diff mbox series

[v11,4/6] arm64: smp: Add arch support for backtrace using pseudo-NMI

Message ID 20230824083012.v11.4.Ie6c132b96ebbbcddbf6954b9469ed40a6960343c@changeid (mailing list archive)
State New, archived
Headers show
Series arm64: Add IPI for backtraces / kgdb; try to use NMI for some IPIs | expand

Commit Message

Doug Anderson Aug. 24, 2023, 3:30 p.m. UTC
Enable arch_trigger_cpumask_backtrace() support on arm64. This enables
things much like they are enabled on arm32 (including some of the
funky logic around NR_IPI, nr_ipi, and MAX_IPI) but with the
difference that, unlike arm32, we'll try to enable the backtrace to
use pseudo-NMI.

NOTE: this patch is a squash of the little bit of code adding the
ability to mark an IPI to try to use pseudo-NMI plus the little bit of
code to hook things up for kgdb. This approach was decided upon in the
discussion of v9 [1].

This patch depends on commit 36759e343ff9 ("nmi_backtrace: allow
excluding an arbitrary CPU") since that commit changed the prototype
of arch_trigger_cpumask_backtrace(), which this patch implements.

[1] https://lore.kernel.org/r/ZORY51mF4alI41G1@FVFF77S0Q05N

Co-developed-by: Sumit Garg <sumit.garg@linaro.org>
Signed-off-by: Sumit Garg <sumit.garg@linaro.org>
Co-developed-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
---

Changes in v11:
- Adjust comment about NR_IPI/MAX_IPI.
- Don't use confusing "backed by" idiom in comment.
- Made arm64_backtrace_ipi() static.

Changes in v10:
- Backtrace now directly supported in smp.c
- Squash backtrace into patch adding support for pseudo-NMI IPIs.

Changes in v9:
- Added comments that we might not be using NMI always.
- Fold in v8 patch #10 ("Fallback to a regular IPI if NMI isn't enabled")
- Moved header file out of "include" since it didn't need to be there.
- Remove arm64_supports_nmi()
- Renamed "NMI IPI" to "debug IPI" since it might not be backed by NMI.
- arch_trigger_cpumask_backtrace() no longer returns bool

Changes in v8:
- Removed "#ifdef CONFIG_SMP" since arm64 is always SMP
- debug_ipi_setup() and debug_ipi_teardown() no longer take cpu param

 arch/arm64/include/asm/irq.h |  3 ++
 arch/arm64/kernel/smp.c      | 86 +++++++++++++++++++++++++++++++-----
 2 files changed, 78 insertions(+), 11 deletions(-)

Comments

Stephen Boyd Aug. 25, 2023, 10:27 p.m. UTC | #1
Quoting Douglas Anderson (2023-08-24 08:30:30)
> diff --git a/arch/arm64/include/asm/irq.h b/arch/arm64/include/asm/irq.h
> index fac08e18bcd5..50ce8b697ff3 100644
> --- a/arch/arm64/include/asm/irq.h
> +++ b/arch/arm64/include/asm/irq.h
> @@ -6,6 +6,9 @@
>
>  #include <asm-generic/irq.h>
>
> +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu);

Some nits, but otherwise

Reviewed-by: Stephen Boyd <swboyd@chromium.org>

> +#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
> +
>  struct pt_regs;
>
>  int set_handle_irq(void (*handle_irq)(struct pt_regs *));
> diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> index a5848f1ef817..c8896cbc5327 100644
> --- a/arch/arm64/kernel/smp.c
> +++ b/arch/arm64/kernel/smp.c
> @@ -72,12 +73,18 @@ enum ipi_msg_type {
>         IPI_CPU_CRASH_STOP,
>         IPI_TIMER,
>         IPI_IRQ_WORK,
> -       NR_IPI
> +       NR_IPI,
> +       /*
> +        * Any enum >= NR_IPI and < MAX_IPI is special and not tracable
> +        * with trace_ipi_*
> +        */
> +       IPI_CPU_BACKTRACE = NR_IPI,
> +       MAX_IPI
>  };
>
>  static int ipi_irq_base __read_mostly;
>  static int nr_ipi __read_mostly = NR_IPI;
> -static struct irq_desc *ipi_desc[NR_IPI] __read_mostly;
> +static struct irq_desc *ipi_desc[MAX_IPI] __read_mostly;

Side note: it would be nice to mark ipi_desc as __ro_after_init. Same
for nr_ipi and ipi_irq_base.

