Linux driver-core infrastructure
 help / color / mirror / Atom feed
From: "Onur Özkan" <work@onurozkan.dev>
To: Beata Michalska <beata.michalska@arm.com>
Cc: Miguel Ojeda <ojeda@kernel.org>,
	Danilo Krummrich <dakr@kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	"Rafael J . Wysocki" <rafael@kernel.org>,
	Boqun Feng <boqun@kernel.org>, Gary Guo <gary@garyguo.net>,
	Bjorn Roy Baron <bjorn3_gh@protonmail.com>,
	Benno Lossin <lossin@kernel.org>,
	Andreas Hindborg <a.hindborg@kernel.org>,
	Alice Ryhl <aliceryhl@google.com>,
	Trevor Gross <tmgross@umich.edu>,
	Daniel Almeida <daniel.almeida@collabora.com>,
	Boris Brezillon <boris.brezillon@collabora.com>,
	rust-for-linux@vger.kernel.org, driver-core@lists.linux.dev,
	linux-kernel@vger.kernel.org
Subject: Re: [RFC PATCH 1/3] rust: Add runtime PM support
Date: Sun, 17 May 2026 10:20:54 +0300	[thread overview]
Message-ID: <20260517072420.38902-1-work@onurozkan.dev> (raw)
In-Reply-To: <20260514150957.3501924-2-beata.michalska@arm.com>

Hi Beata,

Thank you for working on this. This will eventually be needed for the
tyr GPU reset implementation.

On Thu, 14 May 2026 17:09:03 +0200
Beata Michalska <beata.michalska@arm.com> wrote:

> Introduce a Rust abstraction for Linux runtime PM, centered around a
> `PMContext` ownign a device reference and modeling the runtime PM
> lifecycle. The context is initialised inactive and transitions into
> active once the runtime PM gets enabled. The drivers might opt in to
> initialize the devices power state prior to enabling the context, which
> is advised at probe, when the actual PM callbacks cannot be fully served.
> Once active, the context provides access to runtime PM operations and
> request modes used to manage device power state transitions.
> 
> Runtime PM callbacks are provided via a PMOps trait. The generated
> dev_pm_ops callbacks retrieve the driver's PMContext from drvdata,
> validate the expected transition, and transfer ownership of the
> driver-defined runtime PM payload to the relevant callback. Once
> the callback completes, ownership of the returned payload is
> transferred back to the PMContext and stored for the next transition.
> The payload returned by the callback is expected to reflect the
> current runtime PM state of the device. On a successful transition,
> the returned payload should represent the new state reached by
> the device. If the transition fails, it is the responsibility of
> the driver to return a payload either rolled back to the previous
> state or left in a state that remains valid for subsequent runtime PM
> transitions. At present, only runtime_suspend and runtime_resume
> callbacks are supported.
> 
> Profile and scoped runtime PM helpers are provided for common usage
> patterns. ResumeScope resumes the device for the lifetime of the scope,
> AwakeScope resumes the device while holding a runtime PM usage reference,
> and RetainScope only increments the runtime PM usage count. References
> are released on drop, and request behaviour is controlled through
> driver-defined Profile values covering the common runtime PM request
> modes.
> 
> Additionally, PMConfig provides common runtime PM configuration options.
> 
> Signed-off-by: Beata Michalska <beata.michalska@arm.com>
> ---
>  rust/bindings/bindings_helper.h |   1 +
>  rust/helpers/helpers.c          |   1 +
>  rust/helpers/pm_runtime.c       |  38 ++
>  rust/kernel/lib.rs              |   1 +
>  rust/kernel/pm.rs               | 924 ++++++++++++++++++++++++++++++++
>  5 files changed, 965 insertions(+)
>  create mode 100644 rust/helpers/pm_runtime.c
>  create mode 100644 rust/kernel/pm.rs
> 
> diff --git a/rust/bindings/bindings_helper.h b/rust/bindings/bindings_helper.h
> index 446dbeaf0866..3a38db22681f 100644
> --- a/rust/bindings/bindings_helper.h
> +++ b/rust/bindings/bindings_helper.h
> @@ -76,6 +76,7 @@
>  #include <linux/pid_namespace.h>
>  #include <linux/platform_device.h>
>  #include <linux/pm_opp.h>
> +#include <linux/pm_runtime.h>
>  #include <linux/poll.h>
>  #include <linux/property.h>
>  #include <linux/pwm.h>
> diff --git a/rust/helpers/helpers.c b/rust/helpers/helpers.c
> index 625921e27dfb..d67c69ebd33e 100644
> --- a/rust/helpers/helpers.c
> +++ b/rust/helpers/helpers.c
> @@ -75,6 +75,7 @@
>  #include "pci.c"
>  #include "pid_namespace.c"
>  #include "platform.c"
> +#include "pm_runtime.c"
>  #include "poll.c"
>  #include "processor.c"
>  #include "property.c"
> diff --git a/rust/helpers/pm_runtime.c b/rust/helpers/pm_runtime.c
> new file mode 100644
> index 000000000000..9b2056110199
> --- /dev/null
> +++ b/rust/helpers/pm_runtime.c
> @@ -0,0 +1,38 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +#include <linux/pm_runtime.h>
> +
> +void rust_helper_pm_runtime_get_noresume(struct device *dev)
> +{
> +	pm_runtime_get_noresume(dev);
> +}
> +
> +void rust_helper_pm_runtime_put_noidle(struct device *dev)
> +{
> +	pm_runtime_put_noidle(dev);
> +}
> +
> +void rust_helper_pm_runtime_mark_last_busy(struct device *dev)
> +{
> +	pm_runtime_mark_last_busy(dev) ;
> +}
> +
> +bool rust_helper_pm_runtime_active(struct device *dev)
> +{
> +	return pm_runtime_active(dev);
> +}
> +
> +void rust_helper_pm_suspend_ignore_children(struct device *dev, bool enable)
> +{
> +	pm_suspend_ignore_children(dev, enable);
> +}
> +
> +int rust_helper_pm_runtime_set_active(struct device *dev)
> +{
> +	return pm_runtime_set_active(dev);
> +}
> +
> +int rust_helper_pm_runtime_set_suspended(struct device *dev)
> +{
> +	return pm_runtime_set_suspended(dev);
> +}

