Re: Deadlocks with recent SMP current

Sławek Żak wrote:
> Martin Blapp <mb_at_imp.ch> writes:
> 
> 
>>Hi,
>>
>>Since yesterday I'm getting complete deadlocks. This time unrelated
>>the servers are nor loaded at all, the just freeze after a while.
>>No break into DDB possible at all.
> 
> 
>     Despite the numerous reports of current problems I'm running 5.2-CURRENT for
>     12 days now without a glitch. Tested overnight with 2 buildworlds running
>     with -j8. It's a 2 processor Xeon server with 4BSD scheduler. Just my .02
>     euro.
> 
>     Another thing is bothering me. In top I see:
> 
> CPU states: 16.2% user,  0.0% nice,  2.5% system,  1.0% interrupt, 80.3% idle
> Mem: 254M Active, 605M Inact, 254M Wired, 16K Cache, 112M Buf, 2622M Free
> Swap: 4096M Total, 4096M Free
> 
>   PID USERNAME PRI NICE   SIZE    RES STATE  C   TIME   WCPU    CPU COMMAND
> 37033 mysql     20    0   265M   178M kserel 0 610:59 124.37% 124.37% mysqld
> 
> 
>     MySQLd is running with libpthread and can go up to 400% without
>     significantly afecting the idle percentage shown by top. I don't see how I
>     could fit 400% into 20% on those two suckers, therefore I ask :)
> 
> /S    

Cpu percentage use is broken for pthread applications..
it's almost impossible to work out the right way of doing it
due to the nature of M:N threading where there is no fixed rule as
to which user thread is run by which kernel thread.

I have ideas but it will probably have to be an aggreagate cpu% for the thread 
group, divided by the number of kernel threads in the group.