 | #include <linux/module.h>
|
| 8 | #include <linux/page_ext.h>
|
| 9 | #include <linux/proc_fs.h>
|
| 10 | #include <linux/seq_buf.h>
|
| 11 | #include <linux/seq_file.h>
|
| 12 | #include <linux/vmalloc.h>
|
| 13 | #include <linux/kmemleak.h>
|
| 14 |
|
| 15 | #define ALLOCINFO_FILE_NAME "allocinfo"
|
| 16 | #define MODULE_ALLOC_TAG_VMAP_SIZE (100000UL * sizeof(struct alloc_tag))
|
| 17 | #define SECTION_START(NAME) (CODETAG_SECTION_START_PREFIX NAME)
|
| 18 | #define SECTION_STOP(NAME) (CODETAG_SECTION_STOP_PREFIX NAME)
|
| 19 |
|
| 20 | #ifdef CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT
|
| 21 | static bool mem_profiling_support = true;
|
| 22 | #else
|
| 23 | static bool mem_profiling_support;
|
| 24 | #endif
|
| 25 |
|
| 26 | static struct codetag_type *alloc_tag_cttype;
|
| 27 |
|
| 28 | DEFINE_PER_CPU(struct alloc_tag_counters, _shared_alloc_tag);
|
| 29 | EXPORT_SYMBOL(_shared_alloc_tag);
|
| 30 |
|
| 31 | DEFINE_STATIC_KEY_MAYBE(CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT,
|
| 32 | mem_alloc_profiling_key);
|
| 33 | EXPORT_SYMBOL(mem_alloc_profiling_key);
|
| 34 |
|
| 35 | DEFINE_STATIC_KEY_FALSE(mem_profiling_compressed);
|
| 36 |
|
| 37 | struct alloc_tag_kernel_section kernel_tags = { NULL, 0 };
|
| 38 | unsigned long alloc_tag_ref_mask;
|
| 39 | int alloc_tag_ref_offs;
|
| 40 |
|
| 41 | struct allocinfo_private {
|
| 42 | struct codetag_iterator iter;
|
| 43 | bool print_header;
|
| 44 | };
|
| 45 |
|
| 46 | static void *allocinfo_start(struct seq_file *m, loff_t *pos)
|
| 47 | {
|
| 48 | struct allocinfo_private *priv;
|
| 49 | struct codetag *ct;
|
| 50 | loff_t node = *pos;
|
| 51 |
|
| 52 | priv = kzalloc(sizeof(*priv), GFP_KERNEL);
|
| 53 | m->private = priv;
|
| 54 | if (!priv)
|
| 55 | return NULL;
|
| 56 |
|
| 57 | priv->print_header = (node == 0);
|
| 58 | codetag_lock_module_list(alloc_tag_cttype, true);
|
| 59 | priv->iter = codetag_get_ct_iter(alloc_tag_cttype);
|
| 60 | while ((ct = codetag_next_ct(&priv->iter)) != NULL && node)
|
| 61 | node--;
|
| 62 |
|
| 63 | return ct ? priv : NULL;
|
| 64 | }
|
| 65 |
|
| 66 | static void *allocinfo_next(struct seq_file *m, void *arg, loff_t *pos)
|
| 67 | {
|
| 68 | struct allocinfo_private *priv = (struct allocinfo_private *)arg;
|
| 69 | struct codetag *ct = codetag_next_ct(&priv->iter);
|
| 70 |
|
| 71 | (*pos)++;
|
| 72 | if (!ct)
|
| 73 | return NULL;
|
| 74 |
|
| 75 | return priv;
|
| 76 | }
|
| 77 |
|
| 78 | static void allocinfo_stop(struct seq_file *m, void *arg)
|
| 79 | {
|
| 80 | struct allocinfo_private *priv = (struct allocinfo_private *)m->private;
|
| 81 |
|
| 82 | if (priv) {
|
| 83 | codetag_lock_module_list(alloc_tag_cttype, false);
|
| 84 | kfree(priv);
|
| 85 | }
|
| 86 | }
|
| 87 |
|
| 88 | static void print_allocinfo_header(struct seq_buf *buf)
|
| 89 | {
|
| 90 |
|
| 91 | seq_buf_printf(buf, "allocinfo - version: 1.0\n");
|
| 92 | seq_buf_printf(buf, "# <size> <calls> <tag info>\n");
|
| 93 | }
|
| 94 |
|
| 95 | static void alloc_tag_to_text(struct seq_buf *out, struct codetag *ct)
|
| 96 | {
|
| 97 | struct alloc_tag *tag = ct_to_alloc_tag(ct);
|
| 98 | struct alloc_tag_counters counter = alloc_tag_read(tag);
|
| 99 | s64 bytes = counter.bytes;
|
| 100 |
|
| 101 | seq_buf_printf(out, "%12lli %8llu ", bytes, counter.calls);
|
| 102 | codetag_to_text(out, ct);
|
| 103 | seq_buf_putc(out, ' ');
|
| 104 | seq_buf_putc(out, '\n');
|
| 105 | }
|
| 106 |
|
| 107 | static int allocinfo_show(struct seq_file *m, void *arg)
|
| 108 | {
|
| 109 | struct allocinfo_private *priv = (struct allocinfo_private *)arg;
