Add coredump by rust-cooker · Pull Request #417 · vivoblueos/kernel

rust-cooker · 2026-05-13T05:41:19Z

Summary

Implement a complete on-device coredump subsystem that generates ELF ET_CORE
files on fatal faults, supporting all 4 kernel architectures (RISC-V 32/64,
AArch64, Cortex-M).

Key components

ELF builder (elf.rs): Hand-written ELF32/ELF64 #[repr(C)] structs,
5-phase builder (header → PHDRs → notes → segments → backfill)
Register capture (arch/): Per-architecture register dumps following
Linux user_regs_struct layout
- RISC-V: 32 × usize (pc, ra, sp, gp, tp, t0-t6, fp, a0-a7, s1-s11)
- AArch64: 34 × 8 bytes (x0-x30, sp, pc, pstate)
- Cortex-M: 18 × 4 bytes (r0-r15, cpsr, orig_r0)
Signal mapping (signal.rs): Maps arch-specific exception codes to POSIX
signals (SIGSEGV, SIGBUS, SIGILL, SIGFPE, SIGTRAP, SIGABRT)
Memory regions (regions.rs): Collects BSS, system stack, and thread
stacks via GlobalQueueVisitor
Three backends (backend.rs):
- MemoryBackend: Static buffer for post-mortem retrieval
- FileBackend: VFS file output (/tmp/blueos.core.<pid>)
- LoggingBackend: Hex-encoded serial/semihosting output
Integration test: Runs in QEMU with .checker assertion validation
Parser tool: coredump_parser.py reassembles hex output → .core ELF
Usage docs: README.md with backend selection, Kconfig options, and tool usage

Kconfig options

ENABLE_COREDUMP (bool) — global switch
COREDUMP_BUF_SIZE (int) — buffer size per platform (64KB Cortex-M,
256KB 32-bit, 4MB 64-bit)
COREDUMP_MEM_BACKEND (bool) — MemoryBackend selection

Code quality

#[inline(never)] on all exception entry points (capture_regs,
capture_current_regs, dump())
Debug/Clone derives on all public types
// SAFETY: comments on every unsafe block
No unwrap() in core logic

Test plan

Compile on all 4 architectures (riscv64, riscv32, aarch64, cortex-m)
Integration test passes in QEMU (debug config)
All 27 board/config defconfigs updated with CONFIG_ENABLE_COREDUMP=y
Release/coverage builds verified

🤖 Generated with Claude Code
EOF
)"

…end support Add full coredump subsystem for post-mortem debugging across all architectures: - ELF32/ELF64 ET_CORE builder with PT_LOAD, PT_NOTE, NT_PRSTATUS/PRPSINFO/SIGINFO - Three compile-time backends: MemoryBackend (static buffer), FileBackend (VFS), LoggingBackend (hex-encoded serial output) - Per-architecture register capture: RISC-V (32 regs), AArch64 (34 regs), ARM Cortex-M (18 regs) - Fault-to-signal mapping: aarch64_ec_to_signo, riscv_mcause_to_signo, arm_cfsr_to_signo - Memory region collection: BSS, system stack, thread stacks via GlobalQueueVisitor - Kconfig-gated: ENABLE_COREDUMP, COREDUMP_BUF_SIZE, COREDUMP_MEM_BACKEND - .coredump_bss linker section to prevent self-collection - FileBackend PID-specific path (/tmp/blueos.core.<pid>) - Integration test (coredump_test.rs) with QEMU checker validation - Host-side hex-to-ELF parser tool (coredump_parser.py) - Coredump usage documentation (README.md) - Coding standards compliance: #[inline(never)] on exception entry points, Debug derives on public types, no unwrap() in core logic Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

Add missing trailing newlines to all 10 coredump .rs files and revert README.md to English per review feedback. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

han-jiang277 · 2026-05-13T06:26:36Z

+/// overflow since thread stacks (64 KB) are much smaller than the
+/// coredump buffer.
+#[link_section = ".coredump_bss"]
+static mut ELF_BUF: [u8; COREDUMP_BUF_SIZE] = [0u8; COREDUMP_BUF_SIZE];


