Hi, since we both seem to be interested in ISDN support out of the box, let us try to find solutions for the situation at hand.
L T Pensch schrieb: > Carl-Daniel wrote: > >> Of the proprietary drivers, all USB drivers could be migrated >> to userspace with libusb as Greg Kroah-Hartman suggested. This >> would allow to keep them proprietary yet keep them out of the >> kernel. Problem solved for them. > > Are you sure? Greg does not know these devices and their drivers, > his statement may be based on some stable knowledge of successful > transition projects from kernel to user space *for some other de- > vices* but simply taking this as a base to judge upon these ISDN > and DSL device is not reasonable. OK, so what are the technical reasons to keep the drivers in the kernel? So far the only thing I saw was high-level talk about userspace drivers being impossible due to... wait, no reason given. You are the first person to mention a possible technical reason. > Do you know for sure that libusb performance meets the strict > requirements imposed by some of the protocols involved? As far > as I understand this game, you will run into problems when trying > to implement the required timing and reduce lethal latencies *in > user space* for these drivers. So the timing is a problem. Unfortunately I can't guarantee that libusb will have the required latencies on a 100 MHz computer. Nor can anybody guarantee latencies below 10 ms even with kernel implementations on very fast machines. However, if you talk to audio professionals, it seems that they really get decent latency in userspace with recent kernels. Latency is a big problem regardless of whether you do it in the kernel or in userspace. It may be more difficult to handle it in userspace, but unless someone comes up with real numbers proving that this can't be done, I claim it is possible. Now if you want to do echo compensation outside of the kernel, you may have a problem. * Looking at ISDN cards for the S0-Bus, I can't see where echo compensation would be needed (except for modem emulation). * Looking at DSL cards, echo compensation may be needed if the card can't handle it in hardware. So it *seems* that ISDN cards should have simple drivers. Please correct me if I'm wrong. > How many context/task transactions are acceptible for such a > monster while doing 6MB/s DSL from kernel to userspace back to > kernel and so forth? 6 MBit/s or 48 MBit/s? > Let's face it... Go to eBay, sell your device to some Windows guys, > if ever, it will take a long time before you can use your hardware > on Linux again. :-( As I stated in another mail in this thread, it seems that many AVM ISDN cards are supported by GPL drivers. Some others probably could be supported by GPL drivers with a subset of features. I would *love* it if somebody from AVM could tell us what's the reason to keep some ISDN drivers proprietary. Maybe we could help finding a solution better than what's possible now. Regards, Carl-Daniel --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
