xref: /linux-6.15/arch/powerpc/kernel/stacktrace.c (revision 3de27dcf)
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_pointer();
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 (tsk == current)
71 		sp = current_stack_pointer();
72 	else
73 		sp = tsk->thread.ksp;
74 
75 	save_context_stack(trace, sp, tsk, 0);
76 }
77 EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
78 
79 void
80 save_stack_trace_regs(struct pt_regs *regs, struct stack_trace *trace)
81 {
82 	save_context_stack(trace, regs->gpr[1], current, 0);
83 }
84 EXPORT_SYMBOL_GPL(save_stack_trace_regs);
85 
86 #ifdef CONFIG_HAVE_RELIABLE_STACKTRACE
87 /*
88  * This function returns an error if it detects any unreliable features of the
89  * stack.  Otherwise it guarantees that the stack trace is reliable.
90  *
91  * If the task is not 'current', the caller *must* ensure the task is inactive.
92  */
93 int
94 save_stack_trace_tsk_reliable(struct task_struct *tsk,
95 				struct stack_trace *trace)
96 {
97 	unsigned long sp;
98 	unsigned long newsp;
99 	unsigned long stack_page = (unsigned long)task_stack_page(tsk);
100 	unsigned long stack_end;
101 	int graph_idx = 0;
102 	bool firstframe;
103 
104 	stack_end = stack_page + THREAD_SIZE;
105 	if (!is_idle_task(tsk)) {
106 		/*
107 		 * For user tasks, this is the SP value loaded on
108 		 * kernel entry, see "PACAKSAVE(r13)" in _switch() and
109 		 * system_call_common()/EXCEPTION_PROLOG_COMMON().
110 		 *
111 		 * Likewise for non-swapper kernel threads,
112 		 * this also happens to be the top of the stack
113 		 * as setup by copy_thread().
114 		 *
115 		 * Note that stack backlinks are not properly setup by
116 		 * copy_thread() and thus, a forked task() will have
117 		 * an unreliable stack trace until it's been
118 		 * _switch()'ed to for the first time.
119 		 */
120 		stack_end -= STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
121 	} else {
122 		/*
123 		 * idle tasks have a custom stack layout,
124 		 * c.f. cpu_idle_thread_init().
125 		 */
126 		stack_end -= STACK_FRAME_OVERHEAD;
127 	}
128 
129 	if (tsk == current)
130 		sp = current_stack_pointer();
131 	else
132 		sp = tsk->thread.ksp;
133 
134 	if (sp < stack_page + sizeof(struct thread_struct) ||
135 	    sp > stack_end - STACK_FRAME_MIN_SIZE) {
136 		return -EINVAL;
137 	}
138 
139 	for (firstframe = true; sp != stack_end;
140 	     firstframe = false, sp = newsp) {
141 		unsigned long *stack = (unsigned long *) sp;
142 		unsigned long ip;
143 
144 		/* sanity check: ABI requires SP to be aligned 16 bytes. */
145 		if (sp & 0xF)
146 			return -EINVAL;
147 
148 		newsp = stack[0];
149 		/* Stack grows downwards; unwinder may only go up. */
150 		if (newsp <= sp)
151 			return -EINVAL;
152 
153 		if (newsp != stack_end &&
154 		    newsp > stack_end - STACK_FRAME_MIN_SIZE) {
155 			return -EINVAL; /* invalid backlink, too far up. */
156 		}
157 
158 		/*
159 		 * We can only trust the bottom frame's backlink, the
160 		 * rest of the frame may be uninitialized, continue to
161 		 * the next.
162 		 */
163 		if (firstframe)
164 			continue;
165 
166 		/* Mark stacktraces with exception frames as unreliable. */
167 		if (sp <= stack_end - STACK_INT_FRAME_SIZE &&
168 		    stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
169 			return -EINVAL;
170 		}
171 
172 		/* Examine the saved LR: it must point into kernel code. */
173 		ip = stack[STACK_FRAME_LR_SAVE];
174 		if (!__kernel_text_address(ip))
175 			return -EINVAL;
176 
177 		/*
178 		 * FIXME: IMHO these tests do not belong in
179 		 * arch-dependent code, they are generic.
180 		 */
181 		ip = ftrace_graph_ret_addr(tsk, &graph_idx, ip, NULL);
182 #ifdef CONFIG_KPROBES
183 		/*
184 		 * Mark stacktraces with kretprobed functions on them
185 		 * as unreliable.
186 		 */
187 		if (ip == (unsigned long)kretprobe_trampoline)
188 			return -EINVAL;
189 #endif
190 
191 		if (trace->nr_entries >= trace->max_entries)
192 			return -E2BIG;
193 		if (!trace->skip)
194 			trace->entries[trace->nr_entries++] = ip;
195 		else
196 			trace->skip--;
197 	}
198 	return 0;
199 }
200 EXPORT_SYMBOL_GPL(save_stack_trace_tsk_reliable);
201 #endif /* CONFIG_HAVE_RELIABLE_STACKTRACE */
202 
203 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_NMI_IPI)
204 static void handle_backtrace_ipi(struct pt_regs *regs)
205 {
206 	nmi_cpu_backtrace(regs);
207 }
208 
209 static void raise_backtrace_ipi(cpumask_t *mask)
210 {
211 	unsigned int cpu;
212 
213 	for_each_cpu(cpu, mask) {
214 		if (cpu == smp_processor_id())
215 			handle_backtrace_ipi(NULL);
216 		else
217 			smp_send_safe_nmi_ipi(cpu, handle_backtrace_ipi, 5 * USEC_PER_SEC);
218 	}
219 
220 	for_each_cpu(cpu, mask) {
221 		struct paca_struct *p = paca_ptrs[cpu];
222 
223 		cpumask_clear_cpu(cpu, mask);
224 
225 		pr_warn("CPU %d didn't respond to backtrace IPI, inspecting paca.\n", cpu);
226 		if (!virt_addr_valid(p)) {
227 			pr_warn("paca pointer appears corrupt? (%px)\n", p);
228 			continue;
229 		}
230 
231 		pr_warn("irq_soft_mask: 0x%02x in_mce: %d in_nmi: %d",
232 			p->irq_soft_mask, p->in_mce, p->in_nmi);
233 
234 		if (virt_addr_valid(p->__current))
235 			pr_cont(" current: %d (%s)\n", p->__current->pid,
236 				p->__current->comm);
237 		else
238 			pr_cont(" current pointer corrupt? (%px)\n", p->__current);
239 
240 		pr_warn("Back trace of paca->saved_r1 (0x%016llx) (possibly stale):\n", p->saved_r1);
241 		show_stack(p->__current, (unsigned long *)p->saved_r1);
242 	}
243 }
244 
245 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
246 {
247 	nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace_ipi);
248 }
249 #endif /* defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_NMI_IPI) */
250