Re: High interrupt load on VIA C3 machine

Bruce Cran wrote:
> I recently upgraded to 7-CURRENT on my VIA EPIA router, and have found 
> that there's a constant interrupt load of around 15%.   My dmesg 
> reports the CPU as:
>
> CPU: VIA C3 Nehemiah+RNG+AES (533.36-MHz 686-class CPU)
>  Origin = "CentaurHauls"  Id = 0x698  Stepping = 8
>  
> Features=0x381b93f<FPU,VME,DE,PSE,TSC,MSR,CX8,SEP,MTRR,PGE,CMOV,PAT,MMX,FXSR,SSE> 
>
> real memory  = 132055040 (125 MB)
> avail memory = 119701504 (114 MB)
>
> uname -a:
> FreeBSD router.draftnet 7.0-CURRENT FreeBSD 7.0-CURRENT #4: Fri Aug 31 
> 07:04:22 BST 2007     
> brucec_at_router.draftnet:/usr/obj/usr/src/sys/ROUTER  i386
>
> The first few lines of top -S are:
> last pid:  1394;  load averages:  0.08,  0.12,  
> 0.21                                                                                 
> up 0+01:59:10  16:13:59
> 57 processes:  3 running, 39 sleeping, 15 waiting
> CPU states:  3.1% user,  0.0% nice,  4.8% system, 17.0% interrupt, 
> 75.1% idle
> Mem: 7360K Active, 6116K Inact, 12M Wired, 48K Cache, 9504K Buf, 90M Free
> Swap: 2048M Total, 2048M Free
>
>  PID USERNAME  THR PRI NICE   SIZE    RES STATE    TIME   WCPU COMMAND
>   11 root        1 171 ki31     0K     8K RUN    100:01 81.98% idle
>   12 root        1 -32    -     0K     8K WAIT    12:31  7.96% swi4: 
> clock sio
> 1394 brucec      1  46    0  3512K  1744K RUN      0:01  4.98% top
>  586 root        1  44    0  5016K  2516K RUN      2:39  0.00% ppp
> 1044 root        1  44    0  3176K   980K select   0:51  0.00% powerd
>   28 root        1 -68    -     0K     8K WAIT     0:45  0.00% irq11: 
> vr0 dc3
>   30 root        1 -68    -     0K     8K WAIT     0:32  0.00% irq12: dc1
>
> While there's a high interrupt load, vmstat -i doesn't show anything 
> wrong:
>
> interrupt                          total       rate
> irq0: clk                         718003         99
> irq4: sio0                           897          0
> irq5: dc0                             31          0
> irq8: rtc                         919064        127
> irq10: dc2 uhci0+                     31          0
> irq11: vr0 dc3                     63548          8
> irq12: dc1                         62120          8
> irq14: ata0                         1555          0
> irq15: ata1                          581          0
> Total                            1765830        245
>
> I had thought of using hwpmc to find out what was happening, but 
> unfortunately VIA CPUs aren't supported yet.  Is there another way I 
> can find out what's going on?

This appears to be an issue with powerd/cpufreq - disabling powerd 
reduces the interrupt load to a couple of percent at most, and the clock 
interrupt task now only accumulates CPU time very slowly (previously it 
was using 7% CPU all the time).

--
Bruce Cran