Tracking Issue for "C-unwind ABI", RFC 2945 #74990

nikomatsakis · 2020-07-31T21:11:17Z

cratelyn · 2020-07-31T22:03:37Z

Hello! I would be interested in helping implement this RFC. 🙋

nikomatsakis · 2020-08-01T13:59:27Z

Hey @katie-martin-fastly, that's great news! I'm going to assign you to the issue for now:

@rustbot assign @katie-martin-fastly

One thing to mention is that most Rust compiler development discussion takes place on the rust-lang zulip, and this project in particular is chatting in #project-ffi-unwind. I or @Amanieu are probably reasonable people to ping with questions. Also, for general "getting started" tips on building and testing the Rust compiler, the rustc-dev-guide is your friend. Also, rust-analyzer works well for IDE support (e.g., with vscode), though it requires a small bit of configuration since rustc doesn't build with cargo but rather the x.py script. (I couldn't find any documentation on that, I'll ping a few others and see...)

That said, in terms of mentoring and getting started, I think the idea is going to be to adapt an existing mechanism that the compiler offers. We have this attribute #[unwind(allowed)] which allows users to mark functions as permitting unwinding -- that mechanism is basically going to be deprecated and replaced with this new ABI, though I think it'd be best to add the ABI before trying to adapt the unwind attribute code. Still, searching for code related to that attribute will give you a good idea of what needs to change.

I can do more mentoring instructions later but, as a starting point, I updated the head post with some of the major goals of the RFC (see the "Implementation notes") section. Also here are some tips to get you started in terms of reading into the rustc code:

The Abi enum lists out all the known ABIs (there may be more similar enums, not sure). We'll want to extend it with C-unwind and perhaps other unwind variants. One way to do this might be to change the C variant to C { unwind: bool } or something like that, or maybe just add a Cunwind, not sure.
The UnwindAttr type records what kind of unwind attribute is placed on a particular function (is unwinding allowed? does it force an abort?)
The find_unwind_attr function checks for what unwind attribute is placed on a particular function.
The should_abort_on_panic function is used to decide when a function ought to, well, abort if a panic occurs (as the name suggests).
The [can_unwind] field of this struct seems to be used to set the LLVM attribute "nounwind". I'm not quite sure where it gets set, but I have to stop now because it's Saturday :P so I'll leave that as an exercise to the reader.

### Overview This commit begins the implementation work for RFC 2945. For more information, see the rendered RFC [1] and tracking issue [2]. A boolean `unwind` payload is added to the `C`, `System`, `Stdcall`, and `Thiscall` variants, marking whether unwinding across FFI boundaries is acceptable. The cases where each of these variants' `unwind` member is true correspond with the `C-unwind`, `system-unwind`, `stdcall-unwind`, and `thiscall-unwind` ABI strings introduced in RFC 2945 [3]. ### Feature Gate and Unstable Book This commit adds a `c_unwind` feature gate for the new ABI strings. Tests for this feature gate are included in `src/test/ui/c-unwind/`, which ensure that this feature gate works correctly for each of the new ABIs. A new language features entry in the unstable book is added as well. ### Further Work To Be Done This commit does not proceed to implement the new unwinding ABIs, and is intentionally scoped specifically to *defining* the ABIs and their feature flag. ### One Note on Test Churn This will lead to some test churn, in re-blessing hash tests, as the deleted comment in `src/librustc_target/spec/abi.rs` mentioned, because we can no longer guarantee the ordering of the `Abi` variants. While this is a downside, this decision was made bearing in mind that RFC 2945 states the following, in the "Other `unwind` Strings" section [3]: > More unwind variants of existing ABI strings may be introduced, > with the same semantics, without an additional RFC. Adding a new variant for each of these cases, rather than specifying a payload for a given ABI, would quickly become untenable, and make working with the `Abi` enum prone to mistakes. This approach encodes the unwinding information *into* a given ABI, to account for the future possibility of other `-unwind` ABI strings. ### Footnotes [1]: https://github.com/rust-lang/rfcs/blob/master/text/2945-c-unwind-abi.md [2]: rust-lang#74990 [3]: https://github.com/rust-lang/rfcs/blob/master/text/2945-c-unwind-abi.md#other-unwind-abi-strings

