Ilija Kocho wrote: > John Dallaway wrote:
> >Sergei Gavrikov wrote: > >>Hello guys, may be I miss something but I thought that any > >>Ethernet eCos driver is enough abstract thing to manage ETH L2 and > >>that does not depend (well, depends a bit) on any next layer, > >>e.g., a TCP/IP implementation (RedBoot TCP/IP, *BSD, lwIP* stacks) > >>even if the driver uses another channel (SPI) to get a memory > >>access to MAC buffers. I talk about generic io/eth/* stuff > >>and..., well some kind of a future devs/eth/mc/spi/* eth_drv_.* > >>routines, for example. > > > >In theory, of course, you are correct. The network abstractions > >should ensure compatibility between any ethernet device driver and > >any TCP/IP stack. But in practice, testing can reveal all manner of > >issues which no-one was expecting. Hello Guys, Agreed. I just thought about those `eth_hwr_funs' for new ENC chip: http://ecos.sourceware.org/docs-latest/ref/io-eth-drv-generic1.html and then for start a test-path would be one from io/eth `-- src |-- lwip |-- net |-- **newlwip** `-- stand_alone Of course, that will be a choice of hardware keepers. But, looking on the picture I see why Alex suggested to go by a way without asterisks, because, "...testing can reveal all manner of issues which no-one was expecting" :-) > This is our first eCos driver of this kind and we take the > abstraction and stack independence for granted. It may save us some > effort and help bring the driver sooner if somebody more experienced > points out potential potholes. Ilija, which hardware (board) do you plan to use in a development? Is that STM3210E-EVAL? Certainly, if target has not a lot of RAM, then choices can be `lwip' derivatives then. On the other hand, two new fresh projects (yours enc_eth and Simon's lwip-1.3.1) would help each other in a self testing. Keep up the good work. Regards, Sergei
