refactor(axvm): add per-VM interrupt fabric by dweeqhd · Pull Request #1273 · rcore-os/tgoskits

dweeqhd · 2026-06-16T07:35:15Z

背景

这是 #1258 之后 Issue #595 的下一个 stacked PR。PR #1258 已经合并，因此本 PR 现在已直接 rebase 到上游 dev。

范围

仅实现阶段 5：建立每 VM 一个 InterruptFabric，并将其连接到设备构造。

添加架构无关的 VM interrupt fabric 和 IRQ resolver 实现。
将 fabric 存储在 VM 拥有的不可变状态中，且不引入 AxVM 引用环。
通过由当前 VM 的 fabric 支撑的 DeviceBuildContext 构建设备。
保留现有架构特定的生产中断路径，供后续迁移 PR 使用。
添加针对模式校验、事件转发、VM 隔离、factory 集成和所有权生命周期的聚焦测试。

评审边界

本 PR 有意排除：

RISC-V、x86、AArch64 和 LoongArch backend 迁移。
命名或多输出 IRQ 配置。
CI workflow 修复、本地实现记录和规划文档。

验证

git diff --check upstream/dev...HEAD
cargo fmt --all -- --check
cargo test --locked --offline -p virtualization-tests --test axvm_irq 无法在当前离线缓存中运行，因为合并后的上游 dev 现在依赖 aic8800 -> aes，而 aes 本地不存在。
cargo test --locked -p virtualization-tests --test axvm_irq 通过：5/5 个测试。

Part of #595.

mai-team-app

审阅总结

变更概述：本 PR 为 AxVM 引入了每 VM 一个 InterruptFabric 的中断路由层，将 IRQ 解析与 VM 所有权解耦，避免了 AxVM -> device -> IRQ -> AxVM 引用环。具体包括：

新增 InterruptFabric 结构体（irq/mod.rs），实现 IrqResolver trait。
重构 AxVM::init()，引入 init_with_factories() 方法，通过 DeviceBuildContext 使用 fabric 构建设备。
将 inner_const.call_once() 移至 vcpu 初始化之后——现在循环直接使用局部 &vcpu_list 而非依赖未完全初始化的 inner_const。
添加了 5 个聚焦测试：模式校验、事件排序、设备 IRQ 发射、RAII 生命周期、VM 隔离。

影响评估：变更限于 axvm crate 和 virtualization-tests，不影响现有架构特定的中断生产路径。init() 的默认行为通过 InterruptFabric::new(mode) 保持向后兼容。init_with_factories() 为后续架构迁移 PR 提供了扩展点。

CI 状态：所有 50 个 CI check 均为 skipped（PR 仅触及 virtualization 目录，CI workflow 可能有路径过滤器）。这不是 PR 本身的问题。

本地验证结果：

✅ cargo fmt --check -p axvm -p virtualization-tests 通过
✅ git diff --check origin/dev...HEAD 无空白错误
✅ cargo clippy -p axvm -p virtualization-tests --tests 无警告
✅ cargo test --locked -p virtualization-tests --test axvm_irq 全部 5 个测试通过

代码质量：

设计清晰，InterruptFabric 不引用包含 VM，杜绝了引用环
错误处理得当，NoIrq 模式和缺少 backend 的错误信息明确
Send + Sync 边界由 IrqSink: Send + Sync 自动满足
无新增 unsafe 代码

前期评审意见：无已有评审意见。

LGTM，批准合并。

Powered by mimo-v2.5-pro

ZR233

发现一个需要先处理的合入阻断点：当前 rcore-os/tgoskits 的 dev 已经到 1309a4d0c3667adc8cb999c7e76c899b4767cc4d，而该 PR 仍基于 da33641af8cf45d0e79512934a66d17617361e28。我在临时 worktree 里从当前 origin/dev 执行 git merge --no-commit 0731bef97571abb77d8d2d09a213148afb2f9e60，合并虽然没有文本冲突，但会让 Cargo.lock 产生 116 行变更；除了新增 virtualization-tests 对 ax-plat/axvm 的依赖外，还会把当前 dev 中已有的多处 windows-sys/windows-targets 解析回退到旧版本，并删除 #1277 已引入的相关 lockfile 条目。这些 lockfile 回退和本 PR 的 axvm interrupt fabric 变更无关，直接合入会覆盖当前 dev 的依赖解析结果。

请先 rebase 到最新 dev，然后重新生成/更新 Cargo.lock，确保 lockfile 只包含本 PR 新增测试依赖所需的变化。

功能代码本身我没有发现阻断问题：InterruptFabric 不持有 AxVM，默认 init() 仍只注册 Dummy factory，现有架构相关设备继续走 legacy fallback；显式 factory path 通过 DeviceBuildContext 使用 VM-local fabric，和本 PR 范围一致。

本地验证已跑过：

git diff --check origin/dev...HEAD
cargo fmt --all -- --check
cargo metadata --locked --format-version 1 --no-deps
cargo test --locked -p virtualization-tests --test axvm_irq（5/5 通过）

mai-team-app

审阅总结

变更概述：本 PR 为 AxVM 引入每 VM 一个 InterruptFabric 的中断路由层（Issue #595 阶段 5），将 IRQ 解析与 VM 所有权解耦，杜绝 AxVM -> device -> IRQ -> AxVM 引用环。

具体变更：

新增 InterruptFabric（irq/mod.rs），实现 IrqResolver trait，仅持有 mode 和可选的 Arc<dyn IrqSink>。
新增 AxVM::init_with_factories()，通过 DeviceBuildContext 使用 fabric 构建设备。
将 inner_const.call_once() 移至 vcpu 初始化之后——修复了原代码在 vcpu setup 前就 move vcpu_list 的问题，现在 setup 循环使用局部 &vcpu_list。
5 个聚焦测试：模式校验、事件排序、设备 IRQ 发射、RAII 生命周期、VM 隔离。

评审边界：仅 axvm crate 和 virtualization-tests，不影响现有架构特定中断路径。init() 默认行为通过 InterruptFabric::new(mode) 保持向后兼容。

CI 状态：8 个 check 中 3 个 success（Detect changed paths、Cancel stale CI runs、Run sync-lint / run_container），4 个 skipped（路径过滤），1 个 in_progress（Check formatting / run_host）。无任何 CI 失败。

本地验证结果：

✅ cargo fmt --all -- --check 通过
✅ git diff --check origin/dev...HEAD 无空白错误
✅ cargo clippy -p axvm -p virtualization-tests --tests -- -D warnings 无警告
✅ cargo test --locked -p virtualization-tests --test axvm_irq 5/5 测试通过

前期评审意见：

ZR233 之前 REQUEST_CHANGES 要求 rebase 以修复 Cargo.lock 回退问题。当前 PR 已基于最新 dev（3598a85...），merge-base 一致，Cargo.lock diff 仅 2 行新增（ax-plat/axvm 依赖）。该问题已解决。
mai-team-app[bot] 的 APPROVE 基于旧 commit 0731bef...，当前 HEAD 已更新。

代码质量：

设计清晰：InterruptFabric 不持有 VM 引用，断开引用环
call_once 后移是正确的安全性改进：避免 setup 失败时 inner_const 处于不一致状态
错误信息明确（NoIrq 模式、缺少 backend）
无新增 unsafe 代码
测试覆盖全面，包含隔离性和 RAII 验证

LGTM，批准合并。

Powered by mimo-v2.5-pro

ZR233

复审当前 head 075b8530e14f15fcfeea17c5895ebaac39bcf8da。\n\n上一轮阻塞的 stale Cargo.lock 回退已经解决：当前相对 origin/dev 的 lockfile 只为 virtualization-tests 新增 ax-plat 和 axvm 两个 dev 依赖，没有再回退无关依赖版本。旧 blocker 已不再适用。\n\n实现上，InterruptFabric 只保存 VM interrupt mode 和可选 IrqSink，实现 IrqResolver 后通过 DeviceBuildContext 传给设备 factory，不持有 AxVM，因此不会引入 AxVM -> device -> IRQ -> AxVM 引用环；默认 init() 仍注册 builtin factories 并使用与 VM 配置匹配的 fabric，架构后端迁移也保持在本 PR 边界之外。\n\n本轮验证：\n- git merge-tree --write-tree origin/dev refs/remotes/origin/pr/1273：通过。\n- git diff --check origin/dev...refs/remotes/origin/pr/1273：通过。\n- [patch.crates-io] 检查：未发现。\n- cargo test --locked -p virtualization-tests --test axvm_irq：通过，5/5。\n\nCI 当前 head 上 formatting、sync-lint、clippy、Axvisor/ArceOS/Starry 多个相关 host/container job 已通过；最后查询时仅 Starry loongarch64 container 仍 pending，且该 pending 不覆盖本 PR 的 axvm interrupt fabric 核心路径。\n\n重复/重叠分析：#1265 涉及 serial IRQ model，与本 PR 的 VM-local interrupt fabric/device factory 接线不同；#1062/#984/#1125 不是同一功能面。无新的阻塞问题，批准。

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

ZR233

复审当前 head b3711ee9177d75906ac78756146775692811e6d8。

这个 PR 为 axvm 引入 per-VM InterruptFabric：每个 VM 在初始化时创建自己的 interrupt fabric，把 VM 的 interrupt mode 和可选 IrqSink 封装成 IrqResolver，再通过 DeviceBuildContext 传给设备 factory。它的作用是让设备注入中断时只依赖 VM 局部的 IRQ fabric，而不是持有 AxVM 本身，从而避免 AxVM -> device -> IRQ -> AxVM 引用环，并保证相同 IRQ 号在不同 VM 之间隔离。新增 virtualization-tests/tests/axvm_irq.rs 覆盖 fabric 后端接线、事件顺序、无 IRQ 后端时的拒绝路径、VM 间隔离和 drop 后释放后端等核心语义。

本地验证：git merge-tree --write-tree origin/dev HEAD 通过；git diff --check origin/dev...HEAD 通过；未发现 [patch.crates-io]；cargo test --locked -p virtualization-tests --test axvm_irq 通过 5/5；cargo xtask clippy --package axvm 通过 8/8 feature 组合，包括 base、4-level-ept、fs、host-fs、plat-dyn、sstc、svm、vmx。

CI 状态方面，当前 head 的 formatting、sync-lint、run-host clippy、ArceOS/Axvisor 相关 QEMU/board jobs 和 Starry loongarch64 QEMU 已通过；最后查询时仍有部分全仓库 Starry/board 矩阵 pending 或 expected skip，但这些不覆盖本 PR 的 axvm IRQ fabric 核心路径。本 PR 的核心风险已经由本地 package test 和 axvm 多 feature clippy 覆盖。

重复/重叠方面，#1265 关注 serial IRQ model，与这里的 VM-local interrupt fabric/device factory 接线互补；其它开放 PR 未实现同一 axvm IRQ fabric 抽象。无新的阻塞问题，批准。

