fix(mm/oom): staged OOM killer fixes — victim marking, direct reclaim, observability#1995
Open
sparkzky wants to merge 4 commits into
Open
fix(mm/oom): staged OOM killer fixes — victim marking, direct reclaim, observability#1995sparkzky wants to merge 4 commits into
sparkzky wants to merge 4 commits into
Conversation
…, observability Phase 1 (bug fixes + observability): - Fix wait_until_recoverable generation guard dead code (Some(_) => true) - Add OOM kill counter (OOM_KILL_COUNT atomic) wired to /proc/vmstat oom_kill - Add /proc/[pid]/oom_score read-only file (normalized [0,1000]) - Fix hardcoded order:1 → order:0 in pagefault OOM context Phase 2 (victim forward-progress guarantee, prevents OOM livelock): - Add ProcessFlags::OOM_VICTIM (equivalent to Linux TIF_MEMDIE) - Mark all tasks sharing victim mm with OOM_VICTIM before SIGKILL delivery - Clear OOM_VICTIM in exit path via exit_oom_victim() - Skip already-marked victims in should_skip_candidate() Phase 3 (direct reclaim before kill): - Insert PageReclaimer::shrink_list(64) before pagefault_out_of_memory() (matching Linux try_to_free_pages before out_of_memory) - Use tried_direct_reclaim flag to bound to one reclaim attempt per fault Test: - Add test_proc_oom_score and test_vmstat_oom_kill_counter to user/apps/c_unitest/test_oom_killer.c Verified: nix develop -c make kernel + make fmt (cargo fmt + clippy) pass. Signed-off-by: sparkzky <sparkhhhhhhhhhh@outlook.com>
github-actions
Bot
added
Bug fix
Member
Author
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ff53205ed7
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+343
to
+346
| pub fn current_is_oom_victim() -> bool { | ||
| ProcessManager::current_pcb() | ||
| .flags() | ||
| .contains(ProcessFlags::OOM_VICTIM) |
There was a problem hiding this comment.
我检查了全仓，current_is_oom_victim() 只有这里定义而没有任何分配路径调用；因此在真实 OOM 下，SIGKILL 后的受害者并不会获得注释中说的内存储备，退出/拆 mm 仍可能在需要页表或内核对象分配时失败，而且 should_skip_candidate() 已经会跳过带该标志的进程，可能导致继续杀其他任务而不是让已选受害者前进。请把这个标志用于页帧/内核分配的 reserve 判断，或不要依赖它作为前进保障。
Useful? React with 👍 / 👎.
Comment on lines
+522
to
+529
| if !tried_direct_reclaim { | ||
| tried_direct_reclaim = true; | ||
| drop(space_guard); | ||
| crate::mm::page::PageReclaimer::shrink_list( | ||
| <crate::mm::allocator::page_frame::PageFrameCount>::new(64), | ||
| ); | ||
| flags |= FaultFlags::FAULT_FLAG_TRIED; | ||
| continue; |
There was a problem hiding this comment.
当 /proc/sys/vm/oom_fault_inject 配置为 fail_times=1（test_current_is_victim 和新增的 test_vmstat_oom_kill_counter 都这样用）时，第一次缺页会由 should_inject_fault_oom() 直接返回 VM_FAULT_OOM 并消耗唯一一次失败；这里先执行 direct reclaim 后 continue，第二次缺页不再 OOM，于是子进程会走到 _exit(3) 而不是 SIGKILL，oom_kill 也不会增加。需要让注入的 OOM 跳过这次 reclaim，或把测试注入次数调到经过 reclaim 后仍会触发 killer。
Useful? React with 👍 / 👎.
sparkzky
changed the title
P1: Wire OOM_VICTIM into page-frame allocator (oom.rs:343 / page_frame.rs:349) - allocate_page_frames now retries up to 1000 times when the current task is an OOM victim, preventing livelock when the victim is stuck in an uninterruptible allocation during exit. - Also wakes the page reclaim thread to accelerate memory availability. - This eliminates the previous dead_code warning on current_is_oom_victim. P2: Skip direct reclaim for fault-injection OOM (fault.rs:521 / oom.rs:349) - Add is_fault_inject_target() to check if current task matches the oom_fault_inject config. - When true, skip the shrink_list(64) direct-reclaim step so the injection is consumed by the killer path rather than a reclaim-then-retry cycle. - Without this, fail_times=1 injections were consumed by reclaim, causing test_current_is_victim and test_vmstat_oom_kill_counter to fail. Verified: make kernel + make fmt (cargo fmt + clippy) pass, zero warnings. Signed-off-by: sparkzky <sparkhhhhhhhhhh@outlook.com>
…GENTS.md - Add Nix develop/yolo command pointer to '常见命令' section - Add WHERE TO LOOK table with 3 non-intuitive paths: - Running DragonOS -> docs/introduction/develop_nix.md - OOM Killer -> kernel/src/mm/oom.rs - Post-refactor process management -> state.rs / exit.rs (mod.rs split in DragonOS-Community#1993) Signed-off-by: sparkzky <sparkhhhhhhhhhh@outlook.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Phase 1 (bug fixes + observability):
Phase 2 (victim forward-progress guarantee, prevents OOM livelock):
Phase 3 (direct reclaim before kill):
Test:
Verified: nix develop -c make kernel + make fmt (cargo fmt + clippy) pass.
ref #1976