You should add __rust_helper prefixes to helper functions.

> diff --git a/rust/kernel/lib.rs b/rust/kernel/lib.rs
> index b72b2fbe046d..7e51e497e2f0 100644
> --- a/rust/kernel/lib.rs
> +++ b/rust/kernel/lib.rs
> @@ -105,6 +105,7 @@
>  pub mod pci;
>  pub mod pid_namespace;
>  pub mod platform;
> +pub mod pm;
>  pub mod prelude;
>  pub mod print;
>  pub mod processor;
> diff --git a/rust/kernel/pm.rs b/rust/kernel/pm.rs
> new file mode 100644
> index 000000000000..7bdffa727c71
> --- /dev/null
> +++ b/rust/kernel/pm.rs
> @@ -0,0 +1,924 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +//! Rust Runtime Power Management abstraction.
> +//!
> +//! C header: [`include/linux/pm_runtime.h`](srctree/include/linux/pm_runtime.h)
> +
> +use crate::{
> +    bindings, device,
> +    error::{to_result, VTABLE_DEFAULT_ERROR},
> +    macros::paste,
> +    prelude::*,
> +    sync::aref::ARef,
> +    sync::atomic::{
> +        ordering::{Acquire, Full, Release},
> +        Atomic,
> +    },
> +    sync::Arc,
> +    sync::{new_mutex, Mutex},
> +};

We started using vertical format on imports like:

	use crate::{
		bindings,
		device,
		error::{
			to_result,
			VTABLE_DEFAULT_ERROR,
			//
		},
		...

to avoid conflict issues.

