| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 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/string_choices.h> |
| 13 | #include <linux/vmalloc.h> |
| 14 | #include <linux/kmemleak.h> |
| 15 | |
| 16 | #define ALLOCINFO_FILE_NAME "allocinfo" |
| 17 | #define MODULE_ALLOC_TAG_VMAP_SIZE (100000UL * sizeof(struct alloc_tag)) |
| 18 | #define SECTION_START(NAME) (CODETAG_SECTION_START_PREFIX NAME) |
| 19 | #define SECTION_STOP(NAME) (CODETAG_SECTION_STOP_PREFIX NAME) |
| 20 | |
| 21 | #ifdef CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT |
| 22 | static bool mem_profiling_support = true; |
| 23 | #else |
| 24 | static bool mem_profiling_support; |
| 25 | #endif |
| 26 | |
| 27 | static struct codetag_type *alloc_tag_cttype; |
| 28 | |
| 29 | #ifdef CONFIG_ARCH_MODULE_NEEDS_WEAK_PER_CPU |
| 30 | DEFINE_PER_CPU(struct alloc_tag_counters, _shared_alloc_tag); |
| 31 | EXPORT_SYMBOL(_shared_alloc_tag); |
| 32 | #endif |
| 33 | |
| 34 | DEFINE_STATIC_KEY_MAYBE(CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT, |
| 35 | mem_alloc_profiling_key); |
| 36 | EXPORT_SYMBOL(mem_alloc_profiling_key); |
| 37 | |
| 38 | DEFINE_STATIC_KEY_FALSE(mem_profiling_compressed); |
| 39 | |
| 40 | struct alloc_tag_kernel_section kernel_tags = { NULL, 0 }; |
| 41 | unsigned long alloc_tag_ref_mask; |
| 42 | int alloc_tag_ref_offs; |
| 43 | |
| 44 | struct allocinfo_private { |
| 45 | struct codetag_iterator iter; |
| 46 | bool ; |
| 47 | }; |
| 48 | |
| 49 | static void *allocinfo_start(struct seq_file *m, loff_t *pos) |
| 50 | { |
| 51 | struct allocinfo_private *priv; |
| 52 | loff_t node = *pos; |
| 53 | |
| 54 | priv = (struct allocinfo_private *)m->private; |
| 55 | codetag_lock_module_list(cttype: alloc_tag_cttype, lock: true); |
| 56 | if (node == 0) { |
| 57 | priv->print_header = true; |
| 58 | priv->iter = codetag_get_ct_iter(cttype: alloc_tag_cttype); |
| 59 | codetag_next_ct(iter: &priv->iter); |
| 60 | } |
| 61 | return priv->iter.ct ? priv : NULL; |
| 62 | } |
| 63 | |
| 64 | static void *allocinfo_next(struct seq_file *m, void *arg, loff_t *pos) |
| 65 | { |
| 66 | struct allocinfo_private *priv = (struct allocinfo_private *)arg; |
| 67 | struct codetag *ct = codetag_next_ct(iter: &priv->iter); |
| 68 | |
| 69 | (*pos)++; |
| 70 | if (!ct) |
| 71 | return NULL; |
| 72 | |
| 73 | return priv; |
| 74 | } |
| 75 | |
| 76 | static void allocinfo_stop(struct seq_file *m, void *arg) |
| 77 | { |
| 78 | codetag_lock_module_list(cttype: alloc_tag_cttype, lock: false); |
| 79 | } |
| 80 | |
| 81 | static void (struct seq_buf *buf) |
| 82 | { |
| 83 | /* Output format version, so we can change it. */ |
| 84 | seq_buf_printf(s: buf, fmt: "allocinfo - version: 2.0\n" ); |
| 85 | seq_buf_printf(s: buf, fmt: "# <size> <calls> <tag info>\n" ); |
| 86 | } |
| 87 | |
| 88 | static void alloc_tag_to_text(struct seq_buf *out, struct codetag *ct) |
| 89 | { |
| 90 | struct alloc_tag *tag = ct_to_alloc_tag(ct); |
| 91 | struct alloc_tag_counters counter = alloc_tag_read(tag); |
| 92 | s64 bytes = counter.bytes; |
| 93 | |
| 94 | seq_buf_printf(s: out, fmt: "%12lli %8llu " , bytes, counter.calls); |
| 95 | codetag_to_text(out, ct); |
| 96 | if (unlikely(alloc_tag_is_inaccurate(tag))) |
| 97 | seq_buf_printf(s: out, fmt: " accurate:no" ); |
| 98 | seq_buf_putc(s: out, c: ' '); |
| 99 | seq_buf_putc(s: out, c: '\n'); |
| 100 | } |
| 101 | |
| 102 | static int allocinfo_show(struct seq_file *m, void *arg) |
| 103 | { |
| 104 | struct allocinfo_private *priv = (struct allocinfo_private *)arg; |
| 105 | char *bufp; |
| 106 | size_t n = seq_get_buf(m, bufp: &bufp); |
| 107 | struct seq_buf buf; |
| 108 | |
| 109 | seq_buf_init(s: &buf, buf: bufp, size: n); |
| 110 | if (priv->print_header) { |