>
>  static void ipi_setup(int cpu);
>
> @@ -845,6 +852,22 @@ static void __noreturn ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs
>  #endif
>  }
>
> +static void arm64_backtrace_ipi(cpumask_t *mask)
> +{
> +       __ipi_send_mask(ipi_desc[IPI_CPU_BACKTRACE], mask);
> +}
> +
> +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu)

Can this be 'bool exclude_self' instead of int? That matches all other
implementations from what I can tell.

> +{
> +       /*
> +        * NOTE: though nmi_trigger_cpumask_backtrace has "nmi_" in the name,

USe nmi_trigger_cpumask_backtrace() to indicate function.

> +        * nothing about it truly needs to be implemented using an NMI, it's
> +        * just that it's _allowed_ to work with NMIs. If ipi_should_be_nmi()
> +        * returned false our backtrace attempt will just use a regular IPI.
> +        */
> +       nmi_trigger_cpumask_backtrace(mask, exclude_cpu, arm64_backtrace_ipi);
Doug Anderson Aug. 25, 2023, 11:02 p.m. UTC | #2
Hi,

On Fri, Aug 25, 2023 at 3:27 PM Stephen Boyd <swboyd@chromium.org> wrote:
>
> Quoting Douglas Anderson (2023-08-24 08:30:30)
> > diff --git a/arch/arm64/include/asm/irq.h b/arch/arm64/include/asm/irq.h
> > index fac08e18bcd5..50ce8b697ff3 100644
> > --- a/arch/arm64/include/asm/irq.h
> > +++ b/arch/arm64/include/asm/irq.h
> > @@ -6,6 +6,9 @@
> >
> >  #include <asm-generic/irq.h>
> >
> > +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu);
>
> Some nits, but otherwise
>
> Reviewed-by: Stephen Boyd <swboyd@chromium.org>
>
> > +#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
> > +
> >  struct pt_regs;
> >
> >  int set_handle_irq(void (*handle_irq)(struct pt_regs *));
> > diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> > index a5848f1ef817..c8896cbc5327 100644
> > --- a/arch/arm64/kernel/smp.c
> > +++ b/arch/arm64/kernel/smp.c
> > @@ -72,12 +73,18 @@ enum ipi_msg_type {
> >         IPI_CPU_CRASH_STOP,
> >         IPI_TIMER,
> >         IPI_IRQ_WORK,
> > -       NR_IPI
> > +       NR_IPI,
> > +       /*
> > +        * Any enum >= NR_IPI and < MAX_IPI is special and not tracable
> > +        * with trace_ipi_*
> > +        */
> > +       IPI_CPU_BACKTRACE = NR_IPI,
> > +       MAX_IPI
> >  };
> >
> >  static int ipi_irq_base __read_mostly;
> >  static int nr_ipi __read_mostly = NR_IPI;
> > -static struct irq_desc *ipi_desc[NR_IPI] __read_mostly;
> > +static struct irq_desc *ipi_desc[MAX_IPI] __read_mostly;
>
> Side note: it would be nice to mark ipi_desc as __ro_after_init. Same
> for nr_ipi and ipi_irq_base.

I'd rather not change it in this patch since it's a pre-existing and
separate issue, but I can add a patch to the end of the series for
that if I end up spinning it. Otherwise I can send a follow-up patch
for it.


> >  static void ipi_setup(int cpu);
> >
> > @@ -845,6 +852,22 @@ static void __noreturn ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs
> >  #endif
> >  }
> >
> > +static void arm64_backtrace_ipi(cpumask_t *mask)
> > +{
> > +       __ipi_send_mask(ipi_desc[IPI_CPU_BACKTRACE], mask);
> > +}
> > +
> > +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu)
>
> Can this be 'bool exclude_self' instead of int? That matches all other
> implementations from what I can tell.

