[PATCH v2 2/2] arm64: use WFE for long delays

[Will Deacon]

On Thu, Aug 17, 2017 at 09:39:20AM +0100, Julien Thierry wrote:
> The current delay implementation uses the yield instruction, which is a hint
> that it is beneficial to schedule another thread. As this is a hint, it may be
> implemented as a NOP, causing all delays to be busy loops. This is the case for
> many existing CPUs.
> 
> Taking advantage of the generic timer sending periodic events to all cores, we
> can use WFE during delays to reduce power consumption. This is beneficial only
> for delays longer than the period of the timer event stream.
> 
> If timer event stream is not enabled, delays will behave as yield/busy loops.
> 
> Signed-off-by: Julien Thierry <julien.thierry at arm.com>
> Cc: Catalin Marinas <catalin.marinas at arm.com>
> Cc: Will Deacon <will.deacon at arm.com>
> Cc: Mark Rutland <mark.rutland at arm.com>
> Cc: Arnd Bergmann <arnd at arndb.de>
> ---
>  arch/arm64/lib/delay.c      | 25 ++++++++++++++++++++-----
>  include/asm-generic/delay.h |  9 +++++++--
>  2 files changed, 27 insertions(+), 7 deletions(-)
> 
> diff --git a/arch/arm64/lib/delay.c b/arch/arm64/lib/delay.c
> index dad4ec9..ada7005 100644
> --- a/arch/arm64/lib/delay.c
> +++ b/arch/arm64/lib/delay.c
> @@ -24,10 +24,28 @@
>  #include <linux/module.h>
>  #include <linux/timex.h>
> 
> +#include <clocksource/arm_arch_timer.h>
> +
> +#define USECS_TO_CYCLES(TIME_USECS)			\
> +	(xloops_to_cycles(usecs_to_xloops(TIME_USECS)))
> +
> +static inline unsigned long xloops_to_cycles(unsigned long xloops)
> +{
> +	return (xloops * loops_per_jiffy * HZ) >> 32;
> +}
> +
>  void __delay(unsigned long cycles)
>  {
>  	cycles_t start = get_cycles();
> 
> +	if (arch_timer_evtstrm_available()) {
> +		const cycles_t timer_evt_period =
> +			USECS_TO_CYCLES(1000000 / ARCH_TIMER_EVT_STREAM_FREQ);

Given that ARCH_TIMER_EVT_STREAM_FREQ is defined as:

	#define ARCH_TIMER_EVT_STREAM_FREQ      10000   /* 100us */

perhaps it would actually be better to define all of this in terms of
a 100us period.

> +
> +		while (get_cycles() - start + timer_evt_period < cycles)

Please add some brackets here for clarify.

> +			wfe();
> +	}
> +
>  	while ((get_cycles() - start) < cycles)
>  		cpu_relax();
>  }
> @@ -35,16 +53,13 @@ void __delay(unsigned long cycles)
> 
>  inline void __const_udelay(unsigned long xloops)
>  {
> -	unsigned long loops;
> -
> -	loops = xloops * loops_per_jiffy * HZ;
> -	__delay(loops >> 32);
> +	__delay(xloops_to_cycles(xloops));
>  }
>  EXPORT_SYMBOL(__const_udelay);
> 
>  void __udelay(unsigned long usecs)
>  {
> -	__const_udelay(usecs * 0x10C7UL); /* 2**32 / 1000000 (rounded up) */
> +	__const_udelay(usecs_to_xloops(usecs));
>  }
>  EXPORT_SYMBOL(__udelay);
> 
> diff --git a/include/asm-generic/delay.h b/include/asm-generic/delay.h
> index 0f79054..1538e58 100644
> --- a/include/asm-generic/delay.h
> +++ b/include/asm-generic/delay.h
> @@ -10,19 +10,24 @@
>  extern void __const_udelay(unsigned long xloops);
>  extern void __delay(unsigned long loops);
> 
> +/* 0x10c7 is 2**32 / 1000000 (rounded up) */
> +static inline unsigned long usecs_to_xloops(unsigned long usecs)
> +{
> +	return usecs * 0x10C7UL;
> +}

I'm not sure it's worth factoring this out tbh. It's not got lots of use,
and you haven't done one for nsecs so I'd be inclined to leave the generic
code as-is.

Will