1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * mm/debug.c
4  *
5  * mm/ specific debug routines.
6  *
7  */
8 
9 #include <linux/kernel.h>
10 #include <linux/mm.h>
11 #include <linux/trace_events.h>
12 #include <linux/memcontrol.h>
13 #include <trace/events/mmflags.h>
14 #include <linux/migrate.h>
15 #include <linux/page_owner.h>
16 #include <linux/ctype.h>
17 
18 #include "internal.h"
19 #include <trace/events/migrate.h>
20 
21 /*
22  * Define EM() and EMe() so that MIGRATE_REASON from trace/events/migrate.h can
23  * be used to populate migrate_reason_names[].
24  */
25 #undef EM
26 #undef EMe
27 #define EM(a, b)	b,
28 #define EMe(a, b)	b
29 
30 const char *migrate_reason_names[MR_TYPES] = {
31 	MIGRATE_REASON
32 };
33 
34 const struct trace_print_flags pageflag_names[] = {
35 	__def_pageflag_names,
36 	{0, NULL}
37 };
38 
39 const struct trace_print_flags gfpflag_names[] = {
40 	__def_gfpflag_names,
41 	{0, NULL}
42 };
43 
44 const struct trace_print_flags vmaflag_names[] = {
45 	__def_vmaflag_names,
46 	{0, NULL}
47 };
48 
49 #define DEF_PAGETYPE_NAME(_name) [PGTY_##_name - 0xf0] =  __stringify(_name)
50 
51 static const char *page_type_names[] = {
52 	DEF_PAGETYPE_NAME(slab),
53 	DEF_PAGETYPE_NAME(hugetlb),
54 	DEF_PAGETYPE_NAME(offline),
55 	DEF_PAGETYPE_NAME(guard),
56 	DEF_PAGETYPE_NAME(table),
57 	DEF_PAGETYPE_NAME(buddy),
58 	DEF_PAGETYPE_NAME(unaccepted),
59 };
60 
page_type_name(unsigned int page_type)61 static const char *page_type_name(unsigned int page_type)
62 {
63 	unsigned i = (page_type >> 24) - 0xf0;
64 
65 	if (i >= ARRAY_SIZE(page_type_names))
66 		return "unknown";
67 	return page_type_names[i];
68 }
69 
__dump_folio(struct folio * folio,struct page * page,unsigned long pfn,unsigned long idx)70 static void __dump_folio(struct folio *folio, struct page *page,
71 		unsigned long pfn, unsigned long idx)
72 {
73 	struct address_space *mapping = folio_mapping(folio);
74 	int mapcount = atomic_read(&page->_mapcount);
75 	char *type = "";
76 
77 	mapcount = page_mapcount_is_type(mapcount) ? 0 : mapcount + 1;
78 	pr_warn("page: refcount:%d mapcount:%d mapping:%p index:%#lx pfn:%#lx\n",
79 			folio_ref_count(folio), mapcount, mapping,
80 			folio->index + idx, pfn);
81 	if (folio_test_large(folio)) {
82 		pr_warn("head: order:%u mapcount:%d entire_mapcount:%d nr_pages_mapped:%d pincount:%d\n",
83 				folio_order(folio),
84 				folio_mapcount(folio),
85 				folio_entire_mapcount(folio),
86 				folio_nr_pages_mapped(folio),
87 				atomic_read(&folio->_pincount));
88 	}
89 
90 #ifdef CONFIG_MEMCG
91 	if (folio->memcg_data)
92 		pr_warn("memcg:%lx\n", folio->memcg_data);
93 #endif
94 	if (folio_test_ksm(folio))
95 		type = "ksm ";
96 	else if (folio_test_anon(folio))
97 		type = "anon ";
98 	else if (mapping)
99 		dump_mapping(mapping);
100 	BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS + 1);
101 
102 	/*
103 	 * Accessing the pageblock without the zone lock. It could change to
104 	 * "isolate" again in the meantime, but since we are just dumping the
105 	 * state for debugging, it should be fine to accept a bit of
106 	 * inaccuracy here due to racing.