Nope. See the part of the commit message that says:

This patch depends on commit 36759e343ff9 ("nmi_backtrace: allow
excluding an arbitrary CPU") since that commit changed the prototype
of arch_trigger_cpumask_backtrace(), which this patch implements.

> > +{
> > +       /*
> > +        * NOTE: though nmi_trigger_cpumask_backtrace has "nmi_" in the name,
>
> USe nmi_trigger_cpumask_backtrace() to indicate function.

I won't plan on doing an immediate spin for this and for now I'll wait
for additional feedback. If a maintainer is getting ready to land
this, I'm happy to post a new version with this fix or also happy if a
maintainer wants to add the "()" while applying.

-Doug
Stephen Boyd Aug. 25, 2023, 11:23 p.m. UTC | #3
Quoting Doug Anderson (2023-08-25 16:02:46)
> On Fri, Aug 25, 2023 at 3:27 PM Stephen Boyd <swboyd@chromium.org> wrote:
> >
> > Quoting Douglas Anderson (2023-08-24 08:30:30)
> > > diff --git a/arch/arm64/include/asm/irq.h b/arch/arm64/include/asm/irq.h
> > >
> > >  static int ipi_irq_base __read_mostly;
> > >  static int nr_ipi __read_mostly = NR_IPI;
> > > -static struct irq_desc *ipi_desc[NR_IPI] __read_mostly;
> > > +static struct irq_desc *ipi_desc[MAX_IPI] __read_mostly;
> >
> > Side note: it would be nice to mark ipi_desc as __ro_after_init. Same
> > for nr_ipi and ipi_irq_base.
>
> I'd rather not change it in this patch since it's a pre-existing and
> separate issue, but I can add a patch to the end of the series for
> that if I end up spinning it. Otherwise I can send a follow-up patch
> for it.

Of course. Don't change it in this patch.

>
>
> > >  static void ipi_setup(int cpu);
> > >
> > > @@ -845,6 +852,22 @@ static void __noreturn ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs
> > >  #endif
> > >  }
> > >
> > > +static void arm64_backtrace_ipi(cpumask_t *mask)
> > > +{
> > > +       __ipi_send_mask(ipi_desc[IPI_CPU_BACKTRACE], mask);
> > > +}
> > > +
> > > +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu)
> >
> > Can this be 'bool exclude_self' instead of int? That matches all other
> > implementations from what I can tell.
>
> Nope. See the part of the commit message that says:
>
> This patch depends on commit 36759e343ff9 ("nmi_backtrace: allow
> excluding an arbitrary CPU") since that commit changed the prototype
> of arch_trigger_cpumask_backtrace(), which this patch implements.

Ah, ok. Sounds fine then.
Tomohiro Misono (Fujitsu) Aug. 29, 2023, 5:23 a.m. UTC | #4
> -----Original Message-----
> Subject: [PATCH v11 4/6] arm64: smp: Add arch support for backtrace using
> pseudo-NMI
> 
> Enable arch_trigger_cpumask_backtrace() support on arm64. This enables
> things much like they are enabled on arm32 (including some of the
> funky logic around NR_IPI, nr_ipi, and MAX_IPI) but with the
> difference that, unlike arm32, we'll try to enable the backtrace to
> use pseudo-NMI.
> 
> NOTE: this patch is a squash of the little bit of code adding the
> ability to mark an IPI to try to use pseudo-NMI plus the little bit of
> code to hook things up for kgdb. This approach was decided upon in the
> discussion of v9 [1].
> 
> This patch depends on commit 36759e343ff9 ("nmi_backtrace: allow

Hi,
Is this commit ID valid? I believe this commit is not included in the
mainline yet. Other than this,
 Reviewed-by: Misono Tomohiro <misono.tomohiro@fujitsu.com>

