Problems with ath at kern.hz=100

Bengt Ahlgren bengta at sics.se
Tue Feb 10 08:08:50 PST 2009


Ian Smith <smithi at nimnet.asn.au> writes:

> On Mon, 9 Feb 2009, Bengt Ahlgren wrote:
>
>  > I was changing kern.hz to 100 on my IBM X40 (Pentium-M) laptop to see
>  > whether that saves some battery, but ran into problems with ath not
>  > working properly.  At first I thought that my hardware was failing,
>  > but changing hz back to 1000 solved the problem.  I am running
>  > 7.1-RELEASE-p1 and have the following Atheros chip:
>  > 
>  > Feb  9 11:13:10 P142 kernel: ath_hal: 0.9.20.3 (AR5210, AR5211, AR5212, RF5111, RF5112, RF2413, RF5413)
>  > Feb  9 11:13:10 P142 kernel: ath0: <Atheros 5212> mem 0xd0200000-0xd020ffff irq 11 at device 2.0 on pci2
>  > Feb  9 11:13:10 P142 kernel: ath0: [ITHREAD]
>  > Feb  9 11:13:10 P142 kernel: ath0: WARNING: using obsoleted if_watchdog interface
>  > Feb  9 11:13:10 P142 kernel: ath0: Ethernet address: 00:05:4e:4e:1f:c7
>  > Feb  9 11:13:10 P142 kernel: ath0: mac 5.6 phy 4.1 5ghz radio 1.7 2ghz radio 2.3
>
> There's an order of magnitude room to move between 100 and 1000 Hz.  
> You could also try with maybe 250 or 500?

I have been running with hz=250 for a while now without the problem
occuring.

> What speed is your Pentium-M?  Are you using powerd?  If so, what 
> minimum CPU freq does it drop back to (sysctl debug.cpufreq.lowest)?

It is a 1200MHz.  I have debug.cpufreq.lowest=600 in order to only use
EST.  I ran for a while at 1200MHz with no powerd and ath still
stopped sending.

> Sam has talked of a problem with ath (and maybe other) interrupt rates 
> overwhelming slow CPUs to the extent that userland (ie powerd) doesn't 
> get a look-in for a while, as I understand it.  A fast CPU clocked back 
> to less than say 300MHz IS a slow CPU, until powerd gets to crank it up.
>
> You might find that rc.conf setting economy_cx_lowest="LOW" can save you 
> some battery, though this benefit seems somewhat processor-dependent.

I tried with dev.cpu.0.cx_lowest=C1 (at 1200MHz), but with no
difference - the problem is still there.

Bengt


More information about the freebsd-mobile mailing list