IRC logs for #openrisc Saturday, 2014-01-11

--- Log opened Sat Jan 11 00:00:48 2014
_franck_stekern: I just copied this header from openocd just before I commited it...sorry :)08:16
blueCmd_if I want to get myself actual hardware to run openrisc on, is the Ordb2a-ep4ce22 the way to go?09:38
_franck_I would say any Altera board would be better since you don't have access to the JTAG pins on the Ordb2a-ep4ce2211:09
blueCmd__franck_: any recommendatons?11:10
_franck_well I have an old de1 board11:11
_franck_look here: http://www.terasic.com.tw/en/11:12
blueCmd_ah, good old terasic11:12
_franck_the DE2-115 looks ok if you can get academic price11:15
blueCmd_used to have the DE2 board11:16
blueCmd_when I moved I got rid of all the hardware stuff11:16
_franck_I have a de2-7011:17
blueCmd_the  Ordb2a-ep4ce22 looks like a very good size. is it the FPGA JTAG or the OpenRISC jtag that is not available?11:17
_franck_FPGA JTAG is connected to a FTxxx based JTAG interface11:18
_franck_so you can't use Altera tools11:18
blueCmd_appears to be out of stock as well11:26
_franck_I've always wanted to a board myself (because of the success of the Ordb2a-ep4ce22 board)11:27
blueCmd_you forgot a word I think11:31
_franck_*make11:31
_franck_time to eat see ya11:34
blueCmd_have fun11:34
blueCmd_Oh, I saw this just now: http://www.youtube.com/watch?v=o8DReLrNn_k11:38
blueCmd_I'm famous!11:38
rokka:)12:14
rokkawhy?12:16
blueCmd_they talk about my TLS work :P12:23
rokkawhat is TLS12:26
blueCmd_thread-local storage. a thing to allow threads to have .. well, local storage12:27
rokkabtw i have one ordb2a and would like to order more if i could :D12:27
blueCmd_hehe, nice :)12:29
rokkamy application needs a physically small altera fpga board with ethernet, flash and io for connecting other pcb12:30
blueCmd_sounds like what I would want. how do you do debugging on it?12:31
rokkacurrently i can send debug data out with ethernet12:32
rokkaand i use simulating for rtl components12:33
rokkaprogramming the fpga is trivial with the provided virtual machine12:33
blueCmd_ok, cool12:34
rokkavirtual machine has commandline tool to program the fpga or the spi flash12:35
rokkai just compile design with quartusII and transfer svf file to vm and trigger the programming12:35
rokkai think it would be possible to make a proxy that would make it possible to use quartusII programming software12:36
rokkaif somebody knows the protocol for ethernet blaster12:37
gr8why is there so little public interest in open source processors?14:49
olofkgr8: We haven't found out that yet. Otherwise we'll all be rich and famous15:31
gr8ok then why should a non-tech person be interested in open source processors?15:36
gr8that is the central question. it must be stated in the wiki15:45
gr8it is essentially the same question as "why should person A be interested in free software", I think.15:47
stekerngr8: I'm not sure a non-tech person would, or even should for that matter17:52
blueCmd_23 minutes for total system build. that's $0.37 per build. could probably go down to 4 CPUs with about the same time which would end up around $0.2419:43
blueCmd_stekern: how long did it take to run the regression suite again? i forgot how long time it took for me19:43
stekernblueCmd_: for the one linux one against or1ksim? 24h or something like that I think19:51
blueCmd_stekern: ah, sorry - I mean the gcc suite19:54
stekernnewlib?19:54
blueCmd_"make check-c check-c++"19:55
stekernyes, but against a linux target or newlibc target19:55
blueCmd_linux19:55
stekernthat's what I'm trying to ask19:55
stekernyes, that took 24h or something19:55
blueCmd_ok.19:56
blueCmd_should be possible to make that parallel by running a subset of the tests I suppose19:58
stekernyes, that's what jeremybennett did20:00
stekernhis scripts are still available from svn20:00
blueCmd_cool. I got the test suite to work with ssh, which mean no telnet or ftp. the board config is like 10-15 lines now in total20:02
