2010-12-06 22:28:29

by Cliff Wickman

Subject: [PATCH] x86: UV kdump reboot fix

After a crash dump on an SGI Altix UV system the crash kernel fails to
cause a reboot.
Only the reboot_type of BOOT_ACPI works.
Other boot types ...BOOT_EFI, _KBD, _TRIPLE... fail.

The system's BIOS has an EFI layer, so it is a bit of a mystery to me
why BOOT_EFI fails. Can someone explain that?

Diffed against 2.6.37-rc2

Signed-off-by: Cliff Wickman <[email protected]>

---
arch/x86/kernel/reboot.c | 6 ++++++
1 file changed, 6 insertions(+)

Index: linux/arch/x86/kernel/reboot.c
===================================================================
--- linux.orig/arch/x86/kernel/reboot.c
+++ linux/arch/x86/kernel/reboot.c
@@ -10,6 +10,7 @@
#include <linux/dmi.h>
#include <linux/sched.h>
#include <linux/tboot.h>
+#include <linux/crash_dump.h>
#include <acpi/reboot.h>
#include <asm/io.h>
#include <asm/apic.h>
@@ -22,6 +23,7 @@
#include <asm/pci_x86.h>
#include <asm/virtext.h>
#include <asm/cpu.h>
+#include <asm/uv/uv.h>

#ifdef CONFIG_X86_32
# include <linux/ctype.h>
@@ -675,6 +677,10 @@ static void native_machine_restart(char

if (!reboot_force)
machine_shutdown();
+
+ if (is_uv_system() && is_kdump_kernel())
+ reboot_type = BOOT_ACPI;
+
__machine_emergency_restart(0);
}

2010-12-22 11:41:52

by Ingo Molnar

Subject: Re: [PATCH] x86: UV kdump reboot fix

* Cliff Wickman <[email protected]> wrote:

>
> After a crash dump on an SGI Altix UV system the crash kernel fails to
> cause a reboot.
> Only the reboot_type of BOOT_ACPI works.
> Other boot types ...BOOT_EFI, _KBD, _TRIPLE... fail.
>
> The system's BIOS has an EFI layer, so it is a bit of a mystery to me
> why BOOT_EFI fails. Can someone explain that?
>
> Diffed against 2.6.37-rc2
>
> Signed-off-by: Cliff Wickman <[email protected]>
>
> ---
> arch/x86/kernel/reboot.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> Index: linux/arch/x86/kernel/reboot.c
> ===================================================================
> --- linux.orig/arch/x86/kernel/reboot.c
> +++ linux/arch/x86/kernel/reboot.c
> @@ -10,6 +10,7 @@
> #include <linux/dmi.h>
> #include <linux/sched.h>
> #include <linux/tboot.h>
> +#include <linux/crash_dump.h>
> #include <acpi/reboot.h>
> #include <asm/io.h>
> #include <asm/apic.h>
> @@ -22,6 +23,7 @@
> #include <asm/pci_x86.h>
> #include <asm/virtext.h>
> #include <asm/cpu.h>
> +#include <asm/uv/uv.h>
>
> #ifdef CONFIG_X86_32
> # include <linux/ctype.h>
> @@ -675,6 +677,10 @@ static void native_machine_restart(char
>
> if (!reboot_force)
> machine_shutdown();
> +
> + if (is_uv_system() && is_kdump_kernel())
> + reboot_type = BOOT_ACPI;
> +
> __machine_emergency_restart(0);

This kind of is_uv_system() hackery in core x86 code is really unacceptable!

Either add a proper reboot quirk or add platform ops support for UV reboot, a'la:

cfb505a7ebd4: x86: mrst: Add Moorestown specific reboot/shutdown support

In fact all is_uv_system() checkery hacks in core x86 code should be eliminated:

arch/x86/kernel/cpu/common.c: if (is_uv_system())
arch/x86/kernel/smpboot.c: if (is_uv_system())
arch/x86/mm/tlb.c: if (is_uv_system()) {

and proper platform_ops (or machine_ops) should be introduced instead.

Thanks,

Ingo