> +
> +use core::marker::PhantomData;
> +
> +/// Runtime Power Management modes that determine how a particular PM
> +/// transition is to be carried out.
> +/// Corresponds to C Runtime PM flag argument bits:
> +/// - `RPM_ASYNC`
> +/// - `RPM_NOWAIT`
> +/// - `RPM_GET_PUT`
> +/// - `RPM_AUTO`
> +#[derive(Clone, Copy, PartialEq, Eq, Debug)]
> +struct Mode(u32);
> +
> +impl Mode {
> +    /// Synchronous PM operations - default.
> +    #[allow(unused)]
> +    const SYNC: Mode = Mode(0);

This seems to be used in Profile::new, why do we need `#[allow(unused)]` here?

> +    /// Allow asynchronous PM operations.
> +    const ASYNC: Mode = Mode(bindings::RPM_ASYNC);
> +    /// Do not wait for any pending rquests to finish.
> +    const NOWAIT: Mode = Mode(bindings::RPM_NOWAIT);
> +    /// Acquire a runtime-PM usage reference.
> +    const ACQUIRE: Mode = Mode(bindings::RPM_GET_PUT);
> +    /// Use autosuspend.
> +    const AUTO: Mode = Mode(bindings::RPM_AUTO);
> +    /// Additional mode for devices supporting idle states.
> +    const IDLE: Mode = Mode(1 << 16);
> +
> +    const fn join(self, other: Self) -> Self {
> +        Self(self.0 | other.0)
> +    }
> +
> +    fn contains(self, other: Self) -> bool {
> +        (self.0 & other.0) != 0
> +    }
> +}
> +
> +impl core::ops::BitOr for Mode {
> +    type Output = Self;
> +    fn bitor(self, rhs: Self) -> Self::Output {
> +        Self(self.0 | rhs.0)
> +    }
> +}
> +
> +impl core::ops::BitAnd for Mode {
> +    type Output = Self;
> +    fn bitand(self, rhs: Self) -> Self::Output {
> +        Self(self.0 & rhs.0)
> +    }
> +}
> +
> +impl core::ops::Not for Mode {
> +    type Output = Self;
> +    fn not(self) -> Self::Output {
> +        Self(!self.0)
> +    }
> +}
> +
> +impl From<Mode> for core::ffi::c_int {
> +    #[inline]
> +    fn from(mode: Mode) -> core::ffi::c_int {
> +        mode.0 as core::ffi::c_int
> +    }
> +}
> +
> +/// Utility macro for combining multiple request modes
> +macro_rules! mode {
> +    ($mode:expr $(, $args:expr)+ $(,)?) => {{
> +        let mut new_mode = $mode;
> +        $( new_mode = new_mode.join($args); )+
> +        new_mode
> +    }};
> +}
> +
> +/// Runtime power transition scope.
> +pub struct Scope<Tag> {
> +    dev: ARef<device::Device>,
> +    mode: Mode,
> +    _tag: PhantomData<Tag>,
> +}
> +
> +/// Device resumed without incrementing the device's usage count
> +pub struct Resume;
> +/// Device resumed with the devices'a usage count being incremented
> +pub struct Awake;
> +/// Device with increased usage reference
> +pub struct Retain;
> +
> +/// Resumes the device without acquiring the usage reference.
> +/// Note: This does not guarantee the device will be kept active
> +/// for the lifetime of the scope due to potentil pending suspend
> +/// requests.
> +///
> +/// On drop:
> +/// - If `Mode::IDLE`, calls `__pm_runtime_idle()`:
> +///   triggers idle notification before attempting to suspend
> +/// - If `Mode::AUTO`, marks last busy then calls `__pm_runtime_suspend()`.
> +/// - Otherwise calls `__pm_runtime_suspend()`.
> +///
> +/// The guard must be dropped from a context matching the requested transition
> +/// mode: sync vs async.
> +#[must_use = "dropping this guard issues the matching runtime PM release request"]
> +pub struct ResumeScope(Scope<Resume>);
> +
> +/// Acquires a runtime-PM usage reference and keeps the device powered.
> +///
> +/// Requires `Mode::ACQUIRE`. Drop behavior matches `ResumeScope`.
> +/// The guard must be dropped from a context matching the requested transition
> +/// mode: sync vs async.
