Loading...
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
30const char *migrate_reason_names[MR_TYPES] = {
31 MIGRATE_REASON
32};
33
34const struct trace_print_flags pageflag_names[] = {
35 __def_pageflag_names,
36 {0, NULL}
37};
38
39const struct trace_print_flags gfpflag_names[] = {
40 __def_gfpflag_names,
41 {0, NULL}
42};
43
44const 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
51static 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
61static 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
70static 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
123static void __dump_page(const struct page *page)
124{
125 struct folio *foliop, folio;
126 struct page precise;
127 unsigned long head;
128 unsigned long pfn = page_to_pfn(page);
129 unsigned long idx, nr_pages = 1;
130 int loops = 5;
131
132again:
133 memcpy(&precise, page, sizeof(*page));
134 head = precise.compound_head;
135 if ((head & 1) == 0) {
136 foliop = (struct folio *)&precise;
137 idx = 0;
138 if (!folio_test_large(foliop))
139 goto dump;
140 foliop = (struct folio *)page;
141 } else {
142 foliop = (struct folio *)(head - 1);
143 idx = folio_page_idx(foliop, page);
144 }
145
146 if (idx < MAX_FOLIO_NR_PAGES) {
147 memcpy(&folio, foliop, 2 * sizeof(struct page));
148 nr_pages = folio_nr_pages(&folio);
149 foliop = &folio;
150 }
151
152 if (idx > nr_pages) {
153 if (loops-- > 0)
154 goto again;
155 pr_warn("page does not match folio\n");
156 precise.compound_head &= ~1UL;
157 foliop = (struct folio *)&precise;
158 idx = 0;
159 }
160
161dump:
162 __dump_folio(foliop, &precise, pfn, idx);
163}
164
165void dump_page(const struct page *page, const char *reason)
166{
167 if (PagePoisoned(page))
168 pr_warn("page:%p is uninitialized and poisoned", page);
169 else
170 __dump_page(page);
171 if (reason)
172 pr_warn("page dumped because: %s\n", reason);
173 dump_page_owner(page);
174}
175EXPORT_SYMBOL(dump_page);
176
177#ifdef CONFIG_DEBUG_VM
178
179void dump_vma(const struct vm_area_struct *vma)
180{
181 pr_emerg("vma %px start %px end %px mm %px\n"
182 "prot %lx anon_vma %px vm_ops %px\n"
183 "pgoff %lx file %px private_data %px\n"
184 "flags: %#lx(%pGv)\n",
185 vma, (void *)vma->vm_start, (void *)vma->vm_end, vma->vm_mm,
186 (unsigned long)pgprot_val(vma->vm_page_prot),
187 vma->anon_vma, vma->vm_ops, vma->vm_pgoff,
188 vma->vm_file, vma->vm_private_data,
189 vma->vm_flags, &vma->vm_flags);
190}
191EXPORT_SYMBOL(dump_vma);
192
193void dump_mm(const struct mm_struct *mm)
194{
195 pr_emerg("mm %px task_size %lu\n"
196 "mmap_base %lu mmap_legacy_base %lu\n"
197 "pgd %px mm_users %d mm_count %d pgtables_bytes %lu map_count %d\n"
198 "hiwater_rss %lx hiwater_vm %lx total_vm %lx locked_vm %lx\n"
199 "pinned_vm %llx data_vm %lx exec_vm %lx stack_vm %lx\n"
200 "start_code %lx end_code %lx start_data %lx end_data %lx\n"
201 "start_brk %lx brk %lx start_stack %lx\n"
202 "arg_start %lx arg_end %lx env_start %lx env_end %lx\n"
203 "binfmt %px flags %lx\n"
204#ifdef CONFIG_AIO
205 "ioctx_table %px\n"
206#endif
207#ifdef CONFIG_MEMCG
208 "owner %px "
209#endif
210 "exe_file %px\n"
211#ifdef CONFIG_MMU_NOTIFIER
212 "notifier_subscriptions %px\n"
213#endif
214#ifdef CONFIG_NUMA_BALANCING
215 "numa_next_scan %lu numa_scan_offset %lu numa_scan_seq %d\n"
216#endif
217 "tlb_flush_pending %d\n"
218 "def_flags: %#lx(%pGv)\n",
219
220 mm, mm->task_size,
221 mm->mmap_base, mm->mmap_legacy_base,
222 mm->pgd, atomic_read(&mm->mm_users),
223 atomic_read(&mm->mm_count),
224 mm_pgtables_bytes(mm),
225 mm->map_count,
226 mm->hiwater_rss, mm->hiwater_vm, mm->total_vm, mm->locked_vm,
227 (u64)atomic64_read(&mm->pinned_vm),
228 mm->data_vm, mm->exec_vm, mm->stack_vm,
229 mm->start_code, mm->end_code, mm->start_data, mm->end_data,
230 mm->start_brk, mm->brk, mm->start_stack,
231 mm->arg_start, mm->arg_end, mm->env_start, mm->env_end,
232 mm->binfmt, mm->flags,
233#ifdef CONFIG_AIO
234 mm->ioctx_table,
235#endif
236#ifdef CONFIG_MEMCG
237 mm->owner,
238#endif
239 mm->exe_file,
240#ifdef CONFIG_MMU_NOTIFIER
241 mm->notifier_subscriptions,
242#endif
243#ifdef CONFIG_NUMA_BALANCING
244 mm->numa_next_scan, mm->numa_scan_offset, mm->numa_scan_seq,
245#endif
246 atomic_read(&mm->tlb_flush_pending),
247 mm->def_flags, &mm->def_flags
248 );
249}
250EXPORT_SYMBOL(dump_mm);
251
252static bool page_init_poisoning __read_mostly = true;
253
254static int __init setup_vm_debug(char *str)
255{
256 bool __page_init_poisoning = true;
257
258 /*
259 * Calling vm_debug with no arguments is equivalent to requesting
260 * to enable all debugging options we can control.
