diff mbox series

[v3,12/13] rust: sync: reduce stack usage of `UniqueArc::try_new_uninit`

Message ID 20230329223239.138757-13-y86-dev@protonmail.com
State New
Headers
Series Rust pin-init API for pinned initialization of structs |

Commit Message

y86-dev March 29, 2023, 10:33 p.m. UTC 
  From: Benno Lossin <y86-dev@protonmail.com>

`UniqueArc::try_new_uninit` calls `Arc::try_new(MaybeUninit::uninit())`.
This results in the uninitialized memory being placed on the stack,
which may be arbitrarily large due to the generic `T` and thus could
cause a stack overflow for large types.

Change the implementation to use the pin-init API which enables in-place
initialization. In particular it avoids having to first construct and
then move the uninitialized memory from the stack into the final location.

Signed-off-by: Benno Lossin <y86-dev@protonmail.com>
---
 rust/kernel/lib.rs      |  1 -
 rust/kernel/sync/arc.rs | 14 ++++++++++++--
 2 files changed, 12 insertions(+), 3 deletions(-)

--
2.39.2

Comments

Gary Guo March 30, 2023, 11:08 a.m. UTC | #1 
On Wed, 29 Mar 2023 22:33:49 +0000
y86-dev@protonmail.com wrote:

> From: Benno Lossin <y86-dev@protonmail.com>
> 
> `UniqueArc::try_new_uninit` calls `Arc::try_new(MaybeUninit::uninit())`.
> This results in the uninitialized memory being placed on the stack,
> which may be arbitrarily large due to the generic `T` and thus could
> cause a stack overflow for large types.
> 
> Change the implementation to use the pin-init API which enables in-place
> initialization. In particular it avoids having to first construct and
> then move the uninitialized memory from the stack into the final location.
> 
> Signed-off-by: Benno Lossin <y86-dev@protonmail.com>

Reviewed-by: Gary Guo <gary@garyguo.net>

> ---
>  rust/kernel/lib.rs      |  1 -
>  rust/kernel/sync/arc.rs | 14 ++++++++++++--
>  2 files changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/rust/kernel/lib.rs b/rust/kernel/lib.rs
> index 3e2777d26ff5..d9df77132fa2 100644
> --- a/rust/kernel/lib.rs
> +++ b/rust/kernel/lib.rs
> @@ -27,7 +27,6 @@
>  #[cfg(not(CONFIG_RUST))]
>  compile_error!("Missing kernel configuration for conditional compilation");
> 
> -#[allow(unused_extern_crates)]
>  // Allow proc-macros to refer to `::kernel` inside the `kernel` crate (this crate).
>  extern crate self as kernel;
> 
> diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
> index 77a3833cc265..4ed6329a5e5f 100644
> --- a/rust/kernel/sync/arc.rs
> +++ b/rust/kernel/sync/arc.rs
> @@ -18,6 +18,7 @@
>  use crate::{
>      bindings,
>      error::{Error, Result},
> +    init,
>      init::{InPlaceInit, Init, PinInit},
>      types::{ForeignOwnable, Opaque},
>  };
> @@ -29,6 +30,7 @@ use core::{
>      pin::Pin,
>      ptr::NonNull,
>  };
> +use macros::pin_data;
> 
>  /// A reference-counted pointer to an instance of `T`.
>  ///
> @@ -121,6 +123,7 @@ pub struct Arc<T: ?Sized> {
>      _p: PhantomData<ArcInner<T>>,
>  }
> 
> +#[pin_data]
>  #[repr(C)]
>  struct ArcInner<T: ?Sized> {
>      refcount: Opaque<bindings::refcount_t>,
> @@ -501,9 +504,16 @@ impl<T> UniqueArc<T> {
> 
>      /// Tries to allocate a new [`UniqueArc`] instance whose contents are not initialised yet.
>      pub fn try_new_uninit() -> Result<UniqueArc<MaybeUninit<T>>> {
> -        Ok(UniqueArc::<MaybeUninit<T>> {
> +        // INVARIANT: The refcount is initialised to a non-zero value.
> +        let inner = Box::init(init!(ArcInner {
> +            // SAFETY: There are no safety requirements for this FFI call.
> +            refcount: Opaque::new(unsafe { bindings::REFCOUNT_INIT(1) }),
> +            data <- init::uninit(),
> +        }))?;
> +        Ok(UniqueArc {
>              // INVARIANT: The newly-created object has a ref-count of 1.
> -            inner: Arc::try_new(MaybeUninit::uninit())?,
> +            // SAFETY: The pointer from the `Box` is valid.
> +            inner: unsafe { Arc::from_inner(Box::leak(inner).into()) },
>          })
>      }
>  }
> --
> 2.39.2
> 
>
David Laight March 30, 2023, 2:58 p.m. UTC | #2 
From: y86-dev@protonmail.com
> Sent: 29 March 2023 23:34
> 
> `UniqueArc::try_new_uninit` calls `Arc::try_new(MaybeUninit::uninit())`.
> This results in the uninitialized memory being placed on the stack,
> which may be arbitrarily large due to the generic `T` and thus could
> cause a stack overflow for large types.