107 	 */
108 	pr_warn("%sflags: %pGp%s\n", type, &folio->flags,
109 		is_migrate_cma_folio(folio, pfn) ? " CMA" : "");
110 	if (page_has_type(&folio->page))
111 		pr_warn("page_type: %x(%s)\n", folio->page.page_type >> 24,
112 				page_type_name(folio->page.page_type));
113 
114 	print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32,
115 			sizeof(unsigned long), page,
116 			sizeof(struct page), false);
117 	if (folio_test_large(folio))
118 		print_hex_dump(KERN_WARNING, "head: ", DUMP_PREFIX_NONE, 32,
119 			sizeof(unsigned long), folio,
120 			2 * sizeof(struct page), false);
121 }
122 
__dump_page(const struct page * page)123 static void __dump_page(const struct page *page)
124 {
125 	struct folio *foliop, folio;
126 	struct page precise;
127 	unsigned long pfn = page_to_pfn(page);
128 	unsigned long idx, nr_pages = 1;
129 	int loops = 5;
130 
131 again:
132 	memcpy(&precise, page, sizeof(*page));
133 	foliop = page_folio(&precise);
134 	if (foliop == (struct folio *)&precise) {
135 		idx = 0;
136 		if (!folio_test_large(foliop))
137 			goto dump;
138 		foliop = (struct folio *)page;
139 	} else {
140 		idx = folio_page_idx(foliop, page);
141 	}
142 
143 	if (idx < MAX_FOLIO_NR_PAGES) {
144 		memcpy(&folio, foliop, 2 * sizeof(struct page));
145 		nr_pages = folio_nr_pages(&folio);
146 		foliop = &folio;
147 	}
148 
149 	if (idx > nr_pages) {
150 		if (loops-- > 0)
151 			goto again;
152 		pr_warn("page does not match folio\n");
153 		precise.compound_head &= ~1UL;
154 		foliop = (struct folio *)&precise;
155 		idx = 0;
156 	}
157 
158 dump:
159 	__dump_folio(foliop, &precise, pfn, idx);
160 }
161 
dump_page(const struct page * page,const char * reason)162 void dump_page(const struct page *page, const char *reason)
163 {
164 	if (PagePoisoned(page))
165 		pr_warn("page:%p is uninitialized and poisoned", page);
166 	else
167 		__dump_page(page);
168 	if (reason)
169 		pr_warn("page dumped because: %s\n", reason);
170 	dump_page_owner(page);
171 }
172 EXPORT_SYMBOL(dump_page);
173 
174 #ifdef CONFIG_DEBUG_VM
175 
dump_vma(const struct vm_area_struct * vma)176 void dump_vma(const struct vm_area_struct *vma)
177 {
178 	pr_emerg("vma %px start %px end %px mm %px\n"
179 		"prot %lx anon_vma %px vm_ops %px\n"
180 		"pgoff %lx file %px private_data %px\n"
181 		"flags: %#lx(%pGv)\n",
182 		vma, (void *)vma->vm_start, (void *)vma->vm_end, vma->vm_mm,
183 		(unsigned long)pgprot_val(vma->vm_page_prot),
184 		vma->anon_vma, vma->vm_ops, vma->vm_pgoff,
185 		vma->vm_file, vma->vm_private_data,
186 		vma->vm_flags, &vma->vm_flags);
187 }
188 EXPORT_SYMBOL(dump_vma);
189 
dump_mm(const struct mm_struct * mm)190 void dump_mm(const struct mm_struct *mm)
191 {
192 	pr_emerg("mm %px task_size %lu\n"
193 		"mmap_base %lu mmap_legacy_base %lu\n"
194 		"pgd %px mm_users %d mm_count %d pgtables_bytes %lu map_count %d\n"
195 		"hiwater_rss %lx hiwater_vm %lx total_vm %lx locked_vm %lx\n"
196 		"pinned_vm %llx data_vm %lx exec_vm %lx stack_vm %lx\n"
197 		"start_code %lx end_code %lx start_data %lx end_data %lx\n"
