nfe driver 6.2 stable

[Pyun YongHyeon]

On Sat, Sep 29, 2007 at 11:16:18AM -0700, Jos Backus wrote:
 > On Sat, Sep 29, 2007 at 02:02:40PM +0900, Pyun YongHyeon wrote:
 > > On Fri, Sep 28, 2007 at 07:21:47PM -0700, Jos Backus wrote:
 > >  > On Sat, Sep 29, 2007 at 10:28:01AM +0900, Pyun YongHyeon wrote:
 > >  > [snip]
 > >  > > ATM nfe(4)'s interrupt moderation mechanism doesn't seem to work
 > >  > > at all so nfe(4) generates too many interrupts. However I don't
 > >  > > think it wouldn't be major bottleneck of the performance.
 > >  > 
 > >  > Could this be why (on -current) I had to enable polling on nfe0 to get rid of
 > >  > the choppy audio playback I was experiencing?
 > >  > 
 > > 
 > > Probably not. Maybe nfe(4) use shared interrupt.
 > > Check the output of "vmstat -i".
 > 
 > pcm0 and nfe0 share irq23:
 > 
 >     lizzy:~% vmstat -i
 >     interrupt                          total       rate
 >     irq1: atkbd0                      107296          0
 >     irq6: fdc0                             1          0
 >     irq12: psm0                       563936          1
 >     irq22: atapci2                   5293281         11
 >     irq23: pcm0 nfe0                 1225731          2
 >     cpu0: timer                    947413237       2000
 >     Total                          954603482       2015
 >     lizzy:~% 
 > 
 > Both pcm0 and nfe0 are mobo devices. Is there any way to change the assigned
 > interrupts?
 > 

It seems that your NIC doesn't have MSI/MSIX capability. Also the
NIC is LOM version so I guess there is no easy way to change the
interrupt number. ATM polling(4) seems to be the only way to reliably
run nfe(4) under shared interrupt situations. Of course, you may
encounter additional latency from the polling(4) but it normally
wouldn't affect desktop usage patterns. I have a plan that will add
an additional knob that have nfe(4) work more reliably under shared
interrupt environments. However that would happen after branching 7.

-- 
Regards,
Pyun YongHyeon