Mostly sleeping app consuming significant CPU

Hi.

I have a delivered application (32 bit LWL 5.1.2, delivered at level
0), which top(1) claims is steadily using upwards of 25% cpu.  I
profiled the application over a 5 second period[1], and in 94% of the
samples SYSTEM::WAIT-FOR-STACK-WITH-TIMEOUT was at the top of the
stack.  The call tree shows that 75% of the times it was sleeping in
my application loop[2], 19% of the times it was sleeping in my
profiling statement.

So how can something that spends over 90% of its time asleep consume
that much CPU?

This application uses bindings to a C library to read incoming data.
Are FLI calls somehow not counted in the profiler?  The cumulative
statistics show that FLI:BASIC-DEREFERENCE and (SETF
FLI:BASIC-DEREFERENCE) are the clear winners in the "called" column,
with profiling-related functions a distant second.  None of these
functions ever appear on the stack, apparently.

The function that actually reads data is only found on the stack
once.

Has anyone seen similar things?  Any clues as to how to find out what
it's doing with its time?


Footnotes: 
[1]  (hcl:set-up-profiler :package (list-all-packages) :style :tree)
     followed by (hcl:profile (sleep 5))

[2]  The main loop of the application is basically
     (loop
       (read-and-process-all-available-data)
       (sleep 1))

-- 
Mvh/Regards
Peder O. Klingenberg
Netfonds Bank ASA