IRC logs for #openrisc Wednesday, 2013-10-09

--- Log opened Wed Oct 09 00:00:31 2013
-!- Netsplit *.net <-> *.split quits: larks03:22
stekern_franck_: odd, wonder what is so different in your setup06:13
stekernrunning a more recent busybox than one from 2011 is not a bad thing though  ;)06:13
stekernhansfbaier: nice06:34
hansfbaier_franck_: stekern:
hansfbaierstekern: still work in progress, but already quite a bit there, rest tomorrow I hope06:37
hansfbaiergot to do some paid work now06:38
hansfbaierstekern: Tutorial has been tested in xubuntu 12.04 VM06:38
hansfbaierstekern: You might want to get it now, just pushed a new version as I posted the link06:39
stekernok, I think I've already been helped by your DIP-switch explanation06:44
stekernI think I've looked at the switch and not the board when I tried to setup fpga loading from SD-card06:44
hansfbaierstekern: I would add the files you gave me to the tutorial repository, or do you think it would be better to use your links?06:47
stekernno, just add them there06:52
stekernI can add a header to the memloader first if you like ;)06:52
hansfbaierstekern: Yes, that would be nice06:55
hansfbaierstekern: Do you think it would be a good idea to include the ocfb in the HPS? Since the OpenRISC uses it too, I had a crash yesterday, when both tried to access it....07:12
stekernmy motivation for trying it out was two-folded, 1) I counted that connecting it to a completely different system might weed out some bugs in the driver, which it did 2) I wanted to see if it worked07:16
stekernbut since there are some problems with endianness and the transfers from HPS DDR3, perhaps it's not strictly necessary as a step in the guide07:18
stekernbecause that's what I guess you are getting at07:18
stekernbesides, you'd need my wip cleanup of ocfb.c version for it too07:19
hansfbaierstekern: see you later, need rest07:23
stekernme too...07:29
stekernunfortunately it's not time for that now...07:29
stekernI've been insanely tired after the weekend07:29
olofk_Please tell me if you find factual errors, bad language or any other fuck ups. It's been written piece by piece on the little spare time I could find, so it might be a bit messy07:50
xlrogreat write-up olofk_08:08
xlrothe other jeremy from the conference here, btw :)08:09
olofk_xlro: That answered the question I was just about to ask :)08:09
olofk_I never got around to explain in detail how test cases was fed into the simulator in ORPSoCv3. Sorry about that08:10
xlrono worries, I will definitely have a closer look at that at some point - looks like switching to orpsocv3 makes also total sense for our work, especially regarding ease of configurability08:11
olofk_Please let me know how it goes. I'll be happy to answer questions and provide support08:12
xlroatm its a bunch of randomly symlinked include files08:12
xlrook great08:12
xlrowhat I started doing is hacking the testbench to accomodate for simulating on gate level a place&routed design, since this is important for us - at that stage the internal signal names are mainly gone though, which breaks the available monitor08:14
xlroI'd love to do that in a proper form and feed that back08:14
olofk_I have related problems with timing constraints. The only way I can think about is making the synthesis tool spit out a list of renamed elements that can be used for other tools08:16
xlroyes I thought about that too - after DC we also go through encounter though, which alters the names again08:18
olofk_Are the renamings predictable? Can you place syn_keep attributes on the signals that the monitor tries to probe?08:18
olofk_It would still be a problem though, since I guess that the internal tasks that we have for getting GPR contents will be removed after synthesis08:18
olofk_Ahh.. a two stage renaming. That can be pretty messy I guess :)=08:18
olofk_It's good to know these things, as the monitor is one of the things that might need an overhaul in the near future. Would be great to make it more resistent to these kind of problems08:19
xlrogetting the gpr contents is pretty hard indeed I think (in an automated fashion) - but I would totally be happy with just a testbecnh that runs cleanly and dedects when the processor is done, which is not hard (looking on instruction bus)08:20
olofk_I've already made some local changes so that the some of the signal names comes in as ports instead of being hardcoded in the monitor08:20
olofk_Exactly. That feature was what I aimed for first in my stripped down monitor08:20
xlroatm what I do is get rid of ("uncomment" with a custom define) everything that references stuff that is not available anymore08:20
xlrountil it compiles08:21
xlroworks ok so far, but very ugly08:21
olofk_"Automated approach" <-----------------------------------------------------------------------X--> "Manual error-prone method"08:21
olofk_I put an X where you are atm08:21
xlrooh definitely, although there is not much space for error anymore I think because I remove 90% of the monitor basically with this method ;)08:22
xlroright now I'm still here in Cambridge at ARM, but I will continue on this when go back to EPFL end of november - probably one of first things I'll do08:24
olofk_xlro: Do you have something like this?