|
| 110 | char *bufp;
|
| 111 | size_t n = seq_get_buf(m, &bufp);
|
| 112 | struct seq_buf buf;
|
| 113 |
|
| 114 | seq_buf_init(&buf, bufp, n);
|
| 115 | if (priv->print_header) {
|
| 116 | print_allocinfo_header(&buf);
|
| 117 | priv->print_header = false;
|
| 118 | }
|
| 119 | alloc_tag_to_text(&buf, priv->iter.ct);
|
| 120 | seq_commit(m, seq_buf_used(&buf));
|
| 121 | return 0;
|
| 122 | }
|
| 123 |
|
| 124 | static const struct seq_operations allocinfo_seq_op = {
|
| 125 | .start = allocinfo_start,
|
| 126 | .next = allocinfo_next,
|
| 127 | .stop = allocinfo_stop,
|
| 128 | .show = allocinfo_show,
|
| 129 | };
|
| 130 |
|
| 131 | size_t alloc_tag_top_users(struct codetag_bytes *tags, size_t count, bool can_sleep)
|
| 132 | {
|
| 133 | struct codetag_iterator iter;
|
| 134 | struct codetag *ct;
|
| 135 | struct codetag_bytes n;
|
| 136 | unsigned int i, nr = 0;
|
| 137 |
|
| 138 | if (IS_ERR_OR_NULL(alloc_tag_cttype))
|
| 139 | return 0;
|
| 140 |
|
| 141 | if (can_sleep)
|
| 142 | codetag_lock_module_list(alloc_tag_cttype, true);
|
| 143 | else if (!codetag_trylock_module_list(alloc_tag_cttype))
|
| 144 | return 0;
|
| 145 |
|
| 146 | iter = codetag_get_ct_iter(alloc_tag_cttype);
|
| 147 | while ((ct = codetag_next_ct(&iter))) {
|
| 148 | struct alloc_tag_counters counter = alloc_tag_read(ct_to_alloc_tag(ct));
|
| 149 |
|
| 150 | n.ct = ct;
|
| 151 | n.bytes = counter.bytes;
|
| 152 |
|
| 153 | for (i = 0; i < nr; i++)
|
| 154 | if (n.bytes > tags[i].bytes)
|
| 155 | break;
|
| 156 |
|
| 157 | if (i < count) {
|
| 158 | nr -= nr == count;
|
| 159 | memmove(&tags[i + 1],
|
| 160 | &tags[i],
|
| 161 | sizeof(tags[0]) * (nr - i));
|
| 162 | nr++;
|
| 163 | tags[i] = n;
|
| 164 | }
|
| 165 | }
|
| 166 |
|
| 167 | codetag_lock_module_list(alloc_tag_cttype, false);
|
| 168 |
|
| 169 | return nr;
|
| 170 | }
|
| 171 |
|
| 172 | void pgalloc_tag_split(struct folio *folio, int old_order, int new_order)
|
| 173 | {
|
| 174 | int i;
|
| 175 | struct alloc_tag *tag;
|
| 176 | unsigned int nr_pages = 1 << new_order;
|
| 177 |
|
| 178 | if (!mem_alloc_profiling_enabled())
|
| 179 | return;
|
| 180 |
|
| 181 | tag = __pgalloc_tag_get(&folio->page);
|
| 182 | if (!tag)
|
| 183 | return;
|
| 184 |
|
| 185 | for (i = nr_pages; i < (1 << old_order); i += nr_pages) {
|
| 186 | union pgtag_ref_handle handle;
|
| 187 | union codetag_ref ref;
|
| 188 |
|
| 189 | if (get_page_tag_ref(folio_page(folio, i), &ref, &handle)) {
|
| 190 |
|
| 191 | alloc_tag_ref_set(&ref, tag);
|
| 192 | update_page_tag_ref(handle, &ref);
|
| 193 | put_page_tag_ref(handle);
|
| 194 | }
|
| 195 | }
|
| 196 | }
|
| 197 |
|
| 198 | void pgalloc_tag_swap(struct folio *new, struct folio *old)
|
| 199 | {
|
| 200 | union pgtag_ref_handle handle_old, handle_new;
|
| 201 | union codetag_ref ref_old, ref_new;
|
| 202 | struct alloc_tag *tag_old, *tag_new;
|
| 203 |
|
| 204 | if (!mem_alloc_profiling_enabled())
|
| 205 | return;
|
| 206 |
|
| 207 | tag_old = __pgalloc_tag_get(&old->page);
|
| 208 | if (!tag_old)
|
| 209 | return;
|
| 210 | tag_new = __pgalloc_tag_get(&new->page);
|
| 211 | if (!tag_new)
|
| 212 | return;
|
| 213 |
|
| 214 | if (!get_page_tag_ref(&old->page, &ref_old, &handle_old))
|
| 215 | return;
|
| 216 | if (!get_page_tag_ref(&new->page, &ref_new, &handle_new)) {
|
| 217 | put_page_tag_ref(handle_old);
|
| 218 | return;
|
| 219 | }
|
| 220 |
|
| 221 | |
| 222 | * Clear tag references to avoid debug warning when using
|
| 223 | * __alloc_tag_ref_set() with non-empty reference.
