Re: [PATCH v6 09/10] arm64: support cpuidle-haltpoll

[Ankur Arora <ankur.a.arora@oracle.com>]

Okanovic, Haris <harisokn@amazon.com> writes:

> On Mon, 2024-07-29 at 11:02 -0700, Ankur Arora wrote:
>> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
>>
>>
>>
>> Okanovic, Haris <harisokn@amazon.com> writes:
>>
>> > On Fri, 2024-07-26 at 13:21 -0700, Ankur Arora wrote:
>> > > CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
>> > >
>> > >
>> > >
>> > > Add architectural support for cpuidle-haltpoll driver by defining
>> > > arch_haltpoll_*().
>> > >
>> > > Also define ARCH_CPUIDLE_HALTPOLL to allow cpuidle-haltpoll to be
>> > > selected, and given that we have an optimized polling mechanism
>> > > in smp_cond_load*(), select ARCH_HAS_OPTIMIZED_POLL.
>> > >
>> > > smp_cond_load*() are implemented via LDXR, WFE, with LDXR loading
>> > > a memory region in exclusive state and the WFE waiting for any
>> > > stores to it.
>> > >
>> > > In the edge case -- no CPU stores to the waited region and there's no
>> > > interrupt -- the event-stream will provide the terminating condition
>> > > ensuring we don't wait forever, but because the event-stream runs at
>> > > a fixed frequency (configured at 10kHz) we might spend more time in
>> > > the polling stage than specified by cpuidle_poll_time().
>> > >
>> > > This would only happen in the last iteration, since overshooting the
>> > > poll_limit means the governor moves out of the polling stage.
>> > >
>> > > Signed-off-by: Ankur Arora <ankur.a.arora@oracle.com>
>> > > ---
>> > >  arch/arm64/Kconfig                        | 10 ++++++++++
>> > >  arch/arm64/include/asm/cpuidle_haltpoll.h |  9 +++++++++
>> > >  arch/arm64/kernel/cpuidle.c               | 23 +++++++++++++++++++++++
>> > >  3 files changed, 42 insertions(+)
>> > >  create mode 100644 arch/arm64/include/asm/cpuidle_haltpoll.h
>> > >
>> > > diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
>> > > index 5d91259ee7b5..cf1c6681eb0a 100644
>> > > --- a/arch/arm64/Kconfig
>> > > +++ b/arch/arm64/Kconfig
>> > > @@ -35,6 +35,7 @@ config ARM64
>> > >         select ARCH_HAS_MEMBARRIER_SYNC_CORE
>> > >         select ARCH_HAS_NMI_SAFE_THIS_CPU_OPS
>> > >         select ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
>> > > +       select ARCH_HAS_OPTIMIZED_POLL
>> > >         select ARCH_HAS_PTE_DEVMAP
>> > >         select ARCH_HAS_PTE_SPECIAL
>> > >         select ARCH_HAS_HW_PTE_YOUNG
>> > > @@ -2376,6 +2377,15 @@ config ARCH_HIBERNATION_HEADER
>> > >  config ARCH_SUSPEND_POSSIBLE
>> > >         def_bool y
>> > >
>> > > +config ARCH_CPUIDLE_HALTPOLL
>> > > +       bool "Enable selection of the cpuidle-haltpoll driver"
>> > > +       default n
>> > > +       help
>> > > +         cpuidle-haltpoll allows for adaptive polling based on
>> > > +         current load before entering the idle state.
>> > > +
>> > > +         Some virtualized workloads benefit from using it.
>> > > +
>> > >  endmenu # "Power management options"
>> > >
>> > >  menu "CPU Power Management"
>> > > diff --git a/arch/arm64/include/asm/cpuidle_haltpoll.h b/arch/arm64/include/asm/cpuidle_haltpoll.h
>> > > new file mode 100644
>> > > index 000000000000..65f289407a6c
>> > > --- /dev/null
>> > > +++ b/arch/arm64/include/asm/cpuidle_haltpoll.h
>> > > @@ -0,0 +1,9 @@
>> > > +/* SPDX-License-Identifier: GPL-2.0 */
>> > > +#ifndef _ARCH_HALTPOLL_H
>> > > +#define _ARCH_HALTPOLL_H
>> > > +
>> > > +static inline void arch_haltpoll_enable(unsigned int cpu) { }
>> > > +static inline void arch_haltpoll_disable(unsigned int cpu) { }
>> > > +
>> > > +bool arch_haltpoll_want(bool force);
>> > > +#endif
>> > > diff --git a/arch/arm64/kernel/cpuidle.c b/arch/arm64/kernel/cpuidle.c
>> > > index f372295207fb..334df82a0eac 100644
>> > > --- a/arch/arm64/kernel/cpuidle.c
>> > > +++ b/arch/arm64/kernel/cpuidle.c
>> > > @@ -72,3 +72,26 @@ __cpuidle int acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
>> > >                                              lpi->index, state);
>> > >  }
>> > >  #endif
>> > > +
>> > > +#if IS_ENABLED(CONFIG_HALTPOLL_CPUIDLE)
>> > > +
>> > > +#include <asm/cpuidle_haltpoll.h>
>> > > +
>> > > +bool arch_haltpoll_want(bool force)
>> > > +{
>> > > +       /*
>> > > +        * Enabling haltpoll requires two things:
>> > > +        *
>> > > +        * - Event stream support to provide a terminating condition to the
>> > > +        *   WFE in the poll loop.
>> > > +        *
>> > > +        * - KVM support for arch_haltpoll_enable(), arch_haltpoll_enable().
>> >
>> > typo: "arch_haltpoll_enable" and "arch_haltpoll_enable"
>> >
>> > > +        *
>> > > +        * Given that the second is missing, allow haltpoll to only be force
>> > > +        * loaded.
>> > > +        */
>> > > +       return (arch_timer_evtstrm_available() && false) || force;
>> >
>> > This should always evaluate false without force. Perhaps you meant
>> > something like this?
>> >
>> > ```
>> > -       return (arch_timer_evtstrm_available() && false) || force;
>> > +       return arch_timer_evtstrm_available() || force;
>> > ```
>>
>> No. This was intentional. As I meniton in the comment above, right now
>> the KVM support is missing. Which means that the guest has no way to
>> tell the host to not poll as part of host haltpoll.
>>
>> Until that is available, only allow force loading.
>
> I see, arm64's kvm is missing the poll control mechanism.
>
> I'll follow-up your changes with a patch for AWS Graviton; still seeing
> the same performance gains.

Excellent. Could you Cc me when you send out your changes?

> Tested-by: Haris Okanovic <harisokn@amazon.com>

Thanks!

--
ankur