| 111 | print_allocinfo_header(buf: &buf); |
| 112 | priv->print_header = false; |
| 113 | } |
| 114 | alloc_tag_to_text(out: &buf, ct: priv->iter.ct); |
| 115 | seq_commit(m, num: seq_buf_used(s: &buf)); |
| 116 | return 0; |
| 117 | } |
| 118 | |
| 119 | static const struct seq_operations allocinfo_seq_op = { |
| 120 | .start = allocinfo_start, |
| 121 | .next = allocinfo_next, |
| 122 | .stop = allocinfo_stop, |
| 123 | .show = allocinfo_show, |
| 124 | }; |
| 125 | |
| 126 | size_t alloc_tag_top_users(struct codetag_bytes *tags, size_t count, bool can_sleep) |
| 127 | { |
| 128 | struct codetag_iterator iter; |
| 129 | struct codetag *ct; |
| 130 | struct codetag_bytes n; |
| 131 | unsigned int i, nr = 0; |
| 132 | |
| 133 | if (IS_ERR_OR_NULL(ptr: alloc_tag_cttype)) |
| 134 | return 0; |
| 135 | |
| 136 | if (can_sleep) |
| 137 | codetag_lock_module_list(cttype: alloc_tag_cttype, lock: true); |
| 138 | else if (!codetag_trylock_module_list(cttype: alloc_tag_cttype)) |
| 139 | return 0; |
| 140 | |
| 141 | iter = codetag_get_ct_iter(cttype: alloc_tag_cttype); |
| 142 | while ((ct = codetag_next_ct(iter: &iter))) { |
| 143 | struct alloc_tag_counters counter = alloc_tag_read(ct_to_alloc_tag(ct)); |
| 144 | |
| 145 | n.ct = ct; |
| 146 | n.bytes = counter.bytes; |
| 147 | |
| 148 | for (i = 0; i < nr; i++) |
| 149 | if (n.bytes > tags[i].bytes) |
| 150 | break; |
| 151 | |
| 152 | if (i < count) { |
| 153 | nr -= nr == count; |
| 154 | memmove(&tags[i + 1], |
| 155 | &tags[i], |
| 156 | sizeof(tags[0]) * (nr - i)); |
| 157 | nr++; |
| 158 | tags[i] = n; |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | codetag_lock_module_list(cttype: alloc_tag_cttype, lock: false); |
| 163 | |
| 164 | return nr; |
| 165 | } |
| 166 | |
| 167 | void pgalloc_tag_split(struct folio *folio, int old_order, int new_order) |
| 168 | { |
| 169 | int i; |
| 170 | struct alloc_tag *tag; |
| 171 | unsigned int nr_pages = 1 << new_order; |
| 172 | |
| 173 | if (!mem_alloc_profiling_enabled()) |
| 174 | return; |
| 175 | |
| 176 | tag = __pgalloc_tag_get(&folio->page); |
| 177 | if (!tag) |
| 178 | return; |
| 179 | |
| 180 | for (i = nr_pages; i < (1 << old_order); i += nr_pages) { |
| 181 | union pgtag_ref_handle handle; |
| 182 | union codetag_ref ref; |
| 183 | |
| 184 | if (get_page_tag_ref(folio_page(folio, i), &ref, &handle)) { |
| 185 | /* Set new reference to point to the original tag */ |
| 186 | alloc_tag_ref_set(&ref, tag); |
| 187 | update_page_tag_ref(handle, &ref); |
| 188 | put_page_tag_ref(handle); |
| 189 | } |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | void pgalloc_tag_swap(struct folio *new, struct folio *old) |
| 194 | { |
| 195 | union pgtag_ref_handle handle_old, handle_new; |
| 196 | union codetag_ref ref_old, ref_new; |
| 197 | struct alloc_tag *tag_old, *tag_new; |
| 198 | |
| 199 | if (!mem_alloc_profiling_enabled()) |
| 200 | return; |
| 201 | |
| 202 | tag_old = __pgalloc_tag_get(&old->page); |
| 203 | if (!tag_old) |
| 204 | return; |
| 205 | tag_new = __pgalloc_tag_get(&new->page); |
| 206 | if (!tag_new) |
| 207 | return; |
| 208 | |
| 209 | if (!get_page_tag_ref(&old->page, &ref_old, &handle_old)) |
| 210 | return; |
| 211 | if (!get_page_tag_ref(&new->page, &ref_new, &handle_new)) { |
| 212 | put_page_tag_ref(handle_old); |
| 213 | return; |
| 214 | } |
| 215 | |
| 216 | /* |
| 217 | * Clear tag references to avoid debug warning when using |
| 218 | * __alloc_tag_ref_set() with non-empty reference. |
| 219 | */ |
| 220 | set_codetag_empty(&ref_old); |
| 221 | set_codetag_empty(&ref_new); |
| 222 | |
| 223 | /* swap tags */ |
| 224 | __alloc_tag_ref_set(&ref_old, tag_new); |
| 225 | update_page_tag_ref(handle_old, &ref_old); |
| 226 | __alloc_tag_ref_set(&ref_new, tag_old); |
