New interrupt code slows hyperthreading down

[John Baldwin]

On 07-Nov-2003 Jens Rehsack wrote:
> Hi,
> 
> I recompiled my system today and when it came up again,
> it was terrible slow. Using top I've seen, that there're
> around 25% cpu-time is used to handle interrupts.
> The kernel was configured using SMP ('cause it's a HTT
> enabled CPU) and APIC. Setting machdep.hlt_logical_cpus
> to 1 didn't change anything. Simply getting a few mails
> (around 30) takes about 20 minutes. Most of time while
> getting the mails my mozilla was in "*Giant" state,
> what shouldn't be good, should it?
> 
> I've recompiled the kernel without SMP and APIC and
> now the system's behaviour is more "normal". Is the
> behaviour of the new interrupt code better on real
> multiprocessor systems?

Can you do a 'vmstat -i' under your SMP kernel to see where
all the interrupts are coming from?  It sounds like a device
is interrupt storming.  There has been report of this so far
with fwohci(4).

-- 

John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve!"  -  http://www.FreeBSD.org/