Alan, thanks for your response.
> > =========================== REBOOT =====================================
> Which HCD modules are loaded when you reboot? Only uhci-hcd?
Yes. Only uhci-hcd.
> > # hdparm -t /dev/hda
> >
> > /dev/hda:
> > Timing buffered disk reads: 44 MB in 3.13 seconds = 14.05 MB/sec
> > # modprobe ehci-hcd
> > ========= CPU GOES TO 100% =============================================
> > # hdparm -t /dev/hda
> >
> > /dev/hda:
> > Timing buffered disk reads: 172 MB in 3.02 seconds = 56.87 MB/sec
> So loading ehci-hcd _improves_ disk performance.
And lird ist starting to consume all the processor-power. This is where
the PCI BusMastering takes place when CPU has no time to do so? This is
the only explanation I have for this. It looks like the lircd gets hung during
some polling of the lirc_i2c kernel-module. (I'm using lirc_i2c modules from
cvs and debian/stable lircd 0.7.1pre2-2)
It's also responsible for having 14.05MB/sec at first. When I unload them I'm
back at 36MB/s when only uhci-hcd is loaded. when ehci-hcd is loaded too
22MB/s.
> > # rmmod ehci-hcd
> > # hdparm -t /dev/hda
> >
> > /dev/hda:
> > Timing buffered disk reads: 170 MB in 3.01 seconds = 56.44 MB/sec
> And then unloading it doesn't reduce performance. How's that for service?
> :-) The difference between this state and the original state is a
> mystery, however.
Still 100%CPU -> BusMastering?
> > # /etc/init.d/lirc stop
> > Stopping lirc daemon: lircmd lircd.
> What made you think of stopping lircd?
It was the only thing I changed after my last test and "top" shows that
it's responsible for that.
> > ========== CPU GOES BACK TO NORMAL (<<10%) =============================
> > ~# hdparm -t /dev/hda
> >
> > /dev/hda:
> > Timing buffered disk reads: 104 MB in 3.02 seconds = 34.39 MB/sec
> Stopping it did reduce disk performance. Odd.
This is "normal operation". The maximum I got out of the system before.
When loading ehci-hcd 22MB/s is normal operation. Strange? We already know
that.
> > =========================== REBOOT =====================================
> > # hdparm -t /dev/hda
> >
> > /dev/hda:
> > Timing buffered disk reads: 44 MB in 3.06 seconds = 14.40 MB/sec
> > # rmmod uhci-hcd
> > ========= CPU GOES TO 100% =============================================
> > # hdparm -t /dev/hda
> >
> > /dev/hda:
> > Timing buffered disk reads: 172 MB in 3.02 seconds = 56.87 MB/sec
> Unloading uhci-hcd also improved performance. Is this because you have a
> high-speed USB device that is controlled by uhci-hcd only when ehci-hcd
> isn't loaded?
Yes. Already described above.
> > # /etc/init.d/lirc stop
> > Stopping lirc daemon: lircmd lircd.
> > ========= CPU GOES BACK TO NORMAL (<<10%) ==============================
> > # hdparm -t /dev/hda
> >
> > /dev/hda:
> > Timing buffered disk reads: 112 MB in 3.02 seconds = 37.08 MB/sec
>
> Got to admit, I don't understand why stopping lircd should reduce your
> disk performance. Does restarting it cause the performance to improve
> again?
No. It never gets back until a new reboot and the same procedure.
So --- Are there any hints where to insert some debug-code? Currently
using 2.6.15 kernel. Took a first look on the VIA-PCI-IDE handling code.
But it looks that by spec there's only a bad AdressSetupTiming:
/linux-2.6.15/drivers/ide/pci/via82cxxx.c: 85
{ "vt8235", PCI_DEVICE_ID_VIA_8235,
0x00, 0x2f, VIA_UDMA_133 | VIA_BAD_AST },
I have not done hardware programming for the last 7 years (and it was
just very basic programming that time). So me digging
through the code is not very effective ;-)
I'm thankful for every help - even if you just redirect me to an
other person to ask.
Helmut
--
My GNUpg fingerprint http://www.gnupg.org
4563 F4FB 0B7E 8698 53CD 00E9 E319 35BD 6A91 1656
-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems? Stop! Download the new AJAX search engine that makes
searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
http://ads.osdn.com/?ad_id=7637&alloc_id=16865&op=click
_______________________________________________
linux-usb-devel@lists.sourceforge.net
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
