xref: /linux/lib/alloc_tag.c (revision d3c82f618a9c2b764b7651afe16594ffeb50ade9)
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/vmalloc.h>
13 
14 #define ALLOCINFO_FILE_NAME		"allocinfo"
15 #define MODULE_ALLOC_TAG_VMAP_SIZE	(100000UL * sizeof(struct alloc_tag))
16 #define SECTION_START(NAME)		(CODETAG_SECTION_START_PREFIX NAME)
17 #define SECTION_STOP(NAME)		(CODETAG_SECTION_STOP_PREFIX NAME)
18 
19 #ifdef CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT
20 static bool mem_profiling_support = true;
21 #else
22 static bool mem_profiling_support;
23 #endif
24 
25 static struct codetag_type *alloc_tag_cttype;
26 
27 DEFINE_PER_CPU(struct alloc_tag_counters, _shared_alloc_tag);
28 EXPORT_SYMBOL(_shared_alloc_tag);
29 
30 DEFINE_STATIC_KEY_MAYBE(CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT,
31 			mem_alloc_profiling_key);
32 EXPORT_SYMBOL(mem_alloc_profiling_key);
33 
34 DEFINE_STATIC_KEY_FALSE(mem_profiling_compressed);
35 
36 struct alloc_tag_kernel_section kernel_tags = { NULL, 0 };
37 unsigned long alloc_tag_ref_mask;
38 int alloc_tag_ref_offs;
39 
40 struct allocinfo_private {
41 	struct codetag_iterator iter;
42 	bool print_header;
43 };
44 
allocinfo_start(struct seq_file * m,loff_t * pos)45 static void *allocinfo_start(struct seq_file *m, loff_t *pos)
46 {
47 	struct allocinfo_private *priv;
48 	struct codetag *ct;
49 	loff_t node = *pos;
50 
51 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
52 	m->private = priv;
53 	if (!priv)
54 		return NULL;
55 
56 	priv->print_header = (node == 0);
57 	codetag_lock_module_list(alloc_tag_cttype, true);
58 	priv->iter = codetag_get_ct_iter(alloc_tag_cttype);
59 	while ((ct = codetag_next_ct(&priv->iter)) != NULL && node)
60 		node--;
61 
62 	return ct ? priv : NULL;
63 }
64 
allocinfo_next(struct seq_file * m,void * arg,loff_t * pos)65 static void *allocinfo_next(struct seq_file *m, void *arg, loff_t *pos)
66 {
67 	struct allocinfo_private *priv = (struct allocinfo_private *)arg;
68 	struct codetag *ct = codetag_next_ct(&priv->iter);
69 
70 	(*pos)++;
71 	if (!ct)
72 		return NULL;
73 
74 	return priv;
75 }
76 
allocinfo_stop(struct seq_file * m,void * arg)77 static void allocinfo_stop(struct seq_file *m, void *arg)
78 {
79 	struct allocinfo_private *priv = (struct allocinfo_private *)m->private;
80 
81 	if (priv) {
82 		codetag_lock_module_list(alloc_tag_cttype, false);
83 		kfree(priv);
84 	}
85 }
86 
print_allocinfo_header(struct seq_buf * buf)87 static void print_allocinfo_header(struct seq_buf *buf)
88 {
89 	/* Output format version, so we can change it. */
90 	seq_buf_printf(buf, "allocinfo - version: 1.0\n");
91 	seq_buf_printf(buf, "#     <size>  <calls> <tag info>\n");
92 }
93 
alloc_tag_to_text(struct seq_buf * out,struct codetag * ct)94 static void alloc_tag_to_text(struct seq_buf *out, struct codetag *ct)
95 {
96 	struct alloc_tag *tag = ct_to_alloc_tag(ct);
97 	struct alloc_tag_counters counter = alloc_tag_read(tag);
98 	s64 bytes = counter.bytes;
99 
100 	seq_buf_printf(out, "%12lli %8llu ", bytes, counter.calls);
101 	codetag_to_text(out, ct);
102 	seq_buf_putc(out, ' ');
103 	seq_buf_putc(out, '\n');
104 }
105 
allocinfo_show(struct seq_file * m,void * arg)106 static int allocinfo_show(struct seq_file *m, void *arg)
