xref: /linux/lib/alloc_tag.c (revision 0471440c8061e9a7c0fd292c7c6598d6304efd53)
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 print_header;
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(alloc_tag_cttype, true);
56 	if (node == 0) {
57 		priv->print_header = true;
58 		priv->iter = codetag_get_ct_iter(alloc_tag_cttype);
59 		codetag_next_ct(&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(&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(alloc_tag_cttype, false);
79 }
80 
81 static void print_allocinfo_header(struct seq_buf *buf)
82 {
83 	/* Output format version, so we can change it. */
84 	seq_buf_printf(buf, "allocinfo - version: 1.0\n");
85 	seq_buf_printf(buf, "#     <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(out, "%12lli %8llu ", bytes, counter.calls);
95 	codetag_to_text(out, ct);
96 	seq_buf_putc(out, ' ');
97 	seq_buf_putc(out, '\n');
98 }
99 
100 static int allocinfo_show(struct seq_file *m, void *arg)
101 {
102 	struct allocinfo_private *priv = (struct allocinfo_private *)arg;
103 	char *bufp;
104 	size_t n = seq_get_buf(m, &bufp);
105 	struct seq_buf buf;
106 
107 	seq_buf_init(&buf, bufp, n);
108 	if (priv->print_header) {
109 		print_allocinfo_header(&buf);
110 		priv->print_header = false;
111 	}
112 	alloc_tag_to_text(&buf, priv->iter.ct);
113 	seq_commit(m, seq_buf_used(&buf));
114 	return 0;
115 }
116 
117 static const struct seq_operations allocinfo_seq_op = {
118 	.start	= allocinfo_start,
119 	.next	= allocinfo_next,
120 	.stop	= allocinfo_stop,
121 	.show	= allocinfo_show,
122 };
123 
124 size_t alloc_tag_top_users(struct codetag_bytes *tags, size_t count, bool can_sleep)
125 {
126 	struct codetag_iterator iter;
127 	struct codetag *ct;
128 	struct codetag_bytes n;
129 	unsigned int i, nr = 0;
130 
131 	if (IS_ERR_OR_NULL(alloc_tag_cttype))
132 		return 0;
133 
134 	if (can_sleep)
135 		codetag_lock_module_list(alloc_tag_cttype, true);
136 	else if (!codetag_trylock_module_list(alloc_tag_cttype))
137 		return 0;
138 
139 	iter = codetag_get_ct_iter(alloc_tag_cttype);
140 	while ((ct = codetag_next_ct(&iter))) {
141 		struct alloc_tag_counters counter = alloc_tag_read(ct_to_alloc_tag(ct));
142 
143 		n.ct	= ct;
144 		n.bytes = counter.bytes;
145 
146 		for (i = 0; i < nr; i++)
147 			if (n.bytes > tags[i].bytes)
148 				break;
149 
150 		if (i < count) {
151 			nr -= nr == count;
152 			memmove(&tags[i + 1],
153 				&tags[i],
154 				sizeof(tags[0]) * (nr - i));
155 			nr++;
156 			tags[i] = n;
157 		}
158 	}
159 
160 	codetag_lock_module_list(alloc_tag_cttype, false);
161 
162 	return nr;
163 }
164 
165 void pgalloc_tag_split(struct folio *folio, int old_order, int new_order)
166 {
167 	int i;
168 	struct alloc_tag *tag;
169 	unsigned int nr_pages = 1 << new_order;
170 
171 	if (!mem_alloc_profiling_enabled())
172 		return;
173 
174 	tag = __pgalloc_tag_get(&folio->page);
175 	if (!tag)
176 		return;
177 
178 	for (i = nr_pages; i < (1 << old_order); i += nr_pages) {
179 		union pgtag_ref_handle handle;
180 		union codetag_ref ref;
181 
182 		if (get_page_tag_ref(folio_page(folio, i), &ref, &handle)) {
183 			/* Set new reference to point to the original tag */
184 			alloc_tag_ref_set(&ref, tag);
185 			update_page_tag_ref(handle, &ref);
186 			put_page_tag_ref(handle);
187 		}
188 	}
189 }
190 
191 void pgalloc_tag_swap(struct folio *new, struct folio *old)
192 {
193 	union pgtag_ref_handle handle_old, handle_new;
194 	union codetag_ref ref_old, ref_new;
195 	struct alloc_tag *tag_old, *tag_new;
196 
197 	if (!mem_alloc_profiling_enabled())
198 		return;
199 
200 	tag_old = __pgalloc_tag_get(&old->page);
201 	if (!tag_old)
202 		return;
203 	tag_new = __pgalloc_tag_get(&new->page);
204 	if (!tag_new)
205 		return;
206 
207 	if (!get_page_tag_ref(&old->page, &ref_old, &handle_old))
208 		return;
209 	if (!get_page_tag_ref(&new->page, &ref_new, &handle_new)) {
210 		put_page_tag_ref(handle_old);
211 		return;
212 	}
213 
214 	/*
215 	 * Clear tag references to avoid debug warning when using
216 	 * __alloc_tag_ref_set() with non-empty reference.
