From: Joel Fernandes <joelagnelf@nvidia.com>

Move NMI nesting tracking from the preempt_count bits to a separate per-CPU
counter (nmi_nesting). This is to free up the NMI bits in the preempt_count,
allowing those bits to be repurposed for other uses.  This also has the benefit
of tracking more than 16-levels deep if there is ever a need.

Reduce multiple bits in preempt_count for NMI tracking. Reduce NMI_BITS
from 3 to 1, using it only to detect if we're in an NMI.

Suggested-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Joel Fernandes <joelaf@google.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
---
 include/linux/hardirq.h | 16 ++++++++++++----
 include/linux/preempt.h | 13 +++++++++----
 kernel/softirq.c        |  2 ++
 3 files changed, 23 insertions(+), 8 deletions(-)

diff --git a/include/linux/hardirq.h b/include/linux/hardirq.h
index d57cab4d4c06f..cc06bda52c3e5 100644
--- a/include/linux/hardirq.h
+++ b/include/linux/hardirq.h
@@ -10,6 +10,8 @@
 #include <linux/vtime.h>
 #include <asm/hardirq.h>
 
+DECLARE_PER_CPU(unsigned int, nmi_nesting);
+
 extern void synchronize_irq(unsigned int irq);
 extern bool synchronize_hardirq(unsigned int irq);
 
@@ -102,14 +104,16 @@ void irq_exit_rcu(void);
  */
 
 /*
- * nmi_enter() can nest up to 15 times; see NMI_BITS.
+ * nmi_enter() can nest - nesting is tracked in a per-CPU counter.
  */
 #define __nmi_enter()						\
 	do {							\
 		lockdep_off();					\
 		arch_nmi_enter();				\
-		BUG_ON(in_nmi() == NMI_MASK);			\
-		__preempt_count_add(NMI_OFFSET + HARDIRQ_OFFSET);	\
+		BUG_ON(__this_cpu_read(nmi_nesting) == UINT_MAX);	\
+		__this_cpu_inc(nmi_nesting);			\
+		__preempt_count_add(HARDIRQ_OFFSET);		\
+		preempt_count_set(preempt_count() | NMI_MASK);	\
 	} while (0)
 
 #define nmi_enter()						\
@@ -124,8 +128,12 @@ void irq_exit_rcu(void);
 
 #define __nmi_exit()						\
 	do {							\
+		unsigned int nesting;				\
 		BUG_ON(!in_nmi());				\
-		__preempt_count_sub(NMI_OFFSET + HARDIRQ_OFFSET);	\
+		__preempt_count_sub(HARDIRQ_OFFSET);		\
+		nesting = __this_cpu_dec_return(nmi_nesting);	\
+		if (!nesting)					\
+			__preempt_count_sub(NMI_OFFSET);	\
 		arch_nmi_exit();				\
 		lockdep_on();					\
 	} while (0)
diff --git a/include/linux/preempt.h b/include/linux/preempt.h
index 94ebdd98b7a94..860769a717c10 100644
--- a/include/linux/preempt.h
+++ b/include/linux/preempt.h
@@ -18,6 +18,8 @@
  * - bits 0-7 are the preemption count (max preemption depth: 256)
  * - bits 8-15 are the softirq count (max # of softirqs: 256)
  * - bits 16-23 are the hardirq disable count (max # of hardirq disable: 256)
+ * - bits 24-27 are the hardirq count (max # of hardirqs: 16)
+ * - bit 28 is the NMI flag (no nesting count, tracked separately)
  *
  * The hardirq count could in theory be the same as the number of
  * interrupts in the system, but we run all interrupt handlers with
@@ -25,18 +27,21 @@
  * there are a few palaeontologic drivers which reenable interrupts in
  * the handler, so we need more than one bit here.
  *
+ * NMI nesting depth is tracked in a separate per-CPU variable
+ * (nmi_nesting) to save bits in preempt_count.
+ *
  *         PREEMPT_MASK:	0x000000ff
  *         SOFTIRQ_MASK:	0x0000ff00
  * HARDIRQ_DISABLE_MASK:	0x00ff0000
- *         HARDIRQ_MASK:	0x07000000
- *             NMI_MASK:	0x38000000
+ *         HARDIRQ_MASK:	0x0f000000
+ *             NMI_MASK:	0x10000000
  * PREEMPT_NEED_RESCHED:	0x80000000
  */
 #define PREEMPT_BITS	8
 #define SOFTIRQ_BITS	8
 #define HARDIRQ_DISABLE_BITS	8
-#define HARDIRQ_BITS	3
-#define NMI_BITS	3
+#define HARDIRQ_BITS	4
+#define NMI_BITS	1
 
 #define PREEMPT_SHIFT	0
 #define SOFTIRQ_SHIFT	(PREEMPT_SHIFT + PREEMPT_BITS)
diff --git a/kernel/softirq.c b/kernel/softirq.c
index 77198911b8dd4..af47ea23aba3b 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -88,6 +88,8 @@ EXPORT_PER_CPU_SYMBOL_GPL(hardirqs_enabled);
 EXPORT_PER_CPU_SYMBOL_GPL(hardirq_context);
 #endif
 
+DEFINE_PER_CPU(unsigned int, nmi_nesting);
+
 /*
  * SOFTIRQ_OFFSET usage:
  *
-- 
2.51.1