107 {
108 	struct allocinfo_private *priv = (struct allocinfo_private *)arg;
109 	char *bufp;
110 	size_t n = seq_get_buf(m, &bufp);
111 	struct seq_buf buf;
112 
113 	seq_buf_init(&buf, bufp, n);
114 	if (priv->print_header) {
115 		print_allocinfo_header(&buf);
116 		priv->print_header = false;
117 	}
118 	alloc_tag_to_text(&buf, priv->iter.ct);
119 	seq_commit(m, seq_buf_used(&buf));
120 	return 0;
121 }
122 
123 static const struct seq_operations allocinfo_seq_op = {
124 	.start	= allocinfo_start,
125 	.next	= allocinfo_next,
126 	.stop	= allocinfo_stop,
127 	.show	= allocinfo_show,
128 };
129 
alloc_tag_top_users(struct codetag_bytes * tags,size_t count,bool can_sleep)130 size_t alloc_tag_top_users(struct codetag_bytes *tags, size_t count, bool can_sleep)
131 {
132 	struct codetag_iterator iter;
133 	struct codetag *ct;
134 	struct codetag_bytes n;
135 	unsigned int i, nr = 0;
136 
137 	if (can_sleep)
138 		codetag_lock_module_list(alloc_tag_cttype, true);
139 	else if (!codetag_trylock_module_list(alloc_tag_cttype))
140 		return 0;
141 
142 	iter = codetag_get_ct_iter(alloc_tag_cttype);
143 	while ((ct = codetag_next_ct(&iter))) {
144 		struct alloc_tag_counters counter = alloc_tag_read(ct_to_alloc_tag(ct));
145 
146 		n.ct	= ct;
147 		n.bytes = counter.bytes;
148 
149 		for (i = 0; i < nr; i++)
150 			if (n.bytes > tags[i].bytes)
151 				break;
152 
153 		if (i < count) {
154 			nr -= nr == count;
155 			memmove(&tags[i + 1],
156 				&tags[i],
157 				sizeof(tags[0]) * (nr - i));
158 			nr++;
159 			tags[i] = n;
160 		}
161 	}
162 
163 	codetag_lock_module_list(alloc_tag_cttype, false);
164 
165 	return nr;
166 }
167 
pgalloc_tag_split(struct folio * folio,int old_order,int new_order)168 void pgalloc_tag_split(struct folio *folio, int old_order, int new_order)
169 {
170 	int i;
171 	struct alloc_tag *tag;
172 	unsigned int nr_pages = 1 << new_order;
173 
174 	if (!mem_alloc_profiling_enabled())
175 		return;
176 
177 	tag = __pgalloc_tag_get(&folio->page);
178 	if (!tag)
179 		return;
180 
181 	for (i = nr_pages; i < (1 << old_order); i += nr_pages) {
182 		union pgtag_ref_handle handle;
183 		union codetag_ref ref;
184 
185 		if (get_page_tag_ref(folio_page(folio, i), &ref, &handle)) {
186 			/* Set new reference to point to the original tag */
187 			alloc_tag_ref_set(&ref, tag);
188 			update_page_tag_ref(handle, &ref);
189 			put_page_tag_ref(handle);
190 		}
191 	}
192 }
193 
pgalloc_tag_swap(struct folio * new,struct folio * old)194 void pgalloc_tag_swap(struct folio *new, struct folio *old)
195 {
196 	union pgtag_ref_handle handle_old, handle_new;
197 	union codetag_ref ref_old, ref_new;
198 	struct alloc_tag *tag_old, *tag_new;
199 
200 	if (!mem_alloc_profiling_enabled())
201 		return;
202 
203 	tag_old = __pgalloc_tag_get(&old->page);
204 	if (!tag_old)
205 		return;
206 	tag_new = __pgalloc_tag_get(&new->page);
207 	if (!tag_new)
208 		return;
209 
210 	if (!get_page_tag_ref(&old->page, &ref_old, &handle_old))
211 		return;
212 	if (!get_page_tag_ref(&new->page, &ref_new, &handle_new)) {
213 		put_page_tag_ref(handle_old);
214 		return;
215 	}
216 
217 	/*
218 	 * Clear tag references to avoid debug warning when using
219 	 * __alloc_tag_ref_set() with non-empty reference.
