[PATCH v2] x86, UV: make kdump avoid stack dumps

From: Cliff Wickman
Date: Mon Aug 09 2010 - 17:09:52 EST



This replaces Version 1 of this patch, which broke the build when
CONFIG_KEXEC and CONFIG_CRASH_DUMP were configured off. In that case
the storage for the 'in_crash_kexec' flag was never built.

This v2 was submitted on 7/22, but apparently never seen (probably too close
in date and name to v1, sent on 7/20).

This version defines that flag as 0 if CONFIG_KEXEC is not set.
The patch is tested with all combinations of those two options.


UV NMI callback's should not write stack dumps when a kdump is to be written.

When invoking the crash kernel to write a dump, kdump_nmi_shootdown_cpus()
uses NMI's to get all the cpu's to save their register context and halt.

But the NMI interrupt handler runs a callback list. This patch sets a flag
to prevent any of those callbacks from interfering with the halt of the cpu.

For UV, which currently has the only callback to which this is relevant, the
uv_handle_nmi() callback should not do dumping of stacks.

The 'in_crash_kexec' flag is defined as an extern in kdebug.h firstly
because x2apic_uv_x.c includes it. Secondly because some future callback
might need the flag to know that it should not enter the debugger.
(Such a scenario was in fact present in the 2.6.32 kernel, SuSE distribution,
where a call to kdb needed to be avoided.)

Diffed against 2.6.35

Signed-off-by: Cliff Wickman <cpw@xxxxxxx>
---
arch/x86/include/asm/kdebug.h | 6 ++++++
arch/x86/kernel/apic/x2apic_uv_x.c | 4 ++++
arch/x86/kernel/crash.c | 3 +++
3 files changed, 13 insertions(+)

Index: 100722.linux-tip/arch/x86/kernel/apic/x2apic_uv_x.c
===================================================================
--- 100722.linux-tip.orig/arch/x86/kernel/apic/x2apic_uv_x.c
+++ 100722.linux-tip/arch/x86/kernel/apic/x2apic_uv_x.c
@@ -604,6 +604,10 @@ int uv_handle_nmi(struct notifier_block
{
if (reason != DIE_NMI_IPI)
return NOTIFY_OK;
+
+ if (in_crash_kexec)
+ /* do nothing if entering the crash kernel */
+ return NOTIFY_OK;
/*
* Use a lock so only one cpu prints at a time
* to prevent intermixed output.
Index: 100722.linux-tip/arch/x86/kernel/crash.c
===================================================================
--- 100722.linux-tip.orig/arch/x86/kernel/crash.c
+++ 100722.linux-tip/arch/x86/kernel/crash.c
@@ -28,6 +28,8 @@
#include <asm/reboot.h>
#include <asm/virtext.h>

+int in_crash_kexec;
+
#if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)

static void kdump_nmi_callback(int cpu, struct die_args *args)
@@ -61,6 +63,7 @@ static void kdump_nmi_callback(int cpu,

static void kdump_nmi_shootdown_cpus(void)
{
+ in_crash_kexec = 1;
nmi_shootdown_cpus(kdump_nmi_callback);

disable_local_APIC();
Index: 100722.linux-tip/arch/x86/include/asm/kdebug.h
===================================================================
--- 100722.linux-tip.orig/arch/x86/include/asm/kdebug.h
+++ 100722.linux-tip/arch/x86/include/asm/kdebug.h
@@ -33,5 +33,11 @@ extern void __show_regs(struct pt_regs *
extern void show_regs(struct pt_regs *regs);
extern unsigned long oops_begin(void);
extern void oops_end(unsigned long, struct pt_regs *, int signr);
+#ifdef CONFIG_KEXEC
+extern int in_crash_kexec;
+#else
+/* no crash dump is ever in progress if no crash kernel can be kexec'd */
+#define in_crash_kexec 0
+#endif

#endif /* _ASM_X86_KDEBUG_H */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/