Re: Gnome Pilot Causing USB Drive problems (version 2.0.10)

[Nigel Metheringham <Nigel Metheringham dev intechnology co uk>]

On Wed, 2004-09-08 at 06:46 -0400, David A. Desrosiers wrote:
> > Won't help in this case.  Its not when gnome-pilotd is talking to a 
> > USB palm device, but when gnome-pilotd polling to see if a palm 
> > device is attached.  At that point (unless you have just hit hotsync 
> > on the palm) the visor module isn't loaded.
> 
>  	That depends on your system configuration. I've never yet run 
> the visor driver on a sytem where it wasn't already loaded. hotplug 
> causes FAR too many problems with Palm devices, and that is exactly 
> why we strongly recommend disabling it if you use a USB-connected 
> Palm.

OK, I have now managed to reproduce this problem with pretty much a bare
minimum around it.

If a program uses poll() and opens/stat/reads /proc/bus/usb/devices then
USB transfers appear to stutter/fail to USB Mass Storage or other bulk
USB users (I have a USB mouse which is unaffected).  Use of poll() or
read or stat of the device file alone is not effective - you need all
three components.

I did not have the visor module loaded during these tests, so thats not
involved.  The poll was looking at input from a FH opened on /dev/null
(so it wasn't doing much).  I did see freezes on the open/read
of /proc/bus/usb/devices while this was happening - 4 or 5 second hangs.

If anyone wants my bit of perl which I am using as a test load generator
to show this problem then drop me a line - its only 30 or so lines.

I suspect that if gpilod was modded to drop the fstat() of
the /proc/bus/usb/devices file after opening it then the problem would
go away - I may try doing this later.

	Nigel.
-- 
[ Nigel Metheringham           Nigel Metheringham InTechnology co uk ]
[ - Comments in this message are my own and not ITO opinion/policy - ]