"make: write error: stdout" on parallel builds

[eocanha]

- Release of Ubuntu: 18.04.1 LTS (Kubuntu).
- Package version: linux-image-4.15.0-44-generic, linux-image-4.15.0-45-generic.
- Expected outcome: Large parallel make build completing properly.
- What happened instead: Build stops with "make: write error: stdout".

Full explanation:

Since the update to kernel 4.15.0-44 (linux-image-4.15.0-44-generic) in Kubuntu Bionic I'm experiencing issues in my ThinkPad P52 laptop during large parallel builds of source code (eg: a buildroot full rebuild for a work project). Note that this reports isn't about kernel builds, but about large project builds in general.

I don't think it's relevant, but I'm running those builds in an schroot with the same Ubuntu Bionic distro I'm also using as main host distro.

The specific error I get is this (in Spanish):

  make[4]: error al escribir: stdout

I call Buildroot's make like this from KDE Konsole with "infinite history buffer" enabled (uses a tmp file to store the logs):

  V=1 make ...some stuff... 2>&1

However, the error is reproducible in Konsole even if I call make without options and also if I use "limited (in-memory) history buffer" as well as "no history buffer". Buildroot autoselects a parallelization level (-j) automatically. In my case, I'm using a Xeon E-2176 CPU with 12 threads.

Searching on the Internet I found a thread[1] where people from the GNU Make project say that the error may be coming from the C runtime library (and I suspect originating from the kernel).

I can NOT reproduce the issue on linux-image-4.15.0-43-generic (selected from grub at boot), but I experience it with linux-image-4.15.0-44-generic and linux-image-4.15.0-45-generic. That's why I'm reporting this bug against the linux package.

Interestingly enough, I can NOT reproduce the issue on linux-image-4.15.0-45-generic (haven't tried previous versions) when building from xterm instead of Konsole. Unfortunately, this laptop has a 4K screen and xterm isn't an option for me.

If I build redirecting stdout and stderr to different files to avoid terminal output and then watching the stdout file with tail -f, the tail command eventually finishes although the build continues and I can keep watching it by running tail again. An infinite build loop never finishes with this redirection (error not reproducible). Same results with the 2>&1 redirection to a single file.

All this looks a bit like a poltergeist, but the only thing I'm sure about is that it all started with the upgrade from 4.15.0-43 to -44 and it continues on -45.

Thank you.

[1] http://gnu-make.2324884.n4.nabble.com/Spurious-quot-write-error-quot-on-parallel-builds-td15695.html

[Sebastian Urban (surban--)]

I am seeing the same problem on a fully updated Ubuntu system running Linux 4.15.0-45.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Mark Owen (mowen01)]

I'm running 16.04 and seeing the same problem on Linux 4.15.0-45. I can also confirm that when booting with 4.15.0-43 I no longer see the problem.

[Patrick Doyle (wpdster)]

I have this problem as well on 16.04LTS with kernel 4.15.0-45-generic #48.

[Leo Soares (leojrfs)]

affects me

Ubuntu 18.04.2 LTS: 4.15.0-45-generic

[R. Diez (rdiezmail-ubuntu)]

I have recently started experiencing the same problem.

Ubuntu 16.04.5 LTS

Kernel 4.15.0-45-generic #48~16.04.1-Ubuntu SMP Tue Jan 29 18:03:48 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

[R. Diez (rdiezmail-ubuntu)]

This problem happens with both the standard GNU Make 4.1 that comes with Ubuntu 16.04.5, and with a self-compiled GNU Make 4.2.1.

The error only happens if make is writing to stdout. If you redirect to a file like this:

make ... blah... blah... 2>&1 >"$HOME/Build.log"

Then it runs fine.

[R. Diez (rdiezmail-ubuntu)]

It is not GNU Make itself. If I pipe GNU Make's output to tee, then it is tee that fails:

tee: 'standard output': Resource temporarily unavailable

[Leo Soares (leojrfs)]

I can confirm that this does not happen via ssh.

[Scott Kazimour (scottkaz)]

This happens for me with 18.04, kernel 4.15.0-45-generic.

It does not happen with 4.15.0-43-generic.

Running on xterm (as opposed to terminator) it still happens, but much later in the build.

[Per Gregers Bilse (bilse)]

I can confirm this. I noticed after the upgrade from 4.15.0-43-generic, but at first wasn't sure if I had messed up my build; I downgraded to 43 and forgot about it until the system upgraded again yesterday. I'm now on 45, and my build fails as before, an excerpt:

/bin/sh: line 0: echo: write error: Resource temporarily unavailable
mv ../../bcmdrivers/broadcom/char/pktflow/bcm963138/fcache.o_tmp ../../bcmdrivers/broadcom/char/pktflow/bcm963138/fcache.o
/bin/sh: line 0: echo: write error: Resource temporarily unavailable
scripts/Makefile.build:258: recipe for target '/home/bilse/bcm_50203/bcmdrivers/broadcom/char/adsl/bcm963138/BcmAdslCore.o' failed
make[4]: *** [/home/bilse/bcm_50203/bcmdrivers/broadcom/char/adsl/bcm963138/BcmAdslCore.o] Error 1
make[4]: *** Waiting for unfinished jobs....
[...]
make[4]: write error: stdout
scripts/Makefile.build:423: recipe for target '../../bcmdrivers/broadcom/char/pktflow/bcm963138' failed
make[3]: *** [../../bcmdrivers/broadcom/char/pktflow/bcm963138] Error 1
scripts/Makefile.build:423: recipe for target '../../bcmdrivers/broadcom/char/adsl/bcm963138' failed
make[3]: *** [../../bcmdrivers/broadcom/char/adsl/bcm963138] Error 2
make[3]: write error: stdout
Makefile:1201: recipe for target '../../bcmdrivers' failed
make[2]: *** [../../bcmdrivers] Error 1
make[2]: *** wait: No child processes. Stop.
make[2]: write error: stdout
Makefile:476: recipe for target 'inner_kernelbuild' failed
make[1]: *** [inner_kernelbuild] Error 1
make[1]: Leaving directory '/home/bilse/bcm_50203'
Makefile:473: recipe for target 'kernelbuild' failed
make: *** [kernelbuild] Error 2
bilse@ub40:~/bcm_50203$ uname -a
Linux ub40 4.15.0-45-generic #48~16.04.1-Ubuntu SMP Tue Jan 29 18:03:48 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
bilse@ub40:~/bcm_50203$

[Jburgess777 (jburgess777)]

This looks to me like it might be a duplicate of: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1813873

[gmaka (gmaka)]

This is happening to me, started in the last 10 days or so:
$ uname -r -v
4.15.0-45-generic #48~16.04.1-Ubuntu SMP Tue Jan 29 18:03:48 UTC 2019

[gmaka (gmaka)]

I can confirm that moving back to 4.15.0-43-generic avoids this issue.

[Éric Hoffman (ehoffman-videotron)]

I have yet to see any issue (i.e. it's working fine) with 4.15.0.46 (I used xenial-proposed to upgrade).

As the rest of you, I had constant issue with this in 4.15.0-45

So yes, maybe it is related to the tty bug. This does seem to be consistent with the behavior I was seeing with current kernel. For example, I had many 'resources busy' and other output-related messages on make output.

And, doing:

  make -j8 2>&1 | tee outfile.txt

I often saw the console just stop outpointing any text (although the compilation continued, and completed, in the background).

So, Jburgess777 you may be correct about the duplicate...

Regards,
Eric

[_zibuyu (zibuyujun)]

Sorry, operation error. The problem is not solved.