IRC logs for #openrisc Saturday, 2014-07-26

--- Log opened Sat Jul 26 00:00:36 2014
stekernblueCmd: can't it just be that the ethoc emulation in qemu is broken?07:03
stekernolofk: you sometimes want to be able to write to a full buffer because that means you can ommit the logic that prevents ut07:04
stekernI can't come up with a case where you *actually* want to write to the full buffer though07:05
olofkYes, I was also thinking that it might be easier on the control logic to allow writes, but otoh this messes up the pointers, so we probably have to do a clear/reset to get back to a known state07:27
olofkAnd the only extra logic in the fifo would be ram_we = fifo_we & !full07:28
stekernyes, but iirc, that can become a pretty critical path07:29
stekernthat's why you have things like almost_full signals, that assert when there are 1-2 slots left07:31
olofktrue. Critical path is probably the most valid argument here07:31
olofkBut in practice, the user of the FIFO will implement this check anyway08:47
blueCmdstekern: I guess, but I don't understand why some ssh sessions continue working09:14
blueCmdafter a weeks fixing09:18
blueCmd63 failures where the lockups potentially cause 1109:18
olofkblueCmd: For the complete gcc suite?09:19
blueCmdolofk: that's just gcc09:19
blueCmdnot g++09:20
blueCmdor ada and stuff09:20
olofkah ok, but still. Wasn't it like >200 a few days ago?09:20
blueCmdthat's why it took a week :P09:23
blueCmdonly one failure in gcc though, rest in glibc09:23
blueCmdonly one bug was in gcc though, the rest was in glibc*09:25
olofkany news on the gcc contributors btw?09:28
blueCmdolofk: not really, as it stands if we cannot reach Jungsook Yang we might need to rewrite or scrap part of the floating point stuff in gcc, not sure how to do that09:30
blueCmdalso I need to get hold of a guy which I apparently need to call09:30
blueCmdI'm going to send out 'Please fill in this form and send it in' things to the developers that has answered on monday09:31
olofkYeah, it's good to get that done at least. juliusb and I promised a few years ago to go out on a holy crusade to hunt down the last contributors09:32
blueCmdwe should find someone that wants to port Java/Haskell/Go to openrisc09:48
blueCmdi can be interested in go and java (in that order) but that will take time until I'll get around to it I think09:49
olofkHas anyone ever come across a program that is developed on windows but is supposed to work on linux, that actually works on linux?10:28
ysionneauyou mean portable code? or a binary you run with wine?10:41
olofkysionneau: Source code that you can compile on linux10:42
olofkThe code I'm looking at turns out to be full of Windows assumptions10:42
ysionneauI guess Linux or BSD developers are more used to portability concerns10:43
olofkLike looking for include files in d:\projects\whatever10:44
ysionneauis it a student project? :)10:44
olofkAnd windows-specific types10:44
ysionneauseems like whatever you read saying "it works on linux" is just plain wrong advertisment =)10:45
olofkhaha. No, it's actually a commercial project10:45
olofkYes. This doesn't work on linux at all I would say10:45
olofkI give up now10:46
stekernI feel like I'm leaving on the edge of technology, just installed one of these hdmi-cable thingies between my ws and monitor instead of the old trusty VGA-cable11:04
blueCmdyes, something is wrong with ethoc I think11:05
blueCmdI just got a lockup and interrupts continued for everything except ethoc11:05
blueCmdwhich sounds reverse to what I had the other day11:06
blueCmdmaybe it's some grander error11:06
stekernhmm, wonder if I really need to implement some phy emulation in my MAC<->MAC bridge11:08
stekernwhere the emulation would just be the standard mdio regs11:09
stekernchanging to hdmi cable actually made more of a difference than I had thought11:10
stekernespecially with my bad eyesight, I thought I wouldn't be able to notice much11:11
stekernhansfbaier: sublime is working fine, at least all the features I've implemented. missing are filters.11:57
blueCmdNick Krause..11:57
blueCmdHe just doesn't stop11:58
poke53281blueCmd: I have managed to mount the filesystem without errors. So far, nothing unusal.15:49
olofkDoes anyone else keep getting old OpenCores forum posts e-mailed?18:07
stekernno, but I don't get any opencores forum posts e-mailed18:59
olofkI'm not sure, but I think they are coming in chronologically. Just got one from Feb 20 2014 now19:10
