On Thu, Feb 23, 2017 at 09:36:52PM +0800, Xunlei Pang wrote: > We met an issue for kdump: after kdump kernel boots up, > and there comes a broadcasted mce in first kernel, the > other cpus remaining in first kernel will enter the old > mce handler of first kernel, then timeout and panic due > to MCE synchronization, finally reset the kdump cpus. > > This patch lets cpus stay quiet after nmi_shootdown_cpus(), > so after kdump boots, cpus remaining in 1st kernel should > not do anything except clearing MCG_STATUS. This is useful > for kdump to let vmcore dumping perform as hard as it can. Ok, I went and rewrote the text to make it more succinct, to the point and correct spelling and formatting. Tony, ACK? --- >From 2d76fdd4044b4659bb8746948b986e3f4eb75e22 Mon Sep 17 00:00:00 2001 From: Xunlei Pang <xlpang@xxxxxxxxxx> Date: Thu, 23 Feb 2017 21:36:52 +0800 Subject: [PATCH] x86/mce: Handle broadcasted MCE gracefully with kexec When we are about to kexec a crash kernel and right then and there a broadcasted MCE fires while we're still in first kernel and while the other CPUs remain in a holding pattern, the #MC handler of the first kernel will timeout and then panic due to never completing MCE synchronization. Handle this in a similar way to as when the CPUs are offlined when that broadcasted MCE happens. Suggested-by: Borislav Petkov <bp at alien8.de> Signed-off-by: Xunlei Pang <xlpang at redhat.com> Cc: Naoya Horiguchi <n-horiguchi at ah.jp.nec.com> Cc: Tony Luck <tony.luck at intel.com> Cc: kexec at lists.infradead.org Cc: linux-edac <linux-edac at vger.kernel.org> Cc: x86-ml <x86 at kernel.org> Link: http://lkml.kernel.org/r/1487857012-9059-1-git-send-email-xlpang at redhat.com [ Boris: rewrote commit message and comments. ] Signed-off-by: Borislav Petkov <bp at suse.de> --- arch/x86/include/asm/reboot.h | 1 + arch/x86/kernel/cpu/mcheck/mce.c | 18 ++++++++++++++++-- arch/x86/kernel/reboot.c | 5 +++-- 3 files changed, 20 insertions(+), 4 deletions(-) diff --git a/arch/x86/include/asm/reboot.h b/arch/x86/include/asm/reboot.h index 2cb1cc253d51..fc62ba8dce93 100644 --- a/arch/x86/include/asm/reboot.h +++ b/arch/x86/include/asm/reboot.h @@ -15,6 +15,7 @@ struct machine_ops { }; extern struct machine_ops machine_ops; +extern int crashing_cpu; void native_machine_crash_shutdown(struct pt_regs *regs); void native_machine_shutdown(void); diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c index 8e9725c607ea..177472ace838 100644 --- a/arch/x86/kernel/cpu/mcheck/mce.c +++ b/arch/x86/kernel/cpu/mcheck/mce.c @@ -49,6 +49,7 @@ #include <asm/tlbflush.h> #include <asm/mce.h> #include <asm/msr.h> +#include <asm/reboot.h> #include "mce-internal.h" @@ -1127,9 +1128,22 @@ void do_machine_check(struct pt_regs *regs, long error_code) * on Intel. */ int lmce = 1; + int cpu = smp_processor_id(); - /* If this CPU is offline, just bail out. */ - if (cpu_is_offline(smp_processor_id())) { + /* + * Cases where we avoid rendezvous handler timeout: + * 1) If this CPU is offline. + * + * 2) If crashing_cpu was set, e.g. we're entering kdump and we need to + * skip those CPUs which remain looping in the 1st kernel - see + * crash_nmi_callback(). + * + * Note: there still is a small window between kexec-ing and the new, + * kdump kernel establishing a new #MC handler where a broadcasted MCE + * might not get handled properly. + */ + if (cpu_is_offline(cpu) || + (crashing_cpu != -1 && crashing_cpu != cpu)) { u64 mcgstatus; mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS); diff --git a/arch/x86/kernel/reboot.c b/arch/x86/kernel/reboot.c index e244c19a2451..d3718cc5edbf 100644 --- a/arch/x86/kernel/reboot.c +++ b/arch/x86/kernel/reboot.c @@ -749,10 +749,11 @@ void machine_crash_shutdown(struct pt_regs *regs) #endif +/* This is the CPU performing the emergency shutdown work. */ +int crashing_cpu = -1; + #if defined(CONFIG_SMP) -/* This keeps a track of which one is crashing cpu. */ -static int crashing_cpu; static nmi_shootdown_cb shootdown_callback; static atomic_t waiting_for_crash_ipi; -- 2.11.0 -- Regards/Gruss, Boris. Good mailing practices for 400: avoid top-posting and trim the reply.
