On Tue, Jan 17, 2012 at 08:50:30AM +0100, Jan Müller wrote: > Greg KH wrote: > >On Mon, Jan 16, 2012 at 10:19:46PM +0100, Jan Müller wrote: > >>Dear all, > >> > >>I try adapting a network driver (net/ethernet/xilinx/ll_temac) for > >>my needs, but I'm not sure I understand the issues involved. > > > >What are your needs? Specifically please. > > In a Xilinx FPGA, I have a hardware wrapper around the ll_temac > eth-core. This wrapper keeps the ll_temac busy most of the time. So, > the throughput as seen by the CPU (microblaze) is quite reduced. > (i.e. the driver should be able to cope with that) > > The ll_temac driver from the kernel seems to work fine (for ping, > telnet) but when stress-tested (nc) after a while it 'locks-up'. > More precisely, the tx-buffer does not get cleared by the hw as > expected by the driver. Likely, my interfacing with the wrapper is a > little buggy. I'm a bit lost with debugging, but I think the issue > might be some irqs getting lost. I thought it might be possible to > get the driver working even with buggy hardware (watchdog timer?). > > > And why are you modifying > > this specific one? > > I wanted to keep/reuse all the house-keeping this driver already > does for its hardware (which is basically the same in my case). You should ask all of this on the netdev@xxxxxxxxxxxxxxx mailing list, the developers there can help you out. > >>Maybe someone of you could point me in the right directions? > >> > >>For instance, does the *skb structure passed to the start_xmit > >>function has to be freed by the network driver in any case? I.e. not > >>calling dev_kfree_skb(skb) anywhere in the driver is a bug? > >> > >>I'm a bit confused about the need for locking tx-buffers (for > >>SG-dma). Somewhere (in the comments for another driver) I've read > >>the calls to start_xmit are serialised anyway? And then clearing the > >>dma tx buffer in the tx_irq should be fine because it should not get > >>interrupted? > >> > >>Could you maybe recommend a simple "role model" net/ethernet/ driver > >>I should have a look at? One which implements a locked tx buffer (in > >>case you think that's needed at all) > > > >Have you looked at the network skeleton driver? > > find . -name "*skel*" > > Uuhm, in the kernel sources? Where can I find it? Hm, I thought it used to be there, sorry about that, I don't know where it moved to. greg k-h _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