Regards,
Tomohiro

> excluding an arbitrary CPU") since that commit changed the prototype
> of arch_trigger_cpumask_backtrace(), which this patch implements.
> 
> [1] https://lore.kernel.org/r/ZORY51mF4alI41G1@FVFF77S0Q05N
> 
> Co-developed-by: Sumit Garg <sumit.garg@linaro.org>
> Signed-off-by: Sumit Garg <sumit.garg@linaro.org>
> Co-developed-by: Mark Rutland <mark.rutland@arm.com>
> Signed-off-by: Mark Rutland <mark.rutland@arm.com>
> Signed-off-by: Douglas Anderson <dianders@chromium.org>
> ---
> 
> Changes in v11:
> - Adjust comment about NR_IPI/MAX_IPI.
> - Don't use confusing "backed by" idiom in comment.
> - Made arm64_backtrace_ipi() static.
> 
> Changes in v10:
> - Backtrace now directly supported in smp.c
> - Squash backtrace into patch adding support for pseudo-NMI IPIs.
> 
> Changes in v9:
> - Added comments that we might not be using NMI always.
> - Fold in v8 patch #10 ("Fallback to a regular IPI if NMI isn't enabled")
> - Moved header file out of "include" since it didn't need to be there.
> - Remove arm64_supports_nmi()
> - Renamed "NMI IPI" to "debug IPI" since it might not be backed by NMI.
> - arch_trigger_cpumask_backtrace() no longer returns bool
> 
> Changes in v8:
> - Removed "#ifdef CONFIG_SMP" since arm64 is always SMP
> - debug_ipi_setup() and debug_ipi_teardown() no longer take cpu param
> 
>  arch/arm64/include/asm/irq.h |  3 ++
>  arch/arm64/kernel/smp.c      | 86
> +++++++++++++++++++++++++++++++-----
>  2 files changed, 78 insertions(+), 11 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/irq.h b/arch/arm64/include/asm/irq.h
> index fac08e18bcd5..50ce8b697ff3 100644
> --- a/arch/arm64/include/asm/irq.h
> +++ b/arch/arm64/include/asm/irq.h
> @@ -6,6 +6,9 @@
> 
>  #include <asm-generic/irq.h>
> 
> +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int
> exclude_cpu);
> +#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
> +
>  struct pt_regs;
> 
>  int set_handle_irq(void (*handle_irq)(struct pt_regs *));
> diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> index a5848f1ef817..c8896cbc5327 100644
> --- a/arch/arm64/kernel/smp.c
> +++ b/arch/arm64/kernel/smp.c
> @@ -33,6 +33,7 @@
>  #include <linux/kernel_stat.h>
>  #include <linux/kexec.h>
>  #include <linux/kvm_host.h>
> +#include <linux/nmi.h>
> 
>  #include <asm/alternative.h>
>  #include <asm/atomic.h>
> @@ -72,12 +73,18 @@ enum ipi_msg_type {
>  	IPI_CPU_CRASH_STOP,
>  	IPI_TIMER,
>  	IPI_IRQ_WORK,
> -	NR_IPI
> +	NR_IPI,
> +	/*
> +	 * Any enum >= NR_IPI and < MAX_IPI is special and not tracable
> +	 * with trace_ipi_*
> +	 */
> +	IPI_CPU_BACKTRACE = NR_IPI,
> +	MAX_IPI
>  };
> 
>  static int ipi_irq_base __read_mostly;
>  static int nr_ipi __read_mostly = NR_IPI;
> -static struct irq_desc *ipi_desc[NR_IPI] __read_mostly;
> +static struct irq_desc *ipi_desc[MAX_IPI] __read_mostly;
> 
>  static void ipi_setup(int cpu);
> 
> @@ -845,6 +852,22 @@ static void __noreturn ipi_cpu_crash_stop(unsigned int
> cpu, struct pt_regs *regs
>  #endif
>  }
> 
> +static void arm64_backtrace_ipi(cpumask_t *mask)
> +{
> +	__ipi_send_mask(ipi_desc[IPI_CPU_BACKTRACE], mask);
> +}
> +
> +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int
> exclude_cpu)