220 	 */
221 	set_codetag_empty(&ref_old);
222 	set_codetag_empty(&ref_new);
223 
224 	/* swap tags */
225 	__alloc_tag_ref_set(&ref_old, tag_new);
226 	update_page_tag_ref(handle_old, &ref_old);
227 	__alloc_tag_ref_set(&ref_new, tag_old);
228 	update_page_tag_ref(handle_new, &ref_new);
229 
230 	put_page_tag_ref(handle_old);
231 	put_page_tag_ref(handle_new);
232 }
233 
shutdown_mem_profiling(bool remove_file)234 static void shutdown_mem_profiling(bool remove_file)
235 {
236 	if (mem_alloc_profiling_enabled())
237 		static_branch_disable(&mem_alloc_profiling_key);
238 
239 	if (!mem_profiling_support)
240 		return;
241 
242 	if (remove_file)
243 		remove_proc_entry(ALLOCINFO_FILE_NAME, NULL);
244 	mem_profiling_support = false;
245 }
246 
alloc_tag_sec_init(void)247 void __init alloc_tag_sec_init(void)
248 {
249 	struct alloc_tag *last_codetag;
250 
251 	if (!mem_profiling_support)
252 		return;
253 
254 	if (!static_key_enabled(&mem_profiling_compressed))
255 		return;
256 
257 	kernel_tags.first_tag = (struct alloc_tag *)kallsyms_lookup_name(
258 					SECTION_START(ALLOC_TAG_SECTION_NAME));
259 	last_codetag = (struct alloc_tag *)kallsyms_lookup_name(
260 					SECTION_STOP(ALLOC_TAG_SECTION_NAME));
261 	kernel_tags.count = last_codetag - kernel_tags.first_tag;
262 
263 	/* Check if kernel tags fit into page flags */
264 	if (kernel_tags.count > (1UL << NR_UNUSED_PAGEFLAG_BITS)) {
265 		shutdown_mem_profiling(false); /* allocinfo file does not exist yet */
266 		pr_err("%lu allocation tags cannot be references using %d available page flag bits. Memory allocation profiling is disabled!\n",
267 			kernel_tags.count, NR_UNUSED_PAGEFLAG_BITS);
268 		return;
269 	}
270 
271 	alloc_tag_ref_offs = (LRU_REFS_PGOFF - NR_UNUSED_PAGEFLAG_BITS);
272 	alloc_tag_ref_mask = ((1UL << NR_UNUSED_PAGEFLAG_BITS) - 1);
273 	pr_debug("Memory allocation profiling compression is using %d page flag bits!\n",
274 		 NR_UNUSED_PAGEFLAG_BITS);
275 }
276 
277 #ifdef CONFIG_MODULES
278 
279 static struct maple_tree mod_area_mt = MTREE_INIT(mod_area_mt, MT_FLAGS_ALLOC_RANGE);
280 static struct vm_struct *vm_module_tags;
281 /* A dummy object used to indicate an unloaded module */
282 static struct module unloaded_mod;
283 /* A dummy object used to indicate a module prepended area */
284 static struct module prepend_mod;
285 
286 struct alloc_tag_module_section module_tags;
287 
alloc_tag_align(unsigned long val)288 static inline unsigned long alloc_tag_align(unsigned long val)
289 {
290 	if (!static_key_enabled(&mem_profiling_compressed)) {
291 		/* No alignment requirements when we are not indexing the tags */
292 		return val;
293 	}
294 
295 	if (val % sizeof(struct alloc_tag) == 0)
296 		return val;
297 	return ((val / sizeof(struct alloc_tag)) + 1) * sizeof(struct alloc_tag);
298 }
299 
ensure_alignment(unsigned long align,unsigned int * prepend)300 static bool ensure_alignment(unsigned long align, unsigned int *prepend)
301 {
302 	if (!static_key_enabled(&mem_profiling_compressed)) {
303 		/* No alignment requirements when we are not indexing the tags */
304 		return true;
305 	}
306 
307 	/*
308 	 * If alloc_tag size is not a multiple of required alignment, tag
309 	 * indexing does not work.