|
| 224 |
|
| 225 | set_codetag_empty(&ref_old);
|
| 226 | set_codetag_empty(&ref_new);
|
| 227 |
|
| 228 |
|
| 229 | __alloc_tag_ref_set(&ref_old, tag_new);
|
| 230 | update_page_tag_ref(handle_old, &ref_old);
|
| 231 | __alloc_tag_ref_set(&ref_new, tag_old);
|
| 232 | update_page_tag_ref(handle_new, &ref_new);
|
| 233 |
|
| 234 | put_page_tag_ref(handle_old);
|
| 235 | put_page_tag_ref(handle_new);
|
| 236 | }
|
| 237 |
|
| 238 | static void shutdown_mem_profiling(bool remove_file)
|
| 239 | {
|
| 240 | if (mem_alloc_profiling_enabled())
|
| 241 | static_branch_disable(&mem_alloc_profiling_key);
|
| 242 |
|
| 243 | if (!mem_profiling_support)
|
| 244 | return;
|
| 245 |
|
| 246 | if (remove_file)
|
| 247 | remove_proc_entry(ALLOCINFO_FILE_NAME, NULL);
|
| 248 | mem_profiling_support = false;
|
| 249 | }
|
| 250 |
|
| 251 | void __init alloc_tag_sec_init(void)
|
| 252 | {
|
| 253 | struct alloc_tag *last_codetag;
|
| 254 |
|
| 255 | if (!mem_profiling_support)
|
| 256 | return;
|
| 257 |
|
| 258 | if (!static_key_enabled(&mem_profiling_compressed))
|
| 259 | return;
|
| 260 |
|
| 261 | kernel_tags.first_tag = (struct alloc_tag *)kallsyms_lookup_name(
|
| 262 | SECTION_START(ALLOC_TAG_SECTION_NAME));
|
| 263 | last_codetag = (struct alloc_tag *)kallsyms_lookup_name(
|
| 264 | SECTION_STOP(ALLOC_TAG_SECTION_NAME));
|
| 265 | kernel_tags.count = last_codetag - kernel_tags.first_tag;
|
| 266 |
|
| 267 |
|
| 268 | if (kernel_tags.count > (1UL << NR_UNUSED_PAGEFLAG_BITS)) {
|
| 269 | shutdown_mem_profiling(false);
|
| 270 | pr_err("%lu allocation tags cannot be references using %d available page flag bits. Memory allocation profiling is disabled!\n",
|
| 271 | kernel_tags.count, NR_UNUSED_PAGEFLAG_BITS);
|
| 272 | return;
|
| 273 | }
|
| 274 |
|
| 275 | alloc_tag_ref_offs = (LRU_REFS_PGOFF - NR_UNUSED_PAGEFLAG_BITS);
|
| 276 | alloc_tag_ref_mask = ((1UL << NR_UNUSED_PAGEFLAG_BITS) - 1);
|
| 277 | pr_debug("Memory allocation profiling compression is using %d page flag bits!\n",
|
| 278 | NR_UNUSED_PAGEFLAG_BITS);
|
| 279 | }
|
| 280 |
|
| 281 | #ifdef CONFIG_MODULES
|
| 282 |
|
| 283 | static struct maple_tree mod_area_mt = MTREE_INIT(mod_area_mt, MT_FLAGS_ALLOC_RANGE);
|
| 284 | static struct vm_struct *vm_module_tags;
|
| 285 |
|
| 286 | static struct module unloaded_mod;
|
| 287 |
|
| 288 | static struct module prepend_mod;
|
| 289 |
|
| 290 | struct alloc_tag_module_section module_tags;
|
| 291 |
|
| 292 | static inline unsigned long alloc_tag_align(unsigned long val)
|
| 293 | {
|
| 294 | if (!static_key_enabled(&mem_profiling_compressed)) {
|
| 295 |
|
| 296 | return val;
|
| 297 | }
|
| 298 |
|
| 299 | if (val % sizeof(struct alloc_tag) == 0)
|
| 300 | return val;
|
| 301 | return ((val / sizeof(struct alloc_tag)) + 1) * sizeof(struct alloc_tag);
|
| 302 | }
|
| 303 |
|
| 304 | static bool ensure_alignment(unsigned long align, unsigned int *prepend)
|
| 305 | {
|
| 306 | if (!static_key_enabled(&mem_profiling_compressed)) {
|
| 307 |
|
| 308 | return true;
|
| 309 | }
|
| 310 |
|
| 311 | |
| 312 | * If alloc_tag size is not a multiple of required alignment, tag
|
| 313 | * indexing does not work.
