On Sat, Oct 06, 2001, David Brownell <[EMAIL PROTECTED]> wrote: > > > AFAICT, all we need to do is add SA_SAMPLE_RANDOM to the request_irq call > > > flags in the host controllers. > > > > I think most controllers interrupt every frame. > > UHCI sends a more regular stream of interrupts, since it needs the > software to tend things rather often. OHCI and EHCI tend to use > interrupts only when transfers complete.
This isn't exactly true. In most normal situations, it'll interrupt as often as OHCI and EHCI, when the transfer completes. Only when we fiddle with FSBR and active, but slow/stalled transfers, will it interrupt more often. > > That would be a very poor source of entropy. > > The entropy comes from unpredictabillity, which is more a function > of the USB devices rather than the controller. So keyboard and > mouse code (device level) is a more appropriate place to track > entropy than HCDs (controller level). > > Consider video-over-usb: highly regular interrupt pattern, no > matter what controller is in use. Yet ... if you point the video > at a (powered-up) Lava Lamp, the _image_ will be a rather > good source of entropy. (And patented by SGI as I recall.) > > Oliver's suggestion of device connect/disconnect calls might be > worth looking at. While the interrupts themselves are known > to occur at N frame boundaries, not every frame will have > such an interrupt. Much like USB keyboards/mice. Completely agree. JE _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
