Re: sony dsc t1 & other usb2 issues

[Swift Griggs <swift.griggs%coloradovnet.com@localhost>]

On Fri, 15 Aug 2008, Anders Lindgren wrote:
> I have never been able to use USB2 umass (UHCI/EHCI) in NetBSD; it has 
> simply never worked on any computer I or my friends have ever had,

	Despite having been one of the first OSs with USB support, I have 
had a lot of problems with USB/USB2 as well. I have had the best luck with 
Intel USB controllers, terrible luck with ATI (SB600 mobos), and mixed 
luck with VIA and nVidia controllers. I have also noticed that the file 
systems in use on the USB media seem to make a difference in terms of 
stability. I've had lots of lockups/filesystem-freezes with fat/vfat but 
in many cases I can take the same physical media, put FFS1 on it, and it 
works like a champ with no problems. I've also noticed that fat-based 
devices are excruciatingly slow (usually less than 1mb/s) compared with 
the same device under doze or Linux (usually 3-5 mb/s). This is NOT due to 
caching because I've run tests whereby I fill the entire device up (using 
SD cards and Maxtor external hard disk drives). Mounting async only seems 
to make fat lock up faster or freeze longer. Nonetheless, there are a few 
cases where USB2 & NetBSD perform very nicely. I have a Dell 530s with an 
Intel-based chipset that is smoking-fast with FFS2 and outperforms Linux 
on the same machine by some margin. I've seen fat/vfat slowness on this 
machine, but never lockups.

> although I am happy to say it does appear to work on my new eeepc900.

Indeed it does on mine, too. However, it's painfully slow. I just 
installed -current on it (yesterday's cut) and I can't get it to crank out 
more than 1mb/s using an external Maxtor HDD that I get around 18-25 mb/s 
(at least on big blocked out sync-writes) on other machines. I know the 
eee900 is a bit on the slow side with it's 900Mhz Celeron, but something 
still seems rotten...

> I've seen precisely the behaviour you describe with usb sticks and 
> compact flash cards, but also complete IO deadlocks and system freezes.

Ditto. However, I've seen a lot fewer with NetBSD 4.0 and even less with 
-current. It seems the demons are slowly being exercised.

> Slower OHCI devices (mice, keyboards, wacom tablet etc) however have 
> always worked well for me.

I'd add to that list scanners via libusb/sane and in my case cameras 
accessed via gphoto. I'm a little slow to blame NetBSD when some digital 
camera doesn't work just right since I'm wondering how much time their 
engineers spend getting the UMASS functionality to work outside testing 
the bundled Windoze app.

Thanks,
  Swift
--
Swift Griggs -- <swift.griggs%coloradovnet.com@localhost>
Unix Systems Administrator