198 		"start_brk %lx brk %lx start_stack %lx\n"
199 		"arg_start %lx arg_end %lx env_start %lx env_end %lx\n"
200 		"binfmt %px flags %lx\n"
201 #ifdef CONFIG_AIO
202 		"ioctx_table %px\n"
203 #endif
204 #ifdef CONFIG_MEMCG
205 		"owner %px "
206 #endif
207 		"exe_file %px\n"
208 #ifdef CONFIG_MMU_NOTIFIER
209 		"notifier_subscriptions %px\n"
210 #endif
211 #ifdef CONFIG_NUMA_BALANCING
212 		"numa_next_scan %lu numa_scan_offset %lu numa_scan_seq %d\n"
213 #endif
214 		"tlb_flush_pending %d\n"
215 		"def_flags: %#lx(%pGv)\n",
216 
217 		mm, mm->task_size,
218 		mm->mmap_base, mm->mmap_legacy_base,
219 		mm->pgd, atomic_read(&mm->mm_users),
220 		atomic_read(&mm->mm_count),
221 		mm_pgtables_bytes(mm),
222 		mm->map_count,
223 		mm->hiwater_rss, mm->hiwater_vm, mm->total_vm, mm->locked_vm,
224 		(u64)atomic64_read(&mm->pinned_vm),
225 		mm->data_vm, mm->exec_vm, mm->stack_vm,
226 		mm->start_code, mm->end_code, mm->start_data, mm->end_data,
227 		mm->start_brk, mm->brk, mm->start_stack,
228 		mm->arg_start, mm->arg_end, mm->env_start, mm->env_end,
229 		mm->binfmt, mm->flags,
230 #ifdef CONFIG_AIO
231 		mm->ioctx_table,
232 #endif
233 #ifdef CONFIG_MEMCG
234 		mm->owner,
235 #endif
236 		mm->exe_file,
237 #ifdef CONFIG_MMU_NOTIFIER
238 		mm->notifier_subscriptions,
239 #endif
240 #ifdef CONFIG_NUMA_BALANCING
241 		mm->numa_next_scan, mm->numa_scan_offset, mm->numa_scan_seq,
242 #endif
243 		atomic_read(&mm->tlb_flush_pending),
244 		mm->def_flags, &mm->def_flags
245 	);
246 }
247 EXPORT_SYMBOL(dump_mm);
248 
249 static bool page_init_poisoning __read_mostly = true;
250 
setup_vm_debug(char * str)251 static int __init setup_vm_debug(char *str)
252 {
253 	bool __page_init_poisoning = true;
254 
255 	/*
256 	 * Calling vm_debug with no arguments is equivalent to requesting
257 	 * to enable all debugging options we can control.
258 	 */
259 	if (*str++ != '=' || !*str)
260 		goto out;
261 
262 	__page_init_poisoning = false;
263 	if (*str == '-')
264 		goto out;
265 
266 	while (*str) {
267 		switch (tolower(*str)) {
268 		case'p':
269 			__page_init_poisoning = true;
270 			break;
271 		default:
272 			pr_err("vm_debug option '%c' unknown. skipped\n",
273 			       *str);
274 		}
275 
276 		str++;
277 	}
278 out:
279 	if (page_init_poisoning && !__page_init_poisoning)
280 		pr_warn("Page struct poisoning disabled by kernel command line option 'vm_debug'\n");
281 
282 	page_init_poisoning = __page_init_poisoning;
283 
284 	return 1;
285 }
286 __setup("vm_debug", setup_vm_debug);
287 
page_init_poison(struct page * page,size_t size)288 void page_init_poison(struct page *page, size_t size)
289 {
290 	if (page_init_poisoning)
291 		memset(page, PAGE_POISON_PATTERN, size);
292 }
293 
vma_iter_dump_tree(const struct vma_iterator * vmi)294 void vma_iter_dump_tree(const struct vma_iterator *vmi)
295 {
296 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
297 	mas_dump(&vmi->mas);
298 	mt_dump(vmi->mas.tree, mt_dump_hex);
299 #endif	/* CONFIG_DEBUG_VM_MAPLE_TREE */
300 }
301 
302 #endif		/* CONFIG_DEBUG_VM */
303