|
| 314 |
|
| 315 | if (!IS_ALIGNED(sizeof(struct alloc_tag), align))
|
| 316 | return false;
|
| 317 |
|
| 318 |
|
| 319 | if (*prepend)
|
| 320 | *prepend = alloc_tag_align(*prepend);
|
| 321 |
|
| 322 | return true;
|
| 323 | }
|
| 324 |
|
| 325 | static inline bool tags_addressable(void)
|
| 326 | {
|
| 327 | unsigned long tag_idx_count;
|
| 328 |
|
| 329 | if (!static_key_enabled(&mem_profiling_compressed))
|
| 330 | return true;
|
| 331 |
|
| 332 | tag_idx_count = CODETAG_ID_FIRST + kernel_tags.count +
|
| 333 | module_tags.size / sizeof(struct alloc_tag);
|
| 334 |
|
| 335 | return tag_idx_count < (1UL << NR_UNUSED_PAGEFLAG_BITS);
|
| 336 | }
|
| 337 |
|
| 338 | static bool needs_section_mem(struct module *mod, unsigned long size)
|
| 339 | {
|
| 340 | if (!mem_profiling_support)
|
| 341 | return false;
|
| 342 |
|
| 343 | return size >= sizeof(struct alloc_tag);
|
| 344 | }
|
| 345 |
|
| 346 | static bool clean_unused_counters(struct alloc_tag *start_tag,
|
| 347 | struct alloc_tag *end_tag)
|
| 348 | {
|
| 349 | struct alloc_tag *tag;
|
| 350 | bool ret = true;
|
| 351 |
|
| 352 | for (tag = start_tag; tag <= end_tag; tag++) {
|
| 353 | struct alloc_tag_counters counter;
|
| 354 |
|
| 355 | if (!tag->counters)
|
| 356 | continue;
|
| 357 |
|
| 358 | counter = alloc_tag_read(tag);
|
| 359 | if (!counter.bytes) {
|
| 360 | free_percpu(tag->counters);
|
| 361 | tag->counters = NULL;
|
| 362 | } else {
|
| 363 | ret = false;
|
| 364 | }
|
| 365 | }
|
| 366 |
|
| 367 | return ret;
|
| 368 | }
|
| 369 |
|
| 370 |
|
| 371 | static void clean_unused_module_areas_locked(void)
|
| 372 | {
|
| 373 | MA_STATE(mas, &mod_area_mt, 0, module_tags.size);
|
| 374 | struct module *val;
|
| 375 |
|
| 376 | mas_for_each(&mas, val, module_tags.size) {
|
| 377 | struct alloc_tag *start_tag;
|
| 378 | struct alloc_tag *end_tag;
|
| 379 |
|
| 380 | if (val != &unloaded_mod)
|
| 381 | continue;
|
| 382 |
|
| 383 |
|
| 384 | start_tag = (struct alloc_tag *)(module_tags.start_addr + mas.index);
|
| 385 | end_tag = (struct alloc_tag *)(module_tags.start_addr + mas.last);
|
| 386 | if (clean_unused_counters(start_tag, end_tag))
|
| 387 | mas_erase(&mas);
|
| 388 | }
|
| 389 | }
|
| 390 |
|
| 391 |
|
| 392 | static bool find_aligned_area(struct ma_state *mas, unsigned long section_size,
|
| 393 | unsigned long size, unsigned int prepend, unsigned long align)
|
| 394 | {
|
| 395 | bool cleanup_done = false;
|
| 396 |
|
| 397 | repeat:
|
| 398 |
|
| 399 | if (!mas_empty_area(mas, 0, section_size - 1, prepend + size)) {
|
| 400 | if (IS_ALIGNED(mas->index + prepend, align))
|
| 401 | return true;
|
| 402 |
|
| 403 |
|
| 404 | mas_reset(mas);
|
| 405 | if (!mas_empty_area(mas, 0, section_size - 1,
|
| 406 | size + prepend + align - 1))
|
| 407 | return true;
|
| 408 | }
|
| 409 |
|
| 410 |
|
| 411 | if (!cleanup_done) {
|
| 412 | clean_unused_module_areas_locked();
|
| 413 | cleanup_done = true;
|
| 414 | mas_reset(mas);