Why use the BSS segment for memory space that is never used at runtime?

coredump may not trigger by signal when run in RTOS mode.

xuchang-vivo · 2026-05-13T06:14:20Z

    endchoice
 endmenu # os adapter configuration
+
+rsource "coredump/Kconfig"


Plz also put the path on the _kconfig_files, otherwise the build system can not trace the changes of coredump/Kconfig

xuchang-vivo · 2026-05-13T06:26:11Z

Missing assertion condition, how to determine the failure condition

xuchang-vivo · 2026-05-13T07:18:36Z

Put this functionailty into arch might be a better choice

xuchang-vivo · 2026-05-13T07:27:07Z

+            collect_all_stacks(&mut col);
+        }
+        DumpMode::Full => {
+            collect_all_stacks(&mut col);


When the mode is ALL, will the stack of the current thread definitely be recorded?

xuchang-vivo · 2026-05-13T07:33:57Z

+            hex_buf[n] = b'\n';
+            // Use semihosting eprint which is always available
+            let s = unsafe { core::str::from_utf8_unchecked(&hex_buf[..n + 1]) };
+            semihosting::eprint!("{}", s);


Semihosting is only used under qemu test, and it is not suitable for use here.

xuchang-vivo · 2026-05-13T07:38:21Z

+            let n = hex_encode(chunk, &mut hex_buf);
+            hex_buf[n] = b'\n';
+            // Use semihosting eprint which is always available
+            let s = unsafe { core::str::from_utf8_unchecked(&hex_buf[..n + 1]) };


Calling from_utf8_unchecked on hex_buf will result in undefined behavior.

A better approach is to use write_data8, which is part of Has8bitDataReg. You can refer to the implementation of kearly_printk for reference.

But in fact, using kearly_print is not a good method, as it will mix the output of these two channels together. A separate channel should be provided for it. I suggest leaving this blank and writing a todo or fix note. We can come back to fix it once I find an elegant way to connect to the second serial port channel.

xuchang-vivo · 2026-05-13T07:38:33Z

+    fn finalize(&mut self, mode: DumpMode) {
+        // Emit a trailer line with metadata
+        let size_kb = self.chunks * 32 / 1024;
+        semihosting::eprint!(


rust-cooker and others added 4 commits May 12, 2026 15:00

Add an linker script for binary

aa7a605

Merge branch 'main' into feat/coredump-code-review

4fda468

fix(coredump): add trailing newlines and update README to English

e4fdde8

Add missing trailing newlines to all 10 coredump .rs files and revert README.md to English per review feedback. Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

rust-cooker marked this pull request as draft May 13, 2026 06:01

han-jiang277 reviewed May 13, 2026

View reviewed changes

xuchang-vivo suggested changes May 13, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add coredump#417

Add coredump#417
rust-cooker wants to merge 4 commits into
mainfrom
feat/coredump-code-review

rust-cooker commented May 13, 2026

Uh oh!

han-jiang277 May 13, 2026

Uh oh!

han-jiang277 May 13, 2026

Uh oh!

xuchang-vivo May 13, 2026

Uh oh!

xuchang-vivo May 13, 2026

Uh oh!

xuchang-vivo May 13, 2026 •

edited

Loading

Uh oh!

xuchang-vivo May 13, 2026

Uh oh!

xuchang-vivo May 13, 2026

Uh oh!

xuchang-vivo May 13, 2026

Uh oh!

xuchang-vivo May 13, 2026

Uh oh!

xuchang-vivo May 14, 2026

Uh oh!

xuchang-vivo May 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

rust-cooker commented May 13, 2026

Summary

Key components

Kconfig options

Code quality

Test plan

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

xuchang-vivo May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

xuchang-vivo May 13, 2026 •

edited

Loading