xref: /openbmc/linux/mm/debug.c (revision 78bb17f7) 
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 
20 const 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 
30 const struct trace_print_flags pageflag_names[] = {
31 	__def_pageflag_names,
32 	{0, NULL}
33 };
34 
35 const struct trace_print_flags gfpflag_names[] = {
36 	__def_gfpflag_names,
37 	{0, NULL}
38 };
39 
40 const struct trace_print_flags vmaflag_names[] = {
41 	__def_vmaflag_names,
42 	{0, NULL}
43 };
44 
45 void __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 		/* Corrupt page, cannot call page_mapping */
73 		mapping = page->mapping;
74 		head = page;
75 		compound = false;
76 	} else {
77 		mapping = page_mapping(page);
78 	}
79 
80 	/*
81 	 * Avoid VM_BUG_ON() in page_mapcount().
82 	 * page->_mapcount space in struct page is used by sl[aou]b pages to
83 	 * encode own info.
84 	 */
85 	mapcount = PageSlab(head) ? 0 : page_mapcount(page);
86 
87 	if (compound)
88 		if (hpage_pincount_available(page)) {
89 			pr_warn("page:%px refcount:%d mapcount:%d mapping:%p "
90 				"index:%#lx head:%px order:%u "
91 				"compound_mapcount:%d compound_pincount:%d\n",
92 				page, page_ref_count(head), mapcount,
93 				mapping, page_to_pgoff(page), head,
94 				compound_order(head), compound_mapcount(page),
95 				compound_pincount(page));
96 		} else {
97 			pr_warn("page:%px refcount:%d mapcount:%d mapping:%p "
98 				"index:%#lx head:%px order:%u "
99 				"compound_mapcount:%d\n",
100 				page, page_ref_count(head), mapcount,
101 				mapping, page_to_pgoff(page), head,
102 				compound_order(head), compound_mapcount(page));
103 		}
104 	else
105 		pr_warn("page:%px refcount:%d mapcount:%d mapping:%p index:%#lx\n",
106 			page, page_ref_count(page), mapcount,
107 			mapping, page_to_pgoff(page));
108 	if (PageKsm(page))
109 		type = "ksm ";
110 	else if (PageAnon(page))
111 		type = "anon ";
112 	else if (mapping) {
113 		const struct inode *host;
114 		const struct address_space_operations *a_ops;
115 		const struct hlist_node *dentry_first;
116 		const struct dentry *dentry_ptr;
117 		struct dentry dentry;
118 
119 		/*
120 		 * mapping can be invalid pointer and we don't want to crash
121 		 * accessing it, so probe everything depending on it carefully
122 		 */
123 		if (probe_kernel_read_strict(&host, &mapping->host,
124 						sizeof(struct inode *)) ||
125 		    probe_kernel_read_strict(&a_ops, &mapping->a_ops,
126 				sizeof(struct address_space_operations *))) {
127 			pr_warn("failed to read mapping->host or a_ops, mapping not a valid kernel address?\n");
128 			goto out_mapping;
129 		}
130 
131 		if (!host) {
132 			pr_warn("mapping->a_ops:%ps\n", a_ops);
133 			goto out_mapping;
134 		}
135 
136 		if (probe_kernel_read_strict(&dentry_first,
137 			&host->i_dentry.first, sizeof(struct hlist_node *))) {
138 			pr_warn("mapping->a_ops:%ps with invalid mapping->host inode address %px\n",
139 				a_ops, host);
140 			goto out_mapping;
141 		}
142 
143 		if (!dentry_first) {
144 			pr_warn("mapping->a_ops:%ps\n", a_ops);
145 			goto out_mapping;
146 		}
147 
148 		dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
149 		if (probe_kernel_read_strict(&dentry, dentry_ptr,
150 							sizeof(struct dentry))) {
151 			pr_warn("mapping->aops:%ps with invalid mapping->host->i_dentry.first %px\n",
152 				a_ops, dentry_ptr);
153 		} else {
154 			/*
155 			 * if dentry is corrupted, the %pd handler may still
156 			 * crash, but it's unlikely that we reach here with a
157 			 * corrupted struct page
158 			 */
159 			pr_warn("mapping->aops:%ps dentry name:\"%pd\"\n",
160 								a_ops, &dentry);
161 		}
162 	}
163 out_mapping:
164 	BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS + 1);
165 
166 	pr_warn("%sflags: %#lx(%pGp)%s\n", type, page->flags, &page->flags,
167 		page_cma ? " CMA" : "");
168 
169 hex_only:
170 	print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32,
171 			sizeof(unsigned long), page,
172 			sizeof(struct page), false);
173 	if (head != page)
174 		print_hex_dump(KERN_WARNING, "head: ", DUMP_PREFIX_NONE, 32,
175 			sizeof(unsigned long), head,
176 			sizeof(struct page), false);
177 
178 	if (reason)
179 		pr_warn("page dumped because: %s\n", reason);
180 
181 #ifdef CONFIG_MEMCG