310 	 */
311 	if (!IS_ALIGNED(sizeof(struct alloc_tag), align))
312 		return false;
313 
314 	/* Ensure prepend consumes multiple of alloc_tag-sized blocks */
315 	if (*prepend)
316 		*prepend = alloc_tag_align(*prepend);
317 
318 	return true;
319 }
320 
tags_addressable(void)321 static inline bool tags_addressable(void)
322 {
323 	unsigned long tag_idx_count;
324 
325 	if (!static_key_enabled(&mem_profiling_compressed))
326 		return true; /* with page_ext tags are always addressable */
327 
328 	tag_idx_count = CODETAG_ID_FIRST + kernel_tags.count +
329 			module_tags.size / sizeof(struct alloc_tag);
330 
331 	return tag_idx_count < (1UL << NR_UNUSED_PAGEFLAG_BITS);
332 }
333 
needs_section_mem(struct module * mod,unsigned long size)334 static bool needs_section_mem(struct module *mod, unsigned long size)
335 {
336 	if (!mem_profiling_support)
337 		return false;
338 
339 	return size >= sizeof(struct alloc_tag);
340 }
341 
clean_unused_counters(struct alloc_tag * start_tag,struct alloc_tag * end_tag)342 static bool clean_unused_counters(struct alloc_tag *start_tag,
343 				  struct alloc_tag *end_tag)
344 {
345 	struct alloc_tag *tag;
346 	bool ret = true;
347 
348 	for (tag = start_tag; tag <= end_tag; tag++) {
349 		struct alloc_tag_counters counter;
350 
351 		if (!tag->counters)
352 			continue;
353 
354 		counter = alloc_tag_read(tag);
355 		if (!counter.bytes) {
356 			free_percpu(tag->counters);
357 			tag->counters = NULL;
358 		} else {
359 			ret = false;
360 		}
361 	}
362 
363 	return ret;
364 }
365 
366 /* Called with mod_area_mt locked */
clean_unused_module_areas_locked(void)367 static void clean_unused_module_areas_locked(void)
368 {
369 	MA_STATE(mas, &mod_area_mt, 0, module_tags.size);
370 	struct module *val;
371 
372 	mas_for_each(&mas, val, module_tags.size) {
373 		struct alloc_tag *start_tag;
374 		struct alloc_tag *end_tag;
375 
376 		if (val != &unloaded_mod)
377 			continue;
378 
379 		/* Release area if all tags are unused */
380 		start_tag = (struct alloc_tag *)(module_tags.start_addr + mas.index);
381 		end_tag = (struct alloc_tag *)(module_tags.start_addr + mas.last);
382 		if (clean_unused_counters(start_tag, end_tag))
383 			mas_erase(&mas);
384 	}
385 }
386 
387 /* Called with mod_area_mt locked */
find_aligned_area(struct ma_state * mas,unsigned long section_size,unsigned long size,unsigned int prepend,unsigned long align)388 static bool find_aligned_area(struct ma_state *mas, unsigned long section_size,
389 			      unsigned long size, unsigned int prepend, unsigned long align)
390 {
391 	bool cleanup_done = false;
392 
393 repeat:
394 	/* Try finding exact size and hope the start is aligned */
395 	if (!mas_empty_area(mas, 0, section_size - 1, prepend + size)) {
396 		if (IS_ALIGNED(mas->index + prepend, align))
397 			return true;
398 
399 		/* Try finding larger area to align later */
400 		mas_reset(mas);
401 		if (!mas_empty_area(mas, 0, section_size - 1,
402 				    size + prepend + align - 1))
403 			return true;
404 	}
405 
406 	/* No free area, try cleanup stale data and repeat the search once */
407 	if (!cleanup_done) {
