[PATCH 3/3] reliable stack trace support (i386)

[PATCH 3/3] reliable stack trace support (i386)

[Jan Beulich]

These are the i386-specific pieces to enable reliable stack traces. This is
going to be even more useful once CFI annotations get added to he assembly
code, namely to entry.S (a patch for that had been submitted several times).

Signed-off-by: Jan Beulich <jbeulich@novell.com>

--- linux-2.6.17-rc4/arch/i386/kernel/entry.S	2006-05-16 15:15:29.000000000 +0200
+++ 2.6.17-rc4-unwind-i386/arch/i386/kernel/entry.S	2006-05-16 15:16:46.000000000 +0200
@@ -663,6 +663,35 @@ ENTRY(spurious_interrupt_bug)
 	pushl $do_spurious_interrupt_bug
 	jmp error_code
 
+#ifdef CONFIG_STACK_UNWIND
+ENTRY(arch_unwind_init_running)
+	movl	4(%esp), %edx
+	movl	(%esp), %ecx
+	leal	4(%esp), %eax
+	movl	%ebx, EBX(%edx)
+	xorl	%ebx, %ebx
+	movl	%ebx, ECX(%edx)
+	movl	%ebx, EDX(%edx)
+	movl	%esi, ESI(%edx)
+	movl	%edi, EDI(%edx)
+	movl	%ebp, EBP(%edx)
+	movl	%ebx, EAX(%edx)
+	movl	$__USER_DS, DS(%edx)
+	movl	$__USER_DS, ES(%edx)
+	movl	%ebx, ORIG_EAX(%edx)
+	movl	%ecx, EIP(%edx)
+	movl	12(%esp), %ecx
+	movl	$__KERNEL_CS, CS(%edx)
+	movl	%ebx, EFLAGS(%edx)
+	movl	%eax, OLDESP(%edx)
+	movl	8(%esp), %eax
+	movl	%ecx, 8(%esp)
+	movl	EBX(%edx), %ebx
+	movl	$__KERNEL_DS, OLDSS(%edx)
+	jmpl	*%eax
+ENDPROC(arch_unwind_init_running)
+#endif
+
 .section .rodata,"a"
 #include "syscall_table.S"
 
--- linux-2.6.17-rc4/arch/i386/kernel/process.c	2006-05-16 15:15:29.000000000 +0200
+++ 2.6.17-rc4-unwind-i386/arch/i386/kernel/process.c	2006-05-16 15:16:46.000000000 +0200
@@ -312,7 +312,7 @@ void show_regs(struct pt_regs * regs)
 	cr3 = read_cr3();
 	cr4 = read_cr4_safe();
 	printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
-	show_trace(NULL, &regs->esp);
+	show_trace(NULL, regs, &regs->esp);
 }
 
 /*
--- linux-2.6.17-rc4/arch/i386/kernel/traps.c	2006-05-16 15:15:29.000000000 +0200
+++ 2.6.17-rc4-unwind-i386/arch/i386/kernel/traps.c	2006-05-16 15:42:38.000000000 +0200
@@ -47,7 +47,7 @@
 #include <asm/desc.h>
 #include <asm/i387.h>
 #include <asm/nmi.h>
-
+#include <asm/unwind.h>
 #include <asm/smp.h>
 #include <asm/arch_hooks.h>
 #include <asm/kdebug.h>
@@ -165,14 +165,43 @@ static inline unsigned long print_contex
 	return ebp;
 }
 
-static void show_trace_log_lvl(struct task_struct *task,
+static asmlinkage void show_trace_unwind(struct unwind_frame_info *info, void *log_lvl)
+{
+	int printed = 0; /* nr of entries already printed on current line */
+
+	while (unwind(info) == 0 && UNW_PC(info)) {
+		printed = print_addr_and_symbol(UNW_PC(info), log_lvl, printed);
+		if (arch_unw_user_mode(info))
+			break;
+	}
+	if (printed)
+		printk("\n");
+}
+
+static void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
 			       unsigned long *stack, char *log_lvl)
 {
 	unsigned long ebp;
+	struct unwind_frame_info info;
 
 	if (!task)
 		task = current;
 