xlroI think there is a little more left, but effectively not much more, I think it also has the NOP report still in. I never looked exactly at the code that is effectively left, I basically just have conditional "defines" blocks everywhere where signal names are mentioned that are inside the actual core08:26
xlrosince these are gone08:26
xlroso the actual source fiel is actually bigger :P08:26
xlrobut the code is just not used08:26
xlrothat way I can use the same base for both simulations08:26
xlroand just enable one define08:26
olofk_Yeah, that's a clear win08:27
-!- Powermaniac_ is now known as Powermaniac08:48
olofk_Crap! I forgot to mention Saar Drimer's lightning talk08:56
olofk_Was that just after my orpsocv3 presentation?08:59
hansfbaierolofk_: talks + slides already online ?09:08
hansfbaierstekern: If I get it right, the linux-socfpga kernel does not need to be modified if used without the ocfb, right? What was the make argument to generate the default .config again?09:18
hansfbaierah got it, it's in the README09:19
hansfbaierno, that was OpenRISC.09:20
hansfbaierstekern: Well, actually, if we don't modify the kernel, the only reason to compile it is for USB OTG ....09:21
stekernwhich isn't neither strictly necessary to get started with openrisc on sockit09:30
hansfbaierstekern: How do I get a sane .config?09:37
hansfbaierSupply yours?09:38
hansfbaierI tried cp arch/arm/configs/socfpga_defconfig .config09:38
stekernmake socfpga_defconfig09:38
hansfbaierBut on make it started asking myriads of questions, so that defconfig seems to be out of date, or am I missing something here?09:38
hansfbaierstekern: ah09:38
hansfbaierstekern: still need to set CROSS_COMPILE, right?09:39
stekernperhaps, or at least ARCH09:40
stekernor then not... I can't remember09:40
hansfbaierstekern: I get compile errors if I don't09:43
hansfbaierstekern: it works when I do.09:43
hansfbaierstekern: Where do I get the dts, now, (forgot) use the one from the kernel tree?09:46
hansfbaierI probably could leave it as it is, since there was no further modification (as opposed to the ocfb...)09:47
hansfbaierwhere I needed it09:48
stekernyes, you just need to compile the dtb09:55
stekernwith make dtbs09:56
hansfbaierstekern: Isn't that done automatically, when compiling the kernel? the dtbs are already there...10:12
hansfbaier_franck_: pushed new version. Now you should come until sd card image with custom kernel10:35
hansfbaierstekern: I'm not so shure about the last mkpimage, but it seems to work for me, that's what I grepped out of my history ^10:38
PowermaniacAnyone in here happen to know how to program in Python?10:50
olofk_Have you seen the Kickstarter campaign for an LGPL GPU?
