Thread (24 messages) 24 messages, 5 authors, 2025-05-09

Re: [PATCH v5 4/9] x86/nmi: Assign and register NMI-source vectors

From: Peter Zijlstra <peterz@infradead.org>
Date: 2025-05-07 08:22:41
Also in: kvm, linux-edac, linux-perf-users, linux-pm, lkml

On Tue, May 06, 2025 at 06:21:40PM -0700, Sohil Mehta wrote:
quoted hunk ↗ jump to hunk
diff --git a/arch/x86/include/asm/nmi.h b/arch/x86/include/asm/nmi.h
index f0a577bf7bba..b9beb216f2d0 100644
--- a/arch/x86/include/asm/nmi.h
+++ b/arch/x86/include/asm/nmi.h
@@ -57,6 +57,38 @@ struct nmiaction {
 	u8			source_vector;
 };
 
+/**
+ * NMI-source vectors are used to identify the origin of an NMI and to
+ * route the NMI directly to the appropriate handler.
+ *
+ * On CPUs that support NMI-source reporting with FRED, receiving an NMI
+ * with a valid vector sets the corresponding bit in the NMI-source
+ * bitmap. The bitmap is delivered as FRED event data on the stack.
+ * Multiple NMIs are coalesced in the NMI-source bitmap until the next
+ * NMI delivery.
+ *
+ * If an NMI is received without a vector or beyond the defined range,
+ * the CPU sets bit 0 of the NMI-source bitmap.
+ *
+ * Vector 2 is reserved for external NMIs related to the Local APIC -
+ * LINT1. Some third-party chipsets may send NMI messages with a
+ * hardcoded vector of 2, which would result in bit 2 being set in the
+ * NMI-source bitmap.
+ *
+ * The vectors are in no particular priority order. Add new vector
+ * assignments sequentially in the list below.
+ */
+#define NMIS_VECTOR_NONE	0	/* Reserved - Set for all unidentified sources */
+#define NMIS_VECTOR_TEST	1	/* NMI selftest */
+#define NMIS_VECTOR_EXTERNAL	2	/* Reserved - Match External NMI vector 2 */
+#define NMIS_VECTOR_SMP_STOP	3	/* Panic stop CPU */
+#define NMIS_VECTOR_BT		4	/* CPU backtrace */
+#define NMIS_VECTOR_KGDB	5	/* Kernel debugger */
+#define NMIS_VECTOR_MCE		6	/* MCE injection */
+#define NMIS_VECTOR_PMI		7	/* PerfMon counters */
+
+#define NMIS_VECTORS_MAX	16	/* Maximum number of NMI-source vectors */
Are these really independent NMI vectors, or simply NMI source reporting bits?

Because if they are not NMI vectors, naming them such is confusing.

Specifically, is there a latch per source?
quoted hunk ↗ jump to hunk
diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index be93ec7255bf..a1d672dcb6f0 100644
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -184,6 +184,11 @@ int __register_nmi_handler(unsigned int type, struct nmiaction *action)
 
 	raw_spin_lock_irqsave(&desc->lock, flags);
 
+	WARN_ON_ONCE(action->source_vector >= NMIS_VECTORS_MAX);
+
+	if (!cpu_feature_enabled(X86_FEATURE_NMI_SOURCE) || type != NMI_LOCAL)
+		action->source_vector = 0;
How about:

	WARN_ON_ONCE(type != NMI_LOCAL && action->source_vector);

instead?
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help