Oliver, I sill don't understand what you're trying to say,or how it relates to the structural point I was making: that the batching isn'treally needed (or helpful) given sane USB DMA/transfer queues (as on Linux). > > > If we want to do this for networking we'd need to > align > > > the skbs to that size and build a virtual large > package > > > from a chain of skbs. Is that what you had in > mind? (To which I said "most certainly NO; > > > > It looks like Dave meant that you can submit a large > number of URBs, > > each transferring an arbitrary amount of data. (from anywhere in kernel memory ... which is one of the definitions of "scatter/gether"; and you were claiming wouldn't work without adding curious alignment restrictions (i.e. not all source buffers would work). Think of the network packets (SKBs) as mapping directly to USB transfers (of 1..N packets)... There are MTU restrictions of course, which may be affected by any framing/encapsulation needed to talk to the other end. > Yes, but the device will notice if you send an incomplete > packet. That is, device notices driver bugs... kind of self-evident... :) You appear to be thinking of some particular flavor of framing/encapsulation problem, (details unspecified)... (By the way, it's unclear what kind of packet you mean here. USB packet, part of a USB transfer, which holds either (a) cleanest case, a single Ethernet-ish network packet, or (b workaround for some kind of implementation restriction: a batch of such ethernet-ish packets. > As the device is probably optimized for batching, this maybe a problem. If you postulate optimization for (b) you've made a self-fulfilling prophecy. Likewise for (a), which by observation is currently more common. Recall that I was highlighting the fact that batching (b) is a workaround ... one that Linux hasn't needed at all, in the past couple decades... and that since batching involves superfluous copying of data packets, it wastes resources (like battery power, indirectly via CPU cycles). > -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
