Re: [PATCH v3 1/4] arm64: smp: Introduce a new IPI as IPI_CALL_NMI_FUNC

[Sumit Garg]

On Thu, 3 Sep 2020 at 22:06, Marc Zyngier <maz@xxxxxxxxxx> wrote:
>
> On 2020-09-03 13:05, Sumit Garg wrote:
> > Introduce a new inter processor interrupt as IPI_CALL_NMI_FUNC that
> > can be invoked to run special handlers in NMI context. One such handler
> > example is kgdb_nmicallback() which is invoked in order to round up
> > CPUs
> > to enter kgdb context.
> >
> > As currently pseudo NMIs are supported on specific arm64 platforms
> > which
> > incorporates GICv3 or later version of interrupt controller. In case a
> > particular platform doesn't support pseudo NMIs, IPI_CALL_NMI_FUNC will
> > act as a normal IPI which can still be used to invoke special handlers.
> >
> > Signed-off-by: Sumit Garg <sumit.garg@xxxxxxxxxx>
> > ---
> >  arch/arm64/include/asm/smp.h |  1 +
> >  arch/arm64/kernel/smp.c      | 11 +++++++++++
> >  2 files changed, 12 insertions(+)
> >
> > diff --git a/arch/arm64/include/asm/smp.h
> > b/arch/arm64/include/asm/smp.h
> > index 2e7f529..e85f5d5 100644
> > --- a/arch/arm64/include/asm/smp.h
> > +++ b/arch/arm64/include/asm/smp.h
> > @@ -89,6 +89,7 @@ extern void secondary_entry(void);
> >
> >  extern void arch_send_call_function_single_ipi(int cpu);
> >  extern void arch_send_call_function_ipi_mask(const struct cpumask
> > *mask);
> > +extern void arch_send_call_nmi_func_ipi_mask(const struct cpumask
> > *mask);
> >
> >  #ifdef CONFIG_ARM64_ACPI_PARKING_PROTOCOL
> >  extern void arch_send_wakeup_ipi_mask(const struct cpumask *mask);
> > diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> > index b6bde26..1b4c07c 100644
> > --- a/arch/arm64/kernel/smp.c
> > +++ b/arch/arm64/kernel/smp.c
> > @@ -74,6 +74,7 @@ enum ipi_msg_type {
> >       IPI_TIMER,
> >       IPI_IRQ_WORK,
> >       IPI_WAKEUP,
> > +     IPI_CALL_NMI_FUNC,
> >       NR_IPI
> >  };
> >
> > @@ -793,6 +794,7 @@ static const char *ipi_types[NR_IPI]
> > __tracepoint_string = {
> >       S(IPI_TIMER, "Timer broadcast interrupts"),
> >       S(IPI_IRQ_WORK, "IRQ work interrupts"),
> >       S(IPI_WAKEUP, "CPU wake-up interrupts"),
> > +     S(IPI_CALL_NMI_FUNC, "NMI function call interrupts"),
> >  };
> >
> >  static void smp_cross_call(const struct cpumask *target, unsigned int
> > ipinr);
> > @@ -840,6 +842,11 @@ void arch_irq_work_raise(void)
> >  }
> >  #endif
> >
> > +void arch_send_call_nmi_func_ipi_mask(const struct cpumask *mask)
> > +{
> > +     smp_cross_call(mask, IPI_CALL_NMI_FUNC);
> > +}
> > +
> >  static void local_cpu_stop(void)
> >  {
> >       set_cpu_online(smp_processor_id(), false);
> > @@ -932,6 +939,10 @@ static void do_handle_IPI(int ipinr)
> >               break;
> >  #endif
> >
> > +     case IPI_CALL_NMI_FUNC:
> > +             /* nop, IPI handlers for special features can be added here. */
> > +             break;
> > +
> >       default:
> >               pr_crit("CPU%u: Unknown IPI message 0x%x\n", cpu, ipinr);
> >               break;
>
> I'm really not keen on adding more IPIs to the SMP code. One of the
> main reasons for using these SGIs as normal IRQs was to make them
> "requestable" from non-arch code as if they were standard percpu
> interrupts.
>
> What prevents you from moving that all the way to the kgdb code?
>

Since we only have one SGI left (SGI7) which this patch reserves for
NMI purposes, I am not sure what value add do you see to make it
requestable from non-arch code.

Also, allocating SGI7 entirely to kgdb would not allow us to leverage
it for NMI backtrace support via magic sysrq. But with current
implementation we should be able to achieve that.

Moreover, irq ids for normal interrupts are assigned via DT/ACPI, how
do you envision this for SGIs?

-Sumit

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