Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 */
6#include <linux/sched/debug.h>
7#include <linux/kallsyms.h>
8#include <linux/kprobes.h>
9#include <linux/uaccess.h>
10#include <linux/hardirq.h>
11#include <linux/kdebug.h>
12#include <linux/export.h>
13#include <linux/ptrace.h>
14#include <linux/kexec.h>
15#include <linux/sysfs.h>
16#include <linux/bug.h>
17#include <linux/nmi.h>
18
19#include <asm/cpu_entry_area.h>
20#include <asm/stacktrace.h>
21
22static const char * const exception_stack_names[] = {
23 [ ESTACK_DF ] = "#DF",
24 [ ESTACK_NMI ] = "NMI",
25 [ ESTACK_DB ] = "#DB",
26 [ ESTACK_MCE ] = "#MC",
27 [ ESTACK_VC ] = "#VC",
28 [ ESTACK_VC2 ] = "#VC2",
29};
30
31const char *stack_type_name(enum stack_type type)
32{
33 BUILD_BUG_ON(N_EXCEPTION_STACKS != 6);
34
35 if (type == STACK_TYPE_TASK)
36 return "TASK";
37
38 if (type == STACK_TYPE_IRQ)
39 return "IRQ";
40
41 if (type == STACK_TYPE_SOFTIRQ)
42 return "SOFTIRQ";
43
44 if (type == STACK_TYPE_ENTRY) {
45 /*
46 * On 64-bit, we have a generic entry stack that we
47 * use for all the kernel entry points, including
48 * SYSENTER.
49 */
50 return "ENTRY_TRAMPOLINE";
51 }
52
53 if (type >= STACK_TYPE_EXCEPTION && type <= STACK_TYPE_EXCEPTION_LAST)
54 return exception_stack_names[type - STACK_TYPE_EXCEPTION];
55
56 return NULL;
57}
58
59/**
60 * struct estack_pages - Page descriptor for exception stacks
61 * @offs: Offset from the start of the exception stack area
62 * @size: Size of the exception stack
63 * @type: Type to store in the stack_info struct
64 */
65struct estack_pages {
66 u32 offs;
67 u16 size;
68 u16 type;
69};
70
71#define EPAGERANGE(st) \
72 [PFN_DOWN(CEA_ESTACK_OFFS(st)) ... \
73 PFN_DOWN(CEA_ESTACK_OFFS(st) + CEA_ESTACK_SIZE(st) - 1)] = { \
74 .offs = CEA_ESTACK_OFFS(st), \
75 .size = CEA_ESTACK_SIZE(st), \
76 .type = STACK_TYPE_EXCEPTION + ESTACK_ ##st, }
77
78/*
79 * Array of exception stack page descriptors. If the stack is larger than
80 * PAGE_SIZE, all pages covering a particular stack will have the same
81 * info. The guard pages including the not mapped DB2 stack are zeroed
82 * out.
83 */
84static const
85struct estack_pages estack_pages[CEA_ESTACK_PAGES] ____cacheline_aligned = {
86 EPAGERANGE(DF),
87 EPAGERANGE(NMI),
88 EPAGERANGE(DB),
89 EPAGERANGE(MCE),
90 EPAGERANGE(VC),
91 EPAGERANGE(VC2),
92};
93
94static __always_inline bool in_exception_stack(unsigned long *stack, struct stack_info *info)
95{
96 unsigned long begin, end, stk = (unsigned long)stack;
97 const struct estack_pages *ep;
98 struct pt_regs *regs;
99 unsigned int k;
100
101 BUILD_BUG_ON(N_EXCEPTION_STACKS != 6);
102
103 begin = (unsigned long)__this_cpu_read(cea_exception_stacks);
104 /*
105 * Handle the case where stack trace is collected _before_
106 * cea_exception_stacks had been initialized.
107 */
108 if (!begin)
109 return false;
110
111 end = begin + sizeof(struct cea_exception_stacks);
112 /* Bail if @stack is outside the exception stack area. */
113 if (stk < begin || stk >= end)
114 return false;
115
116 /* Calc page offset from start of exception stacks */
117 k = (stk - begin) >> PAGE_SHIFT;
118 /* Lookup the page descriptor */
119 ep = &estack_pages[k];
120 /* Guard page? */
121 if (!ep->size)
122 return false;
123
124 begin += (unsigned long)ep->offs;
125 end = begin + (unsigned long)ep->size;
126 regs = (struct pt_regs *)end - 1;
127
128 info->type = ep->type;
129 info->begin = (unsigned long *)begin;
130 info->end = (unsigned long *)end;
131 info->next_sp = (unsigned long *)regs->sp;
132 return true;
133}
134
135static __always_inline bool in_irq_stack(unsigned long *stack, struct stack_info *info)
136{
137 unsigned long *end = (unsigned long *)this_cpu_read(pcpu_hot.hardirq_stack_ptr);
138 unsigned long *begin;
139
140 /*
141 * @end points directly to the top most stack entry to avoid a -8
142 * adjustment in the stack switch hotpath. Adjust it back before
143 * calculating @begin.
