interrupt: Support no-std pre-v6 ARM targets #28
Conversation
taiki-e
added
the
O-arm
taiki-e
force-pushed
the
armv4t-interrupt
branch
2 times, most recently
from
5937598 to
b1ff9ec
Compare
|
I think it is okay that we only disable the I bit because in some environments the F bit seems not to be able to be changed. Also, it seems that only IRQ is used in GBA. |
taiki-e
force-pushed
the
armv4t-interrupt
branch
from
b1ff9ec to
58fa110
Compare
taiki-e
force-pushed
the
armv4t-interrupt
branch
from
58fa110 to
ec78759
Compare
taiki-e
force-pushed
the
armv4t-interrupt
branch
from
ec78759 to
a5b7f88
Compare
|
bors r+ Test (with mGBA) passed locally. (I also tested that the implementation around disabling interrupts is working using the gba crate example)
|
|
Your readme should likely note that this does not work in User mode. |
|
https://github.com/taiki-e/portable-atomic#optional-cfg
|
|
Oh, i only looked at the lines that the PR diff highlighted in green XD |
| macro_rules! atomic_load { | ||
| ($name:ident, $int_ty:ty, $atomic_ty:ident) => { | ||
| #[no_mangle] | ||
| #[cfg_attr(all(not(windows), not(target_vendor = "apple")), linkage = "weak")] | ||
| pub unsafe extern "C" fn $name(dst: *mut $int_ty, ordering: c_int) -> $int_ty { | ||
| unsafe { | ||
| let guard = FFIPanicGuard; | ||
| let res = mem::transmute( | ||
| (*dst.cast::<portable_atomic::$atomic_ty>()).load(c_load_ordering(ordering)), | ||
| ); | ||
| mem::forget(guard); | ||
| res | ||
| } | ||
| } | ||
| }; | ||
| } | ||
| atomic_load!(__atomic_load_1, u8, AtomicU8); | ||
| atomic_load!(__atomic_load_2, u16, AtomicU16); | ||
| atomic_load!(__atomic_load_4, u32, AtomicU32); |
There was a problem hiding this comment.
Btw, this is needed due to LLVM 15's change (rust-lang/rust#99668). Other GBA users seem to be affected as well (rust-lang/rust#100619). Passing +atomics-32 target feature is another way to solve this.
There was a problem hiding this comment.
Filed the fix to upstream: rust-lang/rust#100621
|
I just wanted to say that this is really cool and much appreciated! As for running mGBA in "headless" mode, you can take a look at how agb compiles and uses its mgba-test-runner. |
|
Thanks @Jinxit! I will look into mgba-test-runner. |
Pass +atomics-32 feature for {arm,thumb}v4t-none-eabi
Similar to rust-lang@89582e8, but for ARMv4t.
Pre-v6 ARM target does not have atomic CAS, except for Linux and Android where atomic CAS is provided by compiler-builtins. So, there is a similar issue as thumbv6m.
I have confirmed that enabling the `atomics-32` target feature fixes the problem in the project affected by this issue. (taiki-e/portable-atomic#28 (comment))
Closes rust-lang#100619
r? `@nikic`
cc `@Lokathor`
Pass +atomics-32 feature for {arm,thumb}v4t-none-eabi
Similar to rust-lang@89582e8, but for ARMv4t.
Pre-v6 ARM target does not have atomic CAS, except for Linux and Android where atomic CAS is provided by compiler-builtins. So, there is a similar issue as thumbv6m.
I have confirmed that enabling the `atomics-32` target feature fixes the problem in the project affected by this issue. (taiki-e/portable-atomic#28 (comment))
Closes rust-lang#100619
r? ``@nikic``
cc ``@Lokathor``
Closes #26
This currently disables only IRQs. This is explained in the document on safety requirements:
I have a few questions:
Should we also disable FIQs? The emulation of atomic operation on Linux seemed to disable only IRQs, but they may have disabled FIQs globally.EDIT: see interrupt: Support no-std pre-v6 ARM targets #28 (comment)