> +#[must_use = "dropping this guard releases its runtime PM hold"]
> +pub struct AwakeScope(Scope<Awake>);
> +
> +/// Prevents the device from getting suspended by holding the usage reference
> +/// count.
> +///
> +/// On drop, calls `pm_runtime_put_noidle()`.
> +#[must_use = "dropping this guard releases its runtime PM hold"]
> +pub struct RetainScope(Scope<Retain>);
> +
> +impl ResumeScope {
> +    fn new(dev: ARef<device::Device>, mode: Mode) -> Result<Self> {
> +        if mode.contains(Mode::ACQUIRE) {
> +            // Mode::ACQUIRE is intended to be used with Awake scope
> +            // Avoid mixing the modes.
> +            return Err(EINVAL);
> +        }
> +
> +        // Mode::IDLE is internal so strip it of before passing further
> +        Request::resume(&dev, mode & !Mode::IDLE).map(|()| {
> +            Self(Scope::<Resume> {
> +                dev: dev.clone(),
> +                mode,
> +                _tag: PhantomData,
> +            })
> +        })
> +    }
> +}
> +
> +impl Drop for ResumeScope {
> +    fn drop(&mut self) {
> +        let scope_mode = self.0.mode & !Mode::IDLE;
> +
> +        match self.0.mode {
> +            mode if mode.contains(Mode::IDLE) => {
> +                Request::idle(&self.0.dev, scope_mode & Mode::ASYNC);
> +            }
> +            mode if mode.contains(Mode::AUTO) => {
> +                Request::mark_last_busy(&self.0.dev);
> +                let _ = Request::suspend(&self.0.dev, scope_mode).inspect_err(|_| {
> +                    dev_err!(self.0.dev.as_ref(), "Failed to suspend device\n");
> +                });
> +            }
> +            _ => {
> +                let _ = Request::suspend(&self.0.dev, scope_mode).inspect_err(|_| {
> +                    dev_err!(self.0.dev.as_ref(), "Failed to suspend device\n");
> +                });
> +            }
> +        }
> +    }
> +}
> +
> +impl AwakeScope {
> +    fn new(dev: ARef<device::Device>, mode: Mode) -> Result<Self> {
> +        if !mode.contains(Mode::ACQUIRE) {
> +            return Err(EINVAL);
> +        }
> +        // Mode::IDLE is internal so strip it of before passing further
> +        Request::resume(&dev, mode & !Mode::IDLE)
> +            .inspect_err(|_| Request::put_noidle(&dev))
> +            .map(|()| {
> +                Self(Scope::<Awake> {
> +                    dev: dev.clone(),
> +                    mode,
> +                    _tag: PhantomData,
> +                })
> +            })
> +    }
> +}
> +
> +impl Drop for AwakeScope {
> +    fn drop(&mut self) {
> +        let scope_mode = self.0.mode & !Mode::IDLE;
> +        match self.0.mode {
> +            mode if mode.contains(Mode::IDLE) => {
> +                Request::idle(&self.0.dev, scope_mode);
> +            }
> +            mode if mode.contains(Mode::AUTO) => {
> +                Request::mark_last_busy(&self.0.dev);
> +                let _ = Request::suspend(&self.0.dev, scope_mode).inspect_err(|_| {
> +                    dev_err!(self.0.dev.as_ref(), "Failed to suspend device\n");
> +                });
> +            }
> +            _ => {
> +                let _ = Request::suspend(&self.0.dev, scope_mode).inspect_err(|_| {
> +                    dev_err!(self.0.dev.as_ref(), "Failed to suspend device\n");
> +                });
> +            }
> +        }
> +    }
> +}
> +
> +impl RetainScope {
> +    fn new(dev: ARef<device::Device>) -> Result<Self> {
> +        Request::get_noresume(&dev);
> +        Ok(Self(Scope::<Retain> {
> +            dev,
> +            mode: Mode(0),

This is Mode::SYNC.

Regards,
Onur

[...]

