xref: /linux-6.15/arch/x86/kernel/dumpstack_64.c (revision 586bc5cc)
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/bug.h>
14 #include <linux/nmi.h>
15 #include <linux/sysfs.h>
16 
17 #include <asm/stacktrace.h>
18 
19 #include "dumpstack.h"
20 
21 
22 static char x86_stack_ids[][8] = {
23 		[DEBUG_STACK - 1] = "#DB",
24 		[NMI_STACK - 1] = "NMI",
25 		[DOUBLEFAULT_STACK - 1] = "#DF",
26 		[STACKFAULT_STACK - 1] = "#SS",
27 		[MCE_STACK - 1] = "#MC",
28 #if DEBUG_STKSZ > EXCEPTION_STKSZ
29 		[N_EXCEPTION_STACKS ...
30 			N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
31 #endif
32 	};
33 
34 int x86_is_stack_id(int id, char *name)
35 {
36 	return x86_stack_ids[id - 1] == name;
37 }
38 
39 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
40 					unsigned *usedp, char **idp)
41 {
42 	unsigned k;
43 
44 	/*
45 	 * Iterate over all exception stacks, and figure out whether
46 	 * 'stack' is in one of them:
47 	 */
48 	for (k = 0; k < N_EXCEPTION_STACKS; k++) {
49 		unsigned long end = per_cpu(orig_ist, cpu).ist[k];
50 		/*
51 		 * Is 'stack' above this exception frame's end?
52 		 * If yes then skip to the next frame.
53 		 */
54 		if (stack >= end)
55 			continue;
56 		/*
57 		 * Is 'stack' above this exception frame's start address?
58 		 * If yes then we found the right frame.
59 		 */
60 		if (stack >= end - EXCEPTION_STKSZ) {
61 			/*
62 			 * Make sure we only iterate through an exception
63 			 * stack once. If it comes up for the second time
64 			 * then there's something wrong going on - just
65 			 * break out and return NULL:
66 			 */
67 			if (*usedp & (1U << k))
68 				break;
69 			*usedp |= 1U << k;
70 			*idp = x86_stack_ids[k];
71 			return (unsigned long *)end;
72 		}
73 		/*
74 		 * If this is a debug stack, and if it has a larger size than
75 		 * the usual exception stacks, then 'stack' might still
76 		 * be within the lower portion of the debug stack:
77 		 */
78 #if DEBUG_STKSZ > EXCEPTION_STKSZ
79 		if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
80 			unsigned j = N_EXCEPTION_STACKS - 1;
81 
82 			/*
83 			 * Black magic. A large debug stack is composed of
84 			 * multiple exception stack entries, which we
85 			 * iterate through now. Dont look:
86 			 */
87 			do {
88 				++j;
89 				end -= EXCEPTION_STKSZ;
90 				x86_stack_ids[j][4] = '1' +
91 						(j - N_EXCEPTION_STACKS);
92 			} while (stack < end - EXCEPTION_STKSZ);
93 			if (*usedp & (1U << j))
94 				break;
95 			*usedp |= 1U << j;
96 			*idp = x86_stack_ids[j];
97 			return (unsigned long *)end;
98 		}
99 #endif
100 	}
101 	return NULL;
102 }
103 
104 static inline int
105 in_irq_stack(unsigned long *stack, unsigned long *irq_stack,
106 	     unsigned long *irq_stack_end)
107 {
108 	return (stack >= irq_stack && stack < irq_stack_end);
109 }
110 
111 /*
112  * We are returning from the irq stack and go to the previous one.
113  * If the previous stack is also in the irq stack, then bp in the first
114  * frame of the irq stack points to the previous, interrupted one.
115  * Otherwise we have another level of indirection: We first save
116  * the bp of the previous stack, then we switch the stack to the irq one
117  * and save a new bp that links to the previous one.
118  * (See save_args())
119  */
120 static inline unsigned long
121 fixup_bp_irq_link(unsigned long bp, unsigned long *stack,
122 		  unsigned long *irq_stack, unsigned long *irq_stack_end)
123 {
124 #ifdef CONFIG_FRAME_POINTER
125 	struct stack_frame *frame = (struct stack_frame *)bp;
126 
127 	if (!in_irq_stack(stack, irq_stack, irq_stack_end))
128 		return (unsigned long)frame->next_frame;
129 #endif
130 	return bp;
131 }
132 
133 /*
134  * x86-64 can have up to three kernel stacks:
135  * process stack
136  * interrupt stack
137  * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
138  */
139 
140 void dump_trace(struct task_struct *task, struct pt_regs *regs,
141 		unsigned long *stack, unsigned long bp,
142 		const struct stacktrace_ops *ops, void *data)
143 {
144 	const unsigned cpu = get_cpu();
145 	unsigned long *irq_stack_end =
146 		(unsigned long *)per_cpu(irq_stack_ptr, cpu);
147 	unsigned used = 0;
148 	struct thread_info *tinfo;
149 	int graph = 0;
150 
151 	if (!task)
152 		task = current;
153 
154 	if (!stack) {
155 		unsigned long dummy;
156 		stack = &dummy;
157 		if (task && task != current)
158 			stack = (unsigned long *)task->thread.sp;
159 	}
160 
161 #ifdef CONFIG_FRAME_POINTER
162 	if (!bp) {
163 		if (task == current) {
164 			/* Grab bp right from our regs */
165 			get_bp(bp);
166 		} else {
167 			/* bp is the last reg pushed by switch_to */
168 			bp = *(unsigned long *) task->thread.sp;
