#milkymist IRC log for Monday, 2011-06-20

wolfspraulaw_: do you understand the L19 = 0ohm fix Werner found on Friday?01:34
wolfspraulI've added it to rc2 known issues at http://en.qi-hardware.com/wiki/Milkymist_One_RC2_Known_Issues01:35
wolfspraulI guess we still have to update the schematics?01:35
wolfspraulsince I need to fix my board somehow, do we have a placement drawing somewhere?01:35
aw_rc2 L19 we mounted with wrong bead with its DC resistor has 0.1 ohm resistor not as rc1 0.01 ohm resistor.01:36
aw_so current rc3 incoming parts already with correct type of bead with 0.01 ohm01:38
wolfspraul0.01 ohm same as 0 ohm? are we sure this is enough for the fix?01:39
aw_a resistor's value defined as 0hm, need to consider its tolerance, so for example, a 1% tolerance define as  0.01 ohm.01:39
aw_wolfspraul, no the answer is that i am not sure!01:40
aw_since sebasiten's rc2 board he had have that problem, but i don't, you may be no, i don't know. so what i want to say is a soldering as "short" is better than using a 1% resistor.01:41
wolfspraulwhy don't we just solder it short in rc3?01:42
aw_thus as i suggested we relayout it as "short" in rc4.01:42
aw_so i susggested to manually soldered all is as "short". :-)01:42
wolfspraulwhat's better? manually solder short 80 units, or 0 ohm resistor?01:43
aw_since it was a nice catch, but nnot just using 0 ohm. short 80 units. :-)01:43
wolfspraulok fine by me01:44
wolfspraulaw_: do we have a placement drawing? where is L19?01:44
aw_becuase currently no on  knew rc1 has this problem potentially existing, so as a better idea that we directly manully solder it as short in rc3.01:44
aw_L19 is located under MIDI RX connector and at the left side of U6 silk screen.01:53
wolfspraulhow did Werner upload this?01:53
wolfsprauland in his tmp folder?01:53
wolfspraullooks like a fairly useful thing to me, maybe we should upload into the wiki and link to it in appropriate places? unfortunately with Altium our process on these things is not so smooth... all needs to be done manually :-)01:54
aw_that one is genereated by his smart "prettygerbv" script under EDA tools project01:55
wolfsprauloh I see. it's created from gerber files?01:56
wolfspraulyes, I found L19, thanks!01:56
aw_no problem I'll have a rc3's pcb wiki page later, so i can link his images. :-)01:56
wolfspraulyes definitely01:56
aw_yes, created from gerber files.01:56
wolfspraulthis one is very nice. only not searchable.01:56
aw_sure sure i'll create a rc3 pcb wiki page. :-)01:57
wolfspraulgood idea to create this kind of thing from gerber files. I like it.01:57
wolfspraulat least that's a standard, so if things can be created out of that - good01:57
aw_then it's easily to find them with real pcb image the I scan them.01:57
wolfspraulaw_: so it should be fairly easy to remove L19 and short it, right?01:58
wolfspraulI will try one day :-)01:58
aw_yes, it's easy for me also I can let smt vendor to short it not me. :-)01:58
aw_btw, i'll still order a 1206 package with ohm +/- 1% resistor today.02:00
wolfspraulwhy? not sure it's needed...02:00
aw_no no, for me just samples only. :-)02:01
aw_so xiangfu's rc2 also have this problem?02:02
aw_wolfspraul, the black one of remoter, you'd order them if they have 100pcs?02:03
wolfspraulaw_: not yet02:06
wolfspraulaw_: yes, xiangfu also has the L19 problem02:06
wolfspraulxiangfu: so you can reproduce the camera-signal-loss bug on your rc2 board, right?02:27
xiangfuwolfspraul: yes. plug, then unplug the camera, the video-in gone.02:33
xiangfuI mean. re-plug the camera, the video-in will not working.02:34
wolfspraulok great02:35
wolfspraullet's try the L19 medicine02:35
wolfspraulyou know where L19 is?02:36
wolfspraulremove and short it :-)02:36
kristianpaulaw: can i guess L19 is easy to remove with a solder iron?02:47
kristianpauli think yes, but was not the same asumption when i was trying to do the same witn a  capacitor in a cell phone board..02:48
wolfspraulkristianpaul: should be easy, yes02:50
awbut you probably need to soldering a big blob on L19's two terminal pads at the same time, then you can soldering(remove) out.02:50
wolfspraulbut maybe you wait until xiangfu tells us what happened to him :-)02:50
wolfspraulhave you seen the actual bug? loosing camera signal, need to reboot m1 to get it back?02:51
kristianpauli realize i need twizzers..02:51
wolfspraulI'm pretty sure it's related to the power supply of the camera. I had this problem for months, but now I cannot reproduce it anymore. And I just switched to a new camera power supply and discarded the old one...02:51
kristianpaulyes that bug is nasty02:51
wolfspraulah ok, so you've seen it02:52
kristianpauli just used to it after a while..02:52
wolfspraulyes, for someone who really wants to use the camera, especially at a real party/event, that's the kidn of thing that can drive you crazy02:52
