xref: /openbmc/linux/arch/powerpc/kernel/stacktrace.c (revision 15e3ae36)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 /*
4  * Stack trace utility functions etc.
5  *
6  * Copyright 2008 Christoph Hellwig, IBM Corp.
7  * Copyright 2018 SUSE Linux GmbH
8  * Copyright 2018 Nick Piggin, Michael Ellerman, IBM Corp.
9  */
10 
11 #include <linux/export.h>
12 #include <linux/kallsyms.h>
13 #include <linux/module.h>
14 #include <linux/nmi.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/sched/task_stack.h>
18 #include <linux/stacktrace.h>
19 #include <asm/ptrace.h>
20 #include <asm/processor.h>
21 #include <linux/ftrace.h>
22 #include <asm/kprobes.h>
23 
24 #include <asm/paca.h>
25 
26 /*
27  * Save stack-backtrace addresses into a stack_trace buffer.
28  */
29 static void save_context_stack(struct stack_trace *trace, unsigned long sp,
30 			struct task_struct *tsk, int savesched)
31 {
32 	for (;;) {
33 		unsigned long *stack = (unsigned long *) sp;
34 		unsigned long newsp, ip;
35 
36 		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
37 			return;
38 
39 		newsp = stack[0];
40 		ip = stack[STACK_FRAME_LR_SAVE];
41 
42 		if (savesched || !in_sched_functions(ip)) {
43 			if (!trace->skip)
44 				trace->entries[trace->nr_entries++] = ip;
45 			else
46 				trace->skip--;
47 		}
48 
49 		if (trace->nr_entries >= trace->max_entries)
50 			return;
51 
52 		sp = newsp;
53 	}
54 }
55 
56 void save_stack_trace(struct stack_trace *trace)
57 {
58 	unsigned long sp;
59 
60 	sp = current_stack_frame();
61 
62 	save_context_stack(trace, sp, current, 1);
63 }
64 EXPORT_SYMBOL_GPL(save_stack_trace);
65 
66 void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
67 {
68 	unsigned long sp;
69 
70 	if (!try_get_task_stack(tsk))
71 		return;
72 
73 	if (tsk == current)
74 		sp = current_stack_frame();
75 	else
76 		sp = tsk->thread.ksp;
77 
78 	save_context_stack(trace, sp, tsk, 0);
79 
80 	put_task_stack(tsk);
81 }
82 EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
83 
84 void
85 save_stack_trace_regs(struct pt_regs *regs, struct stack_trace *trace)
86 {
87 	save_context_stack(trace, regs->gpr[1], current, 0);
88 }
89 EXPORT_SYMBOL_GPL(save_stack_trace_regs);
90 
91 #ifdef CONFIG_HAVE_RELIABLE_STACKTRACE
92 /*
93  * This function returns an error if it detects any unreliable features of the
94  * stack.  Otherwise it guarantees that the stack trace is reliable.
95  *
96  * If the task is not 'current', the caller *must* ensure the task is inactive.
97  */
98 static int __save_stack_trace_tsk_reliable(struct task_struct *tsk,
99 					   struct stack_trace *trace)
100 {
101 	unsigned long sp;
102 	unsigned long newsp;
103 	unsigned long stack_page = (unsigned long)task_stack_page(tsk);
104 	unsigned long stack_end;
105 	int graph_idx = 0;
106 	bool firstframe;
107 
108 	stack_end = stack_page + THREAD_SIZE;
109 	if (!is_idle_task(tsk)) {
110 		/*
111 		 * For user tasks, this is the SP value loaded on
112 		 * kernel entry, see "PACAKSAVE(r13)" in _switch() and
113 		 * system_call_common()/EXCEPTION_PROLOG_COMMON().
114 		 *
115 		 * Likewise for non-swapper kernel threads,
116 		 * this also happens to be the top of the stack
117 		 * as setup by copy_thread().
118 		 *
119 		 * Note that stack backlinks are not properly setup by
120 		 * copy_thread() and thus, a forked task() will have
121 		 * an unreliable stack trace until it's been
122 		 * _switch()'ed to for the first time.
123 		 */
124 		stack_end -= STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
125 	} else {
126 		/*
127 		 * idle tasks have a custom stack layout,
128 		 * c.f. cpu_idle_thread_init().