olofk_Looks like they listened to my advice that they should provide a Wishbone interface. I didn't say that they should do it for $600000 though :)11:14
Powermaniacolofk_: No I had not until now. Damn 200,000 goal, rewards better be nice.11:16
olofk_But I don't think they're gonna make it. Should have offered a dev board or something as a backer reward. You don't get a shit for the money unless you pay $5000 in which case you will get one of them to fly over and run their design in modelsim. Not very tempting11:18
olofk_Well, best of luck to them11:19
stekernyou get an usb stick though11:21
PowermaniacWow, I'm just frustrating myself by being stubborn and not giving in till I manage to do it (it being make a simple program that runs the Fibonacci sequence and gives you the result of the nth term the user input)11:28
stekernbut I agree with you olofk_, you only pay for it to be done.11:33
stekernwhich in itself isn't so bad, but probably not enough for people to really get hooked11:34
stekernI think that was part of what made parallella successfull, even if you weren't sure how devoted you'd be to the actual epiphany cpu, you'd still get a pretty darn good devboard anyway11:36
olofk_stekern: Exactly. They asked for ideas on Phoronix a while ago, and I gave them Parallella as an example. People like to get something physical (other than a USB stick)11:40
olofk_They are basically selling the IP rights now11:41
stekernmmm, so the only ones that will be mostly interested in that kind of deal is parties with commercial interest, not end users. And I get a feeling that a large amount of kickstart backers are exactly that, end users.11:43
olofk_Haha. I refreshed the backer page. It first said three backer/$325, and then it flashed and went back to 2 backers/$100 :)11:43
knzI think the guy is not presenting very well11:44
knzit looks more like a hobby project than a visionary project11:44
knz(good afternoon all)11:44
stekernafternoon knz11:45
Powermaniacgood afternoon11:45
olofk_They could at least have presented the option to get a off-the-shelf board with PCIX or something with a preprogrammed FPGA11:45
olofk_hi knz11:45
knzbtw I'm giving a talk on openrisc this afternoon11:45
knzstekern tech check: the only difference between espresso and pronto espresso is the delay slot, right?11:45
olofk_knz: Wow. That's great. Can we see it, or the slides somewhere? Always interesting to see these kind of things11:46
knzno prob11:46
knzhold on11:46
knzI will provide a more "permanent" link later11:47
stekernknz: yes, the only difference between espresso and pronto is the delay slot, as far as I know at least.11:48
knzexcellent thanks11:48
stekernthey are juliusb's babies, so he might want to fill in, but that's the main thing at least11:49
olofk_knz: oohh... I love your template. Haven't started reading though :)11:50
knzolofk_: standard keynote template11:50
olofk_We Linux guys are starved for eye candy :(11:51
stekernmy first thought too ;)11:51
stekernwhat's mgsim?11:52
knzOSX in virtualbox / qemu11:52
knzI was answering olof's comment11:52
knzfor eye cand on linux, run osx and keynote in virtualbox :)11:52
knzmgsim is our simulation framework11:52
stekernah, ok11:53
knzit's for system design space exploiration of many-core systems and custom cache protocols for large shared memory systems11:53
knzbut the flagship item is our own custom core micro-arch :)11:53
knzmore news at orconf 201411:53
olofk_knz: That's a really nice introduction to OpenRISC. A few number that could be corrected (like mainline Linux was in 3.1, not 3.8), but no biggies11:55
knzah thx11:55
knzwill fix straight away11:55
olofk_knz: I love that you are already planning a presentation for orconf2014. That's the spirit!11:56
stekernand we're not going mainline gcc for 4.911:56
knzof course11:56
knzstekern: I thought that's what joern was saying11:57
olofk_knz: We discussed it in more detail last year. The big issue is that FSF require us to give them ownership of the code, and there are many contributors that need to be tracked down for that11:58
stekernhmm, maybe he was speaking about something else11:58
stekernknz: I think poke53282 would prefer this link to his jor1k simulator:
olofk_And at least one of them might refuse which makes it complicated because we have to find out which parts to rewrite then11:58
knzstekern: thx11:59