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

…on suite (#1125) * test(starry): add nix-smoke case and namespace-focused syscall regressions 新增 Nix smoke 测试用例（`nix-smoke/qemu-x86_64.toml`），在 Alpine StarryOS guest 中安装 Nix 2.31.5、验证 `nix --version`、然后用 `--option sandbox true` 执行最小本地 derivation 构建（预期输出 `NIX_LOCAL_BUILD_OK`）。当前 sandbox 构建阶段挂死（ NIX-SMOKE-004），根因是 `MntNamespace` 为 256 字节桩，未实现 mount namespace 隔离。新增三个聚焦 syscall 回归测试： - `test-proc-self-ns-mnt`：探测 `/proc/self/ns/mnt` 的 stat/open/fstat/read - `test-proc-self-ns-mnt-readlink`：探测 readlink 返回值与 errno - `test-unshare-basic`：探测 unshare(CLONE_NEWNS) 等 flags 的返回值新增 `rust-hello` 用例用于 Rust 程序 app-test 流程验证。清理 `test-capget`、`test-capset`、`test-raw-msg-peek` 中导致 grouped syscall runner 预构建失败（apk: Symbolic link loop）的 CMakeLists.txt 和 test_framework.h。 * chore(nix): update flake with rust-overlay input and newer nixpkgs nixpkgs 更新到 2026-05-28，新增 rust-overlay 输入（oxalica），为后续 Rust 工具链版本管理预留 overlay 集成点。 * feat(starry-kernel): implement PID namespace orphan reaping with subreaper chain and zap 问题 ---- 沙箱化 nix-build 在 Nix store output lock 处等待直至超时中断 (NIX-SMOKE-006/-007)。代码审计发现 `starry-process/src/process.rs` 中 PID namespace init 退出时 `exit()` 直接 return，子进程未 reparent；无 `zap_pid_ns_processes()` 实现；无 `find_new_reaper()` 子 reaper 链；`PR_SET_CHILD_SUBREAPER` 返回 EINVAL。修复 (5 个文件) -------------- 1. components/starry-process/src/process.rs - 添加 is_child_subreaper 字段、find_new_reaper() 沿父链查找最近 subreaper - exit() 使用 find_new_reaper() 替代硬编码 INIT_PROC 2. os/StarryOS/axnsproxy/src/pid.rs - PidNamespace 添加 init_global_tid 追踪 ns init 进程 3. os/StarryOS/kernel/src/syscall/task/clone.rs - clone(CLONE_NEWPID) 和 unshare-prep 两处设置 init_global_tid 4. os/StarryOS/kernel/src/syscall/task/ctl.rs - 实现 PR_SET_CHILD_SUBREAPER (36) 和 PR_GET_CHILD_SUBREAPER (37) 5. os/StarryOS/kernel/src/task/ops.rs - do_exit 中内联 zap_pid_ns_processes: PID ns init 退出时 SIGKILL 命名空间内所有剩余进程诊断与测试 ---------- - D5: strace 不可用 (x86_64 ptrace 缺 PTRACE_GETREGS/SETREGS) - D6: ns-process-diag standalone PASS, unshare CLONE_NEWPID 确认活跃 - D7: kill(-pgid, SIGKILL) 跨 session 边界工作正常 - 新增 C 测试: test-ns-orphan-reap (clone3 CLONE_NEWPID), test-pgkill-session - 新增 shell 诊断: ns-process-diag/qemu-x86_64.toml 验证 ---- - starry-process clippy: 1/1 PASS - starry-kernel clippy: 12/12 PASS - cargo fmt: OK - syscall full suite: PASS (164.48s, 无回归) - ns-process-diag standalone: PASS * feat(test-suit): add grouped nix-prereqs case and merge nix-smoke subcases - Create test-nix-prereqs grouped QEMU case with 11 kernel regression tests serving as prerequisites for nix-smoke: ext4-inode-unique, fcntl-lock-lifecycle, flock-cloexec, lock-close-range-cloexec, nix-builder-lifecycle, open-unlink-write, pidfd-poll-exit, pipe-poll-close, proc-wchan, pty-master-close, pty-openpt - Merge standalone nix-smoke-nosandbox into nix-smoke grouped subcase, removing the separate case directory - Exclude original standalone test directories via .git/info/exclude as they are superseded by the grouped case - Kernel fixes enabling all tests: pipe poll close (POLLIN|POLLHUP), pidfd exit notification, PTY master-close line discipline, rsext4 open-unlink file retention, /proc/<pid>/wchan labels * fix(clippy): sync FAT DirNodeOps::create signature, clean up AGENTS.md - Remove _uid: u32, _gid: u32 from fat/dir.rs DirNodeOps::create signature to match the reverted 4-param trait API. - Remove local proxy tool references (Clash Verge Rev, proxychains4-ng) from AGENTS.md - not appropriate for project-wide instructions. - Remove SpecKit marker referencing uncommitted plan.md. * feat(vfs): merge PR #1097 uid/gid with mount namespace infrastructure Integrate both feature branches that touched the same VFS files: - Our mount-ns: clone_tree, children, MountNamespace, unshare_mount_namespace - PR #1097 uid/gid: DirNodeOps::create trait, create_dir, symlink ownership Changes: - VFS layer (mount.rs, dir.rs, fs.rs): keep mount-ns infrastructure, manually apply uid/gid from PR #1097 (trait signature, _with_owner calls, cred.fsuid/fsgid passthrough in open_file) - Kernel pseudofs/syscall files (cgroup, dir, tmp, mod, ctl, memfd): restored from upstream/dev which already has uid/gid support - rsext4 backend: dev baseline + truncate_inode open-unlink fix + .lock orphan preservation + _with_owner exports - FAT/lwext4 backends, lockdep test: restored from upstream/dev Verified: test-nix-prereqs (12/12), nix-smoke, test-vfs-ownership all PASS in CI-like Podman container. * style: apply cargo fmt to fix CI formatting check * fix(flock): correct dead-OFD pruning threshold from <=1 to <1 The `try_flock_once` function prunes stale same-pid flock entries whose OFD is dead. The threshold `weak.strong_count() <= 1` was too aggressive: when a file has exactly one open fd (the normal case without dup), strong_count==1 and the entry was incorrectly pruned, breaking flock(LOCK_EX) exclusive exclusion. Change to `< 1` — only prune when the Weak cannot be upgraded (strong_count==0), matching Linux `locks_flock_remove_dead()` semantics. The lower-bound check at the top of try_flock_once already removes entries with strong_count==0 via `entries.retain(|e| e.weak.strong_count() != 0)`. Fixes: bug-flock-exclusive-exclusion, bug-flock-failed-upgrade * fix(starry-kernel): sync setns mount view * test(starry): restore raw MSG_PEEK regression * fix(starry-kernel): repair wait merge imports * fix(rsext4): implement filename-independent open-unlink lifetime semantics Replace the `.lock`-only hidden-rename workaround with Linux-compatible unlink(2) behavior for all regular files. When the last directory link is removed while the file is still open, the inode stays alive (accessible by inode number through existing fds) and fstat reports st_nlink == 0. - Remove `name.ends_with(".lock")` branch and `.starry-orphan-lock-*` hidden rename from `inode.rs` unlink handler - Temporarily bump i_links_count before rsext4::unlink to prevent immediate inode-free; set back to 0 afterward for correct st_nlink - Expand test-open-unlink-write with comprehensive Linux-semantics coverage: multi-name patterns, O_APPEND, empty-file unlink, two-fd concurrent access, hard-link (nlink>1), directory non-regression, and hidden-orphan readdir() assertions (186 assertions, 0 failures) - All existing read_at/write_at/append/set_len paths already use inode-number operations; no VFS layer changes needed Validated: test-nix-prereqs PASS, nix-smoke PASS in CI-like container. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(rsext4): implement Linux-compatible open-unlink final-close cleanup Remove the link-count bump hack in DirNodeOps::unlink and replace it with proper pending-delete tracking: last-link unlink sets i_links_count to 0 and registers the inode for deferred cleanup; Inode::drop triggers free_inode when the final reference is released. - Make free_inode_with_dtime public as free_inode (rsext4) - Add PendingUnlinkMap (BTreeSet<InodeNumber>) to Ext4State - Rewrite unlink: i_links_count>1 uses normal unlink; last link removes dentry and registers pending-delete, no immediate free - Implement Drop for Inode to trigger free_inode on final close - Strengthen test-open-unlink-write C test with statvfs-based resource reclamation probes, closed-file unlink case, and hardlink final-link zero-nlink verification * fix(ax-fs-ng): use pending unlink registration helper * test(starry): document tcp concurrent readiness check PR #1125 CI failed once in bug-tcp-concurrent-connect after all clients were accepted, with poll(POLLIN) timing out on accepted client 0. The same bugfix case was rerun locally twice in the CI-like Podman container and did not reproduce; both runs passed. Keep a short comment near the poll check so future failures are easier to classify as payload readiness rather than accept queue handling. * fix: use upstream AGENTS.md * <think>Let me analyze this commit to understand what changes are being made: 1. **New Nix-related test infrastructure** for StarryOS: - `build-x86_64-unknown-none.toml` - build config for x86_64 - `nix-nosandbox.sh` - a smoke test for Nix without sandbox - `nix.sh` - a more comprehensive Nix test with sandbox support (diagnostics, polling) - `prebuild.sh` - builds the Nix overlay by extracting from Alpine rootfs - `qemu-x86_64.toml` - QEMU config for running the Nix tests - `test_nix.sh` - test runner that runs the nosandbox test 2. **Fix in `flake.nix`**: Changed `project_root=...` to `export project_root=...` (missing export) 3. **Namespace ID improvements in axnsproxy**: - Added `AtomicU64` for id tracking in `pid.rs`, `user.rs`, `uts.rs` - This appears to be giving namespace proxies unique IDs 4. **stat() implementation for namespace file descriptors** (`nsfd.rs`): - Now returns proper inode number, mode, nlink, blksize instead of default Kstat 5. **Proc pseudofs namespace info** (`proc.rs`): - Changed to show actual namespace IDs instead of just names or levels - `uts:[id]`, `ipc:[id]`, `mnt:[id]`, `pid:[id]` 6. **ext4 filesystem changes** (`fs.rs`, `inode.rs`): - Added `BTreeMap` import - Added a NOTE comment about ref counting responsibility 7. **Starvy build script** (`mod.rs`): - Added `apply_dynamic_x86_64_qemu_boot` call 8. **New test cases for hardlink semantics**: Three tests added to verify refcount behavior during concurrent unlink operations, partial unlink scenarios where one link persists, and immediate resource reclamation when files are closed before deletion. The core theme is establishing Nix as a first-class package manager within the OS environment, with complementary namespace ID tracking and filesystem improvements to support this integration.</think> Add Nix smoke test infrastructure for StarryOS * fix(clippy): collapse nested if in inode.rs, exclude old nix-smoke - Fix clippy::collapsible_if in rsext4/inode.rs - Remove test-suit nix-smoke from git tracking (moved to apps/starry/nix) - Add nix-smoke dir to .git/info/exclude Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * chore: update flake.lock Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * added README for nix app test, also as a explain/note for https://github.com/rcore-os/tgoskits/pull/1125\#issuecomment-4639168301 * fix(starry-kernel): add CLONE_FS support to sys_unshare Nix 下载线程（fetchTarball/fetchurl）在内部调用 unshare(CLONE_FS) 来隔离下载线程的文件系统状态，但 StarryOS 的 sys_unshare 不支持 CLONE_FS（返回 EINVAL）。此修复添加 CLONE_FS 支持。变更： - namespace.rs: 添加 CLONE_FS 到 SUPPORTED_NS_FLAGS，Phase 2 单锁 clone FsContext 断开共享（避免 task-scoped mutex 死锁） - test-nix-prereqs: 新增 test-unshare-fs 回归测试（3 场景）, 标注 "nixpkgs 测试可能用到" - test-nix-prereqs 从 12 增加到 13 个用例，全部通过同时： - apps/starry/nix: 添加 nix-nixpkgs.sh（pkgs.stdenv.mkDerivation 测试框架，当前因 stdenv bootstrap 下载仍需 unshare 而预期 FAIL） - apps/starry/nix: 更新 README/prebuild/qemu TOML/test_nix.sh 验证： - test-nix-prereqs: 13/13 PASSED (24.81s container) - cargo fmt & clippy: clean * fix(starry-kernel): resolve clippy warnings in CLONE_FS unshare * chore(nix): disable nixpkgs test — blocked by Nix download unshare * chore: trigger CI * chore: trigger CI * chore: trigger CI * fix(starry): address PR #1125 review — unshare(CLONE_FS), builder args, docs - namespace.rs: rebind FS_CONTEXT task-local to new Arc on unshare(CLONE_FS) instead of mutating the shared Arc inner — matches Linux private fs_struct semantics after clone(CLONE_FS) sharing - test-unshare-fs.c: rewrite to use real clone(CLONE_FS|CLONE_VM|SIGCHLD) instead of fork() — verify both shared-cwd (before unshare) and isolated-cwd (after unshare) phases - nix-nosandbox.sh: use /bin/mkdir (absolute path) and && (fail-fast) in derivation builder args — fixes false PASS when mkdir is not in PATH - apps/starry/nix/README.md: correct test content table — nix-nixpkgs is deferred (not CI), add explicit 'not planned at this stage' section Fixes: review #4446698115 blocking items 1-4 * fix(starry-nix): narrow success marker and sync PR docs per review #1125 - test_nix.sh: NIX_ALL_TESTS_PASSED → NIX_NOSANDBOX_COMPLETE (nixpkgs skipped) - qemu-x86_64.toml: sync success_regex to NIX_NOSANDBOX_COMPLETE - namespace.rs: cargo fmt blank-line fix - tasks.md: mark all 18 tasks complete - PR #1125 body: fix test count 11→13, update fork→clone(CLONE_FS) description * fix(starry-ns): privatise FsContext before unshare(NEWNS) on shared context on_enter holds a leaked scope.read() guard for the task lifetime. scope.write() would self-deadlock because spin::RwLock does not allow read->write upgrade. Temporarily drop the read lock (force_read_decrement + ActiveScope::set_global) before scope.write(), then re-acquire. - namespace.rs: merge CLONE_NEWNS and CLONE_FS into shared privatise path (want_ns || want_fs) so mount-ns unshare no longer mutates the shared Arc<Mutex<FsContext>> of a clone(CLONE_FS) sharer - test-unshare-mount-ns: add clone(CLONE_FS|CLONE_VM|SIGCHLD) + child unshare(CLONE_NEWNS) + bind-mount isolation regression (UNSHARE_MOUNT_NS_CLONE_ISOLATION_PASSED) 13/13 test-nix-prereqs PASS * docs(nix): sync README file-structure labels with actual test coverage nix-nixpkgs.sh: (CI, requires network) → (deferred, requires mount ns) test_nix.sh: runs nosandbox → nixpkgs → nosandbox only (nixpkgs intentionally skipped) * chore(starry): add nix-prereqs regression test suite and update flake.lock Add grouped regression test cases under test-nix-prereqs covering: - ext4 inode uniqueness after free/realloc - fcntl lock lifecycle across close/fork/exec - flock cloexec inheritance - lock close-range cloexec cleanup - nix builder lifecycle (fork/exec/pipe supervision) - open-unlink-write open-fd persistence - pidfd poll exit notification - pipe poll close EOF readiness - /proc/<pid>/wchan wait-channel labeling - PTY master close and line-discipline draining - PTY posix_openpt grant/unlock - unshare CLONE_FS and mount namespace isolation Update flake.lock: rust-overlay pin to fe64e6b4. * feat(starry-nix): add aarch64 test configs for nix smoke and test-nix-prereqs Add aarch64 build and QEMU configs for the nix app test, and a grouped qemu-aarch64.toml for test-nix-prereqs. All 13 regression tests pass on aarch64 in CI-like podman QEMU validation. - apps/starry/nix/build-aarch64-unknown-none-softfloat.toml - apps/starry/nix/qemu-aarch64.toml - test-suit/starryos/qemu-smp1/test-nix-prereqs/qemu-aarch64.toml - apps/starry/nix/README.md: document aarch64 run command * chore: trigger CI * chore: trigger CI * chore: trigger CI * chore: trigger CI * fix(starry): remove duplicate test-nix-prereqs from old normal/ layout The test-suit/starryos/normal/qemu-smp1/test-nix-prereqs/ directory was a duplicate of test-suit/starryos/qemu-smp1/test-nix-prereqs/. The current Starry test-suit no longer uses the normal/stress first-level grouping, and CI discovers cases under qemu-smp1/ directly. Keeping the old copy would cause local verification and CI to see inconsistent case sets. - Delete 30 tracked files under normal/qemu-smp1/test-nix-prereqs/ - Update apps/starry/nix/README.md to reference the current path ci: trigger * fix(rsext4): correct inodetable_cache field typo in crc_integrity test The test referenced the field as "inodetable_cahce" but the Ext4FileSystem struct defines it as "inodetable_cache". This caused the "Test with std" CI job to fail at compile time. ci: trigger * fix(starry): move pipe close→IN from poll layer to epoll layer The previous fix added "|| closed" to pipe.rs IN reporting to make Nix see EOF via EPOLLIN. This broke poll(2)/select(2) semantics: after draining a closed pipe, Linux reports only POLLHUP (no POLLIN). The regression was caught by test-select-poll-family (poll_pipe_hup test 3) on loongarch64. Move the compatibility to the epoll layer: match_ready_events() now forces IoEvents::IN into the result when HUP is present. This gives EPOLLIN visibility to Nix without affecting poll(2)/select(2) behavior. - pipe.rs: revert IN = occupied > 0 (no || closed) - epoll.rs: match_ready_events adds IN when HUP is set ci: trigger * ci: trigger * fix(starry): remove old normal/ test-suit layout and sync spec paths The test-suit/starryos/normal/ directory contained old-layout diagnostic tests (ns-process-diag, timeout-diagnostics, syscall/test-*) that the current test-suit runner does not discover. CI only finds cases under test-suit/starryos/qemu-smp1/ (no normal/stress first-level grouping). - Delete 16 tracked files under test-suit/starryos/normal/ - Update specs/002-starryos-nix-smoke/plan.md and tasks.md to reference the current qemu-smp1/test-nix-prereqs/ path * fix(axbuild): rename apply_dynamic_x86_64_qemu_boot to apply_dynamic_platform_qemu_boot Upstream renamed the function in scripts/axbuild/src/test/qemu.rs from apply_dynamic_x86_64_qemu_boot to apply_dynamic_platform_qemu_boot. The starry/mod.rs module still had two stale references to the old name, which caused sync-lint CI failure (E0425). ci: trigger * fix(test): add waitpid before poll in pipe-poll-close to avoid scheduler race Test 1 forks a child that writes data to a pipe, but the parent called poll() immediately without waiting for the child to be scheduled. On x86_64 runners the child ran after the parent, causing poll to timeout (return 0) because no data had been written yet. Other architectures (aarch64, riscv64, loongarch) passed because the child was scheduled before the parent reached poll(), but the test was racy by design. Add waitpid(child, NULL, 0) after closing the parent write end and before the poll call so the child is guaranteed to have finished writing. ci: trigger * fix(test): increase PTY read timeout for child stderr to 5s The third read_line_timeout waits for /bin/sh (execed by child) to write CHILD_OK to the PTY. On busy x86_64 CI runners the /bin/sh process took over 1s to start and write, causing poll to timeout and assert FAIL. Extend from 1000ms to 5000ms. ci: trigger * chore(nix): update nixpkgs and rust-overlay inputs * refactor(starry-test): migrate kernel regression tests from test-nix-prereqs to system grouped suite Move 13 kernel regression test subcases (ext4, fcntl/flock, pidfd, pipe, proc, pty, namespace) out of the standalone test-nix-prereqs case into qemu-smp1/system/ following Starry test-suit conventions. Changes: - 13 subcases moved to system/<name>/ with standardized CMakeLists.txt - All install to usr/bin/starry-test-suit (fixed 7 that were at usr/bin) - 3 test_framework.h local copies replaced with ../common/test_framework.h - -static link option preserved where originally present - Deleted standalone test-nix-prereqs/ (qemu-*.toml + old c/ structure) - Updated apps/starry/nix/README.md to use -c qemu-smp1/system The system/CMakeLists.txt auto-discovers subcases via GLOB, and existing qemu-*.toml files already loop over /usr/bin/starry-test-suit/*. No runtime config changes required. Addresses review feedback on PR #1125. * chore(specs): remove stale spec files from tracking, now covered by .git/info/exclude * fix(pty): restore peer close notification after upstream merge Upstream refactor removed TtyRead::closed() and TtyWrite::close(), breaking PTY master-close detection. When the last slave fd closed, the master side was never woken — poll() blocked and read() hung. Restore via shared Arc<AtomicBool> between PtyReader/PtyWriter: - pty.rs: add writer_closed flag to both tuple structs - ldisc.rs: add closed()/close() back to TtyRead/TtyWrite traits, check closed in poll_read() and read() for Passive mode - mod.rs: call writer.close() on last fd close - Fix SimpleReader::poll() to handle partial buffer pushes (lost bytes on lines > BUF_SIZE=80) Verified: cargo xtask starry test qemu --arch x86_64 -c qemu-smp1/system/test-pty-master-close: 19/19 PASS * fix(starry): record poll_wait wchan in poll/ppoll syscalls The poll syscall used raw poll_fn without WaitChannelGuard, so /proc/pid/wchan was empty even when the task was blocked in poll(). Add WaitChannelGuard::set(WaitChannel::PollWait) to match the existing pattern used by poll_io_with_wchan() in future/poll.rs. Verified: cargo xtask starry test qemu --arch x86_64 -c qemu-smp1/system/test-proc-wchan: 37/37 PASS * fix(starry-nix): switch nix-nosandbox to builtins.toFile, bypass builder socketpair gap Replace builtins.derivation (nix-build) with nix eval + builtins.toFile in nix-nosandbox.sh. The derivation builder requires socketpair-based communication between Nix and the builder subprocess, which currently fails on StarryOS after the IRQ-safe deferred notification merge (PR #1278) — the socketpair read path returns immediate EOF. builtins.toFile verifies Nix expression evaluation and store-path write without depending on builder subprocess communication. The full derivation builder workflow is deferred until socketpair poll semantics are aligned with Linux behaviour. Also update README.md to document the test strategy shift and the known builder communication gap. Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai> * seccomp and capablities (#1275) * feat: nsproxy core - ProcessData migration, sys_unshare, clone, user credentials, PID - Add axnsproxy crate with all namespace types - Replace uts_ns with nsproxy in ProcessData - Implement sys_unshare for all 6 namespace flags - Implement clone namespace inheritance - Add user namespace credential filtering - Add PID namespace translation for getpid * feat: IPC ns isolation, network ns filtering, and mount fs extensions - Use (key, ns_id) composite key in shm/msg for IPC namespace isolation - Add in_root_net_ns() in shared syscall::ns module to avoid duplication - Filter eth0 visibility in net/packet/netlink based on network namespace - Extend sys_mount to support proc/sysfs/devtmpfs/devpts - Wire up sys_unshare dispatch and namespace module registration * fix: sys_getppid PID namespace translation, LazyLock unification, remove duplicate test_framework.h - sys_getppid now returns 0 when parent in different PID namespace - Replace spin::Lazy with spin::LazyLock in all axnsproxy files - Remove duplicate test_framework.h, use common version * fix: replace deprecated spin::Lazy with spin::LazyLock * fix: update pid.rs and test CMakeLists, remove temp file * feat(starry-kernel): add overlay filesystem support * fix(starry-kernel): honor overlay whiteout lookup semantics * feat: add seccomp support and related syscall integrations * docs(starry-kernel): document seccomp and capability code * fix(starry-kernel): distinguish seccomp kill actions * fix(starry-kernel): align seccomp and capability tests * fix(starry-kernel): preserve seccomp errno zero * ci: trigger rebuild * ci: trigger rebuild * ci: trigger rebuild * fix(somehal): send x86 helper IPI on IPI vector (#1297) * refactor(ax-runtime): adapt submit-poll fs block irq registration (#1228) * refactor(ax-fs-ng): remove legacy filesystem stack * refactor(ax-fs-ng): unify filesystem consumers * feat(ax-fs-ng): move root and volume discovery into fs crate * feat(rdif-block): add batched completion contracts * feat(ax-fs-ng): add submit-poll block runtime * refactor(ax-runtime): wire fs block runtime to devices * refactor(ax-fs-ng): port ext4 and fat to submit-poll devices * feat(rsext4): add extent run file IO helpers * fix(ax-fs-ng): serialize cached writeback and mmap dirty pages * perf(nvme): use queue depth and completion batching * refactor(ax-fs-ng): task-synchronous block runtime * feat(axhal): expose bootargs facade for filesystem init * refactor(ax-fs-ng): improve request handling and task management in block device operations * refactor(ax-runtime): adapt fs block irq registration * fix(axvisor): import fs extension traits per target * fix(fs): restore arceos fat root initialization * fix(axvisor): enable ext4 rootfs backend * fix(axfs-ng): mount detected root filesystem kind * fix(ci): stabilize arceos host tests * fix(ax-driver): limit rk3568 fifo block requests * fix(axfs-ng): synchronize cached file state across ext4 paths * fix(starry): adapt qemu nvme tests to irq polling * fix(ax-plat-riscv64-sg2002): set cvsd queue inflight limit * Enhance NVMe driver with interrupt support and improve queue handling - Added support for I/O queue interrupts in the NVMe driver, allowing for more efficient interrupt handling. - Introduced a new `Config` struct method to enable interrupt-driven I/O. - Updated NVMe queue handling to use `UnsafeCell` for safe mutable access to submission and completion queues. - Refactored NVMe command submission and completion polling to utilize the new queue handling methods. - Improved the ARM GIC driver tests to conditionally compile based on architecture. - Simplified SPI priority and target configuration logic in the GIC distributor. - Introduced a new `BlockCompletionMode` enum to manage block device completion strategies (polling vs. IRQ-driven). - Enhanced block device handling to support both polling and interrupt-driven modes. - Added smoke and benchmark tests for block I/O to validate performance and correctness. * refactor(ax-fs-ng): consume rdif block devices directly * fix(ax-fs-ng): mount non-root partitions during init * fix(starry): adapt overlayfs to ax-fs-ng vfs * fix(ax-driver): keep Phytium MCI block path polling * refactor(ax-driver): unify sdmmc submit poll irq path * fix(ax-driver): keep Phytium MCI polling by default * fix(phytium-mci): restore safe block polling startup * fix(ax-fs-ng): synchronously advance polling block queues * fix(ax-fs-ng): drain repolled block completions * fix(ax-fs-ng): mount dirty ext4 root read-only * fix(ax-fs-ng): mount partitions over readonly root placeholders * fix(ax-fs-ng): use irq-safe block drain notify * refactor(axvm): add VM interrupt fabric (#1273) Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies. * fix: resolve post-merge build errors and formatting * fix: restore NsFd::Mnt named-field construction in fd_ops ns_open * chore(nix): update rust-overlay in flake.lock * ci: trigger CI * ci: retrigger CI * chore(flake): update rust-overlay lock * fix(axfs-ng): align ext4 error-state mount with Linux semantics Previously, when ext4 was in error state, mount_readonly_fallback called into_inner() on the Jbd2Dev, discarding the entire block cache, then created a fresh Jbd2Dev with an empty cache. Subsequent reads (e.g. loading a guest kernel image during AxVisor VM boot) missed cache entirely, hit the hardware FIFO I/O path on SdmmcBlockDevice, and could hang forever in polling_wait_for_any_key() (device.rs:1258) if the SD card controller was left in a bad state by the failed mount. Fix: - Preserve the Jbd2Dev and its block cache (never call into_inner()) - Replay journal before read-only mount when filesystem is in error state (Linux always replays journal for consistency) - Keep replay_journal=false only for EUCLEAN fallback where replay already failed This matches Linux behaviour where s_bdev survives from mount to umount, and journal replay happens even for read-only mounts. * fix(axfs-ng): add single retry for block reader in partition scan When the SD card HighSpeed switch fails during init (a known intermittent issue on PhytiumPi boards), the controller may be left in a transient bad state. The first MBR read then fails with a Reader error, causing the partition scan to return no volumes, which eventually panics at root.rs:384 (configured root device not found). Add one retry in VolumeReader::read_block. A single re-read is often sufficient for the hardware to recover without a full controller reset. * fix(starry-kernel): drop FD_TABLE lock before release_locks_on_close sys_close_range and sys_dup3 (the latter is what sys_dup2 wraps on x86_64) called release_locks_on_close() while still holding the FD_TABLE write guard. release_locks_on_close() now walks every fd table via fd_tables_contain_file() -> fd_table_file_refs(), which re-acquires FD_TABLE, so the first close/dup that replaces an already-open fd self-deadlocks. That deadlock is hit by every shell pipeline / command substitution: `echo foo | cat`, `$(...)