On 01.08.2019 12:22, Chao Gao wrote:
> @@ -439,12 +440,37 @@ static int do_microcode_update(void *patch)
> return ret;
> }
>
> +static int microcode_nmi_callback(const struct cpu_user_regs *regs, int cpu)
> +{
> + bool print = false;
> +
> + /* The first thread of a core is to load an update. Don't block it. */
> + if ( cpu == cpumask_first(per_cpu(cpu_sibling_mask, cpu)) )
> + return 0;
> +
> + if ( loading_state == LOADING_ENTERED )
> + {
> + cpumask_set_cpu(cpu, &cpu_callin_map);
> + printk(XENLOG_DEBUG "CPU%u enters %s\n", smp_processor_id(),
> __func__);
Here and ...
> + print = true;
> + }
> +
> + while ( loading_state == LOADING_ENTERED )
> + rep_nop();
> +
> + if ( print )
> + printk(XENLOG_DEBUG "CPU%u exits %s\n", smp_processor_id(),
> __func__);
... here - why smp_processor_id() when you can use "cpu"? And what
use is __func__ here?
The rep_nop() above also presumably wants to be cpu_relax() again.
But on the whole I was really hoping for more aggressive disabling
of NMI handling, more like (but of course not quite as heavy as)
the crash path wiring the IDT entry to trap_nop(). Andrew, I'm
curious to learn what you're thinking would be best here.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
