Re: [PATCH 3/6] riscv: optimize send_ipi_single

[Atish Patra <Atish.Patra@wdc.com>]

On Wed, 2019-08-21 at 23:58 +0900, Christoph Hellwig wrote:
> Don't go through send_ipi_mask, but just set the op bit an then pass
> a
> simple generate hartid mask directly to sbi_send_ipi.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>  arch/riscv/kernel/smp.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/riscv/kernel/smp.c b/arch/riscv/kernel/smp.c
> index 2e21669aa068..a3715d621f60 100644
> --- a/arch/riscv/kernel/smp.c
> +++ b/arch/riscv/kernel/smp.c
> @@ -94,7 +94,13 @@ static void send_ipi_mask(const struct cpumask
> *mask, enum ipi_message_type op)
>  
>  static void send_ipi_single(int cpu, enum ipi_message_type op)
>  {
> -	send_ipi_mask(cpumask_of(cpu), op);
> +	int hartid = cpuid_to_hartid_map(cpu);
> +
> +	smp_mb__before_atomic();
> +	set_bit(op, &ipi_data[cpu].bits);
> +	smp_mb__after_atomic();
> +
> +	sbi_send_ipi(cpumask_bits(cpumask_of(hartid)));

The only cost save I see is you don't have to cpumask_clear anymore.
Is there any other cost save ? If not is it worth duplicating the code
?

May be I am being too pedantic here.. :) :)

>  }
>  
>  static inline void clear_ipi(void)

-- 
Regards,
Atish
_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv