This repository has been archived by the owner on Feb 5, 2019. It is now read-only.
x86 interrupt calling convention: only save xmm registers if the target supports SSE #63
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…et supports SSE The existing code always saves the xmm registers for 64-bit targets even if the target doesn't support SSE (which is common for kernels). Thus, the compiler inserts movaps instructions which lead to CPU exceptions when an interrupt handler is invoked. This commit fixes this bug by returning a register set without xmm registers from getCalleeSavedRegs and getCallPreservedMask for such targets. Patch by Philipp Oppermann. Differential Revision: https://reviews.llvm.org/D29959 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@295347 91177308-0d34-0410-b5e6-96231b3b80d8
3 tasks
|
Thanks! |
frewsxcv
added a commit
to frewsxcv/rust
that referenced
this pull request
Mar 2, 2017
…ention, r=nagisa
Add support for the x86-interrupt calling convention
This calling convention can be used for definining interrupt handlers on 32-bit and 64-bit x86 targets. The compiler then uses `iret` instead of `ret` for returning and ensures that all registers are restored to their
original values.
Usage:
```rust
extern "x86-interrupt" fn handler(stack_frame: &ExceptionStackFrame) {…}
```
for interrupts and exceptions without error code and
```rust
extern "x86-interrupt" fn handler_with_err_code(stack_frame: &ExceptionStackFrame,
error_code: u64) {…}
```
for exceptions that push an error code (e.g., page faults or general protection faults). The programmer must ensure that the correct version is used for each interrupt.
For more details see the [LLVM PR][1] and the corresponding [proposal][2].
[1]: https://reviews.llvm.org/D15567
[2]: http://lists.llvm.org/pipermail/cfe-dev/2015-September/045171.html
It is also possible to implement interrupt handlers on x86 through [naked functions](https://github.com/rust-lang/rfcs/blob/master/text/1201-naked-fns.md). In fact, almost all existing Rust OS projects for x86 use naked functions for this, including [Redox](https://github.com/redox-os/kernel/blob/b9793deb59c7650f0805dea96adb6b773ad99336/arch/x86_64/src/lib.rs#L109-L147), [IntermezzOS](https://github.com/intermezzOS/kernel/blob/f959cc18c78b1ba153f3ff7039d9ecc07f397628/interrupts/src/lib.rs#L28-L72), and [blog_os](https://github.com/phil-opp/blog_os/blob/844d739379ffdea6a7ede88365ec6e21a725bbf5/src/interrupts/mod.rs#L49-L64). So support for the `x86-interrupt` calling convention isn't absolutely needed.
However, it has a number of benefits to naked functions:
- **No inline assembly needed**: [Inline assembly](https://doc.rust-lang.org/book/inline-assembly.html) is highly unstable and dangerous. It's pretty easy to mess things up. Also, it uses an arcane syntax and requires that the programmer knows x86 assembly.
- **Higher performance**: A naked wrapper function always saves _all_ registers before calling the Rust function. This isn't needed for a compiler supported calling convention, since the compiler knows which registers are clobbered by the interrupt handler. Thus, only these registers need to be saved and restored.
- **Safer interfaces**: We can write a `set_handler` function that takes a `extern "x86-interrupt" fn(&ExceptionStackFrame)` and the compiler ensures that we always use the right function type for all handler functions. This isn't possible with the `#[naked]` attribute.
- **More convenient**: Instead of writing [tons of assembly boilerplate](https://github.com/redox-os/kernel/blob/b9793deb59c7650f0805dea96adb6b773ad99336/arch/x86_64/src/lib.rs#L109-L147) and desperately trying to improve things [through macros](https://github.com/phil-opp/blog_os/blob/844d739379ffdea6a7ede88365ec6e21a725bbf5/src/interrupts/mod.rs#L17-L92), we can just write [code like this](https://github.com/phil-opp/blog_os/blob/e6a61f9507a4c4fef6fb4e3474bc596391bc97d2/src/interrupts/mod.rs#L85-L89).
- **Naked functions are unreliable**: It is allowed to use Rust code inside a naked function, which sometimes works and sometimes not. For example, [calling a function](https://github.com/redox-os/kernel/blob/b9793deb59c7650f0805dea96adb6b773ad99336/arch/x86_64/src/lib.rs#L132) through Rust code seems to work fine without function prologue, but [code declaring a variable](https://is.gd/NQYXqE) silently adds a prologue even though the function is naked (look at the generated assembly, there is a `movl` instruction before the `nop`).
**Edit**: See the [tracking issue](rust-lang#40180) for an updated list of issues.
Unfortunately, the implementation of the `x86-interrupt` calling convention in LLVM has some issues that make it unsuitable for 64-bit kernels at the moment:
- LLVM always tries to backup the `xmm` registers on 64-bit platforms even if the target doesn't support SSE. This leads to invalid opcode exceptions whenever an interrupt handler is invoked. I submitted a fix to LLVM in [D29959](https://reviews.llvm.org/D29959). The fix is really small (<10 lines), so maybe we could backport it to [Rust's LLVM fork](https://github.com/rust-lang/llvm)?. **Edit**: The fix was merged to LLVM trunk in [rL295347](https://reviews.llvm.org/rL295347). Backported in rust-lang/llvm#63.
- On targets with SSE support, LLVM uses the `movaps` instruction for saving the `xmm` registers, which requires an alignment of 16. For handlers with error codes, however, the stack alignment is only 8, so a alignment exception occurs. This issue is tracked in [bug 26413](https://bugs.llvm.org/show_bug.cgi?id=26413). ~~Unfortunately, I don't know enough about LLVM to fix this.~~ **Edit**: Fix submitted in [D30049](https://reviews.llvm.org/D30049).
This PR adds experimental support for this calling convention under the `abi_x86_interrupt` feature gate. The implementation is very similar to rust-lang#38465 and was surprisingly simple :).
There is no accepted RFC for this change. In fact, the [RFC for interrupt calling convention](rust-lang/rfcs#1275) from 2015 was closed in favor of naked functions. However, the reactions to the recent [PR](rust-lang#38465) for a MSP430 interrupt calling convention were [in favor of experimental interrupt ABIs](rust-lang#38465 (comment)).
- [x] Add compile-fail tests for the feature gate.
- [x] Create tracking issue for the `abi_x86_interrupt` feature (and link it in code). **Edit**: Tracking issue: rust-lang#40180
- [x] Backport [rL295347](https://reviews.llvm.org/rL295347) to Rust's LLVM fork. **Edit**: Done in rust-lang/llvm#63
@tari @steveklabnik @jackpot51 @ticki @hawkw @thepowersgang, you might be interested in this.
frewsxcv
added a commit
to frewsxcv/rust
that referenced
this pull request
Mar 2, 2017
…ention, r=nagisa
Add support for the x86-interrupt calling convention
This calling convention can be used for definining interrupt handlers on 32-bit and 64-bit x86 targets. The compiler then uses `iret` instead of `ret` for returning and ensures that all registers are restored to their
original values.
Usage:
```rust
extern "x86-interrupt" fn handler(stack_frame: &ExceptionStackFrame) {…}
```
for interrupts and exceptions without error code and
```rust
extern "x86-interrupt" fn handler_with_err_code(stack_frame: &ExceptionStackFrame,
error_code: u64) {…}
```
for exceptions that push an error code (e.g., page faults or general protection faults). The programmer must ensure that the correct version is used for each interrupt.
For more details see the [LLVM PR][1] and the corresponding [proposal][2].
[1]: https://reviews.llvm.org/D15567
[2]: http://lists.llvm.org/pipermail/cfe-dev/2015-September/045171.html
It is also possible to implement interrupt handlers on x86 through [naked functions](https://github.com/rust-lang/rfcs/blob/master/text/1201-naked-fns.md). In fact, almost all existing Rust OS projects for x86 use naked functions for this, including [Redox](https://github.com/redox-os/kernel/blob/b9793deb59c7650f0805dea96adb6b773ad99336/arch/x86_64/src/lib.rs#L109-L147), [IntermezzOS](https://github.com/intermezzOS/kernel/blob/f959cc18c78b1ba153f3ff7039d9ecc07f397628/interrupts/src/lib.rs#L28-L72), and [blog_os](https://github.com/phil-opp/blog_os/blob/844d739379ffdea6a7ede88365ec6e21a725bbf5/src/interrupts/mod.rs#L49-L64). So support for the `x86-interrupt` calling convention isn't absolutely needed.
However, it has a number of benefits to naked functions:
- **No inline assembly needed**: [Inline assembly](https://doc.rust-lang.org/book/inline-assembly.html) is highly unstable and dangerous. It's pretty easy to mess things up. Also, it uses an arcane syntax and requires that the programmer knows x86 assembly.
- **Higher performance**: A naked wrapper function always saves _all_ registers before calling the Rust function. This isn't needed for a compiler supported calling convention, since the compiler knows which registers are clobbered by the interrupt handler. Thus, only these registers need to be saved and restored.
- **Safer interfaces**: We can write a `set_handler` function that takes a `extern "x86-interrupt" fn(&ExceptionStackFrame)` and the compiler ensures that we always use the right function type for all handler functions. This isn't possible with the `#[naked]` attribute.
- **More convenient**: Instead of writing [tons of assembly boilerplate](https://github.com/redox-os/kernel/blob/b9793deb59c7650f0805dea96adb6b773ad99336/arch/x86_64/src/lib.rs#L109-L147) and desperately trying to improve things [through macros](https://github.com/phil-opp/blog_os/blob/844d739379ffdea6a7ede88365ec6e21a725bbf5/src/interrupts/mod.rs#L17-L92), we can just write [code like this](https://github.com/phil-opp/blog_os/blob/e6a61f9507a4c4fef6fb4e3474bc596391bc97d2/src/interrupts/mod.rs#L85-L89).
- **Naked functions are unreliable**: It is allowed to use Rust code inside a naked function, which sometimes works and sometimes not. For example, [calling a function](https://github.com/redox-os/kernel/blob/b9793deb59c7650f0805dea96adb6b773ad99336/arch/x86_64/src/lib.rs#L132) through Rust code seems to work fine without function prologue, but [code declaring a variable](https://is.gd/NQYXqE) silently adds a prologue even though the function is naked (look at the generated assembly, there is a `movl` instruction before the `nop`).
**Edit**: See the [tracking issue](rust-lang#40180) for an updated list of issues.
Unfortunately, the implementation of the `x86-interrupt` calling convention in LLVM has some issues that make it unsuitable for 64-bit kernels at the moment:
- LLVM always tries to backup the `xmm` registers on 64-bit platforms even if the target doesn't support SSE. This leads to invalid opcode exceptions whenever an interrupt handler is invoked. I submitted a fix to LLVM in [D29959](https://reviews.llvm.org/D29959). The fix is really small (<10 lines), so maybe we could backport it to [Rust's LLVM fork](https://github.com/rust-lang/llvm)?. **Edit**: The fix was merged to LLVM trunk in [rL295347](https://reviews.llvm.org/rL295347). Backported in rust-lang/llvm#63.
- On targets with SSE support, LLVM uses the `movaps` instruction for saving the `xmm` registers, which requires an alignment of 16. For handlers with error codes, however, the stack alignment is only 8, so a alignment exception occurs. This issue is tracked in [bug 26413](https://bugs.llvm.org/show_bug.cgi?id=26413). ~~Unfortunately, I don't know enough about LLVM to fix this.~~ **Edit**: Fix submitted in [D30049](https://reviews.llvm.org/D30049).
This PR adds experimental support for this calling convention under the `abi_x86_interrupt` feature gate. The implementation is very similar to rust-lang#38465 and was surprisingly simple :).
There is no accepted RFC for this change. In fact, the [RFC for interrupt calling convention](rust-lang/rfcs#1275) from 2015 was closed in favor of naked functions. However, the reactions to the recent [PR](rust-lang#38465) for a MSP430 interrupt calling convention were [in favor of experimental interrupt ABIs](rust-lang#38465 (comment)).
- [x] Add compile-fail tests for the feature gate.
- [x] Create tracking issue for the `abi_x86_interrupt` feature (and link it in code). **Edit**: Tracking issue: rust-lang#40180
- [x] Backport [rL295347](https://reviews.llvm.org/rL295347) to Rust's LLVM fork. **Edit**: Done in rust-lang/llvm#63
@tari @steveklabnik @jackpot51 @ticki @hawkw @thepowersgang, you might be interested in this.
TimNN
pushed a commit
to TimNN/llvm
that referenced
this pull request
Apr 17, 2017
x86 interrupt calling convention: only save xmm registers if the target supports SSE
13 tasks
TimNN
pushed a commit
that referenced
this pull request
Apr 23, 2017
x86 interrupt calling convention: only save xmm registers if the target supports SSE
arielb1
pushed a commit
that referenced
this pull request
Jun 27, 2017
x86 interrupt calling convention: only save xmm registers if the target supports SSE
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Backport rL295347 to fix a problem with the
x86-interruptcalling convention for 64-bit targets without SSE support. Support for thex86-interruptcalling convention is implemented in rust-lang/rust#39832.backport of: rL295347
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@295347 91177308-0d34-0410-b5e6-96231b3b80d8