217 	 */
218 	set_codetag_empty(&ref_old);
219 	set_codetag_empty(&ref_new);
220 
221 	/* swap tags */
222 	__alloc_tag_ref_set(&ref_old, tag_new);
223 	update_page_tag_ref(handle_old, &ref_old);
224 	__alloc_tag_ref_set(&ref_new, tag_old);
225 	update_page_tag_ref(handle_new, &ref_new);
226 
227 	put_page_tag_ref(handle_old);
228 	put_page_tag_ref(handle_new);
229 }
230 
231 static void shutdown_mem_profiling(bool remove_file)
232 {
233 	if (mem_alloc_profiling_enabled())
234 		static_branch_disable(&mem_alloc_profiling_key);
235 
236 	if (!mem_profiling_support)
237 		return;
238 
239 	if (remove_file)
240 		remove_proc_entry(ALLOCINFO_FILE_NAME, NULL);
241 	mem_profiling_support = false;
242 }
243 
244 void __init alloc_tag_sec_init(void)
245 {
246 	struct alloc_tag *last_codetag;
247 
248 	if (!mem_profiling_support)
249 		return;
250 
251 	if (!static_key_enabled(&mem_profiling_compressed))
252 		return;
253 
254 	kernel_tags.first_tag = (struct alloc_tag *)kallsyms_lookup_name(
255 					SECTION_START(ALLOC_TAG_SECTION_NAME));
256 	last_codetag = (struct alloc_tag *)kallsyms_lookup_name(
257 					SECTION_STOP(ALLOC_TAG_SECTION_NAME));
258 	kernel_tags.count = last_codetag - kernel_tags.first_tag;
259 
260 	/* Check if kernel tags fit into page flags */
261 	if (kernel_tags.count > (1UL << NR_UNUSED_PAGEFLAG_BITS)) {
262 		shutdown_mem_profiling(false); /* allocinfo file does not exist yet */
263 		pr_err("%lu allocation tags cannot be references using %d available page flag bits. Memory allocation profiling is disabled!\n",
264 			kernel_tags.count, NR_UNUSED_PAGEFLAG_BITS);
265 		return;
266 	}
267 
268 	alloc_tag_ref_offs = (LRU_REFS_PGOFF - NR_UNUSED_PAGEFLAG_BITS);
269 	alloc_tag_ref_mask = ((1UL << NR_UNUSED_PAGEFLAG_BITS) - 1);
270 	pr_debug("Memory allocation profiling compression is using %d page flag bits!\n",
271 		 NR_UNUSED_PAGEFLAG_BITS);
272 }
273 
274 #ifdef CONFIG_MODULES
275 
276 static struct maple_tree mod_area_mt = MTREE_INIT(mod_area_mt, MT_FLAGS_ALLOC_RANGE);
277 static struct vm_struct *vm_module_tags;
278 /* A dummy object used to indicate an unloaded module */
279 static struct module unloaded_mod;
280 /* A dummy object used to indicate a module prepended area */
281 static struct module prepend_mod;
282 
283 struct alloc_tag_module_section module_tags;
284 
285 static inline unsigned long alloc_tag_align(unsigned long val)
286 {
287 	if (!static_key_enabled(&mem_profiling_compressed)) {
288 		/* No alignment requirements when we are not indexing the tags */
289 		return val;
290 	}
291 
292 	if (val % sizeof(struct alloc_tag) == 0)
293 		return val;
294 	return ((val / sizeof(struct alloc_tag)) + 1) * sizeof(struct alloc_tag);
295 }
296 
297 static bool ensure_alignment(unsigned long align, unsigned int *prepend)
298 {
299 	if (!static_key_enabled(&mem_profiling_compressed)) {
300 		/* No alignment requirements when we are not indexing the tags */
301 		return true;
302 	}
303 
304 	/*
305 	 * If alloc_tag size is not a multiple of required alignment, tag
306 	 * indexing does not work.
