-
Notifications
You must be signed in to change notification settings - Fork 12.7k
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.
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
ARMv5 needs +strict-align #42314
ARMv5 needs +strict-align #42314
Conversation
Without that flag, LLVM generates unaligned memory access instructions, which are not allowed on ARMv5. For example, the 'hello world' example from `cargo --new` failed with: ``` $ ./hello Hello, world! thread 'main' panicked at 'assertion failed: end <= len', src/libcollections/vec.rs:1113 note: Run with `RUST_BACKTRACE=1` for a backtrace. ``` I traced this error back to the following assembler code in `BufWriter::flush_buf`: ``` 6f44: e28d0018 add r0, sp, rust-lang#24 [...] 6f54: e280b005 add fp, r0, rust-lang#5 [...] 7018: e5cd001c strb r0, [sp, rust-lang#28] 701c: e1a0082a lsr r0, sl, rust-lang#16 7020: 03a01001 moveq r1, rust-lang#1 7024: e5cb0002 strb r0, [fp, rust-lang#2] 7028: e1cba0b0 strh sl, [fp] ``` Note that `fp` points to `sp + 29`, so the three `str*`-instructions should fill up a 32bit - value at `sp + 28`, which is later used as the value `n` in `Ok(n) => written += n`. This doesn't work on ARMv5 as the `strh` can't write to the unaligned contents of `fp`, so the upper bits of `n` won't get cleared, leading to the assertion failure in Vec::drain. With `+strict-align`, the code works as expected.
Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @eddyb (or someone else) soon. If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes. Please see the contribution instructions for more information. |
r? @japaric |
📌 Commit 4450807 has been approved by |
According to rust-lang/llvm@4fb2f33, "clang is now responsible for setting strict-align". The same is probably true for rustc. |
ARMv5 needs +strict-align Without that flag, LLVM generates unaligned memory access instructions, which are not allowed on ARMv5. For example, the 'hello world' example from `cargo --new` failed with: ``` $ ./hello Hello, world! thread 'main' panicked at 'assertion failed: end <= len', src/libcollections/vec.rs:1113 note: Run with `RUST_BACKTRACE=1` for a backtrace. ``` I traced this error back to the following assembler code in `BufWriter::flush_buf`: ``` 6f44: e28d0018 add r0, sp, rust-lang#24 [...] 6f54: e280b005 add fp, r0, rust-lang#5 [...] 7018: e5cd001c strb r0, [sp, rust-lang#28] 701c: e1a0082a lsr r0, sl, rust-lang#16 7020: 03a01001 moveq r1, #1 7024: e5cb0002 strb r0, [fp, rust-lang#2] 7028: e1cba0b0 strh sl, [fp] ``` Note that `fp` points to `sp + 29`, so the three `str*`-instructions should fill up a 32bit - value at `sp + 28`, which is later used as the value `n` in `Ok(n) => written += n`. This doesn't work on ARMv5 as the `strh` can't write to the unaligned contents of `fp`, so the upper bits of `n` won't get cleared, leading to the assertion failure in Vec::drain. With `+strict-align`, the code works as expected.
ARMv5 needs +strict-align Without that flag, LLVM generates unaligned memory access instructions, which are not allowed on ARMv5. For example, the 'hello world' example from `cargo --new` failed with: ``` $ ./hello Hello, world! thread 'main' panicked at 'assertion failed: end <= len', src/libcollections/vec.rs:1113 note: Run with `RUST_BACKTRACE=1` for a backtrace. ``` I traced this error back to the following assembler code in `BufWriter::flush_buf`: ``` 6f44: e28d0018 add r0, sp, rust-lang#24 [...] 6f54: e280b005 add fp, r0, rust-lang#5 [...] 7018: e5cd001c strb r0, [sp, rust-lang#28] 701c: e1a0082a lsr r0, sl, rust-lang#16 7020: 03a01001 moveq r1, #1 7024: e5cb0002 strb r0, [fp, rust-lang#2] 7028: e1cba0b0 strh sl, [fp] ``` Note that `fp` points to `sp + 29`, so the three `str*`-instructions should fill up a 32bit - value at `sp + 28`, which is later used as the value `n` in `Ok(n) => written += n`. This doesn't work on ARMv5 as the `strh` can't write to the unaligned contents of `fp`, so the upper bits of `n` won't get cleared, leading to the assertion failure in Vec::drain. With `+strict-align`, the code works as expected.
Without that flag, LLVM generates unaligned memory access instructions, which are not allowed on ARMv5.
For example, the 'hello world' example from
cargo --new
failed with:I traced this error back to the following assembler code in
BufWriter::flush_buf
:Note that
fp
points tosp + 29
, so the threestr*
-instructions should fill up a 32bit - value atsp + 28
, which is later used as the valuen
inOk(n) => written += n
. This doesn't work on ARMv5 as thestrh
can't write to the unaligned contents offp
, so the upper bits ofn
won't get cleared, leading to the assertion failure in Vec::drain.With
+strict-align
, the code works as expected.