Re: [PATCH v2 01/12] preempt: Track NMI nesting to separate per-CPU counter

[Boqun Feng <boqun@kernel.org>]

On Tue, May 26, 2026 at 08:21:37AM -0700, Boqun Feng wrote:
> 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.
> 
> Reduce NMI_BITS from 4 to 1, using it only to detect if we're in an NMI.
> The per-CPU counter currently caps nesting at 15.
> 
> [boqun: Solve Steven Rostedt's comment on the BUG_ON() condition]
> 
> Suggested-by: Boqun Feng <boqun.feng@gmail.com>
> Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
> Signed-off-by: Lyude Paul <lyude@redhat.com>
> Signed-off-by: Boqun Feng <boqun@kernel.org>
> Link: https://patch.msgid.link/20260121223933.1568682-3-lyude@redhat.com
> ---
>  include/linux/hardirq.h                        | 17 +++++++++++++----
>  include/linux/preempt.h                        |  9 +++++++--
>  kernel/softirq.c                               |  2 ++
>  tools/testing/selftests/bpf/bpf_experimental.h |  2 +-
>  4 files changed, 23 insertions(+), 7 deletions(-)
> 
> diff --git a/include/linux/hardirq.h b/include/linux/hardirq.h
> index d57cab4d4c06..1a0360a1000f 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 +104,17 @@ 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);	\
> +		/* Maximum NMI nesting is 15. */		\
> +		BUG_ON(__this_cpu_read(nmi_nesting) >= 15);	\
> +		__this_cpu_inc(nmi_nesting);			\
> +		__preempt_count_add(HARDIRQ_OFFSET);		\
> +		preempt_count_set(preempt_count() | NMI_MASK);	\
>  	} while (0)
>  
>  #define nmi_enter()						\
> @@ -124,8 +129,12 @@ 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);	\

We have an issue here in the following case:

  // nmi_nesting == 1
  __nmi_exit():
    ..
    nesting = __this_cpu_dec_return(nmi_nesting); // <- nesting == 0
    <another NMI comes>
      __nmi_enter()
      // nmi_nesting becomes 1
      __nmi_exit():
        nesting = __this_cpu_dec_return(nmi_nesting); // <- nesting == 0
        if (!nesting)
	  __preempt_count_sub(NMI_OFFSET);
	// NMI_OFFSET bit is 0
    if (!nesting)
       __preempt_count_sub(NMI_OFFSET); // underflow!

I think we need to do:
      
#define __nmi_exit()						\
	do {							\
		unsigned int nesting;				\
		BUG_ON(!in_nmi());				\
		__preempt_count_sub(HARDIRQ_OFFSET);		\
		nesting = __this_cpu_dec_return(nmi_nesting);	\
		if (!nesting)					\
			preempt_count_set(preempt_count() & ~NMI_MASK);	\
		arch_nmi_exit();				\
  		lockdep_on();					\
	} while (0)

@Joel, thoughts?

Similarly, we have this issue in patch #10 as well.

Regards,
Boqun

>  		arch_nmi_exit();				\
>  		lockdep_on();					\
>  	} while (0)
> diff --git a/include/linux/preempt.h b/include/linux/preempt.h
> index d964f965c8ff..586f96688325 100644
> --- a/include/linux/preempt.h
> +++ b/include/linux/preempt.h
> @@ -17,6 +17,8 @@
>   *
>   * - bits 0-7 are the preemption count (max preemption depth: 256)
>   * - bits 8-15 are the softirq count (max # of softirqs: 256)
> + * - bits 16-19 are the hardirq count (max # of hardirqs: 16)
> + * - bit 20 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
> @@ -24,16 +26,19 @@
>   * 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_MASK:	0x000f0000
> - *             NMI_MASK:	0x00f00000
> + *             NMI_MASK:	0x00100000
>   * PREEMPT_NEED_RESCHED:	0x80000000
>   */
>  #define PREEMPT_BITS	8
>  #define SOFTIRQ_BITS	8
>  #define HARDIRQ_BITS	4
> -#define NMI_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 4425d8dce44b..10af5ed859e7 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:
>   *
> diff --git a/tools/testing/selftests/bpf/bpf_experimental.h b/tools/testing/selftests/bpf/bpf_experimental.h
> index 2234bd6bc9d3..2d4256ff471f 100644
> --- a/tools/testing/selftests/bpf/bpf_experimental.h
> +++ b/tools/testing/selftests/bpf/bpf_experimental.h
> @@ -449,7 +449,7 @@ extern int bpf_cgroup_read_xattr(struct cgroup *cgroup, const char *name__str,
>  #define PREEMPT_BITS	8
>  #define SOFTIRQ_BITS	8
>  #define HARDIRQ_BITS	4
> -#define NMI_BITS	4
> +#define NMI_BITS	1
>  
>  #define PREEMPT_SHIFT	0
>  #define SOFTIRQ_SHIFT	(PREEMPT_SHIFT + PREEMPT_BITS)
> -- 
> 2.50.1 (Apple Git-155)
>