> +{
> +	/*
> +	 * NOTE: though nmi_trigger_cpumask_backtrace has "nmi_" in the
> name,
> +	 * nothing about it truly needs to be implemented using an NMI, it's
> +	 * just that it's _allowed_ to work with NMIs. If ipi_should_be_nmi()
> +	 * returned false our backtrace attempt will just use a regular IPI.
> +	 */
> +	nmi_trigger_cpumask_backtrace(mask, exclude_cpu,
> arm64_backtrace_ipi);
> +}
> +
>  /*
>   * Main handler for inter-processor interrupts
>   */
> @@ -888,6 +911,14 @@ static void do_handle_IPI(int ipinr)
>  		break;
>  #endif
> 
> +	case IPI_CPU_BACKTRACE:
> +		/*
> +		 * NOTE: in some cases this _won't_ be NMI context. See the
> +		 * comment in arch_trigger_cpumask_backtrace().
> +		 */
> +		nmi_cpu_backtrace(get_irq_regs());
> +		break;
> +
>  	default:
>  		pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
>  		break;
> @@ -909,6 +940,19 @@ static void smp_cross_call(const struct cpumask *target,
> unsigned int ipinr)
>  	__ipi_send_mask(ipi_desc[ipinr], target);
>  }
> 
> +static bool ipi_should_be_nmi(enum ipi_msg_type ipi)
> +{
> +	if (!system_uses_irq_prio_masking())
> +		return false;
> +
> +	switch (ipi) {
> +	case IPI_CPU_BACKTRACE:
> +		return true;
> +	default:
> +		return false;
> +	}
> +}
> +
>  static void ipi_setup(int cpu)
>  {
>  	int i;
> @@ -916,8 +960,14 @@ static void ipi_setup(int cpu)
>  	if (WARN_ON_ONCE(!ipi_irq_base))
>  		return;
> 
> -	for (i = 0; i < nr_ipi; i++)
> -		enable_percpu_irq(ipi_irq_base + i, 0);
> +	for (i = 0; i < nr_ipi; i++) {
> +		if (ipi_should_be_nmi(i)) {
> +			prepare_percpu_nmi(ipi_irq_base + i);
> +			enable_percpu_nmi(ipi_irq_base + i, 0);
> +		} else {
> +			enable_percpu_irq(ipi_irq_base + i, 0);
> +		}
> +	}
>  }
> 
>  #ifdef CONFIG_HOTPLUG_CPU
> @@ -928,8 +978,14 @@ static void ipi_teardown(int cpu)
>  	if (WARN_ON_ONCE(!ipi_irq_base))
>  		return;
> 
> -	for (i = 0; i < nr_ipi; i++)
> -		disable_percpu_irq(ipi_irq_base + i);
> +	for (i = 0; i < nr_ipi; i++) {
> +		if (ipi_should_be_nmi(i)) {
> +			disable_percpu_nmi(ipi_irq_base + i);
> +			teardown_percpu_nmi(ipi_irq_base + i);
> +		} else {
> +			disable_percpu_irq(ipi_irq_base + i);
> +		}
> +	}
>  }
>  #endif
> 
> @@ -937,15 +993,23 @@ void __init set_smp_ipi_range(int ipi_base, int n)
>  {
>  	int i;
> 
> -	WARN_ON(n < NR_IPI);
> -	nr_ipi = min(n, NR_IPI);
> +	WARN_ON(n < MAX_IPI);
> +	nr_ipi = min(n, MAX_IPI);
> 
>  	for (i = 0; i < nr_ipi; i++) {
>  		int err;
> 
> -		err = request_percpu_irq(ipi_base + i, ipi_handler,
> -					 "IPI", &cpu_number);
> -		WARN_ON(err);
> +		if (ipi_should_be_nmi(i)) {
> +			err = request_percpu_nmi(ipi_base + i, ipi_handler,
> +						 "IPI", &cpu_number);
> +			WARN(err, "Could not request IPI %d as NMI,
> err=%d\n",
> +			     i, err);
> +		} else {
> +			err = request_percpu_irq(ipi_base + i, ipi_handler,
> +						 "IPI", &cpu_number);
> +			WARN(err, "Could not request IPI %d as IRQ,
> err=%d\n",
> +			     i, err);
> +		}
> 
>  		ipi_desc[i] = irq_to_desc(ipi_base + i);
>  		irq_set_status_flags(ipi_base + i, IRQ_HIDDEN);
> --
> 2.42.0.rc1.204.g551eb34607-goog
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Doug Anderson Aug. 29, 2023, 4:03 p.m. UTC | #5
Hi,

