[PATCH x86/nmi] Upgrade NMI backtrace stall checks

[Paul E. McKenney]

Commit 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 <leitao@xxxxxxxxxx>
Signed-off-by: Breno Leitao <leitao@xxxxxxxxxx>
Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxxxxx>
Cc: Borislav Petkov <bp@xxxxxxxxx>
Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
Cc: <x86@xxxxxxxxxx>

diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index 6e738ad474dcf..d6c9fe9b20c4a 100644
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -582,7 +582,7 @@ 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.				*/
@@ -630,22 +630,26 @@ 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));
 	}
 }