408 		clean_unused_module_areas_locked();
409 		cleanup_done = true;
410 		mas_reset(mas);
411 		goto repeat;
412 	}
413 
414 	return false;
415 }
416 
vm_module_tags_populate(void)417 static int vm_module_tags_populate(void)
418 {
419 	unsigned long phys_end = ALIGN_DOWN(module_tags.start_addr, PAGE_SIZE) +
420 				 (vm_module_tags->nr_pages << PAGE_SHIFT);
421 	unsigned long new_end = module_tags.start_addr + module_tags.size;
422 
423 	if (phys_end < new_end) {
424 		struct page **next_page = vm_module_tags->pages + vm_module_tags->nr_pages;
425 		unsigned long old_shadow_end = ALIGN(phys_end, MODULE_ALIGN);
426 		unsigned long new_shadow_end = ALIGN(new_end, MODULE_ALIGN);
427 		unsigned long more_pages;
428 		unsigned long nr = 0;
429 
430 		more_pages = ALIGN(new_end - phys_end, PAGE_SIZE) >> PAGE_SHIFT;
431 		while (nr < more_pages) {
432 			unsigned long allocated;
433 
434 			allocated = alloc_pages_bulk_node(GFP_KERNEL | __GFP_NOWARN,
435 				NUMA_NO_NODE, more_pages - nr, next_page + nr);
436 
437 			if (!allocated)
438 				break;
439 			nr += allocated;
440 		}
441 
442 		if (nr < more_pages ||
443 		    vmap_pages_range(phys_end, phys_end + (nr << PAGE_SHIFT), PAGE_KERNEL,
444 				     next_page, PAGE_SHIFT) < 0) {
445 			/* Clean up and error out */
446 			for (int i = 0; i < nr; i++)
447 				__free_page(next_page[i]);
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((void *)old_shadow_end,
461 						  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((void *)module_tags.start_addr,
471 				new_end - module_tags.start_addr,
472 				KASAN_VMALLOC_PROT_NORMAL);
473 
474 	return 0;
475 }
476 
reserve_module_tags(struct module * mod,unsigned long size,unsigned int prepend,unsigned long align)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(-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)) {
497 		shutdown_mem_profiling(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(-EINVAL);
501 	}
502 
503 	mas_lock(&mas);
504 	if (!find_aligned_area(&mas, section_size, size, prepend, align)) {
505 		ret = ERR_PTR(-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, &prepend_mod);
518 		if (mas_is_err(&mas)) {
519 			ret = ERR_PTR(xa_err(mas.node));
520 			goto unlock;
521 		}
522 		mas.index = offset;
523 		mas.last = offset + size - 1;
524 		mas_store(&mas, mod);
525 		if (mas_is_err(&mas)) {
526 			mas.index = pad_start;
527 			mas_erase(&mas);
528 			ret = ERR_PTR(xa_err(mas.node));
529 		}
530 	} else {
531 		mas.last = offset + size - 1;
532 		mas_store(&mas, mod);
533 		if (mas_is_err(&mas))
534 			ret = ERR_PTR(xa_err(mas.node));
535 	}
536 unlock:
537 	mas_unlock(&mas);
538 
539 	if (IS_ERR(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(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(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(grow_res);
558 		}
559 	}
560 
561 	return (struct alloc_tag *)(module_tags.start_addr + offset);
562 }
563 
release_module_tags(struct module * mod,bool used)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, used ? &unloaded_mod : NULL);
603 	val = mas_prev_range(&mas, 0);
604 	if (val == &prepend_mod)
605 		mas_store(&mas, NULL);