On Mon, Aug 28, 2023 at 10:23 PM Tomohiro Misono (Fujitsu)
<misono.tomohiro@fujitsu.com> wrote:
>
> > -----Original Message-----
> > Subject: [PATCH v11 4/6] arm64: smp: Add arch support for backtrace using
> > pseudo-NMI
> >
> > Enable arch_trigger_cpumask_backtrace() support on arm64. This enables
> > things much like they are enabled on arm32 (including some of the
> > funky logic around NR_IPI, nr_ipi, and MAX_IPI) but with the
> > difference that, unlike arm32, we'll try to enable the backtrace to
> > use pseudo-NMI.
> >
> > NOTE: this patch is a squash of the little bit of code adding the
> > ability to mark an IPI to try to use pseudo-NMI plus the little bit of
> > code to hook things up for kgdb. This approach was decided upon in the
> > discussion of v9 [1].
> >
> > This patch depends on commit 36759e343ff9 ("nmi_backtrace: allow
>
> Hi,
> Is this commit ID valid? I believe this commit is not included in the
> mainline yet. Other than this,
>  Reviewed-by: Misono Tomohiro <misono.tomohiro@fujitsu.com>

Ah, good call. I must have grabbed that git hash before the commit
moved from Andrew Morton's unstable branch to his stable branch. As
far as I can tell, it should be commit 8d539b84f1e3 ("nmi_backtrace:
allow excluding an arbitrary CPU"). ...at least that's what's in
linuxnext today. That also appears to match the commit hash in the
pull request that Andrew just sent to Linus [1].

I'll update the git hash and add your tag in v12, which I'm currently
aiming to send out tomorrow.

[1] https://lore.kernel.org/r/20230828225431.354d3d2d3b80ee5b88e65eb5@linux-foundation.org
diff mbox series

Patch

diff --git a/arch/arm64/include/asm/irq.h b/arch/arm64/include/asm/irq.h
index fac08e18bcd5..50ce8b697ff3 100644
--- a/arch/arm64/include/asm/irq.h
+++ b/arch/arm64/include/asm/irq.h
@@ -6,6 +6,9 @@ 
 
 #include <asm-generic/irq.h>
 
+void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu);
+#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
+
 struct pt_regs;
 
 int set_handle_irq(void (*handle_irq)(struct pt_regs *));
diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
index a5848f1ef817..c8896cbc5327 100644
--- a/arch/arm64/kernel/smp.c
+++ b/arch/arm64/kernel/smp.c
@@ -33,6 +33,7 @@ 
 #include <linux/kernel_stat.h>
 #include <linux/kexec.h>
 #include <linux/kvm_host.h>
+#include <linux/nmi.h>
 
 #include <asm/alternative.h>
 #include <asm/atomic.h>
@@ -72,12 +73,18 @@  enum ipi_msg_type {
 	IPI_CPU_CRASH_STOP,
 	IPI_TIMER,
 	IPI_IRQ_WORK,
-	NR_IPI
+	NR_IPI,
+	/*
+	 * Any enum >= NR_IPI and < MAX_IPI is special and not tracable
+	 * with trace_ipi_*
+	 */
+	IPI_CPU_BACKTRACE = NR_IPI,
+	MAX_IPI
 };
 
 static int ipi_irq_base __read_mostly;
 static int nr_ipi __read_mostly = NR_IPI;
-static struct irq_desc *ipi_desc[NR_IPI] __read_mostly;
+static struct irq_desc *ipi_desc[MAX_IPI] __read_mostly;
 
 static void ipi_setup(int cpu);
 
@@ -845,6 +852,22 @@  static void __noreturn ipi_cpu_crash_stop(unsigned int cpu, struct pt_regs *regs
 #endif
 }
 
