Vuescan does not recognize scanner - USB transfer size limit problem?

[WNorfleet]

Vuescan 8.3.87 does not recognize a Minolta Scan Dual III scanner on a desktop machine with a fresh install of FC5 updated via yum. An exchange of emails with the developer of Vuescan concludes: "This would indicate a kernel problem . . ., probably in the USB subsystem limiting transfers to 32 KBytes." "The problem is that when VueScan tries to load the firmware into the scanner, it tries to transfer 44904 bytes in a single transfer. This is failing." How can this be fixed?

[bill@desk usb]$ uname -a
Linux desk.os.org 2.6.18-1.2257.fc5 #1 Fri Dec 15 16:06:24 EST 2006 i686 athlon i386 GNU/Linux
[bill@desk usb]$ rpm -qa | grep libusb
libusb-0.1.11-2.2

Hal-device-manager and a review of /sys indicate that the scanner is being recognized by the OS. Vuescan also identifies the scanner as a Dual III, but does not initialize it properly. This machine/Vuescan combo works fine with an Epson Perfection 1260 flatbed scanner. A laptop (IBM Thinkpad 600E), also with a fresh install of FC5 updated via yum, works perfectly with the Dual III. So this would seem to be a case of a specific machine not working properly with a specific scanner.

[multescugeorge@]

I for one have only 1 advice: get the latest kernel from www.kernel.org and compile it to match your requirements. If you don't have any more ideeas give it a try, look around the forum on detailes on how to do it (for starters look into these comands: make xconfigure; make bzImage; make modules; make modules_install; make install). I wish you luck.

[WNorfleet]

Connecting the scanner to the motherboard's USB 1.0 ports results in proper function of the Dual III (why, oh why, did I not think to try that sooner!) So the problem seems to be confined to the add-on PCI USB board (1.1 and 2.0).

[WNorfleet]

Where is the USB 2.0 max sector or transfer size limit established in the kernel structure? Does changing it involve hand-editing a module (ehci_hcd?) or adjusting a parameter when configuring the kernel prior to compiling? I've dug around quite a bit for an answer to this, but no luck so far. Either the answer is so well known as to not be worthy of comment, or the Vuescan => Dual III communication is very unusual.

[WNorfleet]

I have downloaded 2.6.20 from kernel.org and compiled it. Same problem. Is the USB 2.0 max sector or transfer size limit established in one of the modules? Can a simple edit of a module and recompile address this problem? Is this a bug that should be reported on bugzilla (redhat or kernel.org), or just a "feature"?

[WNorfleet]

This problem is "fixed" by "rmmod ehci_hcd". Guess that's why ehci is said to be "experimental".

Edit: The problem seems to be card-specific. The card in question is a "Cables To Go USB 2.0 Hi-Speed 5-Port PCI Card", item number C184-29554 at Tiger Direct. Replacing this card with an "IOFuture PU520 High-Speed USB 2.0 PCI Host Controller", item number 15-104-216 at NewEgg, caused this and some other strange USB-related problems to go away.