IRC logs for #openrisc Monday, 2014-12-01

--- Log opened Mon Dec 01 00:00:21 2014
--- Day changed Mon Dec 01 2014
poke53282Finally ;)00:00
poke53282good, looks like I get it down to 13MB, if I remove every file, which is not accessed during a "make ARCH=openrisc"03:29
poke53282So 5MB seems possible when I do a little bit more bash magic.03:31
poke53282Bad idea. It takes more than 3 hours to compile the kernel. At least it works.07:29
olofkpoke53282: How can it take longer time?10:06
olofkI'm having a tricky wishbone issue. Help appreciated. I want to send data on every 2nd clock cycle (from my 32->64 bit upsizer). The only way I can come up with is to keep cyc and stb asserted and set cti to 111.10:08
olofkBut what I really want to do is to let cti be 010 and just mark every other word as valid. Is that even possible?10:09
olofkI thought that lowering stb in the invalid words would work, but reading through the spec, I can't find any evidence that it should work10:10
stekernwhy not?10:10
stekernafaik, that's the only way to do what you are describing10:11
maxpalnHi, does anyone know if Mentor's NucleOS has been ported to the MOR1KX (or OR1200)? - they don't list it on their webpage but they caveat that by saying the custom processor's may be supported but not listed.11:13
stekernnever heard of it11:27
maxpalnstekern: thanks - yeah, me either. But we've had a request. It's an RTOS developed by Mentor Graphics - not sure of it's advantages or why it is required. I am guess the experience of the engineers.11:39
stekernI took a quick peek, seems like I need a company with a revenue >1M$ to get the sources11:41
stekernerr <1M$11:41
maxpalnhmmm, I didn't spot the request for downloading the source. Well, this is a pretty big company they are definitely open to paying for porting work - we are going to propose based on the ORPSOC and some porting of the OS. I know we have a SW contract houses that we've used for this in the past - any other recommendations very welcome.12:07
poke53282olofk: Last week I included some scripts in jor1k to compile a few things like lua, cmatrix and so on. Yesterday I tried to reduce the size of the kernel in order to compile it on jor1k. Unfortunately that is a little bit slower than expected. I hoped, that I can compile withing half an hour.17:38
poke53282Well, I was wrong.17:42
poke53282I can still try to port tcc (Tiny C compiler) to compile it.17:55
olofkmaxpaln: I did some contract work for a company that used Nucleus. No one there could explain why on earth they used that instead of a free RTOS, so I guess it all comes down to legacy code. I think you will have a hard time finding a port since it's proprietary19:38
olofkpoke53282: Ah ok. I see19:38
olofkstekern: Yes, I assumed that fiddling with stb was the way to go, but I can't actually find it anywhere in the spec19:40
olofkAnd honestly, I haven't ever seen anyone using it. My BFM doesn't handle it, so before adding support for that I wanted to check that it's a real use case19:40
olofkAnd also, should the slave keep the ack signal asserted even when stb is lowered?19:41
olofkhmm... observation 3.10 answers the last question19:44
olofkBut that just makes it more tricky19:44
olofkstekern: Or did you mean that setting cti to 111 was the only way to do it?19:46
olofkAhhh. the wishbone spec is so fucking horrible19:55
olofkNow they suddenly introduced a signal called lock_o, that is only used in one place and seems to do exactly what cyc does19:55
olofkok, so now it's suddenly ok to keep ack asserted even if stb is lowered20:04
olofkAt least for bursts. I should probably stop reading now, because this is what I was looking for20:04
olofkAh right. The sel signal isn't allowed to change during a burst. That's a good idea.... said no one EVER!20:18
stekernno, I meant stb20:18
olofkSomeone should just throw wishbone under a bus20:18
olofkstekern: Yes, it seems like the right way to do it20:19
stekernI don't understand the problem, just de-assert stb when you have no request20:19
olofkProblem is that my BFM is cheating a bit there and doesn't look at stb20:20
stekern...and you can of course only de-assert it once the slave has finished the (previous) cycle with an ack (or err)20:20
stekernyour BFM isn't very wishbone compliant then ;)20:20
olofk...or rty. That's another great signal that gets used a lot20:21
olofkNo. Lot's of trial and error to get that thing working at all. The code is a complete mess (which maxpaln discovered some time ago)20:21
olofkRight now the problem is that I only want to write a half word on the first access, and full words later in the burst20:22
stekernrty could maybe be useful... if anything actually supported it ;)20:22
stekernI think my SDL2 support for scummvm is about ready to submit...20:23
olofkI assume that mor1kx has some fancy logic to resend requests on rty :)20:23
olofkstekern: Cool!20:23
stekernno, it doesn't use it20:23
olofkI think I've seen some slave that just hooks up rty to reset, but no masters ever care20:23
olofkBut we must amend the requirement that sel isn't allowed to change during a burst. That's just stupid, and I'm pretty sure that all slaves look at sel for every cycle anyway20:24
olofkI should send out a request for that on the wishbone mailing list. Just a bit worried that it will get lost in all the noise there20:41
stekernmmm, I know what you mean, I had to filter all those mails into a seperate inbox20:57
olofkIs really sel used for reads btw?20:58
stekernI've never used it for that, but I think I've seen slaves use it21:00
stekernwhere the end interface was 16-bit21:00
olofkah ok. Yeah, that would make sense21:01
-!- Netsplit *.net <-> *.split quits: wallento, sb0, vagn, poke53282, jonmasters, ams, hno, knz, mboehnert, Amadiro, (+1 more, use /NETSPLIT to show all of them)21:35
ysangkokstekern: are you skristiansson on github?22:02
ysangkoki'm having an issue with musl-cross22:02
ysangkokseems that i may have an incompatible version of lex/yacc22:03
ysionneau21:58 < olofk> Is really sel used for reads btw? < my minicon sdram controller uses it22:06
stekernysangkok: do you have flex installed on your system?22:13
ysangkokstekern: no but it didn't fail on it, so i disregarded that warning... i'll try with flex now22:14
ysangkokstekern: failed again... with flex22:14
stekernwhat is the error now then?22:16
stekernthe same but without the error?22:16
stekernwithout the warning I meant22:16
ysangkokstekern: i tried also redoing instead of invoking make myself, but the result is the same, less the interleaved stuff from different make targets22:17
ysangkokstekern: it seems i may need to run distclean.. i'll try again
ysangkokmy goal is to try and get the Trinity syscall checker running in jor1k... compilation is too slow in jor1k22:21
ysangkokis there no prebuilt or1k cross-compiler that has the linux specific headers?22:21
ysangkokstekern: it works now! thanks! i think it would be useful with a hint in the would you accept a pull request if i were to submit one detailing this gotcha?23:01
stekernsure, but the is just what came with the original musl-cross23:03
ysangkoktrinity needs the kernel address and a few other parameters :
ysangkokdoes anyone happen to know them? or how to find them?23:04
--- Log closed Tue Dec 02 00:00:53 2014

Generated by 2.15.2 by Marius Gedminas - find it at!