`, the grouped autorun's `$(date +%s)`, etc. all dup2() over stdin/stdout. With the lock wedged, StarryOS never reaches the `root@starry:` prompt, so the whole qemu-smp1/system suite (and the nix app test) time out without running a single case. Collect the removed descriptors, drop the write guard, then release the locks -- matching close_all_fds(), the execve CLOEXEC close path, and close_file_like(), which all already do this. Root cause is branch-local: fd_tables_contain_file() was wired into release_locks_on_close() for flock release but close_range/dup3 were not updated; upstream/dev's release_locks_on_close() does not touch FD_TABLE, so its otherwise-identical close_range/dup3 are safe. Verified locally (CI container) with the exact CI command: cargo xtask starry test qemu --arch x86_64 -> PASS qemu-smp1/system, PASS qemu-smp4/system, 2/2 cases passed. * chore(starry-nix): wip nix qemu config + flake tweaks - apps/starry/nix/qemu-x86_64.toml: switch to grouped test_commands and a shorter timeout instead of shell_init_cmd (matching the grouped runner schema). - flake.nix: drop the hardcoded USTC rust-static mirror (local-CN only, should not live in the upstream flake). - flake.lock: bump rust-overlay. These do not affect the `cargo xtask starry test qemu` CI gate. * fix(starry-kernel): use TtyRead::closed in PTY read/poll for POLLHUP/EOF Completes the PTY peer-close-notification port: TtyRead::closed() and the PtyReader writer-closed flag were defined but never read, which clippy rejected as dead code ("field `1` is never read", "method `closed` is never used"). Wire closed() into the Passive-mode read path: - SimpleReader::closed() delegates to the inner reader. - poll_read() reports readable when the peer writer is closed, so poll() wakes (POLLHUP) instead of blocking forever after the last writer fd drops. - read() returns Ok(0) (EOF) when the buffer is drained and the peer writer is closed, instead of WouldBlock. Verified locally (CI container): cargo xtask clippy --since 93ebe11 -> 107 passed, 0 failed cargo xtask starry test qemu --arch x86_64 -c qemu-smp1/system -> PASS (incl test-pty-master-close) cargo xtask starry app qemu -t nix --arch x86_64 -> NIX_NOSANDBOX_COMPLETE * fix(starry-kernel): call DeviceOps::open on PTY open for peer-close notify The PTY peer-close-notification (Tty open_count + writer.close() on last-fd close) was dead code: add_to_fd only invoked device.open() for block devices with O_EXCL, so Tty::open (open_count++) never ran for PTY character devices. open_count stayed 0, so Tty::close (fired from File::drop) did checked_sub(0) -> None and never called writer.close(); the peer reader thus never saw POLLHUP after the writer's last fd closed. Restore the device.open() hook at the end of add_to_fd (after /dev/ptmx and /dev/tty rewrites) for non-block devices. This pairs each open with the DeviceOps::close from File::drop, so Tty open_count tracks correctly and the slave/master writer.close() fires on the last fd close on every architecture. Block devices keep using the O_EXCL-only hook above (no double open). This is the call the branch had before upstream's tty refactor removed it. test-pty-master-close only runs on loongarch64 (skipped on x86_64), which is why x86_64 CI stayed green while loongarch64 failed with "master poll returns after slave close" timing out. Verified locally (CI container): cargo xtask starry test qemu --arch loongarch64 -c qemu-smp1/system -> PASS (test-pty-master-close now PASS), 1/1 case(s) passed cargo xtask starry test qemu --arch x86_64 -c qemu-smp1/system -> PASS (no regression) cargo xtask clippy --package starry-kernel -> clean * fix(rsext4): flush dirty cache before invalidation after journal commit invalidate_cache() previously dropped dirty cache entries without flushing, which silently discarded metadata modifications made via buffer_mut() that had not yet been routed through the journal. It also left self.active and self.clock dangling, so subsequent buffer()/buffer_mut() calls could observe a stale active slot. On CI's slower runners this manifested as ext4 metadata corruption accumulated across ~42 journal commits during test-ext4-inode-unique (2048 file creates), eventually triggering a wild write that corrupted QEMU's virtio-blk queue state ("wrong value for queue_enable ffffffff"), hanging the starry loongarch64 qemu job. Make invalidate_cache() flush dirty entries first and reset active/clock to 0. Update all four call sites in journal.rs to handle the new Ext4Result return: replay path returns Incomplete on error, umount path expects (fatal), and the write_block/write_blocks auto-commit paths log-and-continue. Verified locally in podman (ghcr.io/rcore-os/tgoskits-container:latest, QEMU 10.2.1) with the full CI command: tg-xtask starry test qemu --arch loongarch64 - smp1/system: 319 tests PASS (205s) including test-ext4-inode-unique - smp1/tty-console-input-burst: PASS - smp4/system: 16 tests PASS (24s) Also verified x86_64 nix app: NIX_NOSANDBOX_COMPLETE. Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai> * fix(starry-nix): drop nixpkgs prebuild injection from smoke test The nixpkgs tarball download/injection and the nix-nixpkgs.sh script belong to the separate nixpkgs branch (003-starryos-nixpkgs), not the 002-starryos-nix-smoke smoke test. test_nix.sh only runs the nosandbox phase, so prebuild.sh must not download nixpkgs.tar.gz or install /usr/bin/nix-nixpkgs. Removed: - apps/starry/nix/nix-nixpkgs.sh (003-specific stdenv.mkDerivation script) - prepare_nixpkgs_tarball() and its call from prebuild.sh - /usr/bin/nix-nixpkgs install line from prebuild.sh - commented-out nix-nixpkgs block from test_nix.sh - NIX_NIXPKGS_TEST_FAILED fail_regex from both qemu-*.toml - nix-nixpkgs row and section from README.md Addresses ZR233 review on PR #1125: nixpkgs download/injection and script installation should be deferred until the phase is actually enabled. * ci: retrigger CI (previous run hit transient rootfs 404) * fix(ax-posix-api): drop use-hermit-types feature and Hermit getdents64 branch 删除从 003 分支带入的 use-hermit-types feature 及其对应的 Hermit 风格 sys_getdents64 实现。该 feature 在当前工作区无任何启用路径，仅留下死代码和 cfg 分支噪声。具体改动： - Cargo.toml: 移除 use-hermit-types = [] feature 定义 - fs.rs: 移除所有 #[cfg(not(feature = "use-hermit-types"))] 门控，使 Linux 风格 sys_getdents64 成为无条件实现；删除 #[cfg(feature = "use-hermit-types")] 的 Hermit 分支及其 HermitDirBuffer 相关注释块验证： - cargo check -p ax-posix-api --features fs 通过 - cargo check --features 'fs use-hermit-types' 正确报未知 feature - cargo fmt + cargo clippy -p ax-posix-api 无警告 - podman CI 容器内运行 starry qemu x86_64 测试： - bugfix-bug-ext4-dir-ops: 151 pass / 0 fail（含多轮 getdents 调用、 read-delete loop、rm-rf pattern 等直接验证 sys_getdents64 的场景） - bugfix-bug-dir-cookie-unlink-rmdir: PASS（batched getdents64 unlink rmdir、cookie 唯一性验证） * refactor(starry): remove wchan diagnostics from nix smoke * fix(starry-nix): dynamic qemu-user-static binary and rsext4 page-cache key cleanup 1. prebuild.sh: 根据 $STARRY_ARCH 动态选择 qemu-user-static 二进制，修复 aarch64 架构下硬编码 qemu-x86_64-static 导致 prebuild 失败的问题。 x86_64 / aarch64 均在 CI 容器中端到端验证通过 (NIX_NOSANDBOX_COMPLETE)。 2. inode.rs unlink(): 用 Cell<Option<InodeNumber>> 替换硬编码 None，在 deferred_zero_link 赋值后通过 forget_file_ino.set() 捕获 inode 号，确保 inode 释放时同步调用 forget_cached_file_key 清理 page-cache key。 3. 新增回归测试 test-ext4-unlink-pagecache：预分配 512 个临时文件 → 创建目标文件写旧 payload → 保存 st_ino → 打开fd → unlink → 关闭fd → 释放预分配文件 → 循环创建新文件直到同一 inode 号被复用（ext4 实测 attempt 1 即复用）→ 写入不同 payload → 读取验证不是旧数据。若未触发复用则 FAIL 而非静默 PASS。 4. README 补 aarch64 验证章节，含 podman 命令和输出。 * chore(ci): trigger CI re-run --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai> Co-authored-by: fzg-23 <2216703926@qq.com> Co-authored-by: Antareske <sckel@qq.com> Co-authored-by: 周睿 <34859362+ZR233@users.noreply.github.com> Co-authored-by: Qingya Zhang <1351073251@qq.com>