|
| 415 | goto repeat;
|
| 416 | }
|
| 417 |
|
| 418 | return false;
|
| 419 | }
|
| 420 |
|
| 421 | static int vm_module_tags_populate(void)
|
| 422 | {
|
| 423 | unsigned long phys_end = ALIGN_DOWN(module_tags.start_addr, PAGE_SIZE) +
|
| 424 | (vm_module_tags->nr_pages << PAGE_SHIFT);
|
| 425 | unsigned long new_end = module_tags.start_addr + module_tags.size;
|
| 426 |
|
| 427 | if (phys_end < new_end) {
|
| 428 | struct page **next_page = vm_module_tags->pages + vm_module_tags->nr_pages;
|
| 429 | unsigned long old_shadow_end = ALIGN(phys_end, MODULE_ALIGN);
|
| 430 | unsigned long new_shadow_end = ALIGN(new_end, MODULE_ALIGN);
|
| 431 | unsigned long more_pages;
|
| 432 | unsigned long nr = 0;
|
| 433 |
|
| 434 | more_pages = ALIGN(new_end - phys_end, PAGE_SIZE) >> PAGE_SHIFT;
|
| 435 | while (nr < more_pages) {
|
| 436 | unsigned long allocated;
|
| 437 |
|
| 438 | allocated = alloc_pages_bulk_node(GFP_KERNEL | __GFP_NOWARN,
|
| 439 | NUMA_NO_NODE, more_pages - nr, next_page + nr);
|
| 440 |
|
| 441 | if (!allocated)
|
| 442 | break;
|
| 443 | nr += allocated;
|
| 444 | }
|
| 445 |
|
| 446 | if (nr < more_pages ||
|
| 447 | vmap_pages_range(phys_end, phys_end + (nr << PAGE_SHIFT), PAGE_KERNEL,
|
| 448 | next_page, PAGE_SHIFT) < 0) {
|
| 449 |
|
| 450 | for (int i = 0; i < nr; i++)
|
| 451 | __free_page(next_page[i]);
|
| 452 | return -ENOMEM;
|
| 453 | }
|
| 454 |
|
| 455 | vm_module_tags->nr_pages += nr;
|
| 456 |
|
| 457 | |
| 458 | * Kasan allocates 1 byte of shadow for every 8 bytes of data.
|
| 459 | * When kasan_alloc_module_shadow allocates shadow memory,
|
| 460 | * its unit of allocation is a page.
|
| 461 | * Therefore, here we need to align to MODULE_ALIGN.
|
| 462 |
|
| 463 | if (old_shadow_end < new_shadow_end)
|
| 464 | kasan_alloc_module_shadow((void *)old_shadow_end,
|
| 465 | new_shadow_end - old_shadow_end,
|
| 466 | GFP_KERNEL);
|
| 467 | }
|
| 468 |
|
| 469 | |
| 470 | * Mark the pages as accessible, now that they are mapped.
|
| 471 | * With hardware tag-based KASAN, marking is skipped for
|
| 472 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc().
|
| 473 |
|
| 474 | kasan_unpoison_vmalloc((void *)module_tags.start_addr,
|
| 475 | new_end - module_tags.start_addr,
|
| 476 | KASAN_VMALLOC_PROT_NORMAL);
|
| 477 |
|
| 478 | return 0;
|
| 479 | }
|
| 480 |
|
| 481 | static void *reserve_module_tags(struct module *mod, unsigned long size,
|
| 482 | unsigned int prepend, unsigned long align)
|
| 483 | {
|
| 484 | unsigned long section_size = module_tags.end_addr - module_tags.start_addr;
|
| 485 | MA_STATE(mas, &mod_area_mt, 0, section_size - 1);
|
| 486 | unsigned long offset;
|
| 487 | void *ret = NULL;
|
| 488 |
|
| 489 |
|
| 490 | if (size < sizeof(struct alloc_tag))
|
| 491 | return ERR_PTR(-EINVAL);
|
| 492 |
|
| 493 | |
| 494 | * align is always power of 2, so we can use IS_ALIGNED and ALIGN.
|
| 495 | * align 0 or 1 means no alignment, to simplify set to 1.