…945-c-unwind-abi, r=Amanieu Implement RFC 2945: "C-unwind" ABI ## Implement RFC 2945: "C-unwind" ABI This branch implements [RFC 2945]. The tracking issue for this RFC is rust-lang#74990. The feature gate for the issue is `#![feature(c_unwind)]`. This RFC was created as part of the ffi-unwind project group tracked at rust-lang/lang-team#19. ### Changes Further details will be provided in commit messages, but a high-level overview of the changes follows: * A boolean `unwind` payload is added to the `C`, `System`, `Stdcall`, and `Thiscall` variants, marking whether unwinding across FFI boundaries is acceptable. The cases where each of these variants' `unwind` member is true correspond with the `C-unwind`, `system-unwind`, `stdcall-unwind`, and `thiscall-unwind` ABI strings introduced in RFC 2945 [3]. * This commit adds a `c_unwind` feature gate for the new ABI strings. Tests for this feature gate are included in `src/test/ui/c-unwind/`, which ensure that this feature gate works correctly for each of the new ABIs. A new language features entry in the unstable book is added as well. * We adjust the `rustc_middle::ty::layout::fn_can_unwind` function, used to compute whether or not a `FnAbi` object represents a function that should be able to unwind when `panic=unwind` is in use. * Changes are also made to `rustc_mir_build::build::should_abort_on_panic` so that the function ABI is used to determind whether it should abort, assuming that the `panic=unwind` strategy is being used, and no explicit unwind attribute was provided. [RFC 2945]: https://github.com/rust-lang/rfcs/blob/master/text/2945-c-unwind-abi.md

Stabilise `c_unwind` Fix rust-lang#74990 Fix rust-lang#115285 (that's also where FCP is happening) Marking as draft PR for now due to `compiler_builtins` issues r? ``@Amanieu``

Stabilise `c_unwind` Fix rust-lang#74990 Fix rust-lang#115285 (that's also where FCP is happening) Marking as draft PR for now due to `compiler_builtins` issues r? `@Amanieu`

Begining within Rust 1.81 (rust-lang/rust#74990) Rust will abort when attempting to unwind a panic across an "extern C" boundary. Previously it was technically UB, but it worked and Mockall relied on it. Now, unwinding across such a boundary requires the "extern C-unwind" ABI. Use that ABI in the generated code. However, don't use that ABI when mocking a C variadic function. That doesn't work, due to rust-lang/rust#126836 . Fixes #584

### What Fixes for Rust 1.81: 1. Update semver-checks to 0.35.0. 2. Don't use the `extern` function for invoking a contract function when executing tests. ### Why 1. Semver checks is failing to run on the new version of Rust and needs updating. 2. Rust 1.81.0 changed the behavior of panics on `extern "C"` functions, such that panics cannot be caught by [std::panic::catch_unwind](https://doc.rust-lang.org/std/panic/fn.catch_unwind.html). Contracts expose their functions as an `extern` function that defaults to `extern "C"`. The test environment catches panics inside contract functions and resurfaces them as errors by using catch_unwind. The solution used was to make it so that the test environment never calls through the extern function, and instead calls through a non-extern function that does the same work. This seems like the sanest way to take away any dependence / relationship between the test environment and the Rust C-ABI so that any future changes to the C-ABI don't affect the test environment. An alternative solution would be to mark the contract function as `extern "C-unwind"` so that it retained unwind functionality, but that would continue to bind the SDK test environment to behaviour changes in extern fns. In the solution in this change the Abi qualifier `"C"` has been added. When the Abi qualifier is not specified it defaults to `"C"`, but for the sake of being explicit and removing any doubt from a reader it is now specified. The Rust reference confirms that `"C"` is the default: - https://doc.rust-lang.org/reference/items/functions.html#extern-function-qualifier. More details on this change to Rust 1.81.0 can be found at: - https://blog.rust-lang.org/2024/09/05/Rust-1.81.0.html#abort-on-uncaught-panics-in-extern-c-functions - rust-lang/rust#74990 Close #1332