| 227 | update_page_tag_ref(handle_new, &ref_new); |
| 228 | |
| 229 | put_page_tag_ref(handle_old); |
| 230 | put_page_tag_ref(handle_new); |
| 231 | } |
| 232 | |
| 233 | static void shutdown_mem_profiling(bool remove_file) |
| 234 | { |
| 235 | if (mem_alloc_profiling_enabled()) |
| 236 | static_branch_disable(&mem_alloc_profiling_key); |
| 237 | |
| 238 | if (!mem_profiling_support) |
| 239 | return; |
| 240 | |
| 241 | if (remove_file) |
| 242 | remove_proc_entry(ALLOCINFO_FILE_NAME, NULL); |
| 243 | mem_profiling_support = false; |
| 244 | } |
| 245 | |
| 246 | void __init alloc_tag_sec_init(void) |
| 247 | { |
| 248 | struct alloc_tag *last_codetag; |
| 249 | |
| 250 | if (!mem_profiling_support) |
| 251 | return; |
| 252 | |
| 253 | if (!static_key_enabled(&mem_profiling_compressed)) |
| 254 | return; |
| 255 | |
| 256 | kernel_tags.first_tag = (struct alloc_tag *)kallsyms_lookup_name( |
| 257 | SECTION_START(ALLOC_TAG_SECTION_NAME)); |
| 258 | last_codetag = (struct alloc_tag *)kallsyms_lookup_name( |
| 259 | SECTION_STOP(ALLOC_TAG_SECTION_NAME)); |
| 260 | kernel_tags.count = last_codetag - kernel_tags.first_tag; |
| 261 | |
| 262 | /* Check if kernel tags fit into page flags */ |
| 263 | if (kernel_tags.count > (1UL << NR_UNUSED_PAGEFLAG_BITS)) { |
| 264 | shutdown_mem_profiling(remove_file: false); /* allocinfo file does not exist yet */ |
| 265 | pr_err("%lu allocation tags cannot be references using %d available page flag bits. Memory allocation profiling is disabled!\n" , |
| 266 | kernel_tags.count, NR_UNUSED_PAGEFLAG_BITS); |
| 267 | return; |
| 268 | } |
| 269 | |
| 270 | alloc_tag_ref_offs = (LRU_REFS_PGOFF - NR_UNUSED_PAGEFLAG_BITS); |
| 271 | alloc_tag_ref_mask = ((1UL << NR_UNUSED_PAGEFLAG_BITS) - 1); |
| 272 | pr_debug("Memory allocation profiling compression is using %d page flag bits!\n" , |
| 273 | NR_UNUSED_PAGEFLAG_BITS); |
| 274 | } |
| 275 | |
| 276 | #ifdef CONFIG_MODULES |
| 277 | |
| 278 | static struct maple_tree mod_area_mt = MTREE_INIT(mod_area_mt, MT_FLAGS_ALLOC_RANGE); |
| 279 | static struct vm_struct *vm_module_tags; |
| 280 | /* A dummy object used to indicate an unloaded module */ |
| 281 | static struct module unloaded_mod; |
| 282 | /* A dummy object used to indicate a module prepended area */ |
| 283 | static struct module prepend_mod; |
| 284 | |
| 285 | struct alloc_tag_module_section module_tags; |
| 286 | |
| 287 | static inline unsigned long alloc_tag_align(unsigned long val) |
| 288 | { |
| 289 | if (!static_key_enabled(&mem_profiling_compressed)) { |
| 290 | /* No alignment requirements when we are not indexing the tags */ |
| 291 | return val; |
| 292 | } |
| 293 | |
| 294 | if (val % sizeof(struct alloc_tag) == 0) |
| 295 | return val; |
| 296 | return ((val / sizeof(struct alloc_tag)) + 1) * sizeof(struct alloc_tag); |
| 297 | } |
| 298 | |
| 299 | static bool ensure_alignment(unsigned long align, unsigned int *prepend) |
| 300 | { |
| 301 | if (!static_key_enabled(&mem_profiling_compressed)) { |
| 302 | /* No alignment requirements when we are not indexing the tags */ |
| 303 | return true; |
| 304 | } |
| 305 | |
| 306 | /* |
| 307 | * If alloc_tag size is not a multiple of required alignment, tag |
| 308 | * indexing does not work. |
| 309 | */ |
| 310 | if (!IS_ALIGNED(sizeof(struct alloc_tag), align)) |
| 311 | return false; |
| 312 | |
| 313 | /* Ensure prepend consumes multiple of alloc_tag-sized blocks */ |
| 314 | if (*prepend) |
| 315 | *prepend = alloc_tag_align(val: *prepend); |
| 316 | |
| 317 | return true; |
| 318 | } |
| 319 | |
| 320 | static inline bool tags_addressable(void) |
| 321 | { |
| 322 | unsigned long tag_idx_count; |
| 323 | |
| 324 | if (!static_key_enabled(&mem_profiling_compressed)) |