129 		 */
130 		stack_end -= STACK_FRAME_OVERHEAD;
131 	}
132 
133 	if (tsk == current)
134 		sp = current_stack_frame();
135 	else
136 		sp = tsk->thread.ksp;
137 
138 	if (sp < stack_page + sizeof(struct thread_struct) ||
139 	    sp > stack_end - STACK_FRAME_MIN_SIZE) {
140 		return -EINVAL;
141 	}
142 
143 	for (firstframe = true; sp != stack_end;
144 	     firstframe = false, sp = newsp) {
145 		unsigned long *stack = (unsigned long *) sp;
146 		unsigned long ip;
147 
148 		/* sanity check: ABI requires SP to be aligned 16 bytes. */
149 		if (sp & 0xF)
150 			return -EINVAL;
151 
152 		newsp = stack[0];
153 		/* Stack grows downwards; unwinder may only go up. */
154 		if (newsp <= sp)
155 			return -EINVAL;
156 
157 		if (newsp != stack_end &&
158 		    newsp > stack_end - STACK_FRAME_MIN_SIZE) {
159 			return -EINVAL; /* invalid backlink, too far up. */
160 		}
161 
162 		/*
163 		 * We can only trust the bottom frame's backlink, the
164 		 * rest of the frame may be uninitialized, continue to
165 		 * the next.
166 		 */
167 		if (firstframe)
168 			continue;
169 
170 		/* Mark stacktraces with exception frames as unreliable. */
171 		if (sp <= stack_end - STACK_INT_FRAME_SIZE &&
172 		    stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
173 			return -EINVAL;
174 		}
175 
176 		/* Examine the saved LR: it must point into kernel code. */
177 		ip = stack[STACK_FRAME_LR_SAVE];
178 		if (!__kernel_text_address(ip))
179 			return -EINVAL;
180 
181 		/*
182 		 * FIXME: IMHO these tests do not belong in
183 		 * arch-dependent code, they are generic.
184 		 */
185 		ip = ftrace_graph_ret_addr(tsk, &graph_idx, ip, stack);
186 #ifdef CONFIG_KPROBES
187 		/*
188 		 * Mark stacktraces with kretprobed functions on them
189 		 * as unreliable.
190 		 */
191 		if (ip == (unsigned long)kretprobe_trampoline)
192 			return -EINVAL;
193 #endif
194 
195 		if (trace->nr_entries >= trace->max_entries)
196 			return -E2BIG;
197 		if (!trace->skip)
198 			trace->entries[trace->nr_entries++] = ip;
199 		else
200 			trace->skip--;
201 	}
202 	return 0;
203 }
204 
205 int save_stack_trace_tsk_reliable(struct task_struct *tsk,
206 				  struct stack_trace *trace)
207 {
208 	int ret;
209 
210 	/*
211 	 * If the task doesn't have a stack (e.g., a zombie), the stack is
212 	 * "reliably" empty.
213 	 */
214 	if (!try_get_task_stack(tsk))
215 		return 0;
216 
217 	ret = __save_stack_trace_tsk_reliable(tsk, trace);
218 
219 	put_task_stack(tsk);
220 
221 	return ret;
222 }
223 #endif /* CONFIG_HAVE_RELIABLE_STACKTRACE */
224 
225 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_NMI_IPI)
226 static void handle_backtrace_ipi(struct pt_regs *regs)
227 {
228 	nmi_cpu_backtrace(regs);
229 }
230 
231 static void raise_backtrace_ipi(cpumask_t *mask)
232 {
233 	unsigned int cpu;
234 
235 	for_each_cpu(cpu, mask) {
236 		if (cpu == smp_processor_id())
237 			handle_backtrace_ipi(NULL);
238 		else
239 			smp_send_safe_nmi_ipi(cpu, handle_backtrace_ipi, 5 * USEC_PER_SEC);
240 	}
241 
242 	for_each_cpu(cpu, mask) {
243 		struct paca_struct *p = paca_ptrs[cpu];
244 
245 		cpumask_clear_cpu(cpu, mask);
246 
247 		pr_warn("CPU %d didn't respond to backtrace IPI, inspecting paca.\n", cpu);
248 		if (!virt_addr_valid(p)) {
249 			pr_warn("paca pointer appears corrupt? (%px)\n", p);
250 			continue;
251 		}
252 
253 		pr_warn("irq_soft_mask: 0x%02x in_mce: %d in_nmi: %d",
254 			p->irq_soft_mask, p->in_mce, p->in_nmi);
255 
256 		if (virt_addr_valid(p->__current))
257 			pr_cont(" current: %d (%s)\n", p->__current->pid,
258 				p->__current->comm);
259 		else
260 			pr_cont(" current pointer corrupt? (%px)\n", p->__current);
261 
262 		pr_warn("Back trace of paca->saved_r1 (0x%016llx) (possibly stale):\n", p->saved_r1);
263 		show_stack(p->__current, (unsigned long *)p->saved_r1);
264 	}
265 }
266 
267 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
268 {
269 	nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace_ipi);
270 }
271 #endif /* defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_NMI_IPI) */
272