Does that mean rust is using (the equivalent of) alloca() ?

That is banned for C code in the kernel for any sizes.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
y86-dev March 30, 2023, 3:16 p.m. UTC | #3 
On 30.03.23 16:58, David Laight wrote:
> From: y86-dev@protonmail.com
>> Sent: 29 March 2023 23:34
>>
>> `UniqueArc::try_new_uninit` calls `Arc::try_new(MaybeUninit::uninit())`.
>> This results in the uninitialized memory being placed on the stack,
>> which may be arbitrarily large due to the generic `T` and thus could
>> cause a stack overflow for large types.
>
> Does that mean rust is using (the equivalent of) alloca() ?

No, the compiler knows the size of `T` statically (this is a requirement
for calling `MaybeUninit::uninit()`). It would simply reserve stack
space for a stack variable (like the C compiler would also do) and since
the type is generic and user-specified, it could be larger than the
stack size. The problem is that the Rust compiler does not optimize the
stack variable away. This fix avoids creating a stack variable in the
first place.

--
Cheers,
Benno


>
> That is banned for C code in the kernel for any sizes.
>
> 	David
>
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
> Registration No: 1397386 (Wales)
Alice Ryhl March 31, 2023, 7:05 a.m. UTC | #4 
On 3/30/23 00:33, y86-dev@protonmail.com wrote:
> From: Benno Lossin <y86-dev@protonmail.com>
> 
> `UniqueArc::try_new_uninit` calls `Arc::try_new(MaybeUninit::uninit())`.
> This results in the uninitialized memory being placed on the stack,
> which may be arbitrarily large due to the generic `T` and thus could
> cause a stack overflow for large types.
> 
> Change the implementation to use the pin-init API which enables in-place
> initialization. In particular it avoids having to first construct and
> then move the uninitialized memory from the stack into the final location.
> 
> Signed-off-by: Benno Lossin <y86-dev@protonmail.com>

Reviewed-by: Alice Ryhl <aliceryhl@google.com>
Andreas Hindborg March 31, 2023, 1:03 p.m. UTC | #5 
y86-dev@protonmail.com writes:

> From: Benno Lossin <y86-dev@protonmail.com>
>
> `UniqueArc::try_new_uninit` calls `Arc::try_new(MaybeUninit::uninit())`.
> This results in the uninitialized memory being placed on the stack,
> which may be arbitrarily large due to the generic `T` and thus could
> cause a stack overflow for large types.
>
> Change the implementation to use the pin-init API which enables in-place
> initialization. In particular it avoids having to first construct and
> then move the uninitialized memory from the stack into the final location.
>
> Signed-off-by: Benno Lossin <y86-dev@protonmail.com>
> ---