606 out:
607 	mas_unlock(&mas);
608 }
609 
load_module(struct module * mod,struct codetag * start,struct codetag * stop)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(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 	return 0;
638 }
639 
replace_module(struct module * mod,struct module * new_mod)640 static void replace_module(struct module *mod, struct module *new_mod)
641 {
642 	MA_STATE(mas, &mod_area_mt, 0, module_tags.size);
643 	struct module *val;
644 
645 	mas_lock(&mas);
646 	mas_for_each(&mas, val, module_tags.size) {
647 		if (val != mod)
648 			continue;
649 
650 		mas_store_gfp(&mas, new_mod, GFP_KERNEL);
651 		break;
652 	}
653 	mas_unlock(&mas);
654 }
655 
alloc_mod_tags_mem(void)656 static int __init alloc_mod_tags_mem(void)
657 {
658 	/* Map space to copy allocation tags */
659 	vm_module_tags = execmem_vmap(MODULE_ALLOC_TAG_VMAP_SIZE);
660 	if (!vm_module_tags) {
661 		pr_err("Failed to map %lu bytes for module allocation tags\n",
662 			MODULE_ALLOC_TAG_VMAP_SIZE);
663 		module_tags.start_addr = 0;
664 		return -ENOMEM;
665 	}
666 
667 	vm_module_tags->pages = kmalloc_array(get_vm_area_size(vm_module_tags) >> PAGE_SHIFT,
668 					sizeof(struct page *), GFP_KERNEL | __GFP_ZERO);
669 	if (!vm_module_tags->pages) {
670 		free_vm_area(vm_module_tags);
671 		return -ENOMEM;
672 	}
673 
674 	module_tags.start_addr = (unsigned long)vm_module_tags->addr;
675 	module_tags.end_addr = module_tags.start_addr + MODULE_ALLOC_TAG_VMAP_SIZE;
676 	/* Ensure the base is alloc_tag aligned when required for indexing */
677 	module_tags.start_addr = alloc_tag_align(module_tags.start_addr);
678 
679 	return 0;
680 }
681 
free_mod_tags_mem(void)682 static void __init free_mod_tags_mem(void)
683 {
684 	int i;
685 
686 	module_tags.start_addr = 0;
687 	for (i = 0; i < vm_module_tags->nr_pages; i++)
688 		__free_page(vm_module_tags->pages[i]);
689 	kfree(vm_module_tags->pages);
690 	free_vm_area(vm_module_tags);
691 }
692 
693 #else /* CONFIG_MODULES */
694 
alloc_mod_tags_mem(void)695 static inline int alloc_mod_tags_mem(void) { return 0; }
free_mod_tags_mem(void)696 static inline void free_mod_tags_mem(void) {}
697 
698 #endif /* CONFIG_MODULES */
699 
700 /* See: Documentation/mm/allocation-profiling.rst */
setup_early_mem_profiling(char * str)701 static int __init setup_early_mem_profiling(char *str)
702 {
703 	bool compressed = false;
704 	bool enable;
705 
706 	if (!str || !str[0])
707 		return -EINVAL;
708 
709 	if (!strncmp(str, "never", 5)) {
710 		enable = false;
711 		mem_profiling_support = false;
712 		pr_info("Memory allocation profiling is disabled!\n");
713 	} else {
714 		char *token = strsep(&str, ",");
715 
716 		if (kstrtobool(token, &enable))
717 			return -EINVAL;
718 
719 		if (str) {
720 
721 			if (strcmp(str, "compressed"))
722 				return -EINVAL;
723 
724 			compressed = true;
725 		}
726 		mem_profiling_support = true;
727 		pr_info("Memory allocation profiling is enabled %s compression and is turned %s!\n",
728 			compressed ? "with" : "without", enable ? "on" : "off");
729 	}
730 
731 	if (enable != mem_alloc_profiling_enabled()) {
732 		if (enable)
