On Mon, 11 Dec 2006, Ilyes Gouta wrote: > Hi! > > Thank you very much, Alan, for your answers! > > I'm going to carry out more experiments on the 256 ISOC packets issue > tonight and I'll keep you updated. However, I'm pretty sure that > Linux's USB stack
You shouldn't talk about "Linux's USB stack". The fact is, different drivers are involved depending on whether you use ehci-hcd, ohci-hcd, or uhci-hcd. Each has its own limitations. > can't handle ISOC URBs that would point on a large > data buffer, for example an URB w/ ISOC 256 packets, 3072 bytes each > (so that makes 3072 * 256 bytes, which Windows can handle). Wrong. The size of the data buffer doesn't matter at all. However there _are_ scheduling restrictions in each of the drivers, like how far into the future the driver can schedule ISO transactions. Of course, here you're talking about a 768 KB data buffer. In general you shouldn't count on being able to allocate such a large amount of kernel memory as a single contiguous buffer. But if you can do it, the sheer size won't affect anything -- it will work just as well as an URB containing 256 4-byte packets. > And yes, I'm using usb_set_interface to select an alternate setting > for the interface that's supposed to enable the ISOC transfers. > > I got another question! :) > > Is it possible that a USB 2.0 device (my webcam) won't able to return > any data because the host is submitting too many ISOC URBs in a fast > pace? You're making the same mistake as before. The device doesn't know anything about URBs; all it knows about are packets. It doesn't know or care how fast URBs are submitted; it cares only how fast it is asked to send a packet. If you schedule ISO transactions at the speed given in the endpoint's descriptor then the device should be able to handle the transfers. > Is it possible that the device would stall or reply partially to > the request (for example, won't send/dump the entire picture and would > restart from the beginning)? A device can do anything it wants. But if you use it as it was designed to be used, then it should function properly. Alan Stern ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys - and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV _______________________________________________ [email protected] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
