qemu hangs in pselect syscall

[hyokeun]

I'm using git commit d75aa4372f0414c9960534026a562b0302fcff29 (v2.7.0-rc4) configured with;
    --enable-linux-user \
    --disable-system \
    --disable-tools \
    --disable-guest-agent \
    --static --disable-linux-aio \
    --disable-fdt \
    --without-pixman \
    --disable-blobs \
Stable version (v2.6.0) also have the same problem.

In a chroot environment I ran below command-line to compile some things, different sources each time.
    /usr/bin/qemu-arm -0 /usr/bin/edje_cc /usr/bin/edje_cc -id /home/abuild/rpmbuild/BUILD/org.tizen.browser-1.6.2/services/SimpleUI/images_mob/ -DBROWSER_RESOLUTION_720x1280=1 -DPROFILE_MOBILE=1 /home/abuild/rpmbuild/BUILD/org.tizen.browser-1.6.2/services/SimpleUI/edc/TextPopup_mob.edc /home/abuild/rpmbuild/BUILD/org.tizen.browser-1.6.2/build-tizen/services/SimpleUI/720x1280_TextPopup.edj

Here is back trace with gdb;
#0 safe_syscall_end () at /usr/src/debug/qemu-2.6.94/linux-user/host/i386/safe-syscall.inc.S:78
#1 0x60049370 in safe_pselect6 (nfds=10, readfds=0xffa31b5c, writefds=0xffa31bdc, exceptfds=0xffa31c5c, timeout=0x0, sig=0x0)
    at /usr/src/debug/qemu-2.6.94/linux-user/syscall.c:855
#2 0x6004b2fe in do_select (n=10, rfd_addr=1082122232, wfd_addr=1082122360, efd_addr=1082122488, target_tv_addr=0)
    at /usr/src/debug/qemu-2.6.94/linux-user/syscall.c:1386
#3 0x6005e5ba in do_syscall (cpu_env=0x640d0454, num=142, arg1=10, arg2=1082122232, arg3=1082122360, arg4=1082122488, arg5=0, arg6=1087473216, arg7=0,
    arg8=0) at /usr/src/debug/qemu-2.6.94/linux-user/syscall.c:9690
#4 0x60045def in cpu_loop (env=0x640d0454) at /usr/src/debug/qemu-2.6.94/linux-user/main.c:876
#5 0x60047640 in main (argc=10, argv=0xffa33c84, envp=0xffa33cb0) at /usr/src/debug/qemu-2.6.94/linux-user/main.c:4817

Attached core file taken from gdb. To see the stack frame, you could try;
$ tar -xf reproduced_118_04.tar.bz2; gdb --core core.1823 qemu-arm

And recent strace log for PID 1823(stucked one);
79965 [ 313s] 1823 :0x8e _newselect(10,[9,3,],[],[],NULL)
79966 [ 313s] ==>[pselect6(0xa)=]
79967 [ 313s] [pselect6=0x1]<==
79968 [ 313s] 1823 :0x8e _newselect(10,[9,],[],[],NULL)
79969 [ 313s] 1823 :0x8e => = 0x00000001 ([9,],[],[],NULL)
79970 [ 313s] 1823 :0xfc epoll_wait(3,1082121456,32,0,1082121456,3)
79971 [ 313s] 1823 :0xfc epoll_wait(3,1082121456,32,0,1082121456,3)
79972 [ 313s] 1823 :0xfc => = 0
79973 [ 313s] 1823 :0x3 read(9,0x407fdeec,16)
79974 [ 313s] 1823 :0x3 read(9,0x407fdeec,16)
79975 [ 313s] 1823 :0x3 => = 8
79976 [ 313s] 1823 :0x107 clock_gettime(1,1082122120,0,1082829144,1082827588,0)
79977 [ 313s] 1823 :0x107 clock_gettime(1,1082122120,0,1082829144,1082827588,0)
79978 [ 313s] 1823 :0x107 => = 0
79979 [ 313s] 1823 :0x8e _newselect(10,[9,3,],[],[],NULL)
79980 [ 313s] ==>[pselect6(0xa)=]

I'm using 64-bit Ubuntu with kernel release Linux 3.19.0-25-generic #26~14.04.1-Ubuntu.
Reproducibility is low. One occurrence out of 50+ trials.

[hyokeun (hyokeun-jeon)]

FYI, adding a build log with strace enabled.

[Peter Maydell (pmaydell)]

Can you provide sufficient instructions for me to reproduce this on my machine, please?

[hyokeun (hyokeun-jeon)]

Second part of scratch.armv7l.0.tar.gz.

[hyokeun (hyokeun-jeon)]

Third part of scratch.armv7l.0.tar.gz.

[hyokeun (hyokeun-jeon)]

Dear Peter.

Thank you for the update.
Please find the attached full chroot environment that I used(scratch.armv7l.0.tar.gz, split three parts).
You could try make build with below steps;

$ sudo su
$ echo -1 > /proc/sys/fs/binfmt_misc/arm
$ echo ':arm:M::\x7fELF\x01\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x28\x00:\xff\xff\xff\xff\xff\xff\xff\x00\xff\xff\xff\xff\xff\xff\xff\xff\xfa\xff\xff\xff:/usr/bin/qemu-arm-static:' > /proc/sys/fs/binfmt_misc/register

$ cat scratch.armv7l.0.tar.gz.a* > scratch.armv7l.0.tar.gz; sudo tar -zxf scratch.armv7l.0.tar.gz
$ cd scratch.armv7l.0
$ chroot .
  chroot> cd /home/abuild/rpmbuild/BUILD/org.tizen.browser-1.6.2/build-tizen/services/
  chroot> while :; do make clean; make -j32; done

Reproducibility is ver low but it surely happened.

Thanks.

[Peter Maydell (pmaydell)]

I can't reproduce this with current git master, and I know we fixed a lot of race conditions in linux-user. So I'm going to close this bug -- if it's still a problem for you with new QEMU, please reopen, preferably with a repro case that's more frequent than 1-in-50-or-less.