307 	 */
308 	if (!IS_ALIGNED(sizeof(struct alloc_tag), align))
309 		return false;
310 
311 	/* Ensure prepend consumes multiple of alloc_tag-sized blocks */
312 	if (*prepend)
313 		*prepend = alloc_tag_align(*prepend);
314 
315 	return true;
316 }
317 
318 static inline bool tags_addressable(void)
319 {
320 	unsigned long tag_idx_count;
321 
322 	if (!static_key_enabled(&mem_profiling_compressed))
323 		return true; /* with page_ext tags are always addressable */
324 
325 	tag_idx_count = CODETAG_ID_FIRST + kernel_tags.count +
326 			module_tags.size / sizeof(struct alloc_tag);
327 
328 	return tag_idx_count < (1UL << NR_UNUSED_PAGEFLAG_BITS);
329 }
330 
331 static bool needs_section_mem(struct module *mod, unsigned long size)
332 {
333 	if (!mem_profiling_support)
334 		return false;
335 
336 	return size >= sizeof(struct alloc_tag);
337 }
338 
339 static bool clean_unused_counters(struct alloc_tag *start_tag,
340 				  struct alloc_tag *end_tag)
341 {
342 	struct alloc_tag *tag;
343 	bool ret = true;
344 
345 	for (tag = start_tag; tag <= end_tag; tag++) {
346 		struct alloc_tag_counters counter;
347 
348 		if (!tag->counters)
349 			continue;
350 
351 		counter = alloc_tag_read(tag);
352 		if (!counter.bytes) {
353 			free_percpu(tag->counters);
354 			tag->counters = NULL;
355 		} else {
356 			ret = false;
357 		}
358 	}
359 
360 	return ret;
361 }
362 
363 /* Called with mod_area_mt locked */
364 static void clean_unused_module_areas_locked(void)
365 {
366 	MA_STATE(mas, &mod_area_mt, 0, module_tags.size);
367 	struct module *val;
368 
369 	mas_for_each(&mas, val, module_tags.size) {
370 		struct alloc_tag *start_tag;
371 		struct alloc_tag *end_tag;
372 
373 		if (val != &unloaded_mod)
374 			continue;
375 
376 		/* Release area if all tags are unused */
377 		start_tag = (struct alloc_tag *)(module_tags.start_addr + mas.index);
378 		end_tag = (struct alloc_tag *)(module_tags.start_addr + mas.last);
379 		if (clean_unused_counters(start_tag, end_tag))
380 			mas_erase(&mas);
381 	}
382 }
383 
384 /* Called with mod_area_mt locked */
385 static bool find_aligned_area(struct ma_state *mas, unsigned long section_size,
386 			      unsigned long size, unsigned int prepend, unsigned long align)
387 {
388 	bool cleanup_done = false;
389 
390 repeat:
391 	/* Try finding exact size and hope the start is aligned */
392 	if (!mas_empty_area(mas, 0, section_size - 1, prepend + size)) {
393 		if (IS_ALIGNED(mas->index + prepend, align))
394 			return true;
395 
396 		/* Try finding larger area to align later */
397 		mas_reset(mas);
398 		if (!mas_empty_area(mas, 0, section_size - 1,
399 				    size + prepend + align - 1))
400 			return true;
401 	}
402 
403 	/* No free area, try cleanup stale data and repeat the search once */
404 	if (!cleanup_done) {
