**** BEGIN LOGGING AT Fri Apr 18 02:59:56 2008 **** BEGIN LOGGING AT Fri Apr 18 12:37:07 2008 Apr 18 14:07:32 drath: hey, just out of curiosity, does openocd have strongarm support? Apr 18 15:05:27 prpplague: no, strongarms have no debug unit Apr 18 15:05:45 prpplague: they have a jtag port, but that's limited to boundary scan Apr 18 15:28:33 drath: ahh Apr 18 15:28:53 drath: i knew they had jtag, just didn't know what level of debug was supported Apr 18 15:39:46 Has anyone scanned the instruction space to see if there is anything hidden? Apr 18 15:41:29 the last time we played was with boundary scan a long time agop Apr 18 15:42:39 Hi ben. What sort of transfer rates should I be getting on the oki module through ethernet? Apr 18 15:44:01 I'm using wget from an ftp server and I get about 800k/sec with the default bus timings. I can tighten the bus timings and get 860k/sec but now I'm stuck. Apr 18 15:45:11 (ftp server on local network, tried more than one, wireshake indicates delay is with the oki reply and no obvious protocol problems). Apr 18 15:52:14 2-12-7 is the default bus timing, 1-4-3 seems to work fine but with less than a 10% change, suggests that something else is limiting the speed. Both lights are on, and I take this to assume 100Mbit line rate. The dm9000 app notes suggest with the 16bit bus 140Mbit bandwidth is feasable. Apr 18 16:05:43 ali_as: speed is also controlled by how fast you can process the data and turn it around Apr 18 16:06:29 it sounds like you've hit the cpu limit Apr 18 16:06:34 I'm wgetting to /dev/null. Apr 18 16:06:38 Hmm. Apr 18 16:07:46 Has the thing ever been benched with a data rate over 10Mbit? Apr 18 16:09:54 I've not got any figures Apr 18 16:10:45 No optimisation was done whe you wrote the driver? Apr 18 16:11:42 It isn't just the driver, rember you've also got an IP, TCP and a sockets layer in there as well. Apr 18 16:12:30 Will they move the data round or just bother with the header? Apr 18 16:13:08 This is a bit of a blow as my design goal is 1M/sec through jtag and in/out of ethernet. Apr 18 16:13:49 the IP layer needs to process the packets, do a checksum, then lookup the protocol, pass to TCP which then does another chgecksum, matches it against an open stream, checks to see where the data is within the stream (possibly queueing for more data, etc) Apr 18 16:14:54 and then it may be copied to userspace during the read or write calls Apr 18 16:14:54 Hmm, the dm9000 will do the checksum though? Different checksum? Apr 18 16:15:05 no, the dm9000 does an ethernet layer CRC Apr 18 16:15:34 I'd call crc a checksum but fair enough, ethernet layer. Apr 18 16:15:50 CRCs are generally better than checksums Apr 18 16:17:26 So to even test the speed of the device I need to be writing to the chip directly. Apr 18 16:17:39 pretty much Apr 18 16:18:54 I think I have a new slogan "Embedded linux - your clock cycles at work". Apr 18 16:20:35 Does uclinux allow an easy way of locking out ethernet activity? Apr 18 16:20:58 ie, so it won't try to talk to the chip when I am. Apr 18 16:21:24 no idea. Apr 18 16:21:39 Ok, thanks Ben. Apr 18 16:24:10 * bjdooks has to go, 2.6.25 patchset to finish up Apr 18 16:24:44 Later Ben. **** ENDING LOGGING AT Sat Apr 19 02:59:57 2008