On Thu, 1 Apr 2010, Peter C. Wallace wrote: > Date: Thu, 1 Apr 2010 08:34:34 -0700 (PDT) > From: Peter C. Wallace <p...@mesanet.com> > Reply-To: EMC developers <emc-developers@lists.sourceforge.net> > To: EMC developers <emc-developers@lists.sourceforge.net> > Subject: Re: [Emc-developers] EPP timeout > > On Thu, 1 Apr 2010, Slavko Kocjancic wrote: > >> Date: Thu, 01 Apr 2010 10:25:24 +0200 >> From: Slavko Kocjancic <esla...@gmail.com> >> Reply-To: EMC developers <emc-developers@lists.sourceforge.net> >> To: EMC developers <emc-developers@lists.sourceforge.net> >> Subject: [Emc-developers] EPP timeout >> >> Hello... >> >> As I have near finished driver for EPP boards I have one problem. As computer >> is common staretd with CNC machine off the EPP board respond to al IO with >> timeout. EPP timeout is aprox 10uSec. So if I wan't to write 5 bytes that can >> least over 50uSec. And in that case the computer freeze. The solution seems >> to be to check EPP timeout after every IO. That's works but make big >> overhead. On my EPP board (sem's to be slowest one) the fastest cycle is 2uS >> long. The timeout is on aprox 11uS. But the read from status port (to check >> timeout) least 1.5uS. So if I get that approach then I have (for 3/3 byte IO) >> 6 x 2.0uS 12uS for real IO >> 6 x 1.5uS 9uS overhead to check Timeout >> total 21uS and system got little slower already >> >> Then I realize that checking how long in/out can be faster. >> So if write_board routine take more than predetermined time to execute then >> execution shall be aborted and checked what's realy wrong. >> >> maybe the best way is to check if out/inp routine take more than 5uS. If is >> more then probably is something wrong. >> >> in write_board there is argument PERIOD. I can't find what is this. Time ?!? >> >> I have only RDTSC instruction in mind but afraid to use as can be dissabled >> (privileged) >> >> Can someone help?? >> >> >> >> And here is working (but unfinished) driver hal_eppport.c >> >> >> > EPP timeout is per single byte transfer. You should not get a timeout error > regardless of how long a burst of transfers lasts. If you are getting EPP > timeouts, I would check your hardwares handshaking. Also good to notice the > difference between EPP1.7 and EPP1.9 > > > Peter Wallace > Mesa Electronics >
Oh, I didnt read the top of your message carefully, you are saying that the card is off at startup so you get timeouts and you have this running in a fast base thread so you use up all available CPU time. Maybe your driver can just do a single test I/O cycle each thread until it gets a good transfer (no timeout) and then enable the full data transfer. If it ever gets a timeout when running (check at end of block), you could go back into the "wait for good transfer" mode Peter Wallace Mesa Electronics ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
