Here's one NAF might enjoy.

I'm seeing a bug in the  futex()  system call.  ESAMON reports a guest
running heavy,  using one full IFL.  So we peek into Linux and see a
consumer,  one process leading the pack  (99% of the slice).  An  'strace
-p'  on that process shows ...

futex(0xffffffff54a6ec7c, FUTEX_WAIT, 1, ptrace: umoven: Input/output
error

I think the I/O error is an effect of the trace and not a real error,  but
I could be wrong.  In any case,  this is what  'strace -p'  reports and it
just sits there running flat out.  This would indicate one of several
profiles of bad behaviour for a virtualized Linux,  except that pausing
and resuming the rogue process breaks it out of the state of consuming
CPU.  If I  'kill -STOP'  that process,  wait a few seconds,  and then
either  'kill -CONT'  or  'strace -p'  again,  the run queue drops  (the
"load level" drops)  and ESAMON confirms the guest is off the radar.  Said
process is still there,  and still in  futex()  waiting,  but it's not
spinning.

Thoughts?

-- R;


----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to