kristianpauli tought was software really, never hw..02:52
wolfspraulit could be fixed in software, by implementing a watchdog and hard-resetting just the chip02:52
wolfspraulbut that's why we have small runs, slowly ramping up the product, etc. right now it's easy to remove/short L1902:53
kristianpauloh, interesting02:53
wolfsprauland on future boards of course it will be fixed right away02:53
wolfspraulso I don't think there is a need for a software watchdog02:53
wolfspraulif we had thousands of boards and hard to reach users, of course we would do the software watchdog as well02:53
wolfspraulwell, do you want to try in parallel with xiangfu ?02:54
wolfspraulactually we have very few confirmations of the fix so far, so it's a good thing if people who can reproduce the bug report that the fix indeed fixes their problem02:54
kristianpaulnot know, may be tommorrow, reading some datasheets right now02:54
kristianpaulbut sure i confirm the bug is there in my side too02:54
wolfspraulso far we only have analysis of the root cause, and Werner and Sebastien agree on the agnd/dgnd issue02:55
wolfsprauland Sebastien was able to reproduce the bug consistently, and shorting L19 fixed it02:55
wolfspraulif xiangfu and you can add two more confirmations of the fix - awesome02:55
wolfspraulme or Adam cannot currently reproduce the bug, so we also cannot verify a fix :-)02:55
kristianpauli wish i could fix my ac97 chip as well ;)02:56
wolfspraulyou mean the noise?02:56
kristianpaulah, you mean neither you or adam have a mm1?02:56
wolfspraulis it hard to rework the chip to wolfson?02:56
wolfspraulno we have mm1, of course02:56
wolfspraulbut I just wrote it above :-)02:56
wolfspraulI think the bug first of all needs a leaking power supply to show up.02:56
wolfsprauland as it happens, I just replaced my camera power supply a few days ago, and, of course, discarded the old one02:57
wolfspraulnow I cannot reproduce the problem02:57
wolfspraulAdam is probably using a lab power supply, and he also cannot reproduce it02:57
kristianpaul(rework) well i need a solder statio and some exprimentation in other boards before try do it on my mm1 ;)02:57
wolfspraulbut there is no need for many people to reproduce it, we are quite confident of the analysis and fix, and if xiangfu or you (who both can reproduce the bug), can verify the fix, then perfect, done.02:57
kristianpauloh, i see (power supply)02:58
wolfspraulI forgot which modifications exactly we believe lead to less noice on rc303:00
wolfspraulwas it the wolfson codec?03:00
wolfspraulor some other changes?03:00
kristianpauli think the new codec itself imply other changes..03:03
kristianpaulin any case i dont noticed noise when was at labsurlab, or more people complain about it..03:03
kristianpaulwell i was using an iphone in the line-in03:04
kristianpaulbut now i dont have an iphone to try,03:04
kristianpaulalso in labsurlab the audio out was ampliflied for all the place, surelly that explain why noise was gone..03:05
wolfspraulI think you should try to fix L19 first, if xiangfu confirms it03:05
kristianpaulsure :003:05
wolfspraulthat's easy and a small nasty bug, especially when you try to use m1+camera at a real event03:06
xiangfuwolfspraul: hi03:37
xiangfuafter short the L19, I tried 30 times. everytime I re-plug the camera, the video-in works just fine.03:38
wolfspraulaha, perfect!03:46
wolfspraulxiangfu: thanks a lot for trying, and also we have another rc2 fixed ;-)03:47
xiangfuI also tried re-plug camera very fast, works fine.03:48
wolfspraulsure of course. it should be rock solid. we are making a professional product...03:48
Action: kristianpaul need a nap03:51
wolfspraulkristianpaul: np. did you see xiangfu's fix?03:51
wolfspraulworked perfectly fine03:51
wpwrakwolfspraul: you need to follow the backlog more closely ;-) the gerber converter adapted for M1 is here: http://projects.qi-hardware.com/index.php/p/eda-tools/source/tree/master/fab/m104:23
wpwrakwolfspraul: needed a few tweaks and hard-coding of an offset between the drill file (excellon) and the real gerbers, though.04:24
wolfspraulI saw something, but it didn't register with me how cool it actually was :-)04:48
wolfspraulI like the idea of first trying to extract new perspectives into the design from standard file formats.04:48
wolfspraulwpwrak: you saw that xiangfu made another positive confirmation of the L19 fix? very good.04:49
wpwrakyeah. looks pretty much solved04:49
wolfspraulthat makes it two people who went from being able to repro the bug to not having it anymore.04:49
wolfsprauland kristianpaul will probably/hopefully be the third, at some point04:50
wpwrakaccording to ADI's design guidelines, shorting L19 isn't ideal (the bridge between the ground planes should be underneath the video chip), but it's close enough04:50