405 		clean_unused_module_areas_locked();
406 		cleanup_done = true;
407 		mas_reset(mas);
408 		goto repeat;
409 	}
410 
411 	return false;
412 }
413 
414 static int vm_module_tags_populate(void)
415 {
416 	unsigned long phys_end = ALIGN_DOWN(module_tags.start_addr, PAGE_SIZE) +
417 				 (vm_module_tags->nr_pages << PAGE_SHIFT);
418 	unsigned long new_end = module_tags.start_addr + module_tags.size;
419 
420 	if (phys_end < new_end) {
421 		struct page **next_page = vm_module_tags->pages + vm_module_tags->nr_pages;
422 		unsigned long old_shadow_end = ALIGN(phys_end, MODULE_ALIGN);
423 		unsigned long new_shadow_end = ALIGN(new_end, MODULE_ALIGN);
424 		unsigned long more_pages;
425 		unsigned long nr = 0;
426 
427 		more_pages = ALIGN(new_end - phys_end, PAGE_SIZE) >> PAGE_SHIFT;
428 		while (nr < more_pages) {
429 			unsigned long allocated;
430 
431 			allocated = alloc_pages_bulk_node(GFP_KERNEL | __GFP_NOWARN,
432 				NUMA_NO_NODE, more_pages - nr, next_page + nr);
433 
434 			if (!allocated)
435 				break;
436 			nr += allocated;
437 		}
438 
439 		if (nr < more_pages ||
440 		    vmap_pages_range(phys_end, phys_end + (nr << PAGE_SHIFT), PAGE_KERNEL,
441 				     next_page, PAGE_SHIFT) < 0) {
442 			/* Clean up and error out */
443 			for (int i = 0; i < nr; i++)
444 				__free_page(next_page[i]);
445 			return -ENOMEM;
446 		}
447 
448 		vm_module_tags->nr_pages += nr;
449 
450 		/*
451 		 * Kasan allocates 1 byte of shadow for every 8 bytes of data.
452 		 * When kasan_alloc_module_shadow allocates shadow memory,
453 		 * its unit of allocation is a page.
454 		 * Therefore, here we need to align to MODULE_ALIGN.
455 		 */
456 		if (old_shadow_end < new_shadow_end)
457 			kasan_alloc_module_shadow((void *)old_shadow_end,
458 						  new_shadow_end - old_shadow_end,
459 						  GFP_KERNEL);
460 	}
461 
462 	/*
463 	 * Mark the pages as accessible, now that they are mapped.
464 	 * With hardware tag-based KASAN, marking is skipped for
465 	 * non-VM_ALLOC mappings, see __kasan_unpoison_vmalloc().
466 	 */
467 	kasan_unpoison_vmalloc((void *)module_tags.start_addr,
468 				new_end - module_tags.start_addr,
469 				KASAN_VMALLOC_PROT_NORMAL);
470 
471 	return 0;
472 }
473 
474 static void *reserve_module_tags(struct module *mod, unsigned long size,
475 				 unsigned int prepend, unsigned long align)
476 {
477 	unsigned long section_size = module_tags.end_addr - module_tags.start_addr;
478 	MA_STATE(mas, &mod_area_mt, 0, section_size - 1);
479 	unsigned long offset;
480 	void *ret = NULL;
481 
482 	/* If no tags return error */
483 	if (size < sizeof(struct alloc_tag))
484 		return ERR_PTR(-EINVAL);
485 
486 	/*
487 	 * align is always power of 2, so we can use IS_ALIGNED and ALIGN.
488 	 * align 0 or 1 means no alignment, to simplify set to 1.