| 325 | return true; /* with page_ext tags are always addressable */ |
| 326 | |
| 327 | tag_idx_count = CODETAG_ID_FIRST + kernel_tags.count + |
| 328 | module_tags.size / sizeof(struct alloc_tag); |
| 329 | |
| 330 | return tag_idx_count < (1UL << NR_UNUSED_PAGEFLAG_BITS); |
| 331 | } |
| 332 | |
| 333 | static bool needs_section_mem(struct module *mod, unsigned long size) |
| 334 | { |
| 335 | if (!mem_profiling_support) |
| 336 | return false; |
| 337 | |
| 338 | return size >= sizeof(struct alloc_tag); |
| 339 | } |
| 340 | |
| 341 | static bool clean_unused_counters(struct alloc_tag *start_tag, |
| 342 | struct alloc_tag *end_tag) |
| 343 | { |
| 344 | struct alloc_tag *tag; |
| 345 | bool ret = true; |
| 346 | |
| 347 | for (tag = start_tag; tag <= end_tag; tag++) { |
| 348 | struct alloc_tag_counters counter; |
| 349 | |
| 350 | if (!tag->counters) |
| 351 | continue; |
| 352 | |
| 353 | counter = alloc_tag_read(tag); |
| 354 | if (!counter.bytes) { |
| 355 | free_percpu(pdata: tag->counters); |
| 356 | tag->counters = NULL; |
| 357 | } else { |
| 358 | ret = false; |
| 359 | } |
| 360 | } |
| 361 | |
| 362 | return ret; |
| 363 | } |
| 364 | |
| 365 | /* Called with mod_area_mt locked */ |
| 366 | static void clean_unused_module_areas_locked(void) |
| 367 | { |
| 368 | MA_STATE(mas, &mod_area_mt, 0, module_tags.size); |
| 369 | struct module *val; |
| 370 | |
| 371 | mas_for_each(&mas, val, module_tags.size) { |
| 372 | struct alloc_tag *start_tag; |
| 373 | struct alloc_tag *end_tag; |
| 374 | |
| 375 | if (val != &unloaded_mod) |
| 376 | continue; |
| 377 | |
| 378 | /* Release area if all tags are unused */ |
| 379 | start_tag = (struct alloc_tag *)(module_tags.start_addr + mas.index); |
| 380 | end_tag = (struct alloc_tag *)(module_tags.start_addr + mas.last); |
| 381 | if (clean_unused_counters(start_tag, end_tag)) |
| 382 | mas_erase(mas: &mas); |
| 383 | } |
| 384 | } |
| 385 | |
| 386 | /* Called with mod_area_mt locked */ |
| 387 | static bool find_aligned_area(struct ma_state *mas, unsigned long section_size, |
| 388 | unsigned long size, unsigned int prepend, unsigned long align) |
| 389 | { |
| 390 | bool cleanup_done = false; |
| 391 | |
| 392 | repeat: |
| 393 | /* Try finding exact size and hope the start is aligned */ |
| 394 | if (!mas_empty_area(mas, min: 0, max: section_size - 1, size: prepend + size)) { |
| 395 | if (IS_ALIGNED(mas->index + prepend, align)) |
| 396 | return true; |
| 397 | |
| 398 | /* Try finding larger area to align later */ |
| 399 | mas_reset(mas); |
| 400 | if (!mas_empty_area(mas, min: 0, max: section_size - 1, |
| 401 | size: size + prepend + align - 1)) |
| 402 | return true; |
| 403 | } |
| 404 | |
| 405 | /* No free area, try cleanup stale data and repeat the search once */ |
| 406 | if (!cleanup_done) { |
| 407 | clean_unused_module_areas_locked(); |
| 408 | cleanup_done = true; |
| 409 | mas_reset(mas); |
| 410 | goto repeat; |
| 411 | } |
| 412 | |
| 413 | return false; |
| 414 | } |
| 415 | |
| 416 | static int vm_module_tags_populate(void) |
| 417 | { |
| 418 | unsigned long phys_end = ALIGN_DOWN(module_tags.start_addr, PAGE_SIZE) + |
| 419 | (vm_module_tags->nr_pages << PAGE_SHIFT); |
| 420 | unsigned long new_end = module_tags.start_addr + module_tags.size; |
| 421 | |
| 422 | if (phys_end < new_end) { |
| 423 | struct page **next_page = vm_module_tags->pages + vm_module_tags->nr_pages; |
| 424 | unsigned long old_shadow_end = ALIGN(phys_end, MODULE_ALIGN); |
| 425 | unsigned long new_shadow_end = ALIGN(new_end, MODULE_ALIGN); |
| 426 | unsigned long more_pages; |
| 427 | unsigned long nr = 0; |
| 428 | |
| 429 | more_pages = ALIGN(new_end - phys_end, PAGE_SIZE) >> PAGE_SHIFT; |
| 430 | while (nr < more_pages) { |
| 431 | unsigned long allocated; |