|
| 496 |
|
| 497 | if (!align)
|
| 498 | align = 1;
|
| 499 |
|
| 500 | if (!ensure_alignment(align, &prepend)) {
|
| 501 | shutdown_mem_profiling(true);
|
| 502 | pr_err("%s: alignment %lu is incompatible with allocation tag indexing. Memory allocation profiling is disabled!\n",
|
| 503 | mod->name, align);
|
| 504 | return ERR_PTR(-EINVAL);
|
| 505 | }
|
| 506 |
|
| 507 | mas_lock(&mas);
|
| 508 | if (!find_aligned_area(&mas, section_size, size, prepend, align)) {
|
| 509 | ret = ERR_PTR(-ENOMEM);
|
| 510 | goto unlock;
|
| 511 | }
|
| 512 |
|
| 513 |
|
| 514 | offset = mas.index;
|
| 515 | offset += prepend;
|
| 516 | offset = ALIGN(offset, align);
|
| 517 | if (offset != mas.index) {
|
| 518 | unsigned long pad_start = mas.index;
|
| 519 |
|
| 520 | mas.last = offset - 1;
|
| 521 | mas_store(&mas, &prepend_mod);
|
| 522 | if (mas_is_err(&mas)) {
|
| 523 | ret = ERR_PTR(xa_err(mas.node));
|
| 524 | goto unlock;
|
| 525 | }
|
| 526 | mas.index = offset;
|
| 527 | mas.last = offset + size - 1;
|
| 528 | mas_store(&mas, mod);
|
| 529 | if (mas_is_err(&mas)) {
|
| 530 | mas.index = pad_start;
|
| 531 | mas_erase(&mas);
|
| 532 | ret = ERR_PTR(xa_err(mas.node));
|
| 533 | }
|
| 534 | } else {
|
| 535 | mas.last = offset + size - 1;
|
| 536 | mas_store(&mas, mod);
|
| 537 | if (mas_is_err(&mas))
|
| 538 | ret = ERR_PTR(xa_err(mas.node));
|
| 539 | }
|
| 540 | unlock:
|
| 541 | mas_unlock(&mas);
|
| 542 |
|
| 543 | if (IS_ERR(ret))
|
| 544 | return ret;
|
| 545 |
|
| 546 | if (module_tags.size < offset + size) {
|
| 547 | int grow_res;
|
| 548 |
|
| 549 | module_tags.size = offset + size;
|
| 550 | if (mem_alloc_profiling_enabled() && !tags_addressable()) {
|
| 551 | shutdown_mem_profiling(true);
|
| 552 | pr_warn("With module %s there are too many tags to fit in %d page flag bits. Memory allocation profiling is disabled!\n",
|
| 553 | mod->name, NR_UNUSED_PAGEFLAG_BITS);
|
| 554 | }
|
| 555 |
|
| 556 | grow_res = vm_module_tags_populate();
|
| 557 | if (grow_res) {
|
| 558 | shutdown_mem_profiling(true);
|
| 559 | pr_err("Failed to allocate memory for allocation tags in the module %s. Memory allocation profiling is disabled!\n",
|
| 560 | mod->name);
|
| 561 | return ERR_PTR(grow_res);
|
| 562 | }
|
| 563 | }
|
| 564 |
|
| 565 | return (struct alloc_tag *)(module_tags.start_addr + offset);
|
| 566 | }
|
| 567 |
|
| 568 | static void release_module_tags(struct module *mod, bool used)
|
| 569 | {
|
| 570 | MA_STATE(mas, &mod_area_mt, module_tags.size, module_tags.size);
|
| 571 | struct alloc_tag *start_tag;
|
| 572 | struct alloc_tag *end_tag;
|
| 573 | struct module *val;
|
| 574 |
|
| 575 | mas_lock(&mas);
|
| 576 | mas_for_each_rev(&mas, val, 0)
|
| 577 | if (val == mod)
|
| 578 | break;
|
| 579 |
|
| 580 | if (!val)
|
| 581 | goto out;
|
| 582 |
|
| 583 | if (!used)
|
| 584 | goto release_area;
|
| 585 |
|
| 586 | start_tag = (struct alloc_tag *)(module_tags.start_addr + mas.index);
|
| 587 | end_tag = (struct alloc_tag *)(module_tags.start_addr + mas.last);
|
| 588 | if (!clean_unused_counters(start_tag, end_tag)) {
|
| 589 | struct alloc_tag *tag;
|
| 590 |
|
| 591 | for (tag = start_tag; tag <= end_tag; tag++) {
|
| 592 | struct alloc_tag_counters counter;
|
| 593 |
|
| 594 | if (!tag->counters)
|
| 595 | continue;
|
| 596 |
|
| 597 | counter = alloc_tag_read(tag);
|
| 598 | pr_info("%s:%u module %s func:%s has %llu allocated at module unload\n",
|
| 599 | tag->ct.filename, tag->ct.lineno, tag->ct.modname,
|
| 600 | tag->ct.function, counter.bytes);
|
| 601 | }
|
| 602 | } else {
|
| 603 | used = false;
|
| 604 | }
|
| 605 | release_area:
|
| 606 | mas_store(&mas, used ? &unloaded_mod : NULL);
|
| 607 | val = mas_prev_range(&mas, 0);
|
| 608 | if (val == &prepend_mod)
|
| 609 | mas_store(&mas, NULL);
|
| 610 | out:
|
| 611 | mas_unlock(&mas);
|
| 612 | }
|
| 613 |
|
| 614 | static int load_module(struct module *mod, struct codetag *start, struct codetag *stop)
|
| 615 | {
|
| 616 |
|
| 617 | struct alloc_tag *start_tag;
|
| 618 | struct alloc_tag *stop_tag;
|
| 619 | struct alloc_tag *tag;
|
| 620 |
|
| 621 |
|
| 622 | if (!mod)
|
| 623 | return 0;
|
| 624 |
|
| 625 | start_tag = ct_to_alloc_tag(start);
|
| 626 | stop_tag = ct_to_alloc_tag(stop);
|
| 627 | for (tag = start_tag; tag < stop_tag; tag++) {
|
| 628 | WARN_ON(tag->counters);
|
| 629 | tag->counters = alloc_percpu(struct alloc_tag_counters);
|
| 630 | if (!tag->counters) {
|
| 631 | while (--tag >= start_tag) {
|
| 632 | free_percpu(tag->counters);
|
| 633 | tag->counters = NULL;
|
| 634 | }
|
| 635 | pr_err("Failed to allocate memory for allocation tag percpu counters in the module %s\n",
|
| 636 | mod->name);
|
| 637 | return -ENOMEM;
|
| 638 | }
|
| 639 |
|
| 640 | |
| 641 | * Avoid a kmemleak false positive. The pointer to the counters is stored
|
| 642 | * in the alloc_tag section of the module and cannot be directly accessed.