489 	 */
490 	if (!align)
491 		align = 1;
492 
493 	if (!ensure_alignment(align, &prepend)) {
494 		shutdown_mem_profiling(true);
495 		pr_err("%s: alignment %lu is incompatible with allocation tag indexing. Memory allocation profiling is disabled!\n",
496 			mod->name, align);
497 		return ERR_PTR(-EINVAL);
498 	}
499 
500 	mas_lock(&mas);
501 	if (!find_aligned_area(&mas, section_size, size, prepend, align)) {
502 		ret = ERR_PTR(-ENOMEM);
503 		goto unlock;
504 	}
505 
506 	/* Mark found area as reserved */
507 	offset = mas.index;
508 	offset += prepend;
509 	offset = ALIGN(offset, align);
510 	if (offset != mas.index) {
511 		unsigned long pad_start = mas.index;
512 
513 		mas.last = offset - 1;
514 		mas_store(&mas, &prepend_mod);
515 		if (mas_is_err(&mas)) {
516 			ret = ERR_PTR(xa_err(mas.node));
517 			goto unlock;
518 		}
519 		mas.index = offset;
520 		mas.last = offset + size - 1;
521 		mas_store(&mas, mod);
522 		if (mas_is_err(&mas)) {
523 			mas.index = pad_start;
524 			mas_erase(&mas);
525 			ret = ERR_PTR(xa_err(mas.node));
526 		}
527 	} else {
528 		mas.last = offset + size - 1;
529 		mas_store(&mas, mod);
530 		if (mas_is_err(&mas))
531 			ret = ERR_PTR(xa_err(mas.node));
532 	}
533 unlock:
534 	mas_unlock(&mas);
535 
536 	if (IS_ERR(ret))
537 		return ret;
538 
539 	if (module_tags.size < offset + size) {
540 		int grow_res;
541 
542 		module_tags.size = offset + size;
543 		if (mem_alloc_profiling_enabled() && !tags_addressable()) {
544 			shutdown_mem_profiling(true);
545 			pr_warn("With module %s there are too many tags to fit in %d page flag bits. Memory allocation profiling is disabled!\n",
546 				mod->name, NR_UNUSED_PAGEFLAG_BITS);
547 		}
548 
549 		grow_res = vm_module_tags_populate();
550 		if (grow_res) {
551 			shutdown_mem_profiling(true);
552 			pr_err("Failed to allocate memory for allocation tags in the module %s. Memory allocation profiling is disabled!\n",
553 			       mod->name);
554 			return ERR_PTR(grow_res);
555 		}
556 	}
557 
558 	return (struct alloc_tag *)(module_tags.start_addr + offset);
559 }
560 
561 static void release_module_tags(struct module *mod, bool used)
562 {
563 	MA_STATE(mas, &mod_area_mt, module_tags.size, module_tags.size);
564 	struct alloc_tag *start_tag;
565 	struct alloc_tag *end_tag;
566 	struct module *val;
567 
568 	mas_lock(&mas);
569 	mas_for_each_rev(&mas, val, 0)
570 		if (val == mod)
571 			break;
572 
573 	if (!val) /* module not found */
574 		goto out;
575 
576 	if (!used)
577 		goto release_area;
578 
579 	start_tag = (struct alloc_tag *)(module_tags.start_addr + mas.index);
580 	end_tag = (struct alloc_tag *)(module_tags.start_addr + mas.last);
581 	if (!clean_unused_counters(start_tag, end_tag)) {
582 		struct alloc_tag *tag;
583 
584 		for (tag = start_tag; tag <= end_tag; tag++) {
585 			struct alloc_tag_counters counter;
586 
587 			if (!tag->counters)
588 				continue;
589 
590 			counter = alloc_tag_read(tag);
591 			pr_info("%s:%u module %s func:%s has %llu allocated at module unload\n",
592 				tag->ct.filename, tag->ct.lineno, tag->ct.modname,
593 				tag->ct.function, counter.bytes);
594 		}
595 	} else {
596 		used = false;
597 	}
598 release_area:
599 	mas_store(&mas, used ? &unloaded_mod : NULL);
600 	val = mas_prev_range(&mas, 0);
601 	if (val == &prepend_mod)
602 		mas_store(&mas, NULL);
603 out:
604 	mas_unlock(&mas);
605 }
606 
607 static int load_module(struct module *mod, struct codetag *start, struct codetag *stop)
608 {
609 	/* Allocate module alloc_tag percpu counters */
610 	struct alloc_tag *start_tag;
611 	struct alloc_tag *stop_tag;
612 	struct alloc_tag *tag;
613 
614 	/* percpu counters for core allocations are already statically allocated */
615 	if (!mod)
616 		return 0;
617 
618 	start_tag = ct_to_alloc_tag(start);
619 	stop_tag = ct_to_alloc_tag(stop);
620 	for (tag = start_tag; tag < stop_tag; tag++) {
621 		WARN_ON(tag->counters);
622 		tag->counters = alloc_percpu(struct alloc_tag_counters);
623 		if (!tag->counters) {
624 			while (--tag >= start_tag) {
625 				free_percpu(tag->counters);
626 				tag->counters = NULL;
627 			}
628 			pr_err("Failed to allocate memory for allocation tag percpu counters in the module %s\n",
629 			       mod->name);
630 			return -ENOMEM;
631 		}
632 
633 		/*
634 		 * Avoid a kmemleak false positive. The pointer to the counters is stored
635 		 * in the alloc_tag section of the module and cannot be directly accessed.