… r=Mark-Simulacrum Update `catch_unwind` doc comments for `c_unwind` Updates `catch_unwind` doc comments to indicate that catching a foreign exception _will no longer_ be UB. Instead, there are two possible behaviors, though it is not specified which one an implementation will choose. Nominated for t-lang to confirm that they are okay with making such a promise based on t-opsem FCP, or whether they would like to be included in the FCP. Related: rust-lang#74990, rust-lang#115285, rust-lang/reference#1226

…=Mark-Simulacrum Update `catch_unwind` doc comments for `c_unwind` Updates `catch_unwind` doc comments to indicate that catching a foreign exception _will no longer_ be UB. Instead, there are two possible behaviors, though it is not specified which one an implementation will choose. Nominated for t-lang to confirm that they are okay with making such a promise based on t-opsem FCP, or whether they would like to be included in the FCP. Related: rust-lang#74990, rust-lang#115285, rust-lang/reference#1226

Make destructors on `extern "C"` frames to be executed This would make the example in rust-lang#123231 print "Noisy Drop". I didn't mark this as fixing the issue because the behaviour is yet to be spec'ed. Tracking: - rust-lang#74990

…ulacrum Update `catch_unwind` doc comments for `c_unwind` Updates `catch_unwind` doc comments to indicate that catching a foreign exception _will no longer_ be UB. Instead, there are two possible behaviors, though it is not specified which one an implementation will choose. Nominated for t-lang to confirm that they are okay with making such a promise based on t-opsem FCP, or whether they would like to be included in the FCP. Related: rust-lang/rust#74990, rust-lang/rust#115285, rust-lang/reference#1226

Unwinding across functions with the extern "C" ABI is UB, but previously worked. See: rust-lang/rust#74990 Redo the tests so they don't rely on panics at all, using a reference-counted counter instead. Signed-off-by: Hector Martin <marcan@marcan.st>

Make destructors on `extern "C"` frames to be executed This would make the example in rust-lang#123231 print "Noisy Drop". I didn't mark this as fixing the issue because the behaviour is yet to be spec'ed. Tracking: - rust-lang#74990

Make destructors on `extern "C"` frames to be executed This would make the example in #123231 print "Noisy Drop". I didn't mark this as fixing the issue because the behaviour is yet to be spec'ed. Tracking: - rust-lang/rust#74990