|
| 643 |
|
| 644 | kmemleak_ignore_percpu(tag->counters);
|
| 645 | }
|
| 646 | return 0;
|
| 647 | }
|
| 648 |
|
| 649 | static void replace_module(struct module *mod, struct module *new_mod)
|
| 650 | {
|
| 651 | MA_STATE(mas, &mod_area_mt, 0, module_tags.size);
|
| 652 | struct module *val;
|
| 653 |
|
| 654 | mas_lock(&mas);
|
| 655 | mas_for_each(&mas, val, module_tags.size) {
|
| 656 | if (val != mod)
|
| 657 | continue;
|
| 658 |
|
| 659 | mas_store_gfp(&mas, new_mod, GFP_KERNEL);
|
| 660 | break;
|
| 661 | }
|
| 662 | mas_unlock(&mas);
|
| 663 | }
|
| 664 |
|
| 665 | static int __init alloc_mod_tags_mem(void)
|
| 666 | {
|
| 667 |
|
| 668 | vm_module_tags = execmem_vmap(MODULE_ALLOC_TAG_VMAP_SIZE);
|
| 669 | if (!vm_module_tags) {
|
| 670 | pr_err("Failed to map %lu bytes for module allocation tags\n",
|
| 671 | MODULE_ALLOC_TAG_VMAP_SIZE);
|
| 672 | module_tags.start_addr = 0;
|
| 673 | return -ENOMEM;
|
| 674 | }
|
| 675 |
|
| 676 | vm_module_tags->pages = kmalloc_array(get_vm_area_size(vm_module_tags) >> PAGE_SHIFT,
|
| 677 | sizeof(struct page *), GFP_KERNEL | __GFP_ZERO);
|
| 678 | if (!vm_module_tags->pages) {
|
| 679 | free_vm_area(vm_module_tags);
|
| 680 | return -ENOMEM;
|
| 681 | }
|
| 682 |
|
| 683 | module_tags.start_addr = (unsigned long)vm_module_tags->addr;
|
| 684 | module_tags.end_addr = module_tags.start_addr + MODULE_ALLOC_TAG_VMAP_SIZE;
|
| 685 |
|
| 686 | module_tags.start_addr = alloc_tag_align(module_tags.start_addr);
|
| 687 |
|
| 688 | return 0;
|
| 689 | }
|
| 690 |
|
| 691 | static void __init free_mod_tags_mem(void)
|
| 692 | {
|
| 693 | int i;
|
| 694 |
|
| 695 | module_tags.start_addr = 0;
|
| 696 | for (i = 0; i < vm_module_tags->nr_pages; i++)
|
| 697 | __free_page(vm_module_tags->pages[i]);
|
| 698 | kfree(vm_module_tags->pages);
|
| 699 | free_vm_area(vm_module_tags);
|
| 700 | }
|
| 701 |
|
| 702 | #else
|
| 703 |
|
| 704 | static inline int alloc_mod_tags_mem(void) { return 0; }
|
| 705 | static inline void free_mod_tags_mem(void) {}
|
| 706 |
|
| 707 | #endif
|
| 708 |
|
| 709 |
|
| 710 | static int __init setup_early_mem_profiling(char *str)
|
| 711 | {
|
| 712 | bool compressed = false;
|
| 713 | bool enable;
|
| 714 |
|
| 715 | if (!str || !str[0])
|
| 716 | return -EINVAL;
|
| 717 |
|
| 718 | if (!strncmp(str, "never", 5)) {
|
| 719 | enable = false;
|
| 720 | mem_profiling_support = false;
|
| 721 | pr_info("Memory allocation profiling is disabled!\n");
|
| 722 | } else {
|
| 723 | char *token = strsep(&str, ",");
|
| 724 |
|
| 725 | if (kstrtobool(token, &enable))
|
| 726 | return -EINVAL;
|
| 727 |
|
| 728 | if (str) {
|
| 729 |
|
| 730 | if (strcmp(str, "compressed"))
|
| 731 | return -EINVAL;
|
| 732 |
|
| 733 | compressed = true;
|
| 734 | }
|
| 735 | mem_profiling_support = true;
|
| 736 | pr_info("Memory allocation profiling is enabled %s compression and is turned %s!\n",
|
| 737 | compressed ? "with" : "without", enable ? "on" : "off");
|
| 738 | }
|
| 739 |
|
| 740 | if (enable != mem_alloc_profiling_enabled()) {
|
| 741 | if (enable)
|
| 742 | static_branch_enable(&mem_alloc_profiling_key);
|
| 743 | else
|
| 744 | static_branch_disable(&mem_alloc_profiling_key);
|
| 745 | }
|
| 746 | if (compressed != static_key_enabled(&mem_profiling_compressed)) {