636 		 */
637 		kmemleak_ignore_percpu(tag->counters);
638 	}
639 	return 0;
640 }
641 
642 static void replace_module(struct module *mod, struct module *new_mod)
643 {
644 	MA_STATE(mas, &mod_area_mt, 0, module_tags.size);
645 	struct module *val;
646 
647 	mas_lock(&mas);
648 	mas_for_each(&mas, val, module_tags.size) {
649 		if (val != mod)
650 			continue;
651 
652 		mas_store_gfp(&mas, new_mod, GFP_KERNEL);
653 		break;
654 	}
655 	mas_unlock(&mas);
656 }
657 
658 static int __init alloc_mod_tags_mem(void)
659 {
660 	/* Map space to copy allocation tags */
661 	vm_module_tags = execmem_vmap(MODULE_ALLOC_TAG_VMAP_SIZE);
662 	if (!vm_module_tags) {
663 		pr_err("Failed to map %lu bytes for module allocation tags\n",
664 			MODULE_ALLOC_TAG_VMAP_SIZE);
665 		module_tags.start_addr = 0;
666 		return -ENOMEM;
667 	}
668 
669 	vm_module_tags->pages = kmalloc_array(get_vm_area_size(vm_module_tags) >> PAGE_SHIFT,
670 					sizeof(struct page *), GFP_KERNEL | __GFP_ZERO);
671 	if (!vm_module_tags->pages) {
672 		free_vm_area(vm_module_tags);
673 		return -ENOMEM;
674 	}
675 
676 	module_tags.start_addr = (unsigned long)vm_module_tags->addr;
677 	module_tags.end_addr = module_tags.start_addr + MODULE_ALLOC_TAG_VMAP_SIZE;
678 	/* Ensure the base is alloc_tag aligned when required for indexing */
679 	module_tags.start_addr = alloc_tag_align(module_tags.start_addr);
680 
681 	return 0;
682 }
683 
684 static void __init free_mod_tags_mem(void)
685 {
686 	int i;
687 
688 	module_tags.start_addr = 0;
689 	for (i = 0; i < vm_module_tags->nr_pages; i++)
690 		__free_page(vm_module_tags->pages[i]);
691 	kfree(vm_module_tags->pages);
692 	free_vm_area(vm_module_tags);
693 }
694 
695 #else /* CONFIG_MODULES */
696 
697 static inline int alloc_mod_tags_mem(void) { return 0; }
698 static inline void free_mod_tags_mem(void) {}
699 
700 #endif /* CONFIG_MODULES */
701 
702 /* See: Documentation/mm/allocation-profiling.rst */
703 static int __init setup_early_mem_profiling(char *str)
704 {
705 	bool compressed = false;
706 	bool enable;
707 
708 	if (!str || !str[0])
709 		return -EINVAL;
710 
711 	if (!strncmp(str, "never", 5)) {
712 		enable = false;
713 		mem_profiling_support = false;
714 		pr_info("Memory allocation profiling is disabled!\n");
715 	} else {
716 		char *token = strsep(&str, ",");
717 
718 		if (kstrtobool(token, &enable))
719 			return -EINVAL;
720 
721 		if (str) {
722 
723 			if (strcmp(str, "compressed"))
724 				return -EINVAL;
725 
726 			compressed = true;
727 		}
728 		mem_profiling_support = true;
729 		pr_info("Memory allocation profiling is enabled %s compression and is turned %s!\n",
730 			compressed ? "with" : "without", str_on_off(enable));
731 	}
732 
733 	if (enable != mem_alloc_profiling_enabled()) {
734 		if (enable)
735 			static_branch_enable(&mem_alloc_profiling_key);
736 		else
737 			static_branch_disable(&mem_alloc_profiling_key);
738 	}