| 432 | |
| 433 | allocated = alloc_pages_bulk_node(GFP_KERNEL | __GFP_NOWARN, |
| 434 | NUMA_NO_NODE, more_pages - nr, next_page + nr); |
| 435 | |
| 436 | if (!allocated) |
| 437 | break; |
| 438 | nr += allocated; |
| 439 | } |
| 440 | |
| 441 | if (nr < more_pages || |
| 442 | vmap_pages_range(addr: phys_end, end: phys_end + (nr << PAGE_SHIFT), PAGE_KERNEL, |
| 443 | pages: next_page, PAGE_SHIFT) < 0) { |
| 444 | release_pages_arg arg = { .pages = next_page }; |
| 445 | |
| 446 | /* Clean up and error out */ |
| 447 | release_pages(arg, nr); |
| 448 | return -ENOMEM; |
| 449 | } |
| 450 | |
| 451 | vm_module_tags->nr_pages += nr; |
| 452 | |
| 453 | /* |
| 454 | * Kasan allocates 1 byte of shadow for every 8 bytes of data. |
| 455 | * When kasan_alloc_module_shadow allocates shadow memory, |
| 456 | * its unit of allocation is a page. |
| 457 | * Therefore, here we need to align to MODULE_ALIGN. |
| 458 | */ |
| 459 | if (old_shadow_end < new_shadow_end) |
| 460 | kasan_alloc_module_shadow(addr: (void *)old_shadow_end, |
| 461 | size: new_shadow_end - old_shadow_end, |
| 462 | GFP_KERNEL); |
| 463 | } |
| 464 | |
| 465 | /* |
| 466 | * Mark the pages as accessible, now that they are mapped. |
| 467 | * With hardware tag-based KASAN, marking is skipped for |
| 468 | * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc(). |
| 469 | */ |
| 470 | kasan_unpoison_vmalloc(start: (void *)module_tags.start_addr, |
| 471 | size: new_end - module_tags.start_addr, |
| 472 | KASAN_VMALLOC_PROT_NORMAL); |
| 473 | |
| 474 | return 0; |
| 475 | } |
| 476 | |
| 477 | static void *reserve_module_tags(struct module *mod, unsigned long size, |
| 478 | unsigned int prepend, unsigned long align) |
| 479 | { |
| 480 | unsigned long section_size = module_tags.end_addr - module_tags.start_addr; |
| 481 | MA_STATE(mas, &mod_area_mt, 0, section_size - 1); |
| 482 | unsigned long offset; |
| 483 | void *ret = NULL; |
| 484 | |
| 485 | /* If no tags return error */ |
| 486 | if (size < sizeof(struct alloc_tag)) |
| 487 | return ERR_PTR(error: -EINVAL); |
| 488 | |
| 489 | /* |
| 490 | * align is always power of 2, so we can use IS_ALIGNED and ALIGN. |
| 491 | * align 0 or 1 means no alignment, to simplify set to 1. |
| 492 | */ |
| 493 | if (!align) |
| 494 | align = 1; |
| 495 | |
| 496 | if (!ensure_alignment(align, prepend: &prepend)) { |
| 497 | shutdown_mem_profiling(remove_file: true); |
| 498 | pr_err("%s: alignment %lu is incompatible with allocation tag indexing. Memory allocation profiling is disabled!\n" , |
| 499 | mod->name, align); |
| 500 | return ERR_PTR(error: -EINVAL); |
| 501 | } |
| 502 | |
| 503 | mas_lock(&mas); |
| 504 | if (!find_aligned_area(mas: &mas, section_size, size, prepend, align)) { |
| 505 | ret = ERR_PTR(error: -ENOMEM); |
| 506 | goto unlock; |
| 507 | } |
| 508 | |
| 509 | /* Mark found area as reserved */ |
| 510 | offset = mas.index; |
| 511 | offset += prepend; |
| 512 | offset = ALIGN(offset, align); |
| 513 | if (offset != mas.index) { |
| 514 | unsigned long pad_start = mas.index; |
| 515 | |
| 516 | mas.last = offset - 1; |
| 517 | mas_store(mas: &mas, entry: &prepend_mod); |
| 518 | if (mas_is_err(mas: &mas)) { |
| 519 | ret = ERR_PTR(error: xa_err(entry: mas.node)); |
| 520 | goto unlock; |
| 521 | } |
| 522 | mas.index = offset; |
| 523 | mas.last = offset + size - 1; |
| 524 | mas_store(mas: &mas, entry: mod); |
| 525 | if (mas_is_err(mas: &mas)) { |
| 526 | mas.index = pad_start; |
| 527 | mas_erase(mas: &mas); |
| 528 | ret = ERR_PTR(error: xa_err(entry: mas.node)); |
| 529 | } |
| 530 | } else { |
| 531 | mas.last = offset + size - 1; |
| 532 | mas_store(mas: &mas, entry: mod); |
| 533 | if (mas_is_err(mas: &mas)) |
| 534 | ret = ERR_PTR(error: xa_err(entry: mas.node)); |