144 */
145 end++;
146 begin = end - (IRQ_STACK_SIZE / sizeof(long));
147
148 /*
149 * Due to the switching logic RSP can never be == @end because the
150 * final operation is 'popq %rsp' which means after that RSP points
151 * to the original stack and not to @end.
152 */
153 if (stack < begin || stack >= end)
154 return false;
155
156 info->type = STACK_TYPE_IRQ;
157 info->begin = begin;
158 info->end = end;
159
160 /*
161 * The next stack pointer is stored at the top of the irq stack
162 * before switching to the irq stack. Actual stack entries are all
163 * below that.
164 */
165 info->next_sp = (unsigned long *)*(end - 1);
166
167 return true;
168}
169
170bool noinstr get_stack_info_noinstr(unsigned long *stack, struct task_struct *task,
171 struct stack_info *info)
172{
173 if (in_task_stack(stack, task, info))
174 return true;
175
176 if (task != current)
177 return false;
178
179 if (in_exception_stack(stack, info))
180 return true;
181
182 if (in_irq_stack(stack, info))
183 return true;
184
185 if (in_entry_stack(stack, info))
186 return true;
187
188 return false;
189}
190
191int get_stack_info(unsigned long *stack, struct task_struct *task,
192 struct stack_info *info, unsigned long *visit_mask)
193{
194 task = task ? : current;
195
196 if (!stack)
197 goto unknown;
198
199 if (!get_stack_info_noinstr(stack, task, info))
200 goto unknown;
201
202 /*
203 * Make sure we don't iterate through any given stack more than once.
204 * If it comes up a second time then there's something wrong going on:
205 * just break out and report an unknown stack type.
206 */
207 if (visit_mask) {
208 if (*visit_mask & (1UL << info->type)) {
209 if (task == current)
210 printk_deferred_once(KERN_WARNING "WARNING: stack recursion on stack type %d\n", info->type);
211 goto unknown;
212 }
213 *visit_mask |= 1UL << info->type;
214 }
215
216 return 0;
217
218unknown:
219 info->type = STACK_TYPE_UNKNOWN;
220 return -EINVAL;
221}
1/*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 */
5#include <linux/kallsyms.h>
6#include <linux/kprobes.h>
7#include <linux/uaccess.h>
8#include <linux/hardirq.h>
9#include <linux/kdebug.h>
10#include <linux/module.h>
11#include <linux/ptrace.h>
12#include <linux/kexec.h>
13#include <linux/sysfs.h>
14#include <linux/bug.h>
15#include <linux/nmi.h>
16
17#include <asm/stacktrace.h>
18
19
20#define N_EXCEPTION_STACKS_END \
21 (N_EXCEPTION_STACKS + DEBUG_STKSZ/EXCEPTION_STKSZ - 2)
22
23static char x86_stack_ids[][8] = {
24 [ DEBUG_STACK-1 ] = "#DB",
25 [ NMI_STACK-1 ] = "NMI",
26 [ DOUBLEFAULT_STACK-1 ] = "#DF",
27 [ MCE_STACK-1 ] = "#MC",
28#if DEBUG_STKSZ > EXCEPTION_STKSZ
29 [ N_EXCEPTION_STACKS ...
30 N_EXCEPTION_STACKS_END ] = "#DB[?]"
31#endif
32};
33
34static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
35 unsigned *usedp, char **idp)
36{
37 unsigned k;
38
39 /*
40 * Iterate over all exception stacks, and figure out whether
41 * 'stack' is in one of them:
42 */
43 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
44 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
45 /*
46 * Is 'stack' above this exception frame's end?
47 * If yes then skip to the next frame.
48 */
49 if (stack >= end)
50 continue;
51 /*
52 * Is 'stack' above this exception frame's start address?
53 * If yes then we found the right frame.