+static void arm64_backtrace_ipi(cpumask_t *mask)
+{
+	__ipi_send_mask(ipi_desc[IPI_CPU_BACKTRACE], mask);
+}
+
+void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu)
+{
+	/*
+	 * NOTE: though nmi_trigger_cpumask_backtrace has "nmi_" in the name,
+	 * nothing about it truly needs to be implemented using an NMI, it's
+	 * just that it's _allowed_ to work with NMIs. If ipi_should_be_nmi()
+	 * returned false our backtrace attempt will just use a regular IPI.
+	 */
+	nmi_trigger_cpumask_backtrace(mask, exclude_cpu, arm64_backtrace_ipi);
+}
+
 /*
  * Main handler for inter-processor interrupts
  */
@@ -888,6 +911,14 @@  static void do_handle_IPI(int ipinr)
 		break;
 #endif
 
+	case IPI_CPU_BACKTRACE:
+		/*
+		 * NOTE: in some cases this _won't_ be NMI context. See the
+		 * comment in arch_trigger_cpumask_backtrace().
+		 */
+		nmi_cpu_backtrace(get_irq_regs());
+		break;
+
 	default:
 		pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
 		break;
@@ -909,6 +940,19 @@  static void smp_cross_call(const struct cpumask *target, unsigned int ipinr)
 	__ipi_send_mask(ipi_desc[ipinr], target);
 }
 
+static bool ipi_should_be_nmi(enum ipi_msg_type ipi)
+{
+	if (!system_uses_irq_prio_masking())
+		return false;
+
+	switch (ipi) {
+	case IPI_CPU_BACKTRACE:
+		return true;
+	default:
+		return false;
+	}
+}
+
 static void ipi_setup(int cpu)
 {
 	int i;
@@ -916,8 +960,14 @@  static void ipi_setup(int cpu)
 	if (WARN_ON_ONCE(!ipi_irq_base))
 		return;
 
-	for (i = 0; i < nr_ipi; i++)
-		enable_percpu_irq(ipi_irq_base + i, 0);
+	for (i = 0; i < nr_ipi; i++) {
+		if (ipi_should_be_nmi(i)) {
+			prepare_percpu_nmi(ipi_irq_base + i);
+			enable_percpu_nmi(ipi_irq_base + i, 0);
+		} else {
+			enable_percpu_irq(ipi_irq_base + i, 0);
+		}
+	}
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
@@ -928,8 +978,14 @@  static void ipi_teardown(int cpu)
 	if (WARN_ON_ONCE(!ipi_irq_base))
 		return;
 
-	for (i = 0; i < nr_ipi; i++)
-		disable_percpu_irq(ipi_irq_base + i);
+	for (i = 0; i < nr_ipi; i++) {
+		if (ipi_should_be_nmi(i)) {
+			disable_percpu_nmi(ipi_irq_base + i);
+			teardown_percpu_nmi(ipi_irq_base + i);
+		} else {
+			disable_percpu_irq(ipi_irq_base + i);
+		}
+	}
 }
 #endif
 
@@ -937,15 +993,23 @@  void __init set_smp_ipi_range(int ipi_base, int n)
 {
 	int i;
 
-	WARN_ON(n < NR_IPI);
-	nr_ipi = min(n, NR_IPI);
+	WARN_ON(n < MAX_IPI);
+	nr_ipi = min(n, MAX_IPI);
 
 	for (i = 0; i < nr_ipi; i++) {
 		int err;
 
-		err = request_percpu_irq(ipi_base + i, ipi_handler,
-					 "IPI", &cpu_number);
-		WARN_ON(err);
+		if (ipi_should_be_nmi(i)) {
+			err = request_percpu_nmi(ipi_base + i, ipi_handler,
+						 "IPI", &cpu_number);
+			WARN(err, "Could not request IPI %d as NMI, err=%d\n",
+			     i, err);
+		} else {
+			err = request_percpu_irq(ipi_base + i, ipi_handler,
+						 "IPI", &cpu_number);
+			WARN(err, "Could not request IPI %d as IRQ, err=%d\n",
+			     i, err);
+		}
 
 		ipi_desc[i] = irq_to_desc(ipi_base + i);
 		irq_set_status_flags(ipi_base + i, IRQ_HIDDEN);