olofkNahh.. probably just random. Got one from July 26 now19:15
olofkThe best things about verilog is the excellent string and memory handling19:54
olofkhmm.. looks like I can't simulate mor1kx-generic with modelsim20:27
stekernwhy not?20:27
stekerndid I break something?20:27
olofkHaven't analyzed it yet, but it just hangs20:27
olofkWorks fine with icarus20:27
olofkWeird. It just stops fetching new instructions after this 188:   c1 40 00 00     l.mtspr r0,r0,0x500020:36
stekernwhat's 0x5000?20:38
olofkIt's the or1200-simple testcase from orpsocv220:45
olofkNot sure if it's related, but at the same time, I get an undefined value on du_dat_o from mor1kx20:49
olofkHmm.. modelsim spits out some warnings. Port size mismatch for rdad_i in rf_cappuccino and this one from mor1kx_pic "Non-positive replication multiplier inside concat. Replication will be ignored."20:54
olofkNever seen the second one before20:54
stekernmaybe it sucks and can't understand 0{1'b1}20:55
stekerntry setting the OPTION_PIC_NMI_WIDTH parameter to something > 020:56
stekernlike 2...20:56
olofkThat fixed the last pic warning I think21:02
olofkBut it looks like there's strange things going on in fetch_cappuccino. Shadow register-related perhaps?21:03
stekern*shrugs* can't imagine that'd be related21:04
blueCmdstekern: did you see the new failure logs for gcc?21:05
stekerntry disabling du, and the shadow reg stuff should be disabled21:05
stekernblueCmd: yup21:05
blueCmdjeremy_bennett: - might be interesting for you, the torture ones are probably unrelated but the others should be real failures21:06
stekernolofk: yes21:10
stekernI think I'll need a fake phy21:11
stekernisn't there a dummy phy driver in Linux?21:11
olofkstekern: Setting OPTION_NUM_SHADOW_GPR=0 solved it21:17
olofkLooks like some width isn't made dependent on that21:18
olofkdecode_rfa_adr_i in mor1k_rf_cappuccino isn't adjusted for number of shadow RFs21:24
stekernyeah, it shouldn't21:27
olofkBut it goes straight into rdad_i in rf_ram, which is adjusted for number of RFs21:28
stekerndoes it still work if you enable the DEBUG_UNIT again21:29
olofkThat means wait, but takes less time to write21:29
olofkYep. that works21:30
stekernyes, I was lazy there, but decode_rfa_adr_i shouldn't be adjusted, rdad_i should be padded with zeros21:30
olofkThat's easy enough to fix then. I'll try that21:31
* olofk is warming up to the submodules idea21:31
stekerncan't imagine that modelsim wouldn't do that by default?21:31
olofkDon't really know what it did, but there were a a lot of red lines in the waveform21:32
stekernmaybe it 'x' them...21:32
olofkMight do, since there aren't any drivers21:33
stekernannoying f*cking event driven simulators...21:33
stekernif you are fixing it, you might as well put that stuff in the generate that is above the mor1kx_rf_ram instantiations, and call the signals rfa_rdad and rfb_rdad21:36
stekernand then hand over the patch21:36
stekernor.. doesn't have to be in the generate21:37
* stekern tired21:38
* stekern go sleep now21:38
olofkI changed it to       .rdad_i({{(RF_ADDR_WIDTH-OPTION_RF_ADDR_WIDTH){1'b0}},decode_rfa_adr_i}),21:39
olofkHow's that for readability? :)21:39
olofkSeems to work at least21:39
stekernheh, yes, that's why I want it in a seperate signal ;)21:39
olofkBut not for the 0 case. Yeah, a proper fix would be better21:40
stekern0 case?21:40
olofkRF_NUM_SHADOW_GPR = 021:40
olofk{0{1'b0} is one bit wide apparently21:40
stekernyeah, right, it got grumpy about that too...21:41
stekernstupid modelsim21:41
blueCmdstekern: your port to musl, is it feature complete?21:42
stekernapart from fenv, yes21:45
blueCmdpoke53281: appear to have a lockup in or1ksim now as well22:02
olofkTested with modelsim, icarus and verilator with OPTION_RF_NUM_SHADOW_GPR 0 and 122:29
olofkWorks with 2 as well22:29
poke53281Hmm, which compiler version do you use?22:30
poke53281gcc 4.9 seems to be broken for Linux compilation.22:33
olofkpoke53281: At least with -Os22:37
blueCmdpoke53281: ah right, that might be relevant - I actually read the thread but never thought that it might be related to this :P22:41
blueCmdI don't think it is relevant though22:41
blueCmdwe don't have red-zone and I'm using 4.1022:42
blueCmdfor or1k22:42
olofkHow does reading and writing SPI work from linux userspace?23:23
olofkIs it just presented as a char device?23:23
--- Log closed Sun Jul 27 00:00:37 2014

Generated by 2.15.2 by Marius Gedminas - find it at!