261 */
262 if (*str++ != '=' || !*str)
263 goto out;
264
265 __page_init_poisoning = false;
266 if (*str == '-')
267 goto out;
268
269 while (*str) {
270 switch (tolower(*str)) {
271 case'p':
272 __page_init_poisoning = true;
273 break;
274 default:
275 pr_err("vm_debug option '%c' unknown. skipped\n",
276 *str);
277 }
278
279 str++;
280 }
281out:
282 if (page_init_poisoning && !__page_init_poisoning)
283 pr_warn("Page struct poisoning disabled by kernel command line option 'vm_debug'\n");
284
285 page_init_poisoning = __page_init_poisoning;
286
287 return 1;
288}
289__setup("vm_debug", setup_vm_debug);
290
291void page_init_poison(struct page *page, size_t size)
292{
293 if (page_init_poisoning)
294 memset(page, PAGE_POISON_PATTERN, size);
295}
296
297void vma_iter_dump_tree(const struct vma_iterator *vmi)
298{
299#if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
300 mas_dump(&vmi->mas);
301 mt_dump(vmi->mas.tree, mt_dump_hex);
302#endif /* CONFIG_DEBUG_VM_MAPLE_TREE */
303}
304
305#endif /* CONFIG_DEBUG_VM */
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
20const char *migrate_reason_names[MR_TYPES] = {
21 "compaction",
22 "memory_failure",
23 "memory_hotplug",
24 "syscall_or_cpuset",
25 "mempolicy_mbind",
26 "numa_misplaced",
27 "cma",
28};
29
30const struct trace_print_flags pageflag_names[] = {
31 __def_pageflag_names,
32 {0, NULL}
33};
34
35const struct trace_print_flags gfpflag_names[] = {
36 __def_gfpflag_names,
37 {0, NULL}
38};
39
40const struct trace_print_flags vmaflag_names[] = {
41 __def_vmaflag_names,
42 {0, NULL}
43};
44
45void __dump_page(struct page *page, const char *reason)
46{
47 struct page *head = compound_head(page);
48 struct address_space *mapping;
49 bool page_poisoned = PagePoisoned(page);
50 bool compound = PageCompound(page);
51 /*
52 * Accessing the pageblock without the zone lock. It could change to
53 * "isolate" again in the meantime, but since we are just dumping the
54 * state for debugging, it should be fine to accept a bit of
55 * inaccuracy here due to racing.
56 */
57 bool page_cma = is_migrate_cma_page(page);
58 int mapcount;
59 char *type = "";
60
61 /*
62 * If struct page is poisoned don't access Page*() functions as that
63 * leads to recursive loop. Page*() check for poisoned pages, and calls
64 * dump_page() when detected.
65 */
66 if (page_poisoned) {
67 pr_warn("page:%px is uninitialized and poisoned", page);
68 goto hex_only;
69 }
70
71 if (page < head || (page >= head + MAX_ORDER_NR_PAGES)) {
72 /*
73 * Corrupt page, so we cannot call page_mapping. Instead, do a
74 * safe subset of the steps that page_mapping() does. Caution:
75 * this will be misleading for tail pages, PageSwapCache pages,
76 * and potentially other situations. (See the page_mapping()
77 * implementation for what's missing here.)
