[PATCH] kdump: Use real pt_regs from exception

Makes kexec_crashdump() take a pt_regs * as an argument.  This allows to
get exact register state at the point of the crash.  If we come from direct
panic assertion NULL will be passed and the current registers saved before
crashdump.

This hooks into two places:
die(): check the conditions under which we will panic when calling
do_exit and go there directly with the pt_regs that caused the fatal
fault.

die_nmi(): If we receive an NMI lockup while in the kernel use the
pt_regs and go directly to crash_kexec(). We're probably nested up badly
at this point so this might be the only chance to escape with proper
information.
Signed-off-by: Alexander Nyberg <alexn@telia.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

arch/i386/kernel/crash.c

@@ -100,11 +100,30 @@ static void crash_get_current_regs(struct pt_regs *regs)
 	regs->eip = (unsigned long)current_text_addr();
 }
 
-static void crash_save_self(void)
+/* CPU does not save ss and esp on stack if execution is already
+ * running in kernel mode at the time of NMI occurrence. This code
+ * fixes it.
+ */
+static void crash_setup_regs(struct pt_regs *newregs, struct pt_regs *oldregs)
+{
+	memcpy(newregs, oldregs, sizeof(*newregs));
+	newregs->esp = (unsigned long)&(oldregs->esp);
+	__asm__ __volatile__("xorl %eax, %eax;");
+	__asm__ __volatile__ ("movw %%ss, %%ax;" :"=a"(newregs->xss));
+}
+
+/* We may have saved_regs from where the error came from
+ * or it is NULL if via a direct panic().
+ */
+static void crash_save_self(struct pt_regs *saved_regs)
 {
 	struct pt_regs regs;
 	int cpu;
 	cpu = smp_processor_id();
+
+	if (saved_regs)
+		crash_setup_regs(&regs, saved_regs);
+	else
 		crash_get_current_regs(&regs);
 	crash_save_this_cpu(&regs, cpu);
 }
@@ -124,15 +143,8 @@ static int crash_nmi_callback(struct pt_regs *regs, int cpu)
 		return 1;
 	local_irq_disable();
 
-	/* CPU does not save ss and esp on stack if execution is already
-	 * running in kernel mode at the time of NMI occurrence. This code
-	 * fixes it.
-	 */
 	if (!user_mode(regs)) {
-		memcpy(&fixed_regs, regs, sizeof(*regs));
-		fixed_regs.esp = (unsigned long)&(regs->esp);
-		__asm__ __volatile__("xorl %eax, %eax;");
-		__asm__ __volatile__ ("movw %%ss, %%ax;" :"=a"(fixed_regs.xss));
+		crash_setup_regs(&fixed_regs, regs);
 		regs = &fixed_regs;
 	}
 	crash_save_this_cpu(regs, cpu);
@@ -184,7 +196,7 @@ static void nmi_shootdown_cpus(void)
 }
 #endif
 
-void machine_crash_shutdown(void)
+void machine_crash_shutdown(struct pt_regs *regs)
 {
 	/* This function is only called after the system
 	 * has paniced or is otherwise in a critical state.
@@ -204,5 +216,5 @@ void machine_crash_shutdown(void)
 #if defined(CONFIG_X86_IO_APIC)
 	disable_IO_APIC();
 #endif
-	crash_save_self();
+	crash_save_self(regs);
 }

arch/i386/kernel/traps.c

@@ -27,6 +27,7 @@
 #include <linux/ptrace.h>
 #include <linux/utsname.h>
 #include <linux/kprobes.h>
+#include <linux/kexec.h>
 
 #ifdef CONFIG_EISA
 #include <linux/ioport.h>
@@ -294,6 +295,9 @@ static void handle_BUG(struct pt_regs *regs)
 	printk("Kernel BUG\n");
 }
 
+/* This is gone through when something in the kernel
+ * has done something bad and is about to be terminated.
+*/
 void die(const char * str, struct pt_regs * regs, long err)
 {
 	static struct {
@@ -341,6 +345,10 @@ void die(const char * str, struct pt_regs * regs, long err)
 	bust_spinlocks(0);
 	die.lock_owner = -1;
 	spin_unlock_irq(&die.lock);
+
+	if (kexec_should_crash(current))
+		crash_kexec(regs);
+
 	if (in_interrupt())
 		panic("Fatal exception in interrupt");
 