dweeqhd force-pushed the issue-595-vm-interrupt-fabric branch from 61b7588 to 3e41995 Compare June 16, 2026 08:09

mai-team-app Bot mentioned this pull request Jun 16, 2026

fix(wifi): D80 (V3) EAPOL TX + SDIO bus timeout recovery for AIC8800 on SG2002 #1276

Merged

dweeqhd marked this pull request as ready for review June 16, 2026 11:34

mai-team-app Bot approved these changes Jun 16, 2026

View reviewed changes

ZR233 requested changes Jun 17, 2026

View reviewed changes

dweeqhd force-pushed the issue-595-vm-interrupt-fabric branch 2 times, most recently from 290817c to 075b853 Compare June 17, 2026 07:23

mai-team-app Bot approved these changes Jun 17, 2026

View reviewed changes

ZR233 approved these changes Jun 17, 2026

View reviewed changes

This was referenced Jun 17, 2026

refactor(serial): align IRQ model with dev #1265

Merged

feat(axtask): add work-stealing load balancing for SMP #1016

Open

refactor(axvm): add VM interrupt fabric

b3711ee

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

ZR233 force-pushed the issue-595-vm-interrupt-fabric branch from 075b853 to b3711ee Compare June 18, 2026 01:46

ZR233 approved these changes Jun 18, 2026

View reviewed changes

ZR233 merged commit e3d3df3 into rcore-os:dev Jun 18, 2026
50 checks passed

github-actions Bot mentioned this pull request Jun 18, 2026

chore: release #1263

Merged

silicalet pushed a commit to silicalet/tgoskits that referenced this pull request Jun 18, 2026

refactor(axvm): add VM interrupt fabric (rcore-os#1273)

3c72396

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

dweeqhd mentioned this pull request Jun 18, 2026

refactor(axvm): route RISC-V IRQs through vPLIC backend #1317

Merged

SongShiQ pushed a commit to SongShiQ/tgoskits that referenced this pull request Jun 19, 2026

refactor(axvm): add VM interrupt fabric (rcore-os#1273)

09c6bb6

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

This was referenced Jun 22, 2026

chore: release #1342

Merged

chore: release #1344

Merged

Antareske pushed a commit to Antareske/tgoskits that referenced this pull request Jun 27, 2026

refactor(axvm): add VM interrupt fabric (rcore-os#1273)

bf32277

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

Antareske pushed a commit to Antareske/tgoskits that referenced this pull request Jun 27, 2026

refactor(axvm): add VM interrupt fabric (rcore-os#1273)

4aa75d2

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

luodeb pushed a commit that referenced this pull request Jun 30, 2026

refactor(axvm): add VM interrupt fabric (#1273)

1772235

Rebase onto current dev and keep Cargo.lock limited to the new virtualization-tests dependencies.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

refactor(axvm): add per-VM interrupt fabric#1273

refactor(axvm): add per-VM interrupt fabric#1273
ZR233 merged 1 commit into
rcore-os:devfrom
dweeqhd:issue-595-vm-interrupt-fabric

dweeqhd commented Jun 16, 2026 •

edited

Loading

Uh oh!

mai-team-app Bot left a comment

Uh oh!

ZR233 left a comment

Uh oh!

mai-team-app Bot left a comment

Uh oh!

ZR233 left a comment

Uh oh!

ZR233 left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

dweeqhd commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

背景

范围

评审边界

验证

Uh oh!

mai-team-app Bot left a comment

Choose a reason for hiding this comment

审阅总结

Uh oh!

ZR233 left a comment

Choose a reason for hiding this comment

Uh oh!

mai-team-app Bot left a comment

Choose a reason for hiding this comment

审阅总结

Uh oh!

ZR233 left a comment

Choose a reason for hiding this comment

Uh oh!

ZR233 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dweeqhd commented Jun 16, 2026 •

edited

Loading