knzok time for my talk11:59
olofk_Break a leg11:59
PowermaniacIs that to do with the non-sense about how the FSF has certain terms you have to abide with to get on there officially open source list or whatever the non-sense is?11:59
PowermaniacBye knz!11:59
PowermaniacSorry I was referring to this: "knz: We discussed it in more detail last year. The big issue is that FSF require us to give them ownership of the code, and there are many contributors that need to be tracked down for that"12:08
stekernfor FSF to accept code into mainline gcc, they require that copyrights are re-assigned to them12:12
PowermaniacYeah that is kind of ridiculous...12:12
stekernso, you'll need permission from all copyright holders to do that12:13
PowermaniacWell, I'm not bothered by the second part as much as I am the first.12:13
stekernwell, without the first, the second wouldn't be a problem :)12:14
PowermaniacSeems rather silly, as like if the work is open source anyway they don't need the rights assigned to them...12:14
stekernI tend to agree, projects seem to survive without enforcing such rules...12:15
stekernlays out the reasoning behind it12:20
amsPowermaniac: the FSF has no open source list12:29
Powermaniacams: ?12:30
amsPowermaniac: the FSF does free software, open source is something different.12:31
Powermaniacams: I'm pretty sure they mean free as in freedom not free as in it costs nothing...12:31
amsPowermaniac: yes, and we call it "free software"12:32
amsPowermaniac: open source is a different movement with different goals12:32
Powermaniacams: Hmm okay then12:32
amsPowermaniac: the reason for copyright assignments is so one can enforce the license, if you have 1000 licenseors then you need to track them down when going to court or whatever12:33
Powermaniacams: Ahh right the free-sw defined by RMS12:33
amsPowermaniac: no, free software as defined by the GNU project adn the FSF.12:33
stekernams: sure, I understand the reasoning behind it. but in practice, do projects that doesn't enforce this rule have this problem?12:36
amsstekern: busybox had this problem, linux had it as well, and now requires something similar to copyright assignments12:36
stekernyou are referring to the signed-off?12:37
stekernyeah, but isn't that more of a protection the other way?12:38
amsit is basically a "copyright assignment" where by "signing off" you agree to some terms12:38
amsstekern: hm?12:39
amsThe sign-off is a simple line at the end of the explanation for the12:39
amspatch, which certifies that you wrote it or otherwise have the right to12:39
amspass it on as an open-source patch.12:39
amsblech ,sorry ...12:39
stekernyou promise that the code you submit (or at least it's on your neck if it's not) is entitled to be released under the license conditions the kernel rules under12:40
amsstekern: yup, which is also what the fsf copyright assignment has in its terms... it is more of a contract as well.12:40
stekernso no-one can come and say "that's our code, it wasn't intended to be released under GPL"12:41
amsstekern: yeah, linus was short sigted as usual.. the signed off isn't as strong as what the FSF does, or mozilla, apache etc ..12:41
stekernyes, but the point made in the link I posted was to ensure that if anyone takes the code from the FSF project, FSF can take legal actions against the offender.12:42
amsstekern: it kinda protects linus/linux from idiots like SCO but nothing else12:42
stekernI don't think it was out of short-sightnedness, but rather out of purpose12:43
ams(it also allows linux&linus to in theory make the code non-free software ...)12:43
stekernI don't think he cares about it being "strong" in that sense12:43
amsshurg, linus doesn't care about freedom.12:44
amsanyway ... sorry, offtopic i guess this is..12:44
stekernnah, I believe it's in scope of this channel :)12:45
PowermaniacHa, I see what you did there12:45
amsi don't think that hunting down all the copyright holders should be to hard12:45
PowermaniacLinux can make Linux non-free heh. News to me12:45
PowermaniacAlthough a lot of this stuff is news to me anyway...12:46
stekernbut freedom is always relative, personally I prefer to give away my code with the freedom for people to do whatever they want with it12:46
stekernbut that's just me, of course if I contribute to a project with another agenda, I follow their wishes12:47