| 535 | } |
| 536 | unlock: |
| 537 | mas_unlock(&mas); |
| 538 | |
| 539 | if (IS_ERR(ptr: ret)) |
| 540 | return ret; |
| 541 | |
| 542 | if (module_tags.size < offset + size) { |
| 543 | int grow_res; |
| 544 | |
| 545 | module_tags.size = offset + size; |
| 546 | if (mem_alloc_profiling_enabled() && !tags_addressable()) { |
| 547 | shutdown_mem_profiling(remove_file: true); |
| 548 | pr_warn("With module %s there are too many tags to fit in %d page flag bits. Memory allocation profiling is disabled!\n" , |
| 549 | mod->name, NR_UNUSED_PAGEFLAG_BITS); |
| 550 | } |
| 551 | |
| 552 | grow_res = vm_module_tags_populate(); |
| 553 | if (grow_res) { |
| 554 | shutdown_mem_profiling(remove_file: true); |
| 555 | pr_err("Failed to allocate memory for allocation tags in the module %s. Memory allocation profiling is disabled!\n" , |
| 556 | mod->name); |
| 557 | return ERR_PTR(error: grow_res); |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | return (struct alloc_tag *)(module_tags.start_addr + offset); |
| 562 | } |
| 563 | |
| 564 | static void release_module_tags(struct module *mod, bool used) |
| 565 | { |
| 566 | MA_STATE(mas, &mod_area_mt, module_tags.size, module_tags.size); |
| 567 | struct alloc_tag *start_tag; |
| 568 | struct alloc_tag *end_tag; |
| 569 | struct module *val; |
| 570 | |
| 571 | mas_lock(&mas); |
| 572 | mas_for_each_rev(&mas, val, 0) |
| 573 | if (val == mod) |
| 574 | break; |
| 575 | |
| 576 | if (!val) /* module not found */ |
| 577 | goto out; |
| 578 | |
| 579 | if (!used) |
| 580 | goto release_area; |
| 581 | |
| 582 | start_tag = (struct alloc_tag *)(module_tags.start_addr + mas.index); |
| 583 | end_tag = (struct alloc_tag *)(module_tags.start_addr + mas.last); |
| 584 | if (!clean_unused_counters(start_tag, end_tag)) { |
| 585 | struct alloc_tag *tag; |
| 586 | |
| 587 | for (tag = start_tag; tag <= end_tag; tag++) { |
| 588 | struct alloc_tag_counters counter; |
| 589 | |
| 590 | if (!tag->counters) |
| 591 | continue; |
| 592 | |
| 593 | counter = alloc_tag_read(tag); |
| 594 | pr_info("%s:%u module %s func:%s has %llu allocated at module unload\n" , |
| 595 | tag->ct.filename, tag->ct.lineno, tag->ct.modname, |
| 596 | tag->ct.function, counter.bytes); |
| 597 | } |
| 598 | } else { |
| 599 | used = false; |
| 600 | } |
| 601 | release_area: |
| 602 | mas_store(mas: &mas, entry: used ? &unloaded_mod : NULL); |
| 603 | val = mas_prev_range(mas: &mas, max: 0); |
| 604 | if (val == &prepend_mod) |
| 605 | mas_store(mas: &mas, NULL); |
| 606 | out: |
| 607 | mas_unlock(&mas); |
| 608 | } |
| 609 | |
| 610 | static int load_module(struct module *mod, struct codetag *start, struct codetag *stop) |
| 611 | { |
| 612 | /* Allocate module alloc_tag percpu counters */ |
| 613 | struct alloc_tag *start_tag; |
| 614 | struct alloc_tag *stop_tag; |
| 615 | struct alloc_tag *tag; |
| 616 | |
| 617 | /* percpu counters for core allocations are already statically allocated */ |
| 618 | if (!mod) |
| 619 | return 0; |
| 620 | |
| 621 | start_tag = ct_to_alloc_tag(start); |
| 622 | stop_tag = ct_to_alloc_tag(stop); |
| 623 | for (tag = start_tag; tag < stop_tag; tag++) { |
| 624 | WARN_ON(tag->counters); |
| 625 | tag->counters = alloc_percpu(struct alloc_tag_counters); |
| 626 | if (!tag->counters) { |
| 627 | while (--tag >= start_tag) { |
| 628 | free_percpu(pdata: tag->counters); |
| 629 | tag->counters = NULL; |
| 630 | } |
| 631 | pr_err("Failed to allocate memory for allocation tag percpu counters in the module %s\n" , |
| 632 | mod->name); |
| 633 | return -ENOMEM; |
| 634 | } |
| 635 | |
| 636 | /* |
| 637 | * Avoid a kmemleak false positive. The pointer to the counters is stored |
| 638 | * in the alloc_tag section of the module and cannot be directly accessed. |
