Skip to content

Implement __sync builtins for thumbv6-none-eabi #1050

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
taiki-e wants to merge 6 commits into
base: main
Choose a base branch
from
+413 −138
Open

Implement __sync builtins for thumbv6-none-eabi #1050

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
363382c
Implement __sync builtins for thumbv6-none-eabi
taiki-e Dec 27, 2025
c93feea
Separate __sync builtins definitions into its own file
taiki-e Dec 27, 2025
f989e56
Remove asm_use_dmb helper, rename dmb macro to cp15_barrier
taiki-e Dec 28, 2025
5ccc285
Move arm's atomic-related code to sync module
taiki-e Jan 21, 2026
83d096b
Remove casts and needless rustfmt::skip
taiki-e Jan 21, 2026
e420b6b
Add safety comments for arm_linux.rs and thumbv6k.rs
taiki-e Jan 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 19 additions & 0 deletions .github/workflows/main.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,6 +230,24 @@ jobs:
--target etc/thumbv7em-none-eabi-renamed.json \
-Zbuild-std=core

# FIXME: move this target to test job once https://github.com/rust-lang/rust/pull/150138 merged.
build-thumbv6k:
name: Build thumbv6k
runs-on: ubuntu-24.04
timeout-minutes: 10
steps:
- uses: actions/checkout@v4
- name: Install Rust
run: |
rustup update nightly --no-self-update
rustup default nightly
rustup component add rust-src
- uses: Swatinem/rust-cache@v2
- run: |
cargo build -p compiler_builtins -p libm \
--target etc/thumbv6-none-eabi.json \
-Zbuild-std=core

benchmarks:
name: Benchmarks
timeout-minutes: 20
Expand Down Expand Up @@ -369,6 +387,7 @@ jobs:
needs:
- benchmarks
- build-custom
- build-thumbv6k
- clippy
- extensive
- miri
Expand Down
8 changes: 1 addition & 7 deletions compiler-builtins/src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,7 @@ pub mod float;
pub mod int;
pub mod math;
pub mod mem;
pub mod sync;

// `libm` expects its `support` module to be available in the crate root.
use math::libm_math::support;
Expand All @@ -58,13 +59,6 @@ pub mod aarch64;
#[cfg(all(target_arch = "aarch64", target_feature = "outline-atomics"))]
pub mod aarch64_outline_atomics;

#[cfg(all(
kernel_user_helpers,
any(target_os = "linux", target_os = "android"),
target_arch = "arm"
))]
pub mod arm_linux;

#[cfg(target_arch = "avr")]
pub mod avr;

Expand Down
140 changes: 9 additions & 131 deletions compiler-builtins/src/arm_linux.rs → compiler-builtins/src/sync/arm_linux.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -125,14 +125,16 @@ unsafe fn atomic_cmpxchg<T>(ptr: *mut T, oldval: u32, newval: u32) -> u32 {
let (shift, mask) = get_shift_mask(ptr);

loop {
// FIXME(safety): preconditions review needed
// SAFETY: the caller must guarantee that the pointer is valid for read and write
// and aligned to the element size.
let curval_aligned = unsafe { atomic_load_aligned::<T>(aligned_ptr) };
let curval = extract_aligned(curval_aligned, shift, mask);
if curval != oldval {
return curval;
}
let newval_aligned = insert_aligned(curval_aligned, newval, shift, mask);
// FIXME(safety): preconditions review needed
// SAFETY: the caller must guarantee that the pointer is valid for read and write
// and aligned to the element size.
if unsafe { __kuser_cmpxchg(curval_aligned, newval_aligned, aligned_ptr) } {
return oldval;
}
Expand All @@ -143,7 +145,8 @@ macro_rules! atomic_rmw {
($name:ident, $ty:ty, $op:expr, $fetch:expr) => {
intrinsics! {
pub unsafe extern "C" fn $name(ptr: *mut $ty, val: $ty) -> $ty {
// FIXME(safety): preconditions review needed
// SAFETY: the caller must guarantee that the pointer is valid for read and write
// and aligned to the element size.
unsafe {
atomic_rmw(
ptr,
Expand All @@ -167,140 +170,15 @@ macro_rules! atomic_cmpxchg {
($name:ident, $ty:ty) => {
intrinsics! {
pub unsafe extern "C" fn $name(ptr: *mut $ty, oldval: $ty, newval: $ty) -> $ty {
// FIXME(safety): preconditions review needed
// SAFETY: the caller must guarantee that the pointer is valid for read and write
// and aligned to the element size.
unsafe { atomic_cmpxchg(ptr, oldval as u32, newval as u32) as $ty }
}
}
};
}

