Repository navigation
Collect the kernel crashes cameras recover from, rank them, and pay Club stars for them - #411
Merged
Merged
Conversation
…lub stars for them A camera on OpenIPC firmware keeps the kernel's log across a crash (pstore), and its WebUI offers the owner a crash log -- which nobody upstream ever sees. The lab's gk7205v300 + imx335 had one since 2026-09-28: a NULL dereference in RGN_PutRegion called from the VENC interrupt, which reads as an ISP bug because hi_isp_run was the task it interrupted. - POST /api/v1/crashes takes the WebUI's crashlog_*.tar.gz (service/internal/ crashes, CRASH.md is the contract); a member can send the one they downloaded from /club. The records are parsed -- ARM, MIPS and arm64 backtraces, oops, BUG, WARNING, panic, a failsafe boot loop -- and filed under a signature, the top five frames without offsets, so one bug from many cameras is one row. pstore's Oops#1 and Panic#2 of one crash are one crash, and the same records sent again are one event. - Signatures are ranked by how bad a crash is and how many cameras it hits, more when today's firmware still has it. /crashes lists them publicly: signatures, chips and status, never a log, a camera or a member. - Maintainers triage at /club/crashes or with `openipc crashes status`, and see what reproducing one takes: the chips, sensors and builds it was seen on, the firmware builds matching its kernel, the task and what it interrupted, the warnings and i2c/MMZ trouble before it, and each log with MACs and IP addresses replaced by keyed hashes. The bundle as sent is kept in PostgreSQL and never served; events and bundles are append-only. - Stars (crash_stars, migration 030) are paid by the nightly settlement only: 1 per camera and bug, 3 more to the first reporter once confirmed, 5 when fixed, at most 10 a month. A crash asked for with sysrq pays nothing, and marking a signature bogus takes back what it paid. A camera's crashes are its owner's through the Open Wall's link. The leaderboard and the member's total count the third ledger. - nginx proxies /api/v1/crashes on both vhosts.
PR Summary by QodoCollect and rank camera kernel crashes with Club star rewards
AI Description
Diagram
High-Level Assessment
Files changed (40)
|
Code Review by Qodo
1.
|
![qodo-free-for-open-source-projects[bot]](https://avatars.githubusercontent.com/in/1068000?s=60&v=4){kind=small}
…ner, keep the ledger net through merges - Titles, trace reasons, the command line and the lines before a crash are redacted before they are stored, so a panic message carrying an address reaches neither the public list nor a maintainer's detail. Redaction finds IPv6 addresses and dotted MACs, and looks at an IPv4 address's neighbours instead of matching them, so ip=10.0.0.2:10.0.0.1 loses both. - A SoC or sensor that is not a chip's name is refused (400), never stored. - The same records are one crash per camera: two cameras' identical failsafe notes are two. - MIPS's own "Oops[#1]:" header opens a trace, as do "CPU 0 Unable to handle..." lines. - A crash with a MAC pays the camera's linked owner, whoever sent it, and a member sending from /club may name only a camera of theirs. Cameras are counted by MAC, or by sender address when there is none. - The ledger is net per bug through merges: merging pays nothing again, bogus takes back what the bug and everything merged into it hold, undoing it pays again, and the monthly cap counts net stars. A signature merged into another is not marked bogus alone. - A triage keeps the fields it does not send; the CLI changes only the flags given, and the /club form starts from the merge target. - The daily limits are counted inside the insert, under the client's and the camera's locks.
Member
Author
|
Qodo review addressed in ecdb979. All 13 findings were real; each is fixed and has a test.
|
Thanks for the detailed update. The fixes in |
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
2 participants
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.
A camera on OpenIPC firmware keeps the kernel's log across a crash (pstore,
S98crashlog), and its WebUI offers the owner a crash log. Nobody upstream ever sees it. For example, the lab's gk7205v300 + imx335 has had one since 2026-09-28. It is a NULL dereference inRGN_PutRegion [open_rgn]called from the VENC interrupt (ChnlVeduISR), which reads as an ISP bug becausehi_isp_runwas the task it interrupted. That bundle is now the parser's test fixture.What this adds
POST /api/v1/crashesaccepts the WebUI'scrashlog_*.tar.gz. The contract isservice/internal/crashes/CRASH.md. A signed-in member can also send the one they downloaded, from the new My crash reports section on/club.fn [module], without offsets. One bug from many cameras is one row.Oops#1andPanic#2of one crash count as one crash, and the same records sent again are one event./crashesis public and lists signatures, chips, sensors and status. It never shows a log, a camera or a member./club/crashesis for maintainers, and so isopenipc crashes list|show|status|settle|takedown.crash_bundles, at most 256 KB each) and never served. Events and bundles are append-only, guarded like owner reports; onlytakedownremoves one.crash_stars, a third ledger, migration 030.settle-wall.bogustakes back everything a signature paid.^~ /api/v1/crasheson both vhosts (512k bodies).Not in this PR
S98crashlogadditions (OpenIPC/firmware): ameta.jsonwith versions and a strippedmajestic.yaml, printk timestamps, and keeping the last 3 crashes.Tests
service/run.sh test, includingdeploytest, passes./meand the leaderboard.deploy/nginx/check-config.shpasses.frontend/: lint, typecheck andnpm testpass (707 tests, including the /club crash section). The build produces/crashes,/ru/crashesand/club/crashes.