@@ -570,6 +578,15 @@ void die_nmi (struct pt_regs *regs, const char *msg)
 	console_silent();
 	spin_unlock(&nmi_print_lock);
 	bust_spinlocks(0);
+
+	/* If we are in kernel we are probably nested up pretty bad
+	 * and might aswell get out now while we still can.
+	*/
+	if (!user_mode(regs)) {
+		current->thread.trap_no = 2;
+		crash_kexec(regs);
+	}
+
 	do_exit(SIGSEGV);
 }
 

arch/ppc/kernel/machine_kexec.c

@@ -34,7 +34,7 @@ void machine_shutdown(void)
 	}
 }
 
-void machine_crash_shutdown(void)
+void machine_crash_shutdown(struct pt_regs *regs)
 {
 	if (ppc_md.machine_crash_shutdown) {
 		ppc_md.machine_crash_shutdown();

arch/ppc64/kernel/machine_kexec.c

@@ -34,7 +34,7 @@ note_buf_t crash_notes[NR_CPUS];
  * and if what it will achieve. Letting it be now to compile the code
  * in generic kexec environment
  */
-void machine_crash_shutdown(void)
+void machine_crash_shutdown(struct pt_regs *regs)
 {
 	/* do nothing right now */
 	/* smp_relase_cpus() if we want smp on panic kernel */

arch/s390/kernel/crash.c

@@ -12,6 +12,6 @@
 
 note_buf_t crash_notes[NR_CPUS];
 
-void machine_crash_shutdown(void)
+void machine_crash_shutdown(struct pt_regs *regs)
 {
 }

arch/x86_64/kernel/crash.c

@@ -22,7 +22,7 @@
 
 note_buf_t crash_notes[NR_CPUS];
 
-void machine_crash_shutdown(void)
+void machine_crash_shutdown(struct pt_regs *regs)
 {
 	/* This function is only called after the system
 	 * has paniced or is otherwise in a critical state.

drivers/char/sysrq.c

@@ -100,7 +100,7 @@ static struct sysrq_key_op sysrq_unraw_op = {
 static void sysrq_handle_crashdump(int key, struct pt_regs *pt_regs,
 				struct tty_struct *tty)
 {
-	crash_kexec();
+	crash_kexec(pt_regs);
 }
 static struct sysrq_key_op sysrq_crashdump_op = {
 	.handler	= sysrq_handle_crashdump,

include/linux/kexec.h

@@ -99,7 +99,8 @@ extern asmlinkage long compat_sys_kexec_load(unsigned long entry,
 	unsigned long flags);
 #endif
 extern struct page *kimage_alloc_control_pages(struct kimage *image, unsigned int order);
-extern void crash_kexec(void);
+extern void crash_kexec(struct pt_regs *);
+int kexec_should_crash(struct task_struct *);
 extern struct kimage *kexec_image;
 
 #define KEXEC_ON_CRASH  0x00000001
@@ -123,6 +124,9 @@ extern struct kimage *kexec_image;
 extern struct resource crashk_res;
 
 #else /* !CONFIG_KEXEC */
-static inline void crash_kexec(void) { }
+struct pt_regs;
+struct task_struct;
+static inline void crash_kexec(struct pt_regs *regs) { }
+static inline int kexec_should_crash(struct task_struct *p) { return 0; }
 #endif /* CONFIG_KEXEC */
 #endif /* LINUX_KEXEC_H */

include/linux/reboot.h

@@ -52,7 +52,8 @@ extern void machine_halt(void);
 extern void machine_power_off(void);
 
 extern void machine_shutdown(void);
-extern void machine_crash_shutdown(void);
+struct pt_regs;
+extern void machine_crash_shutdown(struct pt_regs *);
 
 #endif
 

kernel/kexec.c

@@ -18,6 +18,8 @@
 #include <linux/reboot.h>
 #include <linux/syscalls.h>
 #include <linux/ioport.h>
+#include <linux/hardirq.h>
+
 #include <asm/page.h>
 #include <asm/uaccess.h>
 #include <asm/io.h>
@@ -32,6 +34,13 @@ struct resource crashk_res = {
 	.flags = IORESOURCE_BUSY | IORESOURCE_MEM
 };
 
+int kexec_should_crash(struct task_struct *p)
+{
+	if (in_interrupt() || !p->pid || p->pid == 1 || panic_on_oops)
+		return 1;
+	return 0;
+}
+
 /*
  * When kexec transitions to the new kernel there is a one-to-one
  * mapping between physical and virtual addresses.  On processors
@@ -1010,7 +1019,7 @@ asmlinkage long compat_sys_kexec_load(unsigned long entry,
 }
 #endif
 
-void crash_kexec(void)
+void crash_kexec(struct pt_regs *regs)
 {
 	struct kimage *image;
 	int locked;
@@ -1028,7 +1037,7 @@ void crash_kexec(void)
 	if (!locked) {
 		image = xchg(&kexec_crash_image, NULL);
 		if (image) {
-			machine_crash_shutdown();
+			machine_crash_shutdown(regs);
 			machine_kexec(image);
 		}
 		xchg(&kexec_lock, 0);

kernel/panic.c

@@ -83,7 +83,7 @@ NORET_TYPE void panic(const char * fmt, ...)
 	 * everything else.
 	 * Do we want to call this before we try to display a message?
 	 */
-	crash_kexec();
+	crash_kexec(NULL);
 
 #ifdef CONFIG_SMP
 	/*