blueCmd_I'll take a look at jeremybennett's work, hopefully it's applicable20:03
blueCmd_but first I need to find a very eluding race condition in the signal code!20:05
stekernin the linux signal code?20:08
blueCmd_stekern: either in linux or in glibc20:09
stekernare you using jonas latest tree?20:09
blueCmd_I want to run the regression suite to go through the features one by one, but with this bug I cannot even run the suite :)20:09
blueCmd_stekern: yeah20:09
blueCmd_stekern: I have some hacks in it to get glibc to run though20:10
blueCmd_TLS stuff and atomicity operators20:10
blueCmd_This is also a gem that I'm getting from sshd: "Received disconnect from 172.16.0.1: 2: Packet corrupt"20:10
blueCmd_client says "Bad packet length 4009614935." - hah. yeah, that's a fun one.20:13
stekernblueCmd_: have you tried with this applied? http://git.openrisc.net/cgit.cgi/jonas/linux/commit/?h=master-next&id=ee49d953576d9b65ce858077ac407ef2f1a5996e20:15
blueCmd_hm, not explicitly no20:16
blueCmd_I will apply it manually and see if it does nice magic things. it's more code, so it ought to be better - right?20:18
stekernyes, and it is proven to fix broken signal code ;)20:18
stekernbut iirc there was still some issue...20:19
blueCmd_stekern: I have quite a lot of "some issue" issues20:21
stekernah, yes, it broke strace20:21
blueCmd_hah, well I will ask the maintainer to fix that then20:22
blueCmd_blueCmd_: fix it!20:22
blueCmd_blueCmd_: ok20:22
stekernhaha20:22
blueCmd_I'm glad that people are using it though20:23
stekernit was sebastian macke that was tracking the syscall issue down and also him that tested with strace20:24
blueCmd_where was this? I would love to read the story20:25
stekernfor some reason he sent it in priv mail to jonas and me (I was just a bystander though)20:26
stekernI can pastie the most import parts of it20:26
blueCmd_thanks20:27
stekernfirst, this was the irc discussion about it: http://juliusbaxter.net/openrisc-irc/%23openrisc.2013-09-20.log.html20:27
blueCmd_anyway you might just have saved me a couple of days of debugging20:27
blueCmd_it seems to be working....20:28
stekernhttp://juliusbaxter.net/openrisc-irc/%23openrisc.2013-09-21.log.html20:28
blueCmd_which is fricking awesome20:28
blueCmd_stekern: I'm trying to get my employer to allow me to work on openrisc parttime, if this had been during work time you would have saved them money I guess :P20:29
stekern;)20:30
stekernas I said, I'm just the messenger20:30
blueCmd_stekern: still, thanks a lot20:31
stekernnp20:31
stekernif you get strace working with that applied (or find out why it didn't) we probably want to remind jonas about it20:32
stekernreading through the private e-mails, there wasn't much more info in them that isn't said in the commit message20:33
blueCmd_ok. well the progress is a 1000 connection/disconnects now. usually crashed around 30-40.20:34
stekernI got my kernel port project booting to init know, I guess I need to start hacking a linux toolchain...20:37
blueCmd_:D20:37
blueCmd_Do TLS!20:37
blueCmd_it's fun20:37
blueCmd_or something20:38
stekernhaha, I bet20:38
blueCmd_regression suite is actually executing tests now. around 1 sec per test I would say. feels responsive enough20:38
stekernI think a simple static uclibc based one will suffice for now20:39
blueCmd_u is for ugly20:39
blueCmd_I tell you that everybody will use glibc soon!20:39
stekernperfect match for me then ;)20:39
blueCmd_never mind the fact that it's like 100 MB20:39
blueCmd_did sebastian work on xorg, or was that somebody else?20:40
stekernyeah, I know, but at the moment I want to chose the path with least resistance to continue with the linux port20:40
blueCmd_from the irc logs it looks like it20:40
blueCmd_stekern: I wouldn't blame you20:40
blueCmd_stekern: I'm just messing with you20:40
stekernyes, sebastian was the one that got xorg running20:40
stekernI know you are ;)20:40
blueCmd_uclibc?20:40
blueCmd_or what did he use?20:40