| 639 | */ |
| 640 | kmemleak_ignore_percpu(ptr: tag->counters); |
| 641 | } |
| 642 | return 0; |
| 643 | } |
| 644 | |
| 645 | static void replace_module(struct module *mod, struct module *new_mod) |
| 646 | { |
| 647 | MA_STATE(mas, &mod_area_mt, 0, module_tags.size); |
| 648 | struct module *val; |
| 649 | |
| 650 | mas_lock(&mas); |
| 651 | mas_for_each(&mas, val, module_tags.size) { |
| 652 | if (val != mod) |
| 653 | continue; |
| 654 | |
| 655 | mas_store_gfp(mas: &mas, entry: new_mod, GFP_KERNEL); |
| 656 | break; |
| 657 | } |
| 658 | mas_unlock(&mas); |
| 659 | } |
| 660 | |
| 661 | static int __init alloc_mod_tags_mem(void) |
| 662 | { |
| 663 | /* Map space to copy allocation tags */ |
| 664 | vm_module_tags = execmem_vmap(MODULE_ALLOC_TAG_VMAP_SIZE); |
| 665 | if (!vm_module_tags) { |
| 666 | pr_err("Failed to map %lu bytes for module allocation tags\n" , |
| 667 | MODULE_ALLOC_TAG_VMAP_SIZE); |
| 668 | module_tags.start_addr = 0; |
| 669 | return -ENOMEM; |
| 670 | } |
| 671 | |
| 672 | vm_module_tags->pages = kmalloc_array(get_vm_area_size(vm_module_tags) >> PAGE_SHIFT, |
| 673 | sizeof(struct page *), GFP_KERNEL | __GFP_ZERO); |
| 674 | if (!vm_module_tags->pages) { |
| 675 | free_vm_area(area: vm_module_tags); |
| 676 | return -ENOMEM; |
| 677 | } |
| 678 | |
| 679 | module_tags.start_addr = (unsigned long)vm_module_tags->addr; |
| 680 | module_tags.end_addr = module_tags.start_addr + MODULE_ALLOC_TAG_VMAP_SIZE; |
| 681 | /* Ensure the base is alloc_tag aligned when required for indexing */ |
| 682 | module_tags.start_addr = alloc_tag_align(val: module_tags.start_addr); |
| 683 | |
| 684 | return 0; |
| 685 | } |
| 686 | |
| 687 | static void __init free_mod_tags_mem(void) |
| 688 | { |
| 689 | release_pages_arg arg = { .pages = vm_module_tags->pages }; |
| 690 | |
| 691 | module_tags.start_addr = 0; |
| 692 | release_pages(arg, nr: vm_module_tags->nr_pages); |
| 693 | kfree(objp: vm_module_tags->pages); |
| 694 | free_vm_area(area: vm_module_tags); |
| 695 | } |
| 696 | |
| 697 | #else /* CONFIG_MODULES */ |
| 698 | |
| 699 | static inline int alloc_mod_tags_mem(void) { return 0; } |
| 700 | static inline void free_mod_tags_mem(void) {} |
| 701 | |
| 702 | #endif /* CONFIG_MODULES */ |
| 703 | |
| 704 | /* See: Documentation/mm/allocation-profiling.rst */ |
| 705 | static int __init setup_early_mem_profiling(char *str) |
| 706 | { |
| 707 | bool compressed = false; |
| 708 | bool enable; |
| 709 | |
| 710 | if (!str || !str[0]) |
| 711 | return -EINVAL; |
| 712 | |
| 713 | if (!strncmp(str, "never" , 5)) { |
| 714 | enable = false; |
| 715 | mem_profiling_support = false; |
| 716 | pr_info("Memory allocation profiling is disabled!\n" ); |
| 717 | } else { |
| 718 | char *token = strsep(&str, "," ); |
| 719 | |
| 720 | if (kstrtobool(s: token, res: &enable)) |
| 721 | return -EINVAL; |
| 722 | |
| 723 | if (str) { |
| 724 | |
| 725 | if (strcmp(str, "compressed" )) |
| 726 | return -EINVAL; |
| 727 | |
| 728 | compressed = true; |
| 729 | } |
| 730 | mem_profiling_support = true; |
| 731 | pr_info("Memory allocation profiling is enabled %s compression and is turned %s!\n" , |
| 732 | compressed ? "with" : "without" , str_on_off(enable)); |
| 733 | } |
| 734 | |
| 735 | if (enable != mem_alloc_profiling_enabled()) { |
| 736 | if (enable) |
| 737 | static_branch_enable(&mem_alloc_profiling_key); |
| 738 | else |
| 739 | static_branch_disable(&mem_alloc_profiling_key); |
| 740 | } |
| 741 | if (compressed != static_key_enabled(&mem_profiling_compressed)) { |
| 742 | if (compressed) |
| 743 | static_branch_enable(&mem_profiling_compressed); |
| 744 | else |
| 745 | static_branch_disable(&mem_profiling_compressed); |
| 746 | } |
| 747 | |
| 748 | return 0; |
| 749 | } |
| 750 | early_param("sysctl.vm.mem_profiling" , setup_early_mem_profiling); |
| 751 | |
| 752 | static __init bool need_page_alloc_tagging(void) |
| 753 | { |
| 754 | if (static_key_enabled(&mem_profiling_compressed)) |
| 755 | return false; |
| 756 | |
| 757 | return mem_profiling_support; |
| 758 | } |
| 759 | |
| 760 | static __init void init_page_alloc_tagging(void) |
| 761 | { |
| 762 | } |
| 763 | |
| 764 | struct page_ext_operations page_alloc_tagging_ops = { |
| 765 | .size = sizeof(union codetag_ref), |
| 766 | .need = need_page_alloc_tagging, |
| 767 | .init = init_page_alloc_tagging, |
| 768 | }; |
| 769 | EXPORT_SYMBOL(page_alloc_tagging_ops); |
| 770 | |
| 771 | #ifdef CONFIG_SYSCTL |
| 772 | /* |
| 773 | * Not using proc_do_static_key() directly to prevent enabling profiling |
| 774 | * after it was shut down. |
| 775 | */ |
| 776 | static int proc_mem_profiling_handler(const struct ctl_table *table, int write, |
| 777 | void *buffer, size_t *lenp, loff_t *ppos) |
| 778 | { |
| 779 | if (!mem_profiling_support && write) |
| 780 | return -EINVAL; |
| 781 | |
| 782 | return proc_do_static_key(table, write, buffer, lenp, ppos); |
| 783 | } |
| 784 | |
| 785 | |
| 786 | static struct ctl_table memory_allocation_profiling_sysctls[] = { |
| 787 | { |
| 788 | .procname = "mem_profiling" , |
| 789 | .data = &mem_alloc_profiling_key, |
| 790 | #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG |
| 791 | .mode = 0444, |
| 792 | #else |
| 793 | .mode = 0644, |
| 794 | #endif |
| 795 | .proc_handler = proc_mem_profiling_handler, |
| 796 | }, |
| 797 | }; |
| 798 | |
| 799 | static void __init sysctl_init(void) |
| 800 | { |
| 801 | if (!mem_profiling_support) |
| 802 | memory_allocation_profiling_sysctls[0].mode = 0444; |
| 803 | |
| 804 | register_sysctl_init("vm" , memory_allocation_profiling_sysctls); |
| 805 | } |
| 806 | #else /* CONFIG_SYSCTL */ |
| 807 | static inline void sysctl_init(void) {} |
| 808 | #endif /* CONFIG_SYSCTL */ |
| 809 | |
| 810 | static int __init alloc_tag_init(void) |
| 811 | { |
| 812 | const struct codetag_type_desc desc = { |
| 813 | .section = ALLOC_TAG_SECTION_NAME, |
| 814 | .tag_size = sizeof(struct alloc_tag), |
| 815 | #ifdef CONFIG_MODULES |
| 816 | .needs_section_mem = needs_section_mem, |
| 817 | .alloc_section_mem = reserve_module_tags, |
| 818 | .free_section_mem = release_module_tags, |
| 819 | .module_load = load_module, |
| 820 | .module_replaced = replace_module, |
| 821 | #endif |
| 822 | }; |
| 823 | int res; |
| 824 | |
| 825 | sysctl_init(); |
| 826 | |
| 827 | if (!mem_profiling_support) { |
| 828 | pr_info("Memory allocation profiling is not supported!\n" ); |
| 829 | return 0; |
| 830 | } |
| 831 | |
| 832 | if (!proc_create_seq_private(ALLOCINFO_FILE_NAME, mode: 0400, NULL, ops: &allocinfo_seq_op, |
| 833 | state_size: sizeof(struct allocinfo_private), NULL)) { |
| 834 | pr_err("Failed to create %s file\n" , ALLOCINFO_FILE_NAME); |
| 835 | shutdown_mem_profiling(remove_file: false); |
| 836 | return -ENOMEM; |
| 837 | } |
| 838 | |
| 839 | res = alloc_mod_tags_mem(); |
| 840 | if (res) { |
| 841 | pr_err("Failed to reserve address space for module tags, errno = %d\n" , res); |
| 842 | shutdown_mem_profiling(remove_file: true); |
| 843 | return res; |
| 844 | } |
| 845 | |
| 846 | alloc_tag_cttype = codetag_register_type(desc: &desc); |
| 847 | if (IS_ERR(ptr: alloc_tag_cttype)) { |
| 848 | pr_err("Allocation tags registration failed, errno = %pe\n" , alloc_tag_cttype); |
| 849 | free_mod_tags_mem(); |
| 850 | shutdown_mem_profiling(remove_file: true); |
| 851 | return PTR_ERR(ptr: alloc_tag_cttype); |
| 852 | } |
| 853 | |
| 854 | return 0; |
| 855 | } |
| 856 | module_init(alloc_tag_init); |
| 857 | |