54 */
55 if (stack >= end - EXCEPTION_STKSZ) {
56 /*
57 * Make sure we only iterate through an exception
58 * stack once. If it comes up for the second time
59 * then there's something wrong going on - just
60 * break out and return NULL:
61 */
62 if (*usedp & (1U << k))
63 break;
64 *usedp |= 1U << k;
65 *idp = x86_stack_ids[k];
66 return (unsigned long *)end;
67 }
68 /*
69 * If this is a debug stack, and if it has a larger size than
70 * the usual exception stacks, then 'stack' might still
71 * be within the lower portion of the debug stack:
72 */
73#if DEBUG_STKSZ > EXCEPTION_STKSZ
74 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
75 unsigned j = N_EXCEPTION_STACKS - 1;
76
77 /*
78 * Black magic. A large debug stack is composed of
79 * multiple exception stack entries, which we
80 * iterate through now. Dont look:
81 */
82 do {
83 ++j;
84 end -= EXCEPTION_STKSZ;
85 x86_stack_ids[j][4] = '1' +
86 (j - N_EXCEPTION_STACKS);
87 } while (stack < end - EXCEPTION_STKSZ);
88 if (*usedp & (1U << j))
89 break;
90 *usedp |= 1U << j;
91 *idp = x86_stack_ids[j];
92 return (unsigned long *)end;
93 }
94#endif
95 }
96 return NULL;
97}
98
99static inline int
100in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
101 unsigned long *irq_stack_end)
102{
103 return (stack >= irq_stack && stack < irq_stack_end);
104}
105
106static const unsigned long irq_stack_size =
107 (IRQ_STACK_SIZE - 64) / sizeof(unsigned long);
108
109enum stack_type {
110 STACK_IS_UNKNOWN,
111 STACK_IS_NORMAL,
112 STACK_IS_EXCEPTION,
113 STACK_IS_IRQ,
114};
115
116static enum stack_type
117analyze_stack(int cpu, struct task_struct *task, unsigned long *stack,
118 unsigned long **stack_end, unsigned long *irq_stack,
119 unsigned *used, char **id)
120{
121 unsigned long addr;
122
123 addr = ((unsigned long)stack & (~(THREAD_SIZE - 1)));
124 if ((unsigned long)task_stack_page(task) == addr)
125 return STACK_IS_NORMAL;
126
127 *stack_end = in_exception_stack(cpu, (unsigned long)stack,
128 used, id);
129 if (*stack_end)
130 return STACK_IS_EXCEPTION;
131
132 if (!irq_stack)
133 return STACK_IS_NORMAL;
134
135 *stack_end = irq_stack;
136 irq_stack = irq_stack - irq_stack_size;
137
138 if (in_irq_stack(stack, irq_stack, *stack_end))
139 return STACK_IS_IRQ;
140
141 return STACK_IS_UNKNOWN;
142}
143
144/*
145 * x86-64 can have up to three kernel stacks:
146 * process stack
147 * interrupt stack
148 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
149 */
150
151void dump_trace(struct task_struct *task, struct pt_regs *regs,
152 unsigned long *stack, unsigned long bp,
153 const struct stacktrace_ops *ops, void *data)
154{
155 const unsigned cpu = get_cpu();
156 struct thread_info *tinfo;
157 unsigned long *irq_stack = (unsigned long *)per_cpu(irq_stack_ptr, cpu);
158 unsigned long dummy;
159 unsigned used = 0;
160 int graph = 0;
161 int done = 0;
162
163 if (!task)
164 task = current;
165
166 if (!stack) {
167 if (regs)
168 stack = (unsigned long *)regs->sp;
169 else if (task != current)
170 stack = (unsigned long *)task->thread.sp;
171 else
172 stack = &dummy;
173 }
174
175 if (!bp)
176 bp = stack_frame(task, regs);
177 /*
178 * Print function call entries in all stacks, starting at the
179 * current stack address. If the stacks consist of nested
180 * exceptions
181 */
182 tinfo = task_thread_info(task);
183 while (!done) {
184 unsigned long *stack_end;
185 enum stack_type stype;
186 char *id;
187
188 stype = analyze_stack(cpu, task, stack, &stack_end,
189 irq_stack, &used, &id);
190
191 /* Default finish unless specified to continue */
192 done = 1;
193
194 switch (stype) {
195
196 /* Break out early if we are on the thread stack */
197 case STACK_IS_NORMAL:
198 break;
199
200 case STACK_IS_EXCEPTION:
201
202 if (ops->stack(data, id) < 0)
203 break;
204
205 bp = ops->walk_stack(tinfo, stack, bp, ops,
206 data, stack_end, &graph);
207 ops->stack(data, "<EOE>");
208 /*
209 * We link to the next stack via the
210 * second-to-last pointer (index -2 to end) in the
211 * exception stack:
212 */
213 stack = (unsigned long *) stack_end[-2];
214 done = 0;
215 break;
216
217 case STACK_IS_IRQ:
218
219 if (ops->stack(data, "IRQ") < 0)
220 break;
221 bp = ops->walk_stack(tinfo, stack, bp,
222 ops, data, stack_end, &graph);
223 /*
224 * We link to the next stack (which would be
225 * the process stack normally) the last
226 * pointer (index -1 to end) in the IRQ stack:
227 */
228 stack = (unsigned long *) (stack_end[-1]);
229 irq_stack = NULL;
230 ops->stack(data, "EOI");
231 done = 0;
232 break;
233
234 case STACK_IS_UNKNOWN:
235 ops->stack(data, "UNK");
236 break;
237 }
238 }
239
240 /*
241 * This handles the process stack:
242 */
243 bp = ops->walk_stack(tinfo, stack, bp, ops, data, NULL, &graph);
244 put_cpu();
245}
246EXPORT_SYMBOL(dump_trace);
247
248void
249show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
250 unsigned long *sp, unsigned long bp, char *log_lvl)
251{
252 unsigned long *irq_stack_end;
253 unsigned long *irq_stack;
254 unsigned long *stack;
255 int cpu;
256 int i;
257
258 preempt_disable();
259 cpu = smp_processor_id();
260
261 irq_stack_end = (unsigned long *)(per_cpu(irq_stack_ptr, cpu));
262 irq_stack = (unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
263
264 /*
265 * Debugging aid: "show_stack(NULL, NULL);" prints the
266 * back trace for this cpu:
267 */
268 if (sp == NULL) {
269 if (task)
270 sp = (unsigned long *)task->thread.sp;
271 else
272 sp = (unsigned long *)&sp;
273 }
274
275 stack = sp;
276 for (i = 0; i < kstack_depth_to_print; i++) {
277 if (stack >= irq_stack && stack <= irq_stack_end) {
278 if (stack == irq_stack_end) {
279 stack = (unsigned long *) (irq_stack_end[-1]);
280 pr_cont(" <EOI> ");
281 }
282 } else {
283 if (kstack_end(stack))
284 break;
285 }
286 if ((i % STACKSLOTS_PER_LINE) == 0) {
287 if (i != 0)
288 pr_cont("\n");
289 printk("%s %016lx", log_lvl, *stack++);
290 } else
291 pr_cont(" %016lx", *stack++);
292 touch_nmi_watchdog();
293 }
294 preempt_enable();
295
296 pr_cont("\n");
297 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
298}
299
300void show_regs(struct pt_regs *regs)
301{
302 int i;
303 unsigned long sp;
304
305 sp = regs->sp;
306 show_regs_print_info(KERN_DEFAULT);
307 __show_regs(regs, 1);
308
309 /*
310 * When in-kernel, we also print out the stack and code at the
311 * time of the fault..
312 */
313 if (!user_mode(regs)) {
314 unsigned int code_prologue = code_bytes * 43 / 64;
315 unsigned int code_len = code_bytes;
316 unsigned char c;
317 u8 *ip;
318
319 printk(KERN_DEFAULT "Stack:\n");
320 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
321 0, KERN_DEFAULT);
322
323 printk(KERN_DEFAULT "Code: ");
324
325 ip = (u8 *)regs->ip - code_prologue;
326 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
327 /* try starting at IP */
328 ip = (u8 *)regs->ip;
329 code_len = code_len - code_prologue + 1;
330 }
331 for (i = 0; i < code_len; i++, ip++) {
332 if (ip < (u8 *)PAGE_OFFSET ||
333 probe_kernel_address(ip, c)) {
334 pr_cont(" Bad RIP value.");
335 break;
336 }
337 if (ip == (u8 *)regs->ip)
338 pr_cont("<%02x> ", c);
339 else
340 pr_cont("%02x ", c);
341 }
342 }
343 pr_cont("\n");
344}
345
346int is_valid_bugaddr(unsigned long ip)
347{
348 unsigned short ud2;
349
350 if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
351 return 0;
352
353 return ud2 == 0x0b0f;
354}