|
| 747 | if (compressed)
|
| 748 | static_branch_enable(&mem_profiling_compressed);
|
| 749 | else
|
| 750 | static_branch_disable(&mem_profiling_compressed);
|
| 751 | }
|
| 752 |
|
| 753 | return 0;
|
| 754 | }
|
| 755 | early_param("sysctl.vm.mem_profiling", setup_early_mem_profiling);
|
| 756 |
|
| 757 | static __init bool need_page_alloc_tagging(void)
|
| 758 | {
|
| 759 | if (static_key_enabled(&mem_profiling_compressed))
|
| 760 | return false;
|
| 761 |
|
| 762 | return mem_profiling_support;
|
| 763 | }
|
| 764 |
|
| 765 | static __init void init_page_alloc_tagging(void)
|
| 766 | {
|
| 767 | }
|
| 768 |
|
| 769 | struct page_ext_operations page_alloc_tagging_ops = {
|
| 770 | .size = sizeof(union codetag_ref),
|
| 771 | .need = need_page_alloc_tagging,
|
| 772 | .init = init_page_alloc_tagging,
|
| 773 | };
|
| 774 | EXPORT_SYMBOL(page_alloc_tagging_ops);
|
| 775 |
|
| 776 | #ifdef CONFIG_SYSCTL
|
| 777 | static struct ctl_table memory_allocation_profiling_sysctls[] = {
|
| 778 | {
|
| 779 | .procname = "mem_profiling",
|
| 780 | .data = &mem_alloc_profiling_key,
|
| 781 | #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG
|
| 782 | .mode = 0444,
|
| 783 | #else
|
| 784 | .mode = 0644,
|
| 785 | #endif
|
| 786 | .proc_handler = proc_do_static_key,
|
| 787 | },
|
| 788 | };
|
| 789 |
|
| 790 | static void __init sysctl_init(void)
|
| 791 | {
|
| 792 | if (!mem_profiling_support)
|
| 793 | memory_allocation_profiling_sysctls[0].mode = 0444;
|
| 794 |
|
| 795 | register_sysctl_init("vm", memory_allocation_profiling_sysctls);
|
| 796 | }
|
| 797 | #else
|
| 798 | static inline void sysctl_init(void) {}
|
| 799 | #endif
|
| 800 |
|
| 801 | static int __init alloc_tag_init(void)
|
| 802 | {
|
| 803 | const struct codetag_type_desc desc = {
|
| 804 | .section = ALLOC_TAG_SECTION_NAME,
|
| 805 | .tag_size = sizeof(struct alloc_tag),
|
| 806 | #ifdef CONFIG_MODULES
|
| 807 | .needs_section_mem = needs_section_mem,
|
| 808 | .alloc_section_mem = reserve_module_tags,
|
| 809 | .free_section_mem = release_module_tags,
|
| 810 | .module_load = load_module,
|
| 811 | .module_replaced = replace_module,
|
| 812 | #endif
|
| 813 | };
|
| 814 | int res;
|
| 815 |
|
| 816 | sysctl_init();
|
| 817 |
|
| 818 | if (!mem_profiling_support) {
|
| 819 | pr_info("Memory allocation profiling is not supported!\n");
|
| 820 | return 0;
|
| 821 | }
|
| 822 |
|
| 823 | if (!proc_create_seq(ALLOCINFO_FILE_NAME, 0400, NULL, &allocinfo_seq_op)) {
|
| 824 | pr_err("Failed to create %s file\n", ALLOCINFO_FILE_NAME);
|
| 825 | shutdown_mem_profiling(false);
|
| 826 | return -ENOMEM;
|
| 827 | }
|
| 828 |
|
| 829 | res = alloc_mod_tags_mem();
|
| 830 | if (res) {
|
| 831 | pr_err("Failed to reserve address space for module tags, errno = %d\n", res);
|
| 832 | shutdown_mem_profiling(true);
|
| 833 | return res;
|
| 834 | }
|
| 835 |
|
| 836 | alloc_tag_cttype = codetag_register_type(&desc);
|
| 837 | if (IS_ERR(alloc_tag_cttype)) {
|
| 838 | pr_err("Allocation tags registration failed, errno = %ld\n", PTR_ERR(alloc_tag_cttype));
|
| 839 | free_mod_tags_mem();
|
| 840 | shutdown_mem_profiling(true);
|
| 841 | return PTR_ERR(alloc_tag_cttype);
|
| 842 | }
|
| 843 |
|
| 844 | return 0;
|
| 845 | }
|
| 846 | module_init(alloc_tag_init);
|