733 			static_branch_enable(&mem_alloc_profiling_key);
734 		else
735 			static_branch_disable(&mem_alloc_profiling_key);
736 	}
737 	if (compressed != static_key_enabled(&mem_profiling_compressed)) {
738 		if (compressed)
739 			static_branch_enable(&mem_profiling_compressed);
740 		else
741 			static_branch_disable(&mem_profiling_compressed);
742 	}
743 
744 	return 0;
745 }
746 early_param("sysctl.vm.mem_profiling", setup_early_mem_profiling);
747 
need_page_alloc_tagging(void)748 static __init bool need_page_alloc_tagging(void)
749 {
750 	if (static_key_enabled(&mem_profiling_compressed))
751 		return false;
752 
753 	return mem_profiling_support;
754 }
755 
init_page_alloc_tagging(void)756 static __init void init_page_alloc_tagging(void)
757 {
758 }
759 
760 struct page_ext_operations page_alloc_tagging_ops = {
761 	.size = sizeof(union codetag_ref),
762 	.need = need_page_alloc_tagging,
763 	.init = init_page_alloc_tagging,
764 };
765 EXPORT_SYMBOL(page_alloc_tagging_ops);
766 
767 #ifdef CONFIG_SYSCTL
768 static struct ctl_table memory_allocation_profiling_sysctls[] = {
769 	{
770 		.procname	= "mem_profiling",
771 		.data		= &mem_alloc_profiling_key,
772 #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG
773 		.mode		= 0444,
774 #else
775 		.mode		= 0644,
776 #endif
777 		.proc_handler	= proc_do_static_key,
778 	},
779 };
780 
sysctl_init(void)781 static void __init sysctl_init(void)
782 {
783 	if (!mem_profiling_support)
784 		memory_allocation_profiling_sysctls[0].mode = 0444;
785 
786 	register_sysctl_init("vm", memory_allocation_profiling_sysctls);
787 }
788 #else /* CONFIG_SYSCTL */
sysctl_init(void)789 static inline void sysctl_init(void) {}
790 #endif /* CONFIG_SYSCTL */
791 
alloc_tag_init(void)792 static int __init alloc_tag_init(void)
793 {
794 	const struct codetag_type_desc desc = {
795 		.section		= ALLOC_TAG_SECTION_NAME,
796 		.tag_size		= sizeof(struct alloc_tag),
797 #ifdef CONFIG_MODULES
798 		.needs_section_mem	= needs_section_mem,
799 		.alloc_section_mem	= reserve_module_tags,
800 		.free_section_mem	= release_module_tags,
801 		.module_load		= load_module,
802 		.module_replaced	= replace_module,
803 #endif
804 	};
805 	int res;
806 
807 	sysctl_init();
808 
809 	if (!mem_profiling_support) {
810 		pr_info("Memory allocation profiling is not supported!\n");
811 		return 0;
812 	}
813 
814 	if (!proc_create_seq(ALLOCINFO_FILE_NAME, 0400, NULL, &allocinfo_seq_op)) {
815 		pr_err("Failed to create %s file\n", ALLOCINFO_FILE_NAME);
816 		shutdown_mem_profiling(false);
817 		return -ENOMEM;
818 	}
819 
820 	res = alloc_mod_tags_mem();
821 	if (res) {
822 		pr_err("Failed to reserve address space for module tags, errno = %d\n", res);
823 		shutdown_mem_profiling(true);
824 		return res;
825 	}
826 
827 	alloc_tag_cttype = codetag_register_type(&desc);
828 	if (IS_ERR(alloc_tag_cttype)) {
829 		pr_err("Allocation tags registration failed, errno = %ld\n", PTR_ERR(alloc_tag_cttype));
830 		free_mod_tags_mem();
831 		shutdown_mem_profiling(true);
832 		return PTR_ERR(alloc_tag_cttype);
833 	}
834 
835 	return 0;
836 }
837 module_init(alloc_tag_init);
838