Powermaniacstekern: Just if you could actually make decent money doing that...=\12:47
stekerncompanies seem to contribute back to projects for other reasons than just the license nowadays anyway, they see the beneficial of having their code being co-maintained by the communities12:51
olofk_Gaisler is a prime example of this with their dual-licensed Leon core. Individuals and universities use the GPL version and feed back fixes, but they know that no commercial ASIC (or FPGA) wants to touched by GPL code, so they sell them a commercial license for a lot of money12:56
olofk_The price was raised quite recently (rumour alert!), which is one reason why the space industry is looking towards OpenRISC for example12:57
PowermaniacOh really12:57
knzwell not only that13:01
knzgaisler has changed management13:01
knzand the new sales guys are long-toothed pricks13:01
PowermaniacHi again knz, how did it go/13:01
knz(I worked with gaisler on a previous EU-funded project)13:02
Powermaniacknz: I feel even luckier now that I've heard that, as I found this community instead.13:02
knzthe SPARC ISA13:03
knzseriously, guy13:03
knzit can't get more 80s than that13:03
knzsparc register windows, OMG how I hate them13:03
knzand the double loads and stores13:03
PowermaniacDo you mean like the OpenSPARC T1?13:04
olofk_Did you know that both the MicroBlaze and the Leon CPU are from Gothenburg, Sweden?13:07
olofk_There should be a lot of competent CPU guys here, but I haven't found any so far13:08
olofk_Or maybe they are still busy doing 80's CPU designs :)13:09
stekernI didn't know microblaze (although now when you mention it, I think you've mentioned it before)13:10
olofk_stekern: I learned that myself quite recently from Elektroniikkalehti13:12
stekernI think it says on the outdated page of Damjan Lampret that Xilinx considered openrisc before they went onto microblaze13:12
olofk_Ha. That's interesting13:13
stekernI'm happy that they didn't13:13
stekernmake it happen13:13
olofk_I think Lattice was thinking about changing too, but with the renewed interest in lm32, I guess that they are happy to keep that13:13
stekernwas it really on elektroniikkalehti or on elektroniktidningen?13:15
olofk_stekern: Sorry. I should have changed the finnish of that sentence13:16
stekernyou are truly the master of puns ;)13:17
olofk_Happy to be of service :)13:17
stekernI try and try, even in my presentation, but I'll never get to the high grounds you are at ;)13:17
olofk_I think you're a natural.13:18
amsolofk_: competent people in gothenburg_13:19
amsyou moved to the wrong town ... everyone is called Ada and Glenn and have bad jokes.13:19
olofk_Amen that13:20
olofk_stekern: Yeah, Mr. Bilski didn't like my OpenRISC comment :)13:23
stekernI heard the real master of puns on finnish-swedish radio though, they were discussing why paralympics wasn't more receiving more attention, then this guy burst out in finnish-swedsih accent "Ja, men det är ju para lympics"13:29
stekern(sorry all non-swedish speakers about that)13:30
olofk_haha. Was that intentional? :)13:33
stekernI believe so ;)13:34
stekernthere was a pause between the words13:35
knzhi again13:40
knzI got some interesting questions after my talk13:41
knzone of which was: what protection is there against people accidentally implementing openrisc using patented micro-architectural features?13:41
amsknz: First rule of fight^Wpatent club, do not talk about fight^Wpatent club.13:43
olofk_knz: Interesting question that indicates that people were awake during your talk :) . None, I would say.13:43
olofk_We have been worried about threading on ARMs patents when we want to implement 16b operations13:43
knzmy answer was 1) not enough money in the pocket of openrisc users to invest in litigation and 2) openrisc contributors don't care about that13:44
olofk_One thing though, is that Flextronics (who invested heavily in OpenRISC up tto 2005) had lawyers that went through the license, and I guess that the code was analyzed too at that point13:44
knz"the code"13:45
knzwhich code?13:45
olofk_And the peripherals they used. Which is mostly the same we use now13:45
olofk_I don't think there is any trouble with the current or1200 since Samsung and Allwinner would have been easy targets in that case13:45
olofk_But it's uncharted territory13:46
knzok but the quesytion I received was more targeting mor1kx :)13:46