Each `bindgen` release may upgrade the list of Rust targets. For instance, currently, in their master branch [1], the latest ones are: Nightly => { vectorcall_abi: #124485, ptr_metadata: #81513, layout_for_ptr: #69835, }, Stable_1_77(77) => { offset_of: #106655 }, Stable_1_73(73) => { thiscall_abi: #42202 }, Stable_1_71(71) => { c_unwind_abi: #106075 }, Stable_1_68(68) => { abi_efiapi: #105795 }, By default, the highest stable release in their list is used, and users are expected to set one if they need to support older Rust versions (e.g. see [2]). Thus, over time, new Rust features are used by default, and at some point, it is likely that `bindgen` will emit Rust code that requires a Rust version higher than our minimum (or perhaps enabling an unstable feature). Currently, there is no problem because the maximum they have, as seen above, is Rust 1.77.0, and our current minimum is Rust 1.78.0. Therefore, set a Rust target explicitly now to prevent going forward in time too much and thus getting potential build failures at some point. Since we also support a minimum `bindgen` version, and since `bindgen` does not support passing unknown Rust target versions, we need to use the list of our minimum `bindgen` version, rather than the latest. So, since `bindgen` 0.65.1 had this list [3], we need to use Rust 1.68.0: /// Rust stable 1.64 /// * `core_ffi_c` ([Tracking issue](rust-lang/rust#94501)) => Stable_1_64 => 1.64; /// Rust stable 1.68 /// * `abi_efiapi` calling convention ([Tracking issue](rust-lang/rust#65815)) => Stable_1_68 => 1.68; /// Nightly rust /// * `thiscall` calling convention ([Tracking issue](rust-lang/rust#42202)) /// * `vectorcall` calling convention (no tracking issue) /// * `c_unwind` calling convention ([Tracking issue](rust-lang/rust#74990)) => Nightly => nightly; ... /// Latest stable release of Rust pub const LATEST_STABLE_RUST: RustTarget = RustTarget::Stable_1_68; Thus add the `--rust-target 1.68` parameter. Add a comment as well explaining this. An alternative would be to use the currently running (i.e. actual) `rustc` and `bindgen` versions to pick a "better" Rust target version. However, that would introduce more moving parts depending on the user setup and is also more complex to implement. Cc: Christian Poveda <git@pvdrz.com> Cc: Emilio Cobos Álvarez <emilio@crisal.io> Link: https://github.com/rust-lang/rust-bindgen/blob/21c60f473f4e824d4aa9b2b508056320d474b110/bindgen/features.rs#L97-L105 [1] Link: rust-lang/rust-bindgen#2960 [2] Link: https://github.com/rust-lang/rust-bindgen/blob/7d243056d335fdc4537f7bca73c06d01aae24ddc/bindgen/features.rs#L131-L150 [3] Signed-off-by: Miguel Ojeda <ojeda@kernel.org>

Each `bindgen` release may upgrade the list of Rust targets. For instance, currently, in their master branch [1], the latest ones are: Nightly => { vectorcall_abi: #124485, ptr_metadata: #81513, layout_for_ptr: #69835, }, Stable_1_77(77) => { offset_of: #106655 }, Stable_1_73(73) => { thiscall_abi: #42202 }, Stable_1_71(71) => { c_unwind_abi: #106075 }, Stable_1_68(68) => { abi_efiapi: #105795 }, By default, the highest stable release in their list is used, and users are expected to set one if they need to support older Rust versions (e.g. see [2]). Thus, over time, new Rust features are used by default, and at some point, it is likely that `bindgen` will emit Rust code that requires a Rust version higher than our minimum (or perhaps enabling an unstable feature). Currently, there is no problem because the maximum they have, as seen above, is Rust 1.77.0, and our current minimum is Rust 1.78.0. Therefore, set a Rust target explicitly now to prevent going forward in time too much and thus getting potential build failures at some point. Since we also support a minimum `bindgen` version, and since `bindgen` does not support passing unknown Rust target versions, we need to use the list of our minimum `bindgen` version, rather than the latest. So, since `bindgen` 0.65.1 had this list [3], we need to use Rust 1.68.0: /// Rust stable 1.64 /// * `core_ffi_c` ([Tracking issue](rust-lang/rust#94501)) => Stable_1_64 => 1.64; /// Rust stable 1.68 /// * `abi_efiapi` calling convention ([Tracking issue](rust-lang/rust#65815)) => Stable_1_68 => 1.68; /// Nightly rust /// * `thiscall` calling convention ([Tracking issue](rust-lang/rust#42202)) /// * `vectorcall` calling convention (no tracking issue) /// * `c_unwind` calling convention ([Tracking issue](rust-lang/rust#74990)) => Nightly => nightly; ... /// Latest stable release of Rust pub const LATEST_STABLE_RUST: RustTarget = RustTarget::Stable_1_68; Thus add the `--rust-target 1.68` parameter. Add a comment as well explaining this. An alternative would be to use the currently running (i.e. actual) `rustc` and `bindgen` versions to pick a "better" Rust target version. However, that would introduce more moving parts depending on the user setup and is also more complex to implement. Starting with `bindgen` 0.71.0 [4], we will be able to set any future Rust version instead, i.e. we will be able to set here our minimum supported Rust version. Christian implemented it [5] after seeing this patch. Thanks! Cc: stable@vger.kernel.org # 6.12.y only (since older LTSs only support a single `bindgen` version) Cc: Christian Poveda <git@pvdrz.com> Cc: Emilio Cobos Álvarez <emilio@crisal.io> Link: https://github.com/rust-lang/rust-bindgen/blob/21c60f473f4e824d4aa9b2b508056320d474b110/bindgen/features.rs#L97-L105 [1] Link: rust-lang/rust-bindgen#2960 [2] Link: https://github.com/rust-lang/rust-bindgen/blob/7d243056d335fdc4537f7bca73c06d01aae24ddc/bindgen/features.rs#L131-L150 [3] Link: https://github.com/rust-lang/rust-bindgen/blob/main/CHANGELOG.md#0710-2024-12-06 [4] Link: rust-lang/rust-bindgen#2993 [5] Reviewed-by: Alice Ryhl <aliceryhl@google.com> Link: https://lore.kernel.org/r/20241123180323.255997-1-ojeda@kernel.org Signed-off-by: Miguel Ojeda <ojeda@kernel.org>

Each `bindgen` release may upgrade the list of Rust targets. For instance, currently, in their master branch [1], the latest ones are: Nightly => { vectorcall_abi: #124485, ptr_metadata: #81513, layout_for_ptr: #69835, }, Stable_1_77(77) => { offset_of: #106655 }, Stable_1_73(73) => { thiscall_abi: #42202 }, Stable_1_71(71) => { c_unwind_abi: #106075 }, Stable_1_68(68) => { abi_efiapi: #105795 }, By default, the highest stable release in their list is used, and users are expected to set one if they need to support older Rust versions (e.g. see [2]). Thus, over time, new Rust features are used by default, and at some point, it is likely that `bindgen` will emit Rust code that requires a Rust version higher than our minimum (or perhaps enabling an unstable feature). Currently, there is no problem because the maximum they have, as seen above, is Rust 1.77.0, and our current minimum is Rust 1.78.0. Therefore, set a Rust target explicitly now to prevent going forward in time too much and thus getting potential build failures at some point. Since we also support a minimum `bindgen` version, and since `bindgen` does not support passing unknown Rust target versions, we need to use the list of our minimum `bindgen` version, rather than the latest. So, since `bindgen` 0.65.1 had this list [3], we need to use Rust 1.68.0: /// Rust stable 1.64 /// * `core_ffi_c` ([Tracking issue](rust-lang/rust#94501)) => Stable_1_64 => 1.64; /// Rust stable 1.68 /// * `abi_efiapi` calling convention ([Tracking issue](rust-lang/rust#65815)) => Stable_1_68 => 1.68; /// Nightly rust /// * `thiscall` calling convention ([Tracking issue](rust-lang/rust#42202)) /// * `vectorcall` calling convention (no tracking issue) /// * `c_unwind` calling convention ([Tracking issue](rust-lang/rust#74990)) => Nightly => nightly; ... /// Latest stable release of Rust pub const LATEST_STABLE_RUST: RustTarget = RustTarget::Stable_1_68; Thus add the `--rust-target 1.68` parameter. Add a comment as well explaining this. An alternative would be to use the currently running (i.e. actual) `rustc` and `bindgen` versions to pick a "better" Rust target version. However, that would introduce more moving parts depending on the user setup and is also more complex to implement. Starting with `bindgen` 0.71.0 [4], we will be able to set any future Rust version instead, i.e. we will be able to set here our minimum supported Rust version. Christian implemented it [5] after seeing this patch. Thanks! Cc: Christian Poveda <git@pvdrz.com> Cc: Emilio Cobos Álvarez <emilio@crisal.io> Cc: stable@vger.kernel.org # needed for 6.12.y; unneeded for 6.6.y; do not apply to 6.1.y Fixes: c844fa6 ("rust: start supporting several `bindgen` versions") Link: https://github.com/rust-lang/rust-bindgen/blob/21c60f473f4e824d4aa9b2b508056320d474b110/bindgen/features.rs#L97-L105 [1] Link: rust-lang/rust-bindgen#2960 [2] Link: https://github.com/rust-lang/rust-bindgen/blob/7d243056d335fdc4537f7bca73c06d01aae24ddc/bindgen/features.rs#L131-L150 [3] Link: https://github.com/rust-lang/rust-bindgen/blob/main/CHANGELOG.md#0710-2024-12-06 [4] Link: rust-lang/rust-bindgen#2993 [5] Reviewed-by: Alice Ryhl <aliceryhl@google.com> Link: https://lore.kernel.org/r/20241123180323.255997-1-ojeda@kernel.org Signed-off-by: Miguel Ojeda <ojeda@kernel.org>

This was referenced Jul 31, 2020

RFC: 'C-unwind' ABI rust-lang/rfcs#2945

Merged

ffi-unwind rust-lang/lang-team#19

Closed

jonas-schievink added the A-FFI Area: Foreign function interface (FFI) label Jul 31, 2020

rustbot self-assigned this Aug 1, 2020

lopopolo mentioned this issue Aug 25, 2020

Implement mrb_raise with panic artichoke/artichoke#35

Open

z33ky mentioned this issue Sep 8, 2020

Expose sqlite's sqlite3_create_collation() diesel-rs/diesel#2495

Closed

cratelyn mentioned this issue Sep 10, 2020

Implement RFC 2945: "C-unwind" ABI #76570

Merged

Aaron1011 mentioned this issue Sep 19, 2020

Abort instead of unwinding past FFI functions #52652

Closed

nvzqz mentioned this issue Dec 7, 2020

Use "C-unwind" ABI for ObjC exception-throwing functions nvzqz/fruity#7

Open

lopopolo mentioned this issue Jan 10, 2021

Implement mrb_protect and exception raising with panic! and catch_unwind artichoke/artichoke#1029

Closed

bors closed this as completed in 1aaab8b Jun 20, 2024

This was referenced Jun 22, 2024

extern C functions need to be C-unwind in Rust 1.81.0 asomers/mockall#584

Closed

C-variadic functions may not be extern "C-unwind" #126836

Closed

asomers mentioned this issue Jun 22, 2024

Use the C-unwind ABI instead of C when mocking extern C functions asomers/mockall#585

Merged

scovich mentioned this issue Jun 23, 2024

FFI is not prod-safe until c_unwind RFC stabilizes and we can use it delta-io/delta-kernel-rs#113

Open

devsnek mentioned this issue Jul 5, 2024

Panics should not unwind into non-Rust code denoland/rusty_v8#534

Closed

BatmanAoD mentioned this issue Jul 28, 2024

Update catch_unwind doc comments for c_unwind #128321

Merged

This was referenced Aug 31, 2024

panic in a no-unwind function leads to not dropping local variables #123231

Open

Make destructors on extern "C" frames to be executed #129582

Merged

leighmcculloch mentioned this issue Sep 10, 2024

Fixes for Rust 1.81.0 stellar/rs-soroban-sdk#1333

Merged

marcan mentioned this issue Oct 13, 2024

Fix tests on Rust 1.81.0 RustAudio/rust-lv2#126

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Tracking Issue for "C-unwind ABI", RFC 2945 #74990

Tracking Issue for "C-unwind ABI", RFC 2945 #74990

nikomatsakis commented Jul 31, 2020 •

edited by traviscross

Loading

cratelyn commented Jul 31, 2020

nikomatsakis commented Aug 1, 2020

Tracking Issue for "C-unwind ABI", RFC 2945 #74990

Tracking Issue for "C-unwind ABI", RFC 2945 #74990

Comments

nikomatsakis commented Jul 31, 2020 • edited by traviscross Loading

About tracking issues

Steps

Implementation notes

Unresolved Questions

Implementation history

cratelyn commented Jul 31, 2020

nikomatsakis commented Aug 1, 2020

nikomatsakis commented Jul 31, 2020 •

edited by traviscross

Loading