739 	if (compressed != static_key_enabled(&mem_profiling_compressed)) {
740 		if (compressed)
741 			static_branch_enable(&mem_profiling_compressed);
742 		else
743 			static_branch_disable(&mem_profiling_compressed);
744 	}
745 
746 	return 0;
747 }
748 early_param("sysctl.vm.mem_profiling", setup_early_mem_profiling);
749 
750 static __init bool need_page_alloc_tagging(void)
751 {
752 	if (static_key_enabled(&mem_profiling_compressed))
753 		return false;
754 
755 	return mem_profiling_support;
756 }
757 
758 static __init void init_page_alloc_tagging(void)
759 {
760 }
761 
762 struct page_ext_operations page_alloc_tagging_ops = {
763 	.size = sizeof(union codetag_ref),
764 	.need = need_page_alloc_tagging,
765 	.init = init_page_alloc_tagging,
766 };
767 EXPORT_SYMBOL(page_alloc_tagging_ops);
768 
769 #ifdef CONFIG_SYSCTL
770 static struct ctl_table memory_allocation_profiling_sysctls[] = {
771 	{
772 		.procname	= "mem_profiling",
773 		.data		= &mem_alloc_profiling_key,
774 #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG
775 		.mode		= 0444,
776 #else
777 		.mode		= 0644,
778 #endif
779 		.proc_handler	= proc_do_static_key,
780 	},
781 };
782 
783 static void __init sysctl_init(void)
784 {
785 	if (!mem_profiling_support)
786 		memory_allocation_profiling_sysctls[0].mode = 0444;
787 
788 	register_sysctl_init("vm", memory_allocation_profiling_sysctls);
789 }
790 #else /* CONFIG_SYSCTL */
791 static inline void sysctl_init(void) {}
792 #endif /* CONFIG_SYSCTL */
793 
794 static int __init alloc_tag_init(void)
795 {
796 	const struct codetag_type_desc desc = {
797 		.section		= ALLOC_TAG_SECTION_NAME,
798 		.tag_size		= sizeof(struct alloc_tag),
799 #ifdef CONFIG_MODULES
800 		.needs_section_mem	= needs_section_mem,
801 		.alloc_section_mem	= reserve_module_tags,
802 		.free_section_mem	= release_module_tags,
803 		.module_load		= load_module,
804 		.module_replaced	= replace_module,
805 #endif
806 	};
807 	int res;
808 
809 	sysctl_init();
810 
811 	if (!mem_profiling_support) {
812 		pr_info("Memory allocation profiling is not supported!\n");
813 		return 0;
814 	}
815 
816 	if (!proc_create_seq_private(ALLOCINFO_FILE_NAME, 0400, NULL, &allocinfo_seq_op,
817 				     sizeof(struct allocinfo_private), NULL)) {
818 		pr_err("Failed to create %s file\n", ALLOCINFO_FILE_NAME);
819 		shutdown_mem_profiling(false);
820 		return -ENOMEM;
821 	}
822 
823 	res = alloc_mod_tags_mem();
824 	if (res) {
825 		pr_err("Failed to reserve address space for module tags, errno = %d\n", res);
826 		shutdown_mem_profiling(true);
827 		return res;
828 	}
829 
830 	alloc_tag_cttype = codetag_register_type(&desc);
831 	if (IS_ERR(alloc_tag_cttype)) {
832 		pr_err("Allocation tags registration failed, errno = %ld\n", PTR_ERR(alloc_tag_cttype));
833 		free_mod_tags_mem();
834 		shutdown_mem_profiling(true);
835 		return PTR_ERR(alloc_tag_cttype);
836 	}
837 
838 	return 0;
839 }
840 module_init(alloc_tag_init);
841