[Qemu-devel] QEMU temporarily hang

[Uri Lublin]

Hello,

I would like to share with you a QEMU problem I've encountered,
    and hope to get pointers where to look for it.
I am running CVS QEMU, x86_64-softmmu, with vmdk image and redirected monitor
    (unix socket).

If the guest is busy and I rapidly send a LOT of commands to QEMU monitor,
    then eventually QEMU temporarily hangs (vnc and monitor are not responsive)
At that state:
 - QEMU process's cpu usage is around 100%
 - strace shows ONLY clock_gettime(CLOCK_MONOTONIC, {time changing}) = 0
 - pstack shows:
      #0  0x0000003603e0459b in clock_gettime () from /lib64/librt.so.1
      #1  0x0000000000405c9d in get_clock ()
      #2  0x0000000000405d65 in cpu_get_clock ()
      #3  0x000000000044f660 in pit_get_count ()
      #4  0x000000000044f805 in pit_latch_count ()
      #5  0x0000000002450ec8 in code_gen_buffer ()
      #6  0xbe26f6118053c79f in ?? ()
      #7  0x000000000047df6e in cpu_x86_exec ()
      #8  0x000000000040c11a in main_loop ()
      #9  0x000000000040da69 in main ()


After a while (few or more minutes), it comes back to life.
I can shorten my waiting time by sending a SIGUSR2 signal to QEMU.
Why doesn't host_alarm_handler() get called/bring qemu back to life ?

Any comment would be appreciated,
Thanks,
    Uri.