atomic_rmw!(@old __sync_fetch_and_add_1, u8, |a: u8, b: u8| a.wrapping_add(b));
atomic_rmw!(@old __sync_fetch_and_add_2, u16, |a: u16, b: u16| a
.wrapping_add(b));
atomic_rmw!(@old __sync_fetch_and_add_4, u32, |a: u32, b: u32| a
.wrapping_add(b));

atomic_rmw!(@new __sync_add_and_fetch_1, u8, |a: u8, b: u8| a.wrapping_add(b));
atomic_rmw!(@new __sync_add_and_fetch_2, u16, |a: u16, b: u16| a
.wrapping_add(b));
atomic_rmw!(@new __sync_add_and_fetch_4, u32, |a: u32, b: u32| a
.wrapping_add(b));

atomic_rmw!(@old __sync_fetch_and_sub_1, u8, |a: u8, b: u8| a.wrapping_sub(b));
atomic_rmw!(@old __sync_fetch_and_sub_2, u16, |a: u16, b: u16| a
.wrapping_sub(b));
atomic_rmw!(@old __sync_fetch_and_sub_4, u32, |a: u32, b: u32| a
.wrapping_sub(b));

atomic_rmw!(@new __sync_sub_and_fetch_1, u8, |a: u8, b: u8| a.wrapping_sub(b));
atomic_rmw!(@new __sync_sub_and_fetch_2, u16, |a: u16, b: u16| a
.wrapping_sub(b));
atomic_rmw!(@new __sync_sub_and_fetch_4, u32, |a: u32, b: u32| a
.wrapping_sub(b));

atomic_rmw!(@old __sync_fetch_and_and_1, u8, |a: u8, b: u8| a & b);
atomic_rmw!(@old __sync_fetch_and_and_2, u16, |a: u16, b: u16| a & b);
atomic_rmw!(@old __sync_fetch_and_and_4, u32, |a: u32, b: u32| a & b);

atomic_rmw!(@new __sync_and_and_fetch_1, u8, |a: u8, b: u8| a & b);
atomic_rmw!(@new __sync_and_and_fetch_2, u16, |a: u16, b: u16| a & b);
atomic_rmw!(@new __sync_and_and_fetch_4, u32, |a: u32, b: u32| a & b);

atomic_rmw!(@old __sync_fetch_and_or_1, u8, |a: u8, b: u8| a | b);
atomic_rmw!(@old __sync_fetch_and_or_2, u16, |a: u16, b: u16| a | b);
atomic_rmw!(@old __sync_fetch_and_or_4, u32, |a: u32, b: u32| a | b);

atomic_rmw!(@new __sync_or_and_fetch_1, u8, |a: u8, b: u8| a | b);
atomic_rmw!(@new __sync_or_and_fetch_2, u16, |a: u16, b: u16| a | b);
atomic_rmw!(@new __sync_or_and_fetch_4, u32, |a: u32, b: u32| a | b);

atomic_rmw!(@old __sync_fetch_and_xor_1, u8, |a: u8, b: u8| a ^ b);
atomic_rmw!(@old __sync_fetch_and_xor_2, u16, |a: u16, b: u16| a ^ b);
atomic_rmw!(@old __sync_fetch_and_xor_4, u32, |a: u32, b: u32| a ^ b);

atomic_rmw!(@new __sync_xor_and_fetch_1, u8, |a: u8, b: u8| a ^ b);
atomic_rmw!(@new __sync_xor_and_fetch_2, u16, |a: u16, b: u16| a ^ b);
atomic_rmw!(@new __sync_xor_and_fetch_4, u32, |a: u32, b: u32| a ^ b);

atomic_rmw!(@old __sync_fetch_and_nand_1, u8, |a: u8, b: u8| !(a & b));
atomic_rmw!(@old __sync_fetch_and_nand_2, u16, |a: u16, b: u16| !(a & b));
atomic_rmw!(@old __sync_fetch_and_nand_4, u32, |a: u32, b: u32| !(a & b));

atomic_rmw!(@new __sync_nand_and_fetch_1, u8, |a: u8, b: u8| !(a & b));
atomic_rmw!(@new __sync_nand_and_fetch_2, u16, |a: u16, b: u16| !(a & b));
atomic_rmw!(@new __sync_nand_and_fetch_4, u32, |a: u32, b: u32| !(a & b));

atomic_rmw!(@old __sync_fetch_and_max_1, i8, |a: i8, b: i8| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_max_2, i16, |a: i16, b: i16| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_max_4, i32, |a: i32, b: i32| if a > b {
a
} else {
b
});