blueCmd_do you know?20:41
stekernyes, he used uclibc20:41
stekernI think he tried out your glibc stuff too20:41
stekernnot sure how far he went with that though20:41
blueCmd_he would have needed to patch the kernel and code some new stuff, I wouldn't blame him from not using it20:41
blueCmd_I pinged jonas to make a new effort with upstreaming the kernel TLS stuff20:42
blueCmd_so it should become much easier20:42
stekernok, cool20:42
olofk_franck_: You're around? I'm having some problems with your or1k-elf-loader patch21:30
olofkhttp://pastie.org/862464921:35
jeremybennettblueCmd_: Just saw your earlier comment. We ran the GCC 4.5.1 Linux GNU regression in parallel against 16 Linux instances simulated by Or1ksim.21:49
jeremybennettThe scripts should still be in the repo for the old tool chain.21:49
jeremybennettEssentially you get the board description to select the target IP address from a file, ensuring each is used only once.21:50
_franck_olofk: oops, let me see this or1200-basic.elf. I did try vmlinux, uart-simple.elf and a led_blinker.elf of mine21:51
_franck_olofk: samanthafox :)21:51
blueCmd_jeremybennett: cool! i guess this is the repo http://opencores.org/websvn,listing?repname=openrisc ?21:51
jeremybennettFor OpenRISC we had a couple of extra issues. First we were working with a Linux that wasn't reliable at the time. So we had to implement a restart mechanism in the GNU regression parallellism. Second, there is only modest parallelism in the regression testing, so we had to manually split out the parallellism.21:51
jeremybennettblueCmd_: It should still be there. We were still in the SVN world when we did this work.21:52
blueCmd_yes, it's the manual parallelism I'm after21:52
jeremybennettIIRC it took a few hours to run the entire set of tests, but to a large extent that was due to having to retry portions of the parallellism.21:52
jeremybennettIt's exactly the same mechanism used in the GNU Makefiles - we just did a finer grain splitting. 128-way I seem to recall. That way each individual run was quite short, and the retries did not take too long.21:53
blueCmd_I like to think that what I have would allow for quicker and more robust executing, but talk is cheap - I will try to prove it :)21:54
jeremybennettI think we used to find the 16 running instances of Linux were down to 4 or 5 at the end of each run. I imagine the Linux is better now.21:54
jeremybennettGood luck. I came to the conclusion that a lot of very clever minds had gone into this work, and it was easiest to reuse what was there.21:55
jeremybennettWe discussed it informally at the GNU tools conference in London a couple of years ago, and it seemed there was some interest in finer grain parallellism with a restart feature. But once the project finished I didn't have time to take it further.21:56
jeremybennettIt's getting late here, so I'm stopping for the night, but I should be on IRC tomorrow if you want to discuss more.21:56
olofk_franck_: samanthafox has soemthing for you at https://www.dropbox.com/s/gbzvu98qdtx884a/or1200-basic.elf21:57
blueCmd_jeremybennett: Sounds good, thanks again21:57
_franck_olofk: thanks to her21:59
olofk_franck_: Got some warnings saying that phdr_num and shdr_num should be size_t instead of unsigned int21:59
olofkI can amend that to the patch22:00
olofkgot to sleep now.22:00
_franck_ok, thanks for testing it22:01
_franck_good night22:01
olofk_franck_: Just so you know, I'm not sure if those files were compiled with the or32 or the or1k toolchain. Don't know if that makes any difference22:01
_franck_I just tested it now and it works....22:02
_franck_(tested with pure C using main())22:02
_franck_http://pastie.org/862470122:03
_franck_well it seems ok with my local repo (http://pastie.org/8624707), I'll try with a clean copy from github22:07
_franck_it also works with github repos. don't know what to say...22:17
--- Log closed Sun Jan 12 00:00:50 2014

Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!