[RFC PATCH v1 6/9] arm64: unwinder: Add a reliability check in the unwinder based on DWARF CFI

[madvenka at linux.microsoft.com]

From: "Madhavan T. Venkataraman" <madvenka at linux.microsoft.com>

Introduce a reliability flag in struct stackframe. This will be set to false
if the PC does not have valid DWARF rules or if the frame pointer computed
from the DWARF rules does not match the actual frame pointer.

Now that the unwinder can validate the frame pointer, introduce
arch_stack_walk_reliable().

Signed-off-by: Madhavan T. Venkataraman <madvenka at linux.microsoft.com>
---
 arch/arm64/include/asm/stacktrace.h |  6 +++
 arch/arm64/kernel/stacktrace.c      | 69 +++++++++++++++++++++++++++++
 2 files changed, 75 insertions(+)

diff --git a/arch/arm64/include/asm/stacktrace.h b/arch/arm64/include/asm/stacktrace.h
index 6564a01cc085..93adee4219ed 100644
--- a/arch/arm64/include/asm/stacktrace.h
+++ b/arch/arm64/include/asm/stacktrace.h
@@ -5,6 +5,7 @@
 #ifndef __ASM_STACKTRACE_H
 #define __ASM_STACKTRACE_H
 
+#include <linux/dwarf.h>
 #include <linux/percpu.h>
 #include <linux/sched.h>
 #include <linux/sched/task_stack.h>
@@ -35,6 +36,7 @@ struct stack_info {
  * A snapshot of a frame record or fp/lr register values, along with some
  * accounting information necessary for robust unwinding.
  *
+ * @sp:          The sp value (CFA) at the call site of the current function.
  * @fp:          The fp value in the frame record (or the real fp)
  * @pc:          The lr value in the frame record (or the real lr)
  *
@@ -47,8 +49,11 @@ struct stack_info {
  * @prev_type:   The type of stack this frame record was on, or a synthetic
  *               value of STACK_TYPE_UNKNOWN. This is used to detect a
  *               transition from one stack to another.
+ *
+ * @reliable     Stack trace is reliable.
  */
 struct stackframe {
+	unsigned long sp;
 	unsigned long fp;
 	unsigned long pc;
 	DECLARE_BITMAP(stacks_done, __NR_STACK_TYPES);
@@ -57,6 +62,7 @@ struct stackframe {
 #ifdef CONFIG_KRETPROBES
 	struct llist_node *kr_cur;
 #endif
+	bool reliable;
 };
 
 extern int unwind_frame(struct task_struct *tsk, struct stackframe *frame);
diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
index 94f83cd44e50..f9ef7a3e7296 100644
--- a/arch/arm64/kernel/stacktrace.c
+++ b/arch/arm64/kernel/stacktrace.c
@@ -5,6 +5,7 @@
  * Copyright (C) 2012 ARM Ltd.
  */
 #include <linux/kernel.h>
+#include <linux/dwarf.h>
 #include <linux/export.h>
 #include <linux/ftrace.h>
 #include <linux/kprobes.h>
@@ -36,8 +37,22 @@
 void start_backtrace(struct stackframe *frame, unsigned long fp,
 		     unsigned long pc)
 {
+	struct dwarf_rule *rule;
+
+	frame->reliable = true;
 	frame->fp = fp;
 	frame->pc = pc;
+	frame->sp = 0;
+	/*
+	 * Lookup the dwarf rule for PC. If it exists, initialize the SP
+	 * based on the frame pointer passed in.
+	 */
+	rule = dwarf_lookup(pc);
+	if (rule)
+		frame->sp = fp - rule->fp_offset;
+	else
+		frame->reliable = false;
+
 #ifdef CONFIG_KRETPROBES
 	frame->kr_cur = NULL;
 #endif
@@ -67,6 +82,8 @@ int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame)
 {
 	unsigned long fp = frame->fp;
 	struct stack_info info;
+	struct dwarf_rule *rule;
+	unsigned long lookup_pc;
 
 	if (!tsk)
 		tsk = current;
@@ -137,6 +154,32 @@ int notrace unwind_frame(struct task_struct *tsk, struct stackframe *frame)
 		frame->pc = kretprobe_find_ret_addr(tsk, (void *)frame->fp, &frame->kr_cur);
 #endif
 
+	/*
+	 * If it is the last frame, no need to check dwarf.
+	 */
+	if (frame->fp == (unsigned long)task_pt_regs(tsk)->stackframe)
+		return 0;
+
+	if (!frame->reliable) {
+		/*
+		 * The sp value cannot be reliably computed anymore because a
+		 * previous frame was unreliable.
+		 */
+		return 0;
+	}
+	lookup_pc = frame->pc;
+
+	rule = dwarf_lookup(lookup_pc);
+	if (!rule) {
+		frame->reliable = false;
+		return 0;
+	}
+
+	frame->sp += rule->sp_offset;
+	if (frame->fp != (frame->sp + rule->fp_offset)) {
+		frame->reliable = false;
+		return 0;
+	}
 	return 0;
 }
 NOKPROBE_SYMBOL(unwind_frame);
@@ -242,4 +285,30 @@ noinline notrace void arch_stack_walk(stack_trace_consume_fn consume_entry,
 	walk_stackframe(task, &frame, consume_entry, cookie);
 }
 
+noinline int arch_stack_walk_reliable(stack_trace_consume_fn consume_entry,
+			      void *cookie, struct task_struct *task)
+{
+	struct stackframe frame;
+	int ret = 0;
+
+	if (task == current) {
+		start_backtrace(&frame,
+				(unsigned long)__builtin_frame_address(1),
+				(unsigned long)__builtin_return_address(0));
+	} else {
+		start_backtrace(&frame, thread_saved_fp(task),
+				thread_saved_pc(task));
+	}
+
+	while (!ret) {
+		if (!frame.reliable)
+			return -EINVAL;
+		if (!consume_entry(cookie, frame.pc))
+			return -EINVAL;
+		ret = unwind_frame(task, &frame);
+	}
+
+	return ret == -ENOENT ? 0 : -EINVAL;
+}
+
 #endif
-- 
2.25.1