Re: [PATCH 20/37] softirq: Introduce disabled softirq vectors bits

[Sebastian Andrzej Siewior]

On 2019-02-28 18:12:25 [+0100], Frederic Weisbecker wrote:
> --- a/include/linux/interrupt.h
> +++ b/include/linux/interrupt.h
> @@ -474,17 +474,62 @@ enum
…
> +static inline unsigned int local_softirq_pending(void)
> +{
> +     return local_softirq_data() & SOFTIRQ_PENDING_MASK;
> +}
…

I'm still digesting but…

> +static inline void softirq_enabled_set(unsigned int enabled)
> +{
> +     unsigned int data;
> +
> +     data = enabled << SOFTIRQ_ENABLED_SHIFT;
> +     data |= local_softirq_pending();

if an interrupt occurs at this point and invokes
        raise_softirq_irqof() => softirq_pending_set_mask()
        
then we lose the pending bits, correct? This may be invoked from
        local_bh_enable() => local_bh_enable_ip_mask() =>
        local_bh_enable_common() => softirq_enabled_set()

and as far as I can, interrupts are only disabled in the tracing case.

> +     __this_cpu_write(local_softirq_data_ref, data);
> +}

Sebastian