+	if (regs) {
+		if (unwind_init_frame_info(&info, task, regs) == 0) {
+			show_trace_unwind(&info, log_lvl);
+			return;
+		}
+	} else if (task == current) {
+		if (unwind_init_running(&info, show_trace_unwind, log_lvl) == 0)
+			return;
+	} else {
+		if (unwind_init_blocked(&info, task) == 0) {
+			show_trace_unwind(&info, log_lvl);
+			return;
+		}
+	}
+
 	if (task == current) {
 		/* Grab ebp right from our regs */
 		asm ("movl %%ebp, %0" : "=r" (ebp) : );
@@ -193,13 +222,13 @@ static void show_trace_log_lvl(struct ta
 	}
 }
 
-void show_trace(struct task_struct *task, unsigned long * stack)
+void show_trace(struct task_struct *task, struct pt_regs *regs, unsigned long * stack)
 {
-	show_trace_log_lvl(task, stack, "");
+	show_trace_log_lvl(task, regs, stack, "");
 }
 
-static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
-			       char *log_lvl)
+static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
+			       unsigned long *esp, char *log_lvl)
 {
 	unsigned long *stack;
 	int i;
@@ -221,13 +250,13 @@ static void show_stack_log_lvl(struct ta
 		printk("%08lx ", *stack++);
 	}
 	printk("\n%sCall Trace:\n", log_lvl);
-	show_trace_log_lvl(task, esp, log_lvl);
+	show_trace_log_lvl(task, regs, esp, log_lvl);
 }
 
 void show_stack(struct task_struct *task, unsigned long *esp)
 {
 	printk("       ");
-	show_stack_log_lvl(task, esp, "");
+	show_stack_log_lvl(task, NULL, esp, "");
 }
 
 /*
@@ -237,7 +266,7 @@ void dump_stack(void)
 {
 	unsigned long stack;
 
-	show_trace(current, &stack);
+	show_trace(current, NULL, &stack);
 }
 
 EXPORT_SYMBOL(dump_stack);
@@ -280,7 +309,7 @@ void show_registers(struct pt_regs *regs
 		u8 __user *eip;
 
 		printk("\n" KERN_EMERG "Stack: ");
-		show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
+		show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
 
 		printk(KERN_EMERG "Code: ");
 
--- linux-2.6.17-rc4/arch/i386/kernel/vmlinux.lds.S	2006-05-16 15:15:29.000000000 +0200
+++ 2.6.17-rc4-unwind-i386/arch/i386/kernel/vmlinux.lds.S	2006-05-16 15:16:46.000000000 +0200
@@ -64,6 +64,15 @@ SECTIONS
   .data.read_mostly : AT(ADDR(.data.read_mostly) - LOAD_OFFSET) { *(.data.read_mostly) }
   _edata = .;			/* End of data section */
 
+#ifdef CONFIG_STACK_UNWIND
+  . = ALIGN(4);
+  .eh_frame : AT(ADDR(.eh_frame) - LOAD_OFFSET) {
+	__start_unwind = .;
+  	*(.eh_frame)
+	__end_unwind = .;
+  }
+#endif
+
   . = ALIGN(THREAD_SIZE);	/* init_task */
   .data.init_task : AT(ADDR(.data.init_task) - LOAD_OFFSET) {
 	*(.data.init_task)
--- linux-2.6.17-rc4/include/asm-i386/processor.h	2006-05-16 15:15:47.000000000 +0200
+++ 2.6.17-rc4-unwind-i386/include/asm-i386/processor.h	2006-05-16 15:16:46.000000000 +0200
@@ -555,7 +555,7 @@ extern void prepare_to_copy(struct task_
 extern int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags);
 
 extern unsigned long thread_saved_pc(struct task_struct *tsk);
-void show_trace(struct task_struct *task, unsigned long *stack);
+void show_trace(struct task_struct *task, struct pt_regs *regs, unsigned long *stack);
 
 unsigned long get_wchan(struct task_struct *p);
 
