RE: [PATCH v5] cpuidle, ACPI: Evaluate LPI arch_flags for broadcast timer

[Pawandeep Oza (QUIC)]

-----Original Message-----
From: Will Deacon <will@xxxxxxxxxx> 
Sent: Thursday, September 14, 2023 6:22 AM
To: Pawandeep Oza (QUIC) <quic_poza@xxxxxxxxxxx>
Cc: sudeep.holla@xxxxxxx; catalin.marinas@xxxxxxx; rafael@xxxxxxxxxx; lenb@xxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; linux-acpi@xxxxxxxxxxxxxxx
Subject: Re: [PATCH v5] cpuidle, ACPI: Evaluate LPI arch_flags for broadcast timer

On Tue, Sep 12, 2023 at 10:29:33AM -0700, Oza Pawandeep wrote:
> Arm(r) Functional Fixed Hardware Specification defines LPI states, which 
> provide an architectural context loss flags field that can be used to 
> describe the context that might be lost when an LPI state is entered.
> 
> - Core context Lost
>         - General purpose registers.
>         - Floating point and SIMD registers.
>         - System registers, include the System register based
>         - generic timer for the core.
>         - Debug register in the core power domain.
>         - PMU registers in the core power domain.
>         - Trace register in the core power domain.
> - Trace context loss
> - GICR
> - GICD
> 
> Qualcomm's custom CPUs preserves the architectural state, including 
> keeping the power domain for local timers active.
> when core is power gated, the local timers are sufficient to wake the 
> core up without needing broadcast timer.
> 
> The patch fixes the evaluation of cpuidle arch_flags, and moves only 
> to broadcast timer if core context lost is defined in ACPI LPI.
> 
> Reviewed-by: Sudeep Holla <sudeep.holla@xxxxxxx>
> Signed-off-by: Oza Pawandeep <quic_poza@xxxxxxxxxxx>
> 
> diff --git a/arch/arm64/include/asm/acpi.h 
> b/arch/arm64/include/asm/acpi.h index 4d537d56eb84..a30b6e16628d 
> 100644
> --- a/arch/arm64/include/asm/acpi.h
> +++ b/arch/arm64/include/asm/acpi.h
> @@ -9,6 +9,7 @@
>  #ifndef _ASM_ACPI_H
>  #define _ASM_ACPI_H
>  
> +#include <linux/cpuidle.h>
>  #include <linux/efi.h>
>  #include <linux/memblock.h>
>  #include <linux/psci.h>
> @@ -44,6 +45,25 @@
>  
>  #define ACPI_MADT_GICC_TRBE  (offsetof(struct acpi_madt_generic_interrupt, \
>  	trbe_interrupt) + sizeof(u16))
> +/*
> + * Arm(r) Functional Fixed Hardware Specification Version 1.2.
> + * Table 2: Arm Architecture context loss flags  */
> +#define CPUIDLE_CORE_CTXT		BIT(0) /* Core context Lost */
> +
> +#ifndef arch_update_idle_state_flags
> +static __always_inline void _arch_update_idle_state_flags(u32 arch_flags,
> +							unsigned int *sflags)
> +{
> +	if (arch_flags & CPUIDLE_CORE_CTXT)
> +		*sflags |= CPUIDLE_FLAG_TIMER_STOP; } #define 
> +arch_update_idle_state_flags _arch_update_idle_state_flags #endif

Why do you need the #ifndef/endif guards here? I'd have thought you'd _always_ want this definition to be the one used, with the compiler complaining otherwise.

Oza: thanks, Will!, will remove it.

Will