| Summary: | Build race in perf plugin_scsi.o | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Product: | [Build System, Metadata & Runtime] OE-Core | Reporter: | Ross Burton <ross.burton> | ||||||
| Component: | kernel | Assignee: | Bruce Ashfield <bruce.ashfield> | ||||||
| Status: | RESOLVED FIXED | QA Contact: | |||||||
| Severity: | normal | ||||||||
| Priority: | Medium+ | CC: | bruce.ashfield, tom.zanussi | ||||||
| Version: | 2.1 | ||||||||
| Target Milestone: | 2.1 M4 | ||||||||
| Hardware: | All | ||||||||
| OS: | Multiple | ||||||||
| Whiteboard: | |||||||||
| OS type for building Yocto: | --- | Type of Regression: | Regression (Used to work) | ||||||
| Verified: | Documentation change: | No (bug/feature does not impact docs) | |||||||
| Attachments: |
|
||||||||
|
Description
Ross Burton
2016-02-29 15:58:12 UTC
More races in today's AB run (http://errors.yoctoproject.org/Errors/Details/40623/) install: cannot stat '/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-arm64/build/build/tmp/work/qemuarm64-poky-linux/perf/1.0-r9/perf-1.0/plugin_kvm.so': No such file or directory install: cannot stat '/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-arm64/build/build/tmp/work/qemuarm64-poky-linux/perf/1.0-r9/perf-1.0/plugin_scsi.so': No such file or directory install: cannot stat '/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-arm64/build/build/tmp/work/qemuarm64-poky-linux/perf/1.0-r9/perf-1.0/plugin_cfg80211.so': No such file or directory Makefile:264: recipe for target 'install_plugins' failed As before there are multiple LINK lines in the build running in parallel. Two more fails last night on the AB... I've had no luck reproducing this, and am no closer to solving it. Looking at another sstate issue though, and perhaps they are related. I can't replicate on demand but it looks like it may be from the calling of fresh makes from the perf directory to ../traceevents/. If two sub-makes are fired in parallel then they'll be unaware and race over building those sources. (In reply to comment #4) > I can't replicate on demand but it looks like it may be from the calling of > fresh makes from the perf directory to ../traceevents/. If two sub-makes > are fired in parallel then they'll be unaware and race over building those > sources. I've also been looking at it as a parallel build issue. The latest kernel has updates in this area, and I'll scan them to see if there's anything I can backport. I'd rather not go -j1 for this. I'm running perf builds with -j24 in a tight loop, and haven't had this break. So it is time to change tactics. The key may be question posed earlier in the bug. Why is compilation happening during the install rule. In my local tests, that isn't the case. So there's something that is trigger those builds .. and leading to our race. Do the builders use rm_work ? Something else ? Can we get access to the do_compile and do_log of the broken and some working builds ? Mine have zero compilation in the install phase .. again, we need to figure that out as a first step. Created attachment 3093 [details]
do_compile
Created attachment 3094 [details]
do_install
Following from the previous do_compile, this is a do_install which shows compilation happening.
rm_work isn't enabled. The build succeeded without problems.
It seems that I can stop it building in do_install by passing DESTDIR to make all. This might work around the race? (patch sent) Fixed in oe-core 76c473dbe9e6a1eb8bca89f26cf29b41ca18d680. |