78 */
79 unsigned long tmp = (unsigned long)page->mapping;
80
81 if (tmp & PAGE_MAPPING_ANON)
82 mapping = NULL;
83 else
84 mapping = (void *)(tmp & ~PAGE_MAPPING_FLAGS);
85 head = page;
86 compound = false;
87 } else {
88 mapping = page_mapping(page);
89 }
90
91 /*
92 * Avoid VM_BUG_ON() in page_mapcount().
93 * page->_mapcount space in struct page is used by sl[aou]b pages to
94 * encode own info.
95 */
96 mapcount = PageSlab(head) ? 0 : page_mapcount(page);
97
98 pr_warn("page:%p refcount:%d mapcount:%d mapping:%p index:%#lx pfn:%#lx\n",
99 page, page_ref_count(head), mapcount, mapping,
100 page_to_pgoff(page), page_to_pfn(page));
101 if (compound) {
102 if (hpage_pincount_available(page)) {
103 pr_warn("head:%p order:%u compound_mapcount:%d compound_pincount:%d\n",
104 head, compound_order(head),
105 head_mapcount(head),
106 head_pincount(head));
107 } else {
108 pr_warn("head:%p order:%u compound_mapcount:%d\n",
109 head, compound_order(head),
110 head_mapcount(head));
111 }
112 }
113 if (PageKsm(page))
114 type = "ksm ";
115 else if (PageAnon(page))
116 type = "anon ";
117 else if (mapping) {
118 struct inode *host;
119 const struct address_space_operations *a_ops;
120 struct hlist_node *dentry_first;
121 struct dentry *dentry_ptr;
122 struct dentry dentry;
123
124 /*
125 * mapping can be invalid pointer and we don't want to crash
126 * accessing it, so probe everything depending on it carefully
127 */
128 if (get_kernel_nofault(host, &mapping->host) ||
129 get_kernel_nofault(a_ops, &mapping->a_ops)) {
130 pr_warn("failed to read mapping contents, not a valid kernel address?\n");
131 goto out_mapping;
132 }
133
134 if (!host) {
135 pr_warn("aops:%ps\n", a_ops);
136 goto out_mapping;
137 }
138
139 if (get_kernel_nofault(dentry_first, &host->i_dentry.first)) {
140 pr_warn("aops:%ps with invalid host inode %px\n",
141 a_ops, host);
142 goto out_mapping;
143 }
144
145 if (!dentry_first) {
146 pr_warn("aops:%ps ino:%lx\n", a_ops, host->i_ino);
147 goto out_mapping;
148 }
149
150 dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
151 if (get_kernel_nofault(dentry, dentry_ptr)) {
152 pr_warn("aops:%ps with invalid dentry %px\n", a_ops,
153 dentry_ptr);
154 } else {
155 /*
156 * if dentry is corrupted, the %pd handler may still
157 * crash, but it's unlikely that we reach here with a
158 * corrupted struct page
159 */
160 pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n",
161 a_ops, host->i_ino, &dentry);
162 }
163 }
164out_mapping:
165 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS + 1);
166
167 pr_warn("%sflags: %#lx(%pGp)%s\n", type, head->flags, &head->flags,
168 page_cma ? " CMA" : "");
169
170hex_only:
171 print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32,
172 sizeof(unsigned long), page,
173 sizeof(struct page), false);
174 if (head != page)
175 print_hex_dump(KERN_WARNING, "head: ", DUMP_PREFIX_NONE, 32,
176 sizeof(unsigned long), head,
177 sizeof(struct page), false);
178
179 if (reason)
180 pr_warn("page dumped because: %s\n", reason);
181
182#ifdef CONFIG_MEMCG
183 if (!page_poisoned && page->mem_cgroup)
184 pr_warn("page->mem_cgroup:%px\n", page->mem_cgroup);
185#endif
186}
187
188void dump_page(struct page *page, const char *reason)
189{
190 __dump_page(page, reason);
191 dump_page_owner(page);
192}
193EXPORT_SYMBOL(dump_page);
194
195#ifdef CONFIG_DEBUG_VM
196
197void dump_vma(const struct vm_area_struct *vma)
198{
199 pr_emerg("vma %px start %px end %px\n"
200 "next %px prev %px mm %px\n"
201 "prot %lx anon_vma %px vm_ops %px\n"
