Re: [PATCH v5 5/5] arm64: ipi_nmi: Add support for NMI backtrace

From: Sumit Garg
Date: Wed Oct 21 2020 - 07:28:59 EST


On Wed, 21 Oct 2020 at 16:02, Marc Zyngier <maz@xxxxxxxxxx> wrote:
>
> On 2020-10-20 10:13, Sumit Garg wrote:
> > On Mon, 19 Oct 2020 at 17:50, Marc Zyngier <maz@xxxxxxxxxx> wrote:
> >>
> >> On 2020-10-14 12:12, Sumit Garg wrote:
> >> > Enable NMI backtrace support on arm64 using IPI turned as an NMI
> >> > leveraging pseudo NMIs support. It is now possible for users to get a
> >> > backtrace of a CPU stuck in hard-lockup using magic SYSRQ.
> >> >
> >> > Signed-off-by: Sumit Garg <sumit.garg@xxxxxxxxxx>
> >> > ---
> >> > arch/arm64/include/asm/irq.h | 6 ++++++
> >> > arch/arm64/kernel/ipi_nmi.c | 12 +++++++++++-
> >> > 2 files changed, 17 insertions(+), 1 deletion(-)
> >> >
> >> > diff --git a/arch/arm64/include/asm/irq.h
> >> > b/arch/arm64/include/asm/irq.h
> >> > index b2b0c64..e840bf1 100644
> >> > --- a/arch/arm64/include/asm/irq.h
> >> > +++ b/arch/arm64/include/asm/irq.h
> >> > @@ -6,6 +6,12 @@
> >> >
> >> > #include <asm-generic/irq.h>
> >> >
> >> > +#ifdef CONFIG_SMP
> >> > +extern void arch_trigger_cpumask_backtrace(const cpumask_t *mask,
> >> > + bool exclude_self);
> >> > +#define arch_trigger_cpumask_backtrace arch_trigger_cpumask_backtrace
> >> > +#endif
> >> > +
> >> > struct pt_regs;
> >> >
> >> > static inline int nr_legacy_irqs(void)
> >> > diff --git a/arch/arm64/kernel/ipi_nmi.c b/arch/arm64/kernel/ipi_nmi.c
> >> > index e0a9e03..e1dc19d 100644
> >> > --- a/arch/arm64/kernel/ipi_nmi.c
> >> > +++ b/arch/arm64/kernel/ipi_nmi.c
> >> > @@ -9,6 +9,7 @@
> >> > #include <linux/interrupt.h>
> >> > #include <linux/irq.h>
> >> > #include <linux/kgdb.h>
> >> > +#include <linux/nmi.h>
> >> > #include <linux/smp.h>
> >> >
> >> > #include <asm/nmi.h>
> >> > @@ -25,12 +26,21 @@ void arch_send_call_nmi_func_ipi_mask(cpumask_t
> >> > *mask)
> >> > __ipi_send_mask(ipi_desc, mask);
> >> > }
> >> >
> >> > +void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool
> >> > exclude_self)
> >> > +{
> >> > + nmi_trigger_cpumask_backtrace(mask, exclude_self,
> >> > + arch_send_call_nmi_func_ipi_mask);
> >> > +}
> >> > +
> >> > static irqreturn_t ipi_nmi_handler(int irq, void *data)
> >> > {
> >> > unsigned int cpu = smp_processor_id();
> >> >
> >> > - ipi_kgdb_nmicallback(cpu, get_irq_regs());
> >> > + if (nmi_cpu_backtrace(get_irq_regs()))
> >> > + goto out;
> >> >
> >> > + ipi_kgdb_nmicallback(cpu, get_irq_regs());
> >> > +out:
> >> > return IRQ_HANDLED;
> >> > }
> >>
> >> Can't you have *both* a request for a backtrace and a KGDB call?
> >> It really shouldn't be either/or. It also outlines how well shared
> >> interrupts work with edge triggered signalling...
> >
> > Unfortunately, NMIs doesn't seem to support shared mode [1].
>
> You are totally missing the point: shared interrupts *cannot* work
> reliably with edge signalling. What I am saying here is that you need
> implement the sharing yourself in this function.

Ah, I see your point now. Will instead share this IPI among both handlers.

-Sumit

>
> M.
> --
> Jazz is not dead. It just smells funny...