Reviewed-by Andreas Hindborg <a.hindborg@samsung.com>

>  rust/kernel/lib.rs      |  1 -
>  rust/kernel/sync/arc.rs | 14 ++++++++++++--
>  2 files changed, 12 insertions(+), 3 deletions(-)
>
> diff --git a/rust/kernel/lib.rs b/rust/kernel/lib.rs
> index 3e2777d26ff5..d9df77132fa2 100644
> --- a/rust/kernel/lib.rs
> +++ b/rust/kernel/lib.rs
> @@ -27,7 +27,6 @@
>  #[cfg(not(CONFIG_RUST))]
>  compile_error!("Missing kernel configuration for conditional compilation");
>
> -#[allow(unused_extern_crates)]
>  // Allow proc-macros to refer to `::kernel` inside the `kernel` crate (this crate).
>  extern crate self as kernel;
>
> diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
> index 77a3833cc265..4ed6329a5e5f 100644
> --- a/rust/kernel/sync/arc.rs
> +++ b/rust/kernel/sync/arc.rs
> @@ -18,6 +18,7 @@
>  use crate::{
>      bindings,
>      error::{Error, Result},
> +    init,
>      init::{InPlaceInit, Init, PinInit},
>      types::{ForeignOwnable, Opaque},
>  };
> @@ -29,6 +30,7 @@ use core::{
>      pin::Pin,
>      ptr::NonNull,
>  };
> +use macros::pin_data;
>
>  /// A reference-counted pointer to an instance of `T`.
>  ///
> @@ -121,6 +123,7 @@ pub struct Arc<T: ?Sized> {
>      _p: PhantomData<ArcInner<T>>,
>  }
>
> +#[pin_data]
>  #[repr(C)]
>  struct ArcInner<T: ?Sized> {
>      refcount: Opaque<bindings::refcount_t>,
> @@ -501,9 +504,16 @@ impl<T> UniqueArc<T> {
>
>      /// Tries to allocate a new [`UniqueArc`] instance whose contents are not initialised yet.
>      pub fn try_new_uninit() -> Result<UniqueArc<MaybeUninit<T>>> {
> -        Ok(UniqueArc::<MaybeUninit<T>> {
> +        // INVARIANT: The refcount is initialised to a non-zero value.
> +        let inner = Box::init(init!(ArcInner {
> +            // SAFETY: There are no safety requirements for this FFI call.
> +            refcount: Opaque::new(unsafe { bindings::REFCOUNT_INIT(1) }),
> +            data <- init::uninit(),
> +        }))?;
> +        Ok(UniqueArc {
>              // INVARIANT: The newly-created object has a ref-count of 1.
> -            inner: Arc::try_new(MaybeUninit::uninit())?,
> +            // SAFETY: The pointer from the `Box` is valid.
> +            inner: unsafe { Arc::from_inner(Box::leak(inner).into()) },
>          })
>      }
>  }
diff mbox series

Patch

diff --git a/rust/kernel/lib.rs b/rust/kernel/lib.rs
index 3e2777d26ff5..d9df77132fa2 100644
--- a/rust/kernel/lib.rs
+++ b/rust/kernel/lib.rs
@@ -27,7 +27,6 @@ 
 #[cfg(not(CONFIG_RUST))]
 compile_error!("Missing kernel configuration for conditional compilation");

-#[allow(unused_extern_crates)]
 // Allow proc-macros to refer to `::kernel` inside the `kernel` crate (this crate).
 extern crate self as kernel;

diff --git a/rust/kernel/sync/arc.rs b/rust/kernel/sync/arc.rs
index 77a3833cc265..4ed6329a5e5f 100644
--- a/rust/kernel/sync/arc.rs
+++ b/rust/kernel/sync/arc.rs
@@ -18,6 +18,7 @@ 
 use crate::{
     bindings,
     error::{Error, Result},
+    init,
     init::{InPlaceInit, Init, PinInit},
     types::{ForeignOwnable, Opaque},
 };
@@ -29,6 +30,7 @@  use core::{
     pin::Pin,
     ptr::NonNull,
 };
+use macros::pin_data;

 /// A reference-counted pointer to an instance of `T`.
 ///
@@ -121,6 +123,7 @@  pub struct Arc<T: ?Sized> {
     _p: PhantomData<ArcInner<T>>,
 }

+#[pin_data]
 #[repr(C)]
 struct ArcInner<T: ?Sized> {
     refcount: Opaque<bindings::refcount_t>,
@@ -501,9 +504,16 @@  impl<T> UniqueArc<T> {

     /// Tries to allocate a new [`UniqueArc`] instance whose contents are not initialised yet.
     pub fn try_new_uninit() -> Result<UniqueArc<MaybeUninit<T>>> {
-        Ok(UniqueArc::<MaybeUninit<T>> {
+        // INVARIANT: The refcount is initialised to a non-zero value.
+        let inner = Box::init(init!(ArcInner {
+            // SAFETY: There are no safety requirements for this FFI call.
+            refcount: Opaque::new(unsafe { bindings::REFCOUNT_INIT(1) }),
+            data <- init::uninit(),
+        }))?;
+        Ok(UniqueArc {
             // INVARIANT: The newly-created object has a ref-count of 1.
-            inner: Arc::try_new(MaybeUninit::uninit())?,
+            // SAFETY: The pointer from the `Box` is valid.
+            inner: unsafe { Arc::from_inner(Box::leak(inner).into()) },
         })
     }
 }