device pcm kills SMP performance -CURRENT

[Robert Watson]

On Wed, 12 Nov 2003, matthew tillman wrote:

> I have an ACorp 6A815EPD motherboard, Intel i815 chipset, with a pair of
> P3 733s.  My sound card, is a SoundBlaster Live Value. 
>  
> Whenever sound support is compiled in the Kernel, performance takes a
> huge hit, X runs about the same as on a P2 300 or so. But If I boot to a
> kernel that is configured exactly the same, without device pcm, the
> system is very responsive. 
>  
> Booting to a non-SMP kernel in 5.1-Release with or without sound
> support, the system was very fast.  I did not try no sound support with
> SMP, as I thought the issue was only SMP performance in general. 
> 
> I then CVSuped to -CURRENT on Nov 8th, because at the time I didn't
> realise the issue was isolated to the sound system.  However, I still
> have the same issue. 
>  
> I also tried swapping the card into a diffrent PCI slot, but that didn't
> help. 

Just to confirm -- when you upgraded to -CURRENT, did you remove
INVARIANTS and WITNESS from your kernel configuration?  They both add
substantial overhead to locking operations, and while invaluable to
debugging, they can have quite a detrimental affect on performance (see
also UPDATING: Why Is Current Slow?).  5.1-RELEASE has these features
disabled by default, but 5.1-CURRENT has them enabled since there's active
locking work going on.  There's also a set of patches being floated to
improve pcm locking in a variety of ways, but I'm not sure what the status
of those patches is. 

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Network Associates Laboratories