--- linux-2.6.17-rc4/include/asm-i386/unwind.h	1970-01-01 01:00:00.000000000 +0100
+++ 2.6.17-rc4-unwind-i386/include/asm-i386/unwind.h	2006-05-16 15:47:41.000000000 +0200
@@ -0,0 +1,97 @@
+#ifndef _ASM_I386_UNWIND_H
+#define _ASM_I386_UNWIND_H
+
+/*
+ * Copyright (C) 2002-2006 Novell, Inc.
+ *	Jan Beulich <jbeulich@novell.com>
+ */
+
+#include <linux/config.h>
+
+#ifdef CONFIG_STACK_UNWIND
+
+#include <asm/ptrace.h>
+#include <asm/uaccess.h>
+
+struct unwind_frame_info {
+	struct pt_regs regs;
+	struct task_struct *task;
+};
+
+#define UNW_PC(frame) (frame)->regs.eip
+#define UNW_SP(frame) (frame)->regs.esp
+#ifdef CONFIG_FRAME_POINTER
+#define UNW_FP(frame) (frame)->regs.ebp
+#define FRAME_RETADDR_OFFSET 4
+#define FRAME_LINK_OFFSET 0
+#define STACK_BOTTOM(tsk) (((tsk)->thread.esp0 - 1) & ~(THREAD_SIZE - 1))
+#define STACK_TOP(tsk) ((tsk)->thread.esp0)
+#endif
+
+#define UNW_REGISTER_INFO \
+	PTREGS_INFO(eax), \
+	PTREGS_INFO(ecx), \
+	PTREGS_INFO(edx), \
+	PTREGS_INFO(ebx), \
+	PTREGS_INFO(esp), \
+	PTREGS_INFO(ebp), \
+	PTREGS_INFO(esi), \
+	PTREGS_INFO(edi), \
+	PTREGS_INFO(eip)
+
+static inline void
+arch_unw_init_frame_info(struct unwind_frame_info *info, /*const*/ struct pt_regs *regs)
+{
+	if (user_mode_vm(regs))
+		info->regs = *regs;
+	else {
+		memcpy(&info->regs, regs, offsetof(struct pt_regs, esp));
+		info->regs.esp = (unsigned long)&regs->esp;
+		info->regs.xss = __KERNEL_DS;
+	}
+}
+
+static inline void
+arch_unw_init_blocked(struct unwind_frame_info *info)
+{
+	memset(&info->regs, 0, sizeof(info->regs));
+	info->regs.eip = info->task->thread.eip;
+	info->regs.xcs = __KERNEL_CS;
+	__get_user(info->regs.ebp, (long *)info->task->thread.esp);
+	info->regs.esp = info->task->thread.esp;
+	info->regs.xss = __KERNEL_DS;
+	info->regs.xds = __USER_DS;
+	info->regs.xes = __USER_DS;
+}
+
+extern asmlinkage void
+arch_unwind_init_running(struct unwind_frame_info *,
+                         asmlinkage void (*callback)(struct unwind_frame_info *, void *arg),
+                         void *arg);
+
+static inline int
+arch_unw_user_mode(const struct unwind_frame_info *info)
+{
+#if 0 /* This can only work when selector register and EFLAGS saves/restores
+         are properly annotated (and tracked in UNW_REGISTER_INFO). */
+	return user_mode_vm(&info->regs);
+#else
+	return info->regs.eip < PAGE_OFFSET;
+#endif
+}
+
+#else
+
+#define UNW_PC(frame) ((void)(frame), 0)
+
+static inline int
+arch_unw_user_mode(const void *info)
+{
+	return 0;
+}
+
+#endif
+
+#include <asm-generic/unwind.h>
+
+#endif /* _ASM_I386_UNWIND_H */
--- linux-2.6.17-rc4/lib/Kconfig.debug	2006-05-16 15:15:49.000000000 +0200
+++ 2.6.17-rc4-unwind-i386/lib/Kconfig.debug	2006-05-16 15:16:46.000000000 +0200
@@ -199,7 +199,7 @@ config UNWIND_INFO
 config STACK_UNWIND
 	bool "Stack unwind support"
 	depends on UNWIND_INFO
-	depends on X86_64
+	depends on X86
 	help
 	  This enables more precise stack traces, omitting all unrelated
 	  occurrences of pointers into kernel code from the dump.

Re: [PATCH 3/3] reliable stack trace support (i386)

[Ingo Molnar]

* Jan Beulich <jbeulich@novell.com> wrote:

> These are the i386-specific pieces to enable reliable stack traces. This is
> going to be even more useful once CFI annotations get added to he assembly
> code, namely to entry.S (a patch for that had been submitted several times).

> +#define UNW_PC(frame) (frame)->regs.eip
> +#define UNW_SP(frame) (frame)->regs.esp
> +#ifdef CONFIG_FRAME_POINTER
> +#define UNW_FP(frame) (frame)->regs.ebp
> +#define FRAME_RETADDR_OFFSET 4
> +#define FRAME_LINK_OFFSET 0
> +#define STACK_BOTTOM(tsk) (((tsk)->thread.esp0 - 1) & ~(THREAD_SIZE - 1))
> +#define STACK_TOP(tsk) ((tsk)->thread.esp0)
> +#endif

style: i'd suggest to improve readability by aligning these.

