Hello, On 9/1/05, Alan Stern <[EMAIL PROTECTED]> wrote: > On Thu, 1 Sep 2005, Sven Duscha wrote: > > > Hello, > > > > I'm using a Cypress FX2LP EZ-USB development board to design software > > for an USB2 Cy7c68013a-128 FX2LP chip. > > > > I got the device to work with the usbcore-driver and am now > > experimenting with bulk transfers. The easiest approach seemed to use > > functions provided in libusb-0.1.10a (http://libusb.sourceforge.net/) > > to do bulk transfers with usb_bulk_read and usb_bulk_write. > > > > The problem is that in doing so I only get a maximum transfer rate of > > ca. 3000k/sec. That is beyond the limitations of USB1.1 and therefore > > the device seems to work in USB2.0 mode, but still far below the > > limitations of USB2-High speed specifications. > > > > I did tests with different firmwares that were provided with KEIL > > development tools (bulksrc, bulkloop) and ezconsole and none provides > > any higher transfer speeds. > > > > My only guess is that going through generic usbcore-support could be > > the bottle neck. But I'm a bit reluctant to delve into development of > > a device driver for the Cypress FX2LP module. > > > > On the other hand there is hardly any information on the linux support > > for the FX2LP available on the net, even the less from Cypress > > itself. Most drivers and sample applications are for the > > FX-USB1.1-version only. > > The bottleneck isn't usbcore, it's libusb. However you don't need to > write your own device driver. It's possible to do what you want using > ioctl and usbfs. You could even hack the libusb source code to support > it, if you want. > > What you need is to make asynchronous submissions -- that is, don't wait > for an URB to complete before submitting the next one. You should queue > enough data to fill the bulk pipe for 20 ms or more; thereafter each time > one URB completes you submit a new URB. Remember that usbfs has a limit > of 16 KB per URB. High-speed bulk transfers are capable of handling up to > 52 KB/ms, which means you will need to keep at least 1 MB of data (or 64 > URBs) in flight.
I had the same problem (not enough throughput from an FX2 with libusb). Eric Blossom pointed me (http://marc.theaimsgroup.com/?l=linux-usb-devel&m=109899814018128&w=2) to the usrp lib from the gnu software radio project: ftp://ftp.gnu.org/gnu/gnuradio/usrp-0.8.tar.gz I was able to get something like 42 MB/sec bulk transfers from my FX2. But it was dependent on the host controller, some nec chip based hardware only got me 32 MB/sec. There was some discussion whether this could go into libusb or whether the usbfs interface could be extended to support some scatter-gather-list based high-throughput zero-copy mechanism. There seemed to be some issues with proper error handling, memory boundaries, etc. I think there was no real agreement on how, or even if this could go into usbfs/libusb. Maybe this has changed, meanwhile? I'd still be interested in some usrp/fusb like functionality in libusb. I'm afraid my knowledge of dma / sg / memory-boundary related topics is insufficient to help extending usbfs, but I'd be happy to help with implementing and testing some kind of a port of the fusb stuff to libusb. Cheers, Axel ------------------------------------------------------- SF.Net email is Sponsored by the Better Software Conference & EXPO September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf _______________________________________________ linux-usb-devel@lists.sourceforge.net To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