> +
> +impl<T: PMOps, State> PMContext<T, State> {
> +    /// Applies runtime PM configuration options.
> +    ///
> +    /// Options are applied in the order provided. The currently supported
> +    /// options do not report per-option failures.
> +    ///
> +    /// This may be called before or after runtime PM is enabled, depending on
> +    /// the option and the driver's setup sequence.
> +    pub fn apply_config(&self, opts: &[PMConfig]) -> Result {
> +        #[cfg(not(CONFIG_PM))]
> +        let _ = opts;
> +        #[cfg(CONFIG_PM)]
> +        for opt in opts {
> +            match opt {
> +                // SAFETY: `self.dev` is a valid `&ARef<Device>`, so the underlying `Device` is
> +                // guaranteed to be alive and `as_raw()` yields a valid pointer for the
> +                // duration of this call.
> +                PMConfig::IgnoreChildren(v) => unsafe {
> +                    bindings::pm_suspend_ignore_children(self.dev.as_raw(), *v)
> +                },
> +                // SAFETY: `self.dev` is a valid `&ARef<Device>`, so the underlying `Device` is
> +                // guaranteed to be alive and `as_raw()` yields a valid pointer for the
> +                // duration of this call.
> +                PMConfig::NoCallbacks => unsafe {
> +                    bindings::pm_runtime_no_callbacks(self.dev.as_raw())
> +                },
> +                // SAFETY: `self.dev` is a valid `&ARef<Device>`, so the underlying `Device` is
> +                // guaranteed to be alive and `as_raw()` yields a valid pointer for the
> +                // duration of this call.
> +                PMConfig::IrqSafe => unsafe { bindings::pm_runtime_irq_safe(self.dev.as_raw()) },
> +                // SAFETY: `self.dev` is a valid `&ARef<Device>`, so the underlying `Device` is
> +                // guaranteed to be alive and `as_raw()` yields a valid pointer for the
> +                // duration of this call.
> +                PMConfig::AutoSuspend(v) => unsafe {
> +                    bindings::__pm_runtime_use_autosuspend(self.dev.as_raw(), *v);
> +                },
> +                // SAFETY: `self.dev` is a valid `&ARef<Device>`, so the underlying `Device` is
> +                // guaranteed to be alive and `as_raw()` yields a valid pointer for the
> +                // duration of this call.
> +                PMConfig::AutoSuspendDelay(v) => unsafe {
> +                    bindings::pm_runtime_set_autosuspend_delay(self.dev.as_raw(), *v as i32);
> +                    bindings::__pm_runtime_use_autosuspend(self.dev.as_raw(), true);
> +                },
> +            }
> +        }
> +        Ok(())
> +    }
> +}
> -- 
> 2.43.0
> 

  parent reply	other threads:[~2026-05-17  7:24 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-05-14 15:09 [RFC PATCH 0/3] Rust: add runtime PM support Beata Michalska
2026-05-14 15:09 ` [RFC PATCH 1/3] rust: Add " Beata Michalska
2026-05-17  0:08   ` Danilo Krummrich
2026-05-17  7:20   ` Onur Özkan [this message]
2026-05-14 15:09 ` [RFC PATCH 2/3] rust/platform: Add support for runtime PM Beata Michalska
2026-05-14 15:09 ` [RFC PATCH 3/3] [DO NOT MERGE] drm/tyr: wire runtime PM hooks Beata Michalska
2026-05-16 20:35 ` [RFC PATCH 0/3] Rust: add runtime PM support Miguel Ojeda

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260517072420.38902-1-work@onurozkan.dev \
    --to=work@onurozkan.dev \
    --cc=a.hindborg@kernel.org \
    --cc=aliceryhl@google.com \
    --cc=beata.michalska@arm.com \
    --cc=bjorn3_gh@protonmail.com \
    --cc=boqun@kernel.org \
    --cc=boris.brezillon@collabora.com \
    --cc=dakr@kernel.org \
    --cc=daniel.almeida@collabora.com \
    --cc=driver-core@lists.linux.dev \
    --cc=gary@garyguo.net \
    --cc=gregkh@linuxfoundation.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=lossin@kernel.org \
    --cc=ojeda@kernel.org \
    --cc=rafael@kernel.org \
    --cc=rust-for-linux@vger.kernel.org \
    --cc=tmgross@umich.edu \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox