Open
Description
While RISC-V is still an unofficial platform, there are lots of interest to push it to experimental tier(#2876) .
This issue tracks bugs that are closely related to linux riscv64 platform.
Build Errors
- (main, 24.x) node-v24.1.0 fails to build on riscv / gcc-15 node#58485 (should be fixed by deps: V8: cherry-pick 0ce2edb7adfd node#58773)
Bugs
- riscv64: strange behavior of exception handling node#47522. This one is caused by a GCC regression and affects arm64 as well.
- (main, v24.x) [Regression] V8 crashed while building AdGuard Home with webpack riscv-forks/electron#9. Should be fixed by [riscv] Fix check failed in bind_to node#58746
- Node.js 23 wasm related error (reproducer: 389-ds-base) riscv-forks/electron#7. This one also appears to be a GCC specific bug as node compiled with clang is not affected.
- (v22.x, v23.x) [Regression] Zettlr crashes with illegal instruction riscv-forks/electron#6. Fixed by deps: V8: cherry-pick 9ab40592f697 node#56781
- (all versions after v20.x) [Regression] v8 wasm error when running yarn install with yarn v3+ riscv-forks/electron#3. WebAssembly's trap handler implementation uses a lot of virtual memory for guard pages, where risc-v systems with
Sv39only have 256GB virtual memory and would quickly run out of address space when multiple wasm are running.- Disabling trap handler would solve it but also leads to other new bugs like Unreachable error in wasm rollup when building on riscv64 riscv-forks/electron#8
- (all versions from at least v16.x) Sporadic Nodejs segfault on sg2042 revyos/revyos#27. This one only triggers on SG2042, which has 64 cores. It is mysterious as it happens by chance and when gdb is attached, the call stack is already corrupted so the cause is still unknown.
- This one also appears to affect loongarch64.
Performance Regressions
Metadata
Metadata
Assignees
Labels
No labels