169 		}
170 	}
171 #endif
172 
173 	/*
174 	 * Print function call entries in all stacks, starting at the
175 	 * current stack address. If the stacks consist of nested
176 	 * exceptions
177 	 */
178 	tinfo = task_thread_info(task);
179 	for (;;) {
180 		char *id;
181 		unsigned long *estack_end;
182 		estack_end = in_exception_stack(cpu, (unsigned long)stack,
183 						&used, &id);
184 
185 		if (estack_end) {
186 			if (ops->stack(data, id) < 0)
187 				break;
188 
189 			bp = print_context_stack(tinfo, stack, bp, ops,
190 						 data, estack_end, &graph);
191 			ops->stack(data, "<EOE>");
192 			/*
193 			 * We link to the next stack via the
194 			 * second-to-last pointer (index -2 to end) in the
195 			 * exception stack:
196 			 */
197 			stack = (unsigned long *) estack_end[-2];
198 			continue;
199 		}
200 		if (irq_stack_end) {
201 			unsigned long *irq_stack;
202 			irq_stack = irq_stack_end -
203 				(IRQ_STACK_SIZE - 64) / sizeof(*irq_stack);
204 
205 			if (in_irq_stack(stack, irq_stack, irq_stack_end)) {
206 				if (ops->stack(data, "IRQ") < 0)
207 					break;
208 				bp = print_context_stack(tinfo, stack, bp,
209 					ops, data, irq_stack_end, &graph);
210 				/*
211 				 * We link to the next stack (which would be
212 				 * the process stack normally) the last
213 				 * pointer (index -1 to end) in the IRQ stack:
214 				 */
215 				stack = (unsigned long *) (irq_stack_end[-1]);
216 				bp = fixup_bp_irq_link(bp, stack, irq_stack,
217 						       irq_stack_end);
218 				irq_stack_end = NULL;
219 				ops->stack(data, "EOI");
220 				continue;
221 			}
222 		}
223 		break;
224 	}
225 
226 	/*
227 	 * This handles the process stack:
228 	 */
229 	bp = print_context_stack(tinfo, stack, bp, ops, data, NULL, &graph);
230 	put_cpu();
231 }
232 EXPORT_SYMBOL(dump_trace);
233 
234 void
235 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
236 		unsigned long *sp, unsigned long bp, char *log_lvl)
237 {
238 	unsigned long *stack;
239 	int i;
240 	const int cpu = smp_processor_id();
241 	unsigned long *irq_stack_end =
242 		(unsigned long *)(per_cpu(irq_stack_ptr, cpu));
243 	unsigned long *irq_stack =
244 		(unsigned long *)(per_cpu(irq_stack_ptr, cpu) - IRQ_STACK_SIZE);
245 
246 	/*
247 	 * debugging aid: "show_stack(NULL, NULL);" prints the
248 	 * back trace for this cpu.
249 	 */
250 
251 	if (sp == NULL) {
252 		if (task)
253 			sp = (unsigned long *)task->thread.sp;
254 		else
255 			sp = (unsigned long *)&sp;
256 	}
257 
258 	stack = sp;
259 	for (i = 0; i < kstack_depth_to_print; i++) {
260 		if (stack >= irq_stack && stack <= irq_stack_end) {
261 			if (stack == irq_stack_end) {
262 				stack = (unsigned long *) (irq_stack_end[-1]);
263 				printk(" <EOI> ");
264 			}
265 		} else {
266 		if (((long) stack & (THREAD_SIZE-1)) == 0)
267 			break;
268 		}
269 		if (i && ((i % STACKSLOTS_PER_LINE) == 0))
270 			printk("\n%s", log_lvl);
271 		printk(" %016lx", *stack++);
272 		touch_nmi_watchdog();
273 	}
274 	printk("\n");
275 	show_trace_log_lvl(task, regs, sp, bp, log_lvl);
276 }
277 
278 void show_registers(struct pt_regs *regs)
279 {
280 	int i;
281 	unsigned long sp;
282 	const int cpu = smp_processor_id();
283 	struct task_struct *cur = current;
284 
285 	sp = regs->sp;
286 	printk("CPU %d ", cpu);
287 	__show_regs(regs, 1);
288 	printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
289 		cur->comm, cur->pid, task_thread_info(cur), cur);
290 
291 	/*
292 	 * When in-kernel, we also print out the stack and code at the
293 	 * time of the fault..
294 	 */
295 	if (!user_mode(regs)) {
296 		unsigned int code_prologue = code_bytes * 43 / 64;
297 		unsigned int code_len = code_bytes;
298 		unsigned char c;
299 		u8 *ip;
300 
301 		printk(KERN_EMERG "Stack:\n");
302 		show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
303 				regs->bp, KERN_EMERG);
304 
305 		printk(KERN_EMERG "Code: ");
306 
307 		ip = (u8 *)regs->ip - code_prologue;
308 		if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
309 			/* try starting at IP */
310 			ip = (u8 *)regs->ip;
311 			code_len = code_len - code_prologue + 1;
312 		}
313 		for (i = 0; i < code_len; i++, ip++) {
314 			if (ip < (u8 *)PAGE_OFFSET ||
315 					probe_kernel_address(ip, c)) {
316 				printk(" Bad RIP value.");
317 				break;
318 			}
319 			if (ip == (u8 *)regs->ip)
320 				printk("<%02x> ", c);
321 			else
322 				printk("%02x ", c);
323 		}
324 	}
325 	printk("\n");
326 }
327 
328 int is_valid_bugaddr(unsigned long ip)
329 {
330 	unsigned short ud2;
331 
332 	if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
333 		return 0;
334 
335 	return ud2 == 0x0b0f;
336 }
337 
338