wolfspraulwe should start the rc3 errata already04:51
wolfspraulas long as it's just one thing I can remember it :-)04:51
wpwrakhehe :)04:52
lekernelaw, aren't you able to source 0 ohm "resistors" that are just bits of metal?07:49
lekernelthat datasheet you pointed looks more like resistors to make current probes and such07:49
awlekernel, yes, I've not found out a metal one. won't you want to use soldering "short" L19?07:52
lekernelthat looks more work-intensive than just throwing a reel of 0 ohm resistors into the pick and place machine07:52
lekernelbut well... up to you07:53
awsmt factory can do this for me while producing brds.07:54
lekerneli'm planning to do quite a bit of publicity after this run is out, so hopefully they will sell fast07:54
lekernelwhich means, it's better if we can produce them fast (by just re-doing the run 3 design) as well07:54
awwith soldering.07:55
awyeah...we all hope that07:55
lekernelaw, do you happen to know how to use a chip wire bonding machine?08:32
awlekernel, i saw bonding machines with Wolfgang before, but how to use? I don't exactly know too much.08:34
awmy last time to see it with golden bonding machine while producing avt2 run1 with 0.7 mil diameter golden wire.08:36
lekernelpeople at a hackerspace in Paris scavenged one, but need help with using it08:36
awseems that have two catogories on golden and aluminum wires, golden wire it uses an electric arc to heat a platform to heat golden wire.08:38
awthe aluminum i didn't see it carefully. but too bad that here bonding vendors din't allow us to take video on machines though.08:40
awwell...can't help you anything. :-)08:41
lekernelinteresting, they have an ADV7181C as well10:05
lekernelseems it's a drop in replacement of the B, except they have integrated the antialias filter10:06
lekernelhttp://www.analog.com/en/analog-to-digital-converters/video-decoders/adv7181c/products/product.html doesn't say "not recommended" like the B10:06
lekernelit's probably easier to switch to C than to switch to 718010:06
lekernelthe C also supports digitizing VGA (but the software and fpga design won't)10:09
GitHub6[rtems-milkymist] sbourdeauducq pushed 1 new commit to master: http://bit.ly/ingkXz13:05
GitHub6[rtems-milkymist/master] Use __asm__ instead of asm for improved c99-compliance. - Sebastien Bourdeauducq13:05
lekernelah, as always anal ralf doesn't like my patch13:15
lekerneli should offer him this t-shirt for his birthday http://images4.cpcache.com/product/psycology-humor-black/318452434v4_225x225_Front.jpg13:16
Fallenouyes sometimes he's a pain13:29
carlobarhi, i have modified the HPDMC controller to support 16 bit memories. Now memory tests pass and bios is executed, but when i try to read the memory (executing mr), just the the 16 most significant bits are ok. The least significant bits printed are wrong and do not change after rewriting data. Can someone help me to solve that problem? thanks16:19
kristianpaulhow are you rewriting data16:21
carlobarwith mw16:21
kristianpauland from wich addres are you readin at first?16:21
kristianpaulsure, but, the exact command you used16:22
carlobarsdram is at 0x400000000, so i execute something like: mw 0x40000000 0xabadface16:22
lekernelnext time some nerd says M1 is expensive i'll point them to this: http://www.picocomputing.com/E101/16:29
kristianpaulcan tou try read  0xe000103c ?16:29
kristianpaulvalue should be the one you defined in system.v for systemid16:31
carlobari dont have devices on 0xe0000000 address, but csrbrg is at 0x8000000016:32
kristianpaulwhats the address of the system controller core then?16:32
kristianpaulis yor code public?16:32
kristianpaulalso that 16bit suppport :-), i think i have an altera board to test it too16:33
kristianpaulah i see16:33
carlobaryes, but isnt updated16:34
kristianpaulok, but you already tried to read systemid for your SoC?16:35
carlobaryes, i got 0x53334558, that is S3EX in ascii..16:36
kristianpaulso if you implement the bram core for example (is ti on the avnet milkymist port)16:38
kristianpaulcan you repeathat mr and mw commands on it16:38
kristianpaulto be sure you issue is not related witht the HPDMC16:38
kristianpaulsorry, i'm not very related to this core yet to help you more16:38
kristianpaulah, well for easy you could implement a general porpuse register in sysctl, you can confirm readin and writing is OK16:39
kristianpaulbut nice you ported mm1 to this board !16:39
carlobaryes, but now i have to port linux :-D16:40
carlobarwritings, readings on bram are ok16:41
kristianpaulgood !16:41
carlobarthat problem is only on hpdmc.. but memory test on sdram pass..16:42
kristianpaullekernel: (pico) cute board, but indeed a bit pricy16:45
kristianpaularghh, again that nasty cypress controller..16:53
kristianpaula basic device mode support for the mm1 will