olofk_And the license/patent discussions have been more active recently. juliusb invented his own license for mor1kx13:46
knzyeah I saw that13:46
olofk_license != patents, I know13:47
amsi suspect it is as it is with software ...13:47
ams"don't read patents, and hope for the best"13:47
olofk_But as you say, I don't think it's a problem in reality with individuals hacking together new stuff13:47
knzanother thing I discussed with colleagues: the mmu spec is waaaay too complicated13:48
olofk_Something for or2k perhaps? :)13:48
knzBUT, the idea to prefix virtual addresses with a CID is excellent13:48
knzor2k won't have a mmu afaik13:49
knzor so I've understood during the last orconf discussion13:49
stekernams: I think that summarize it pretty well, we try to avoid widely known patents, but we don't go out of our ways to find any we don't know about13:49
olofk_or2k is a floating target. It's vaporware right now, so I think it's unfair to say it won't include a mmu13:49
stekernknz: the MMU chapter is a big mess in the spec...13:49
knzstekern: yes :)13:49
olofk_stekern: Is that fixable with minor changes to the spec and code, or are we doomed?13:50
knzolofk_: I think you could replace the part of the spec entirely13:50
knzeven if you were to redesign the ISA you could "bake in" backward compatibility via exceptions13:50
knzthe idea that exceptions disable MMU translation is also excellent by the way13:50
stekernI really would want to clean it up and bring in changes concluded in this ML thread:
knzit reads like a simplification but I'm no expert enough to judge13:53
stekernolofk_: I think the changes are relatively modest, at least if you look at it with the attitude that it currently contains a lot of cruft that no-one use and likely no-one will ever use it13:55
knzspeaking about cruft...13:56
knzORVDX64: hmmm....?13:56
stekernwell, vector instructions could be interesting, but I don't think anyone reviewed them. I can certainly not comment if they make any sense at all13:59
stekerncompiler research seems to be directed toward that alot at the moment at least, so for that it could be useful.14:00
knzI'm not sure how good it is to make a spec beffore anyone cares to think about an implementation14:01
olofk_knz: Got a spare student you can chain to computer and implement that part?14:01
stekernperhaps that kind of researcher would have preferred it not been defined though...14:01
knzyes :)14:02
olofk_knz: No, I think you're right, and I guess that's part of the agile movement. Don't invent things you're not going to use14:02
knzth good part is to use the same registers in the ISA14:02
knzbut of course that precludes multi-word 32-bit and 64-bit FP14:02
knzseriously the VDX right now seems specialized towards sound and graphics14:03
knzoh well14:04
olofk_I'm off14:23
AmadiroI haven't been following along very closely lately, has the ORPSoC or some other form of or1k made it into ASIC form that you can buy yet?14:46
AmadiroHm, wikipedia lists some instances, but nothing for general availability14:48
knzAmadiro: the "AS" part of ASIC stands for application-specific. Chances are that commodity products you use daily contain some openrisc stuff in them but that you will never know, nor meant to know.15:14
Amadiroknz, right. I was thinking something in the form of an eval board or somesuch, that the end-user can program.15:15
knzwhat you probably mean is, is there any dedicated silicon for an openrisc-based processor15:15
knzthen the answer is, not for end-users I believe15:15
knzbut the reason is more that it's not the primary intended goal of the openrisc project, than that it is not possible15:16
knzcustom silicon means10k upfront investment (at least), sometimes north of 50k15:17
knzit's a bit tough investment for a platform that changes over time :)15:17
knzbut who know, with kickstarter nowadays anything is possible15:17
AmadiroYeah, I realize that. Thought someone may have done it anyway, for some sort of embedded SoC, maybe a high-end microcontroller, something like that.15:18
AmadiroSomething end-user programmable, at any rate15:18
knzI think NXP or Allwinner have standalone chips15:18
knzyou can uy on catlogue15:18
knzmaybe not in small quantities though15:19
AmadiroCan't find anything immediately, but I'll look around, thanks.15:21
--- Log closed Thu Oct 10 00:00:33 2013

Generated by 2.15.2 by Marius Gedminas - find it at!