202 "pgoff %lx file %px private_data %px\n"
203 "flags: %#lx(%pGv)\n",
204 vma, (void *)vma->vm_start, (void *)vma->vm_end, vma->vm_next,
205 vma->vm_prev, vma->vm_mm,
206 (unsigned long)pgprot_val(vma->vm_page_prot),
207 vma->anon_vma, vma->vm_ops, vma->vm_pgoff,
208 vma->vm_file, vma->vm_private_data,
209 vma->vm_flags, &vma->vm_flags);
210}
211EXPORT_SYMBOL(dump_vma);
212
213void dump_mm(const struct mm_struct *mm)
214{
215 pr_emerg("mm %px mmap %px seqnum %llu task_size %lu\n"
216#ifdef CONFIG_MMU
217 "get_unmapped_area %px\n"
218#endif
219 "mmap_base %lu mmap_legacy_base %lu highest_vm_end %lu\n"
220 "pgd %px mm_users %d mm_count %d pgtables_bytes %lu map_count %d\n"
221 "hiwater_rss %lx hiwater_vm %lx total_vm %lx locked_vm %lx\n"
222 "pinned_vm %llx data_vm %lx exec_vm %lx stack_vm %lx\n"
223 "start_code %lx end_code %lx start_data %lx end_data %lx\n"
224 "start_brk %lx brk %lx start_stack %lx\n"
225 "arg_start %lx arg_end %lx env_start %lx env_end %lx\n"
226 "binfmt %px flags %lx core_state %px\n"
227#ifdef CONFIG_AIO
228 "ioctx_table %px\n"
229#endif
230#ifdef CONFIG_MEMCG
231 "owner %px "
232#endif
233 "exe_file %px\n"
234#ifdef CONFIG_MMU_NOTIFIER
235 "notifier_subscriptions %px\n"
236#endif
237#ifdef CONFIG_NUMA_BALANCING
238 "numa_next_scan %lu numa_scan_offset %lu numa_scan_seq %d\n"
239#endif
240 "tlb_flush_pending %d\n"
241 "def_flags: %#lx(%pGv)\n",
242
243 mm, mm->mmap, (long long) mm->vmacache_seqnum, mm->task_size,
244#ifdef CONFIG_MMU
245 mm->get_unmapped_area,
246#endif
247 mm->mmap_base, mm->mmap_legacy_base, mm->highest_vm_end,
248 mm->pgd, atomic_read(&mm->mm_users),
249 atomic_read(&mm->mm_count),
250 mm_pgtables_bytes(mm),
251 mm->map_count,
252 mm->hiwater_rss, mm->hiwater_vm, mm->total_vm, mm->locked_vm,
253 (u64)atomic64_read(&mm->pinned_vm),
254 mm->data_vm, mm->exec_vm, mm->stack_vm,
255 mm->start_code, mm->end_code, mm->start_data, mm->end_data,
256 mm->start_brk, mm->brk, mm->start_stack,
257 mm->arg_start, mm->arg_end, mm->env_start, mm->env_end,
258 mm->binfmt, mm->flags, mm->core_state,
259#ifdef CONFIG_AIO
260 mm->ioctx_table,
261#endif
262#ifdef CONFIG_MEMCG
263 mm->owner,
264#endif
265 mm->exe_file,
266#ifdef CONFIG_MMU_NOTIFIER
267 mm->notifier_subscriptions,
268#endif
269#ifdef CONFIG_NUMA_BALANCING
270 mm->numa_next_scan, mm->numa_scan_offset, mm->numa_scan_seq,
271#endif
272 atomic_read(&mm->tlb_flush_pending),
273 mm->def_flags, &mm->def_flags
274 );
275}
276
277static bool page_init_poisoning __read_mostly = true;
278
279static int __init setup_vm_debug(char *str)
280{
281 bool __page_init_poisoning = true;
282
283 /*
284 * Calling vm_debug with no arguments is equivalent to requesting
285 * to enable all debugging options we can control.
286 */
287 if (*str++ != '=' || !*str)
288 goto out;
289
290 __page_init_poisoning = false;
291 if (*str == '-')
292 goto out;
293
294 while (*str) {
295 switch (tolower(*str)) {
296 case'p':
297 __page_init_poisoning = true;
298 break;
299 default:
300 pr_err("vm_debug option '%c' unknown. skipped\n",
301 *str);
302 }
303
304 str++;
305 }
306out:
307 if (page_init_poisoning && !__page_init_poisoning)
308 pr_warn("Page struct poisoning disabled by kernel command line option 'vm_debug'\n");
309
310 page_init_poisoning = __page_init_poisoning;
311
312 return 1;
313}
314__setup("vm_debug", setup_vm_debug);
315
316void page_init_poison(struct page *page, size_t size)
317{
318 if (page_init_poisoning)
319 memset(page, PAGE_POISON_PATTERN, size);
320}
321EXPORT_SYMBOL_GPL(page_init_poison);
322#endif /* CONFIG_DEBUG_VM */