182 	if (!page_poisoned && page->mem_cgroup)
183 		pr_warn("page->mem_cgroup:%px\n", page->mem_cgroup);
184 #endif
185 }
186 
187 void dump_page(struct page *page, const char *reason)
188 {
189 	__dump_page(page, reason);
190 	dump_page_owner(page);
191 }
192 EXPORT_SYMBOL(dump_page);
193 
194 #ifdef CONFIG_DEBUG_VM
195 
196 void dump_vma(const struct vm_area_struct *vma)
197 {
198 	pr_emerg("vma %px start %px end %px\n"
199 		"next %px prev %px mm %px\n"
200 		"prot %lx anon_vma %px vm_ops %px\n"
201 		"pgoff %lx file %px private_data %px\n"
202 		"flags: %#lx(%pGv)\n",
203 		vma, (void *)vma->vm_start, (void *)vma->vm_end, vma->vm_next,
204 		vma->vm_prev, vma->vm_mm,
205 		(unsigned long)pgprot_val(vma->vm_page_prot),
206 		vma->anon_vma, vma->vm_ops, vma->vm_pgoff,
207 		vma->vm_file, vma->vm_private_data,
208 		vma->vm_flags, &vma->vm_flags);
209 }
210 EXPORT_SYMBOL(dump_vma);
211 
212 void dump_mm(const struct mm_struct *mm)
213 {
214 	pr_emerg("mm %px mmap %px seqnum %llu task_size %lu\n"
215 #ifdef CONFIG_MMU
216 		"get_unmapped_area %px\n"
217 #endif
218 		"mmap_base %lu mmap_legacy_base %lu highest_vm_end %lu\n"
219 		"pgd %px mm_users %d mm_count %d pgtables_bytes %lu map_count %d\n"
220 		"hiwater_rss %lx hiwater_vm %lx total_vm %lx locked_vm %lx\n"
221 		"pinned_vm %llx data_vm %lx exec_vm %lx stack_vm %lx\n"
222 		"start_code %lx end_code %lx start_data %lx end_data %lx\n"
223 		"start_brk %lx brk %lx start_stack %lx\n"
224 		"arg_start %lx arg_end %lx env_start %lx env_end %lx\n"
225 		"binfmt %px flags %lx core_state %px\n"
226 #ifdef CONFIG_AIO
227 		"ioctx_table %px\n"
228 #endif
229 #ifdef CONFIG_MEMCG
230 		"owner %px "
231 #endif
232 		"exe_file %px\n"
233 #ifdef CONFIG_MMU_NOTIFIER
234 		"notifier_subscriptions %px\n"
235 #endif
236 #ifdef CONFIG_NUMA_BALANCING
237 		"numa_next_scan %lu numa_scan_offset %lu numa_scan_seq %d\n"
238 #endif
239 		"tlb_flush_pending %d\n"
240 		"def_flags: %#lx(%pGv)\n",
241 
242 		mm, mm->mmap, (long long) mm->vmacache_seqnum, mm->task_size,
243 #ifdef CONFIG_MMU
244 		mm->get_unmapped_area,
245 #endif
246 		mm->mmap_base, mm->mmap_legacy_base, mm->highest_vm_end,
247 		mm->pgd, atomic_read(&mm->mm_users),
248 		atomic_read(&mm->mm_count),
249 		mm_pgtables_bytes(mm),
250 		mm->map_count,
251 		mm->hiwater_rss, mm->hiwater_vm, mm->total_vm, mm->locked_vm,
252 		(u64)atomic64_read(&mm->pinned_vm),
253 		mm->data_vm, mm->exec_vm, mm->stack_vm,
254 		mm->start_code, mm->end_code, mm->start_data, mm->end_data,
255 		mm->start_brk, mm->brk, mm->start_stack,
256 		mm->arg_start, mm->arg_end, mm->env_start, mm->env_end,
257 		mm->binfmt, mm->flags, mm->core_state,
258 #ifdef CONFIG_AIO
259 		mm->ioctx_table,
260 #endif
261 #ifdef CONFIG_MEMCG
262 		mm->owner,
263 #endif
264 		mm->exe_file,
265 #ifdef CONFIG_MMU_NOTIFIER
266 		mm->notifier_subscriptions,
267 #endif
268 #ifdef CONFIG_NUMA_BALANCING
269 		mm->numa_next_scan, mm->numa_scan_offset, mm->numa_scan_seq,
270 #endif
271 		atomic_read(&mm->tlb_flush_pending),
272 		mm->def_flags, &mm->def_flags
273 	);
274 }
275 
276 static bool page_init_poisoning __read_mostly = true;
277 
278 static int __init setup_vm_debug(char *str)
279 {
280 	bool __page_init_poisoning = true;
281 
282 	/*
283 	 * Calling vm_debug with no arguments is equivalent to requesting
284 	 * to enable all debugging options we can control.
285 	 */
286 	if (*str++ != '=' || !*str)
287 		goto out;
288 
289 	__page_init_poisoning = false;
290 	if (*str == '-')
291 		goto out;
292 
293 	while (*str) {
294 		switch (tolower(*str)) {
295 		case'p':
296 			__page_init_poisoning = true;
297 			break;
298 		default:
299 			pr_err("vm_debug option '%c' unknown. skipped\n",
300 			       *str);
301 		}
302 
303 		str++;
304 	}
305 out:
306 	if (page_init_poisoning && !__page_init_poisoning)
307 		pr_warn("Page struct poisoning disabled by kernel command line option 'vm_debug'\n");
308 
309 	page_init_poisoning = __page_init_poisoning;
310 
311 	return 1;
312 }
313 __setup("vm_debug", setup_vm_debug);
314 
315 void page_init_poison(struct page *page, size_t size)
316 {
317 	if (page_init_poisoning)
318 		memset(page, PAGE_POISON_PATTERN, size);
319 }
320 EXPORT_SYMBOL_GPL(page_init_poison);
321 #endif		/* CONFIG_DEBUG_VM */
322