> +static inline void
> +arch_unw_init_blocked(struct unwind_frame_info *info)
> +{
> +	memset(&info->regs, 0, sizeof(info->regs));
> +	info->regs.eip = info->task->thread.eip;
> +	info->regs.xcs = __KERNEL_CS;
> +	__get_user(info->regs.ebp, (long *)info->task->thread.esp);
> +	info->regs.esp = info->task->thread.esp;
> +	info->regs.xss = __KERNEL_DS;
> +	info->regs.xds = __USER_DS;
> +	info->regs.xes = __USER_DS;

hm, arent you using this from within show_trace()? In that case we 
shouldnt do a __get_user() i think, we might be in an arbitrary context 
...

> +static inline int
> +arch_unw_user_mode(const struct unwind_frame_info *info)
> +{
> +#if 0 /* This can only work when selector register and EFLAGS saves/restores
> +         are properly annotated (and tracked in UNW_REGISTER_INFO). */
> +	return user_mode_vm(&info->regs);
> +#else
> +	return info->regs.eip < PAGE_OFFSET;
> +#endif

same here as for x86_64: is this condition safe? Userspace can provoke 
an EIP of >= PAGE_OFFSET by for example jumping to the vsyscall page.

	Ingo

Re: [PATCH 3/3] reliable stack trace support (i386)

[Andi Kleen]

On Tuesday 16 May 2006 16:22, Jan Beulich wrote:
> These are the i386-specific pieces to enable reliable stack traces. This is
> going to be even more useful once CFI annotations get added to he assembly
> code, namely to entry.S (a patch for that had been submitted several times).

Yes we should merge that. Can you resend it please?

> +static asmlinkage void show_trace_unwind(struct unwind_frame_info *info, void *log_lvl)

static asmlinkage ?

> +#define UNW_PC(frame) (frame)->regs.eip
> +#define UNW_SP(frame) (frame)->regs.esp
> +#ifdef CONFIG_FRAME_POINTER
> +#define UNW_FP(frame) (frame)->regs.ebp
> +#define FRAME_RETADDR_OFFSET 4
> +#define FRAME_LINK_OFFSET 0
> +#define STACK_BOTTOM(tsk) (((tsk)->thread.esp0 - 1) & ~(THREAD_SIZE - 1))
> +#define STACK_TOP(tsk) ((tsk)->thread.esp0)
> +#endif

Same comments as for x86-64

-Andi

Re: [discuss] Re: [PATCH 3/3] reliable stack trace support (i386)

[Andi Kleen]

> > +static inline void
> > +arch_unw_init_blocked(struct unwind_frame_info *info)
> > +{
> > +	memset(&info->regs, 0, sizeof(info->regs));
> > +	info->regs.eip = info->task->thread.eip;
> > +	info->regs.xcs = __KERNEL_CS;
> > +	__get_user(info->regs.ebp, (long *)info->task->thread.esp);
> > +	info->regs.esp = info->task->thread.esp;
> > +	info->regs.xss = __KERNEL_DS;
> > +	info->regs.xds = __USER_DS;
> > +	info->regs.xes = __USER_DS;
> 
> hm, arent you using this from within show_trace()? In that case we 
> shouldnt do a __get_user() i think, we might be in an arbitrary context 

Should be ok. We already do this in some parts of oopsing and the page fault handler 
can handle it as long as there is an exception entry (which is there for this case)

> ...
> 
> > +static inline int
> > +arch_unw_user_mode(const struct unwind_frame_info *info)
> > +{
> > +#if 0 /* This can only work when selector register and EFLAGS saves/restores
> > +         are properly annotated (and tracked in UNW_REGISTER_INFO). */
> > +	return user_mode_vm(&info->regs);
> > +#else
> > +	return info->regs.eip < PAGE_OFFSET;
> > +#endif
> 
> same here as for x86_64: is this condition safe? Userspace can provoke 
> an EIP of >= PAGE_OFFSET by for example jumping to the vsyscall page.

Good point, it's probably not. Especially since we already have a user_mode() macro

-Andi

Re: [PATCH 3/3] reliable stack trace support (i386)

[Jan Beulich]

>> +static asmlinkage void show_trace_unwind(struct unwind_frame_info *info, void *log_lvl)
>
>static asmlinkage ?

Yes, as it's being used as a callback from assembly.

Jan