atomic_rmw!(@old __sync_fetch_and_umax_1, u8, |a: u8, b: u8| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umax_2, u16, |a: u16, b: u16| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umax_4, u32, |a: u32, b: u32| if a > b {
a
} else {
b
});

atomic_rmw!(@old __sync_fetch_and_min_1, i8, |a: i8, b: i8| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_min_2, i16, |a: i16, b: i16| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_min_4, i32, |a: i32, b: i32| if a < b {
a
} else {
b
});

atomic_rmw!(@old __sync_fetch_and_umin_1, u8, |a: u8, b: u8| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umin_2, u16, |a: u16, b: u16| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umin_4, u32, |a: u32, b: u32| if a < b {
a
} else {
b
});

atomic_rmw!(@old __sync_lock_test_and_set_1, u8, |_: u8, b: u8| b);
atomic_rmw!(@old __sync_lock_test_and_set_2, u16, |_: u16, b: u16| b);
atomic_rmw!(@old __sync_lock_test_and_set_4, u32, |_: u32, b: u32| b);

atomic_cmpxchg!(__sync_val_compare_and_swap_1, u8);
atomic_cmpxchg!(__sync_val_compare_and_swap_2, u16);
atomic_cmpxchg!(__sync_val_compare_and_swap_4, u32);
include!("arm_thumb_shared.rs");

intrinsics! {
pub unsafe extern "C" fn __sync_synchronize() {
Expand Down
134 changes: 134 additions & 0 deletions compiler-builtins/src/sync/arm_thumb_shared.rs
View file
Open in desktop
Copy link
Contributor

@tgross35 tgross35 Jan 4, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: could you name this arm_thumb_sync_builtins or similar?

Copy link
Contributor

@tgross35 tgross35 Jan 4, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, I think we may as well group this by functionality like we have for other areas, something like:

src/sync/
    arm_linux.rs
    thumbv6k.rs
    arm_thumb_shared.rs // this file

I'll move the aarch64 builtins there too.

Copy link
Member Author

@taiki-e taiki-e Jan 21, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done in 5ccc285.

Original file line number Diff line number Diff line change
@@ -0,0 +1,134 @@
// Used by both arm_linux.rs and thumbv6k.rs.

// References:
// - https://llvm.org/docs/Atomics.html#libcalls-sync
// - https://gcc.gnu.org/onlinedocs/gcc/_005f_005fsync-Builtins.html
// - https://refspecs.linuxfoundation.org/elf/IA64-SysV-psABI.pdf#page=58

atomic_rmw!(@old __sync_fetch_and_add_1, u8, |a: u8, b: u8| a.wrapping_add(b));
atomic_rmw!(@old __sync_fetch_and_add_2, u16, |a: u16, b: u16| a
.wrapping_add(b));
atomic_rmw!(@old __sync_fetch_and_add_4, u32, |a: u32, b: u32| a
.wrapping_add(b));

atomic_rmw!(@new __sync_add_and_fetch_1, u8, |a: u8, b: u8| a.wrapping_add(b));
atomic_rmw!(@new __sync_add_and_fetch_2, u16, |a: u16, b: u16| a
.wrapping_add(b));
atomic_rmw!(@new __sync_add_and_fetch_4, u32, |a: u32, b: u32| a
.wrapping_add(b));

atomic_rmw!(@old __sync_fetch_and_sub_1, u8, |a: u8, b: u8| a.wrapping_sub(b));
atomic_rmw!(@old __sync_fetch_and_sub_2, u16, |a: u16, b: u16| a
.wrapping_sub(b));
atomic_rmw!(@old __sync_fetch_and_sub_4, u32, |a: u32, b: u32| a
.wrapping_sub(b));

atomic_rmw!(@new __sync_sub_and_fetch_1, u8, |a: u8, b: u8| a.wrapping_sub(b));
atomic_rmw!(@new __sync_sub_and_fetch_2, u16, |a: u16, b: u16| a
.wrapping_sub(b));
atomic_rmw!(@new __sync_sub_and_fetch_4, u32, |a: u32, b: u32| a
.wrapping_sub(b));

atomic_rmw!(@old __sync_fetch_and_and_1, u8, |a: u8, b: u8| a & b);
atomic_rmw!(@old __sync_fetch_and_and_2, u16, |a: u16, b: u16| a & b);
atomic_rmw!(@old __sync_fetch_and_and_4, u32, |a: u32, b: u32| a & b);

atomic_rmw!(@new __sync_and_and_fetch_1, u8, |a: u8, b: u8| a & b);
atomic_rmw!(@new __sync_and_and_fetch_2, u16, |a: u16, b: u16| a & b);
atomic_rmw!(@new __sync_and_and_fetch_4, u32, |a: u32, b: u32| a & b);

atomic_rmw!(@old __sync_fetch_and_or_1, u8, |a: u8, b: u8| a | b);
atomic_rmw!(@old __sync_fetch_and_or_2, u16, |a: u16, b: u16| a | b);
atomic_rmw!(@old __sync_fetch_and_or_4, u32, |a: u32, b: u32| a | b);

atomic_rmw!(@new __sync_or_and_fetch_1, u8, |a: u8, b: u8| a | b);
atomic_rmw!(@new __sync_or_and_fetch_2, u16, |a: u16, b: u16| a | b);
atomic_rmw!(@new __sync_or_and_fetch_4, u32, |a: u32, b: u32| a | b);

atomic_rmw!(@old __sync_fetch_and_xor_1, u8, |a: u8, b: u8| a ^ b);
atomic_rmw!(@old __sync_fetch_and_xor_2, u16, |a: u16, b: u16| a ^ b);
atomic_rmw!(@old __sync_fetch_and_xor_4, u32, |a: u32, b: u32| a ^ b);

atomic_rmw!(@new __sync_xor_and_fetch_1, u8, |a: u8, b: u8| a ^ b);
atomic_rmw!(@new __sync_xor_and_fetch_2, u16, |a: u16, b: u16| a ^ b);
atomic_rmw!(@new __sync_xor_and_fetch_4, u32, |a: u32, b: u32| a ^ b);

atomic_rmw!(@old __sync_fetch_and_nand_1, u8, |a: u8, b: u8| !(a & b));
atomic_rmw!(@old __sync_fetch_and_nand_2, u16, |a: u16, b: u16| !(a & b));
atomic_rmw!(@old __sync_fetch_and_nand_4, u32, |a: u32, b: u32| !(a & b));

atomic_rmw!(@new __sync_nand_and_fetch_1, u8, |a: u8, b: u8| !(a & b));
atomic_rmw!(@new __sync_nand_and_fetch_2, u16, |a: u16, b: u16| !(a & b));
atomic_rmw!(@new __sync_nand_and_fetch_4, u32, |a: u32, b: u32| !(a & b));

atomic_rmw!(@old __sync_fetch_and_max_1, i8, |a: i8, b: i8| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_max_2, i16, |a: i16, b: i16| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_max_4, i32, |a: i32, b: i32| if a > b {
a
} else {
b
});

atomic_rmw!(@old __sync_fetch_and_umax_1, u8, |a: u8, b: u8| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umax_2, u16, |a: u16, b: u16| if a > b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umax_4, u32, |a: u32, b: u32| if a > b {
a
} else {
b
});

atomic_rmw!(@old __sync_fetch_and_min_1, i8, |a: i8, b: i8| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_min_2, i16, |a: i16, b: i16| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_min_4, i32, |a: i32, b: i32| if a < b {
a
} else {
b
});

atomic_rmw!(@old __sync_fetch_and_umin_1, u8, |a: u8, b: u8| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umin_2, u16, |a: u16, b: u16| if a < b {
a
} else {
b
});
atomic_rmw!(@old __sync_fetch_and_umin_4, u32, |a: u32, b: u32| if a < b {
a
} else {
b
});

atomic_rmw!(@old __sync_lock_test_and_set_1, u8, |_: u8, b: u8| b);
atomic_rmw!(@old __sync_lock_test_and_set_2, u16, |_: u16, b: u16| b);
atomic_rmw!(@old __sync_lock_test_and_set_4, u32, |_: u32, b: u32| b);

atomic_cmpxchg!(__sync_val_compare_and_swap_1, u8);
atomic_cmpxchg!(__sync_val_compare_and_swap_2, u16);
atomic_cmpxchg!(__sync_val_compare_and_swap_4, u32);
20 changes: 20 additions & 0 deletions compiler-builtins/src/sync/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,20 @@
#[cfg(all(
kernel_user_helpers,
any(target_os = "linux", target_os = "android"),
target_arch = "arm"
))]
pub mod arm_linux;

// Armv6k supports atomic instructions, but they are unavailable in Thumb mode
// unless Thumb-2 instructions available (v6t2).
// Using Thumb interworking allows us to use these instructions even from Thumb mode
// without Thumb-2 instructions, but LLVM does not implement that processing (as of LLVM 21),
// so we implement it here at this time.
// (`not(target_feature = "mclass")` is unneeded because v6k is not set on thumbv6m.)
#[cfg(all(
target_arch = "arm",
target_feature = "thumb-mode",
target_feature = "v6k",
not(target_feature = "v6t2"),
))]
pub mod thumbv6k;
Loading
Loading