x86/nmi: Accumulate NMI-progress evidence in exc_nmi()

CPUs ignoring NMIs is often a sign of those CPUs going bad, but there
are quite a few other reasons why a CPU might ignore NMIs.  Therefore,
accumulate evidence within exc_nmi() as to what might be preventing a
given CPU from responding to an NMI.

[ paulmck: Apply Peter Zijlstra feedback. ]

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: <x86@kernel.org>
Reviewed-by: Ingo Molnar <mingo@kernel.org>

diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index cec0bfa3bc04fa21fbf741d5c6fca0464956bf1d..701ff27a77ecfbfc110ad63587033353fa636f8a 100644 (file)
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -69,6 +69,15 @@ struct nmi_stats {
        unsigned int unknown;
        unsigned int external;
        unsigned int swallow;
+       unsigned long recv_jiffies;
+       unsigned long idt_seq;
+       unsigned long idt_nmi_seq;
+       unsigned long idt_ignored;
+       atomic_long_t idt_calls;
+       unsigned long idt_seq_snap;
+       unsigned long idt_nmi_seq_snap;
+       unsigned long idt_ignored_snap;
+       long idt_calls_snap;
 };
 
 static DEFINE_PER_CPU(struct nmi_stats, nmi_stats);
@@ -479,12 +488,15 @@ static DEFINE_PER_CPU(unsigned long, nmi_dr7);
 DEFINE_IDTENTRY_RAW(exc_nmi)
 {
        irqentry_state_t irq_state;
+       struct nmi_stats *nsp = this_cpu_ptr(&nmi_stats);
 
        /*
         * Re-enable NMIs right here when running as an SEV-ES guest. This might
         * cause nested NMIs, but those can be handled safely.
         */
        sev_es_nmi_complete();
+       if (IS_ENABLED(CONFIG_NMI_CHECK_CPU))
+               arch_atomic_long_inc(&nsp->idt_calls);
 
        if (IS_ENABLED(CONFIG_SMP) && arch_cpu_is_offline(smp_processor_id()))
                return;
@@ -495,6 +507,11 @@ DEFINE_IDTENTRY_RAW(exc_nmi)
        }
        this_cpu_write(nmi_state, NMI_EXECUTING);
        this_cpu_write(nmi_cr2, read_cr2());
+       if (IS_ENABLED(CONFIG_NMI_CHECK_CPU)) {
+               WRITE_ONCE(nsp->idt_seq, nsp->idt_seq + 1);
+               WARN_ON_ONCE(!(nsp->idt_seq & 0x1));
+               WRITE_ONCE(nsp->recv_jiffies, jiffies);
+       }
 nmi_restart:
 
        /*
@@ -509,8 +526,19 @@ nmi_restart:
 
        inc_irq_stat(__nmi_count);
 
-       if (!ignore_nmis)
+       if (IS_ENABLED(CONFIG_NMI_CHECK_CPU) && ignore_nmis) {
+               WRITE_ONCE(nsp->idt_ignored, nsp->idt_ignored + 1);
+       } else if (!ignore_nmis) {
+               if (IS_ENABLED(CONFIG_NMI_CHECK_CPU)) {
+                       WRITE_ONCE(nsp->idt_nmi_seq, nsp->idt_nmi_seq + 1);
+                       WARN_ON_ONCE(!(nsp->idt_nmi_seq & 0x1));
+               }
                default_do_nmi(regs);
+               if (IS_ENABLED(CONFIG_NMI_CHECK_CPU)) {
+                       WRITE_ONCE(nsp->idt_nmi_seq, nsp->idt_nmi_seq + 1);
+                       WARN_ON_ONCE(nsp->idt_nmi_seq & 0x1);
+               }
+       }
 
        irqentry_nmi_exit(regs, irq_state);
 
@@ -525,6 +553,11 @@ nmi_restart:
 
        if (user_mode(regs))
                mds_user_clear_cpu_buffers();
+       if (IS_ENABLED(CONFIG_NMI_CHECK_CPU)) {
+               WRITE_ONCE(nsp->idt_seq, nsp->idt_seq + 1);
+               WARN_ON_ONCE(nsp->idt_seq & 0x1);
+               WRITE_ONCE(nsp->recv_jiffies, jiffies);
+       }
 }
 
 #if defined(CONFIG_X86_64) && IS_ENABLED(CONFIG_KVM_INTEL)

diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 881c3f84e88a32da05751dad0b54c0fbca5468e7..dad99197d26954c6c81df810f02f142f9a4cff30 100644 (file)
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -1552,6 +1552,17 @@ config TRACE_IRQFLAGS_NMI
        depends on TRACE_IRQFLAGS
        depends on TRACE_IRQFLAGS_NMI_SUPPORT
 
+config NMI_CHECK_CPU
+       bool "Debugging for CPUs failing to respond to backtrace requests"
+       depends on DEBUG_KERNEL
+       depends on X86
+       default n
+       help
+         Enables debug prints when a CPU fails to respond to a given
+         backtrace NMI.  These prints provide some reasons why a CPU
+         might legitimately be failing to respond, for example, if it
+         is offline of if ignore_nmis is set.
+
 config DEBUG_IRQFLAGS
        bool "Debug IRQ flag manipulation"
        help