Re: [RFC PATCH] powerpc/64s/radix: introduce option to disable broadcast tlbie

[Michael Ellerman <mpe@ellerman.id.au>]

Hi Nick,

Just a few comments.

Nicholas Piggin <npiggin@gmail.com> writes:
> diff --git a/arch/powerpc/mm/book3s64/radix_tlb.c b/arch/powerpc/mm/book3s64/radix_tlb.c
> index 71f7fede2fa4..56ceecbd3d5c 100644
> --- a/arch/powerpc/mm/book3s64/radix_tlb.c
> +++ b/arch/powerpc/mm/book3s64/radix_tlb.c
> @@ -285,6 +286,30 @@ static inline void _tlbie_pid(unsigned long pid, unsigned long ric)
>  	asm volatile("eieio; tlbsync; ptesync": : :"memory");
>  }
>  
> +struct tlbiel_pid {
> +	unsigned long pid;
> +	unsigned long ric;
> +};
> +
> +static void do_tlbiel_pid(void *info)
> +{
> +	struct tlbiel_pid *t = info;
> +
> +	if (t->ric == RIC_FLUSH_TLB)
> +		_tlbiel_pid(t->pid, RIC_FLUSH_TLB);
> +	else if (t->ric == RIC_FLUSH_PWC)
> +		_tlbiel_pid(t->pid, RIC_FLUSH_PWC);
> +	else
> +		_tlbiel_pid(t->pid, RIC_FLUSH_ALL);
> +}
> +
> +static inline void _tlbiel_pid_broadcast(const struct cpumask *cpus,
> +				unsigned long pid, unsigned long ric)

Can we call these "multicast" instead of "broadcast"?

I think that's more accurate, and avoids confusion with tlbie which
literally does a broadcast (at least architecturally).

> @@ -524,6 +604,12 @@ static bool mm_needs_flush_escalation(struct mm_struct *mm)
>  	return false;
>  }
>  
> +static bool tlbie_enabled = true;
> +static bool use_tlbie(void)
> +{
> +	return tlbie_enabled;
> +}

No synchronisation, but that's OK. Would probably be good to have
a comment though explaining why.

We could use a static_key but I guess the overhead of a comparison and
branch is in the noise vs the tlbie/tlbiel.

> @@ -1100,3 +1221,13 @@ extern void radix_kvm_prefetch_workaround(struct mm_struct *mm)
>  }
>  EXPORT_SYMBOL_GPL(radix_kvm_prefetch_workaround);
>  #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
> +
> +static int __init radix_tlb_setup(void)
> +{
> +	debugfs_create_bool("tlbie_enabled", 0600,
> +			powerpc_debugfs_root,
> +			&tlbie_enabled);
> +
> +	return 0;
> +}
> +arch_initcall(radix_tlb_setup);

For working around hardware bugs we would want a command line parameter
or other boot time way to flip this. But I guess you're saying because
we haven't converted all uses of tlbie we can't really support that
anyway, and so a runtime switch is sufficient?

cheers