From 3186b61812c007179f1852d8e63a0f0f7dd7c783 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Thu, 7 Mar 2024 17:32:15 -0800 Subject: [PATCH] x86/nmi: Upgrade NMI backtrace stall checks & messages The commit to improve NMI stall debuggability: 344da544f177 ("x86/nmi: Print reasons why backtrace NMIs are ignored") ... has shown value, but widespread use has also identified a few opportunities for improvement. The systems have (as usual) shown far more creativity than that commit's author, demonstrating yet again that failing CPUs can do whatever they want. In addition, the current message format is less friendly than one might like to those attempting to use these messages to identify failing CPUs. Therefore, separately flag CPUs that, during the full time that the stack-backtrace request was waiting, were always in an NMI handler, were never in an NMI handler, or exited one NMI handler. Also, split the message identifying the CPU and the time since that CPU's last NMI-related activity so that a single line identifies the CPU without any other variable information, greatly reducing the processing overhead required to identify repeat-offender CPUs. Co-developed-by: Breno Leitao Signed-off-by: Breno Leitao Signed-off-by: Paul E. McKenney Signed-off-by: Ingo Molnar Cc: Linus Torvalds Link: https://lore.kernel.org/r/ab4d70c8-c874-42dc-b206-643018922393@paulmck-laptop --- arch/x86/kernel/nmi.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-) diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c index 9a5b372c706f..ed163c8c8604 100644 --- a/arch/x86/kernel/nmi.c +++ b/arch/x86/kernel/nmi.c @@ -580,7 +580,7 @@ EXPORT_SYMBOL_GPL(asm_exc_nmi_kvm_vmx); static char *nmi_check_stall_msg[] = { /* */ -/* +--------- nsp->idt_seq_snap & 0x1: CPU is in NMI handler. */ +/* +--------- nmi_seq & 0x1: CPU is currently in NMI handler. */ /* | +------ cpu_is_offline(cpu) */ /* | | +--- nsp->idt_calls_snap != atomic_long_read(&nsp->idt_calls): */ /* | | | NMI handler has been invoked. */ @@ -628,22 +628,26 @@ void nmi_backtrace_stall_check(const struct cpumask *btp) nmi_seq = READ_ONCE(nsp->idt_nmi_seq); if (nsp->idt_nmi_seq_snap + 1 == nmi_seq && (nmi_seq & 0x1)) { msgp = "CPU entered NMI handler function, but has not exited"; - } else if ((nsp->idt_nmi_seq_snap & 0x1) != (nmi_seq & 0x1)) { - msgp = "CPU is handling NMIs"; - } else { - idx = ((nsp->idt_seq_snap & 0x1) << 2) | + } else if (nsp->idt_nmi_seq_snap == nmi_seq || + nsp->idt_nmi_seq_snap + 1 == nmi_seq) { + idx = ((nmi_seq & 0x1) << 2) | (cpu_is_offline(cpu) << 1) | (nsp->idt_calls_snap != atomic_long_read(&nsp->idt_calls)); msgp = nmi_check_stall_msg[idx]; if (nsp->idt_ignored_snap != READ_ONCE(nsp->idt_ignored) && (idx & 0x1)) modp = ", but OK because ignore_nmis was set"; - if (nmi_seq & 0x1) - msghp = " (CPU currently in NMI handler function)"; - else if (nsp->idt_nmi_seq_snap + 1 == nmi_seq) + if (nsp->idt_nmi_seq_snap + 1 == nmi_seq) msghp = " (CPU exited one NMI handler function)"; + else if (nmi_seq & 0x1) + msghp = " (CPU currently in NMI handler function)"; + else + msghp = " (CPU was never in an NMI handler function)"; + } else { + msgp = "CPU is handling NMIs"; } - pr_alert("%s: CPU %d: %s%s%s, last activity: %lu jiffies ago.\n", - __func__, cpu, msgp, modp, msghp, j - READ_ONCE(nsp->recv_jiffies)); + pr_alert("%s: CPU %d: %s%s%s\n", __func__, cpu, msgp, modp, msghp); + pr_alert("%s: last activity: %lu jiffies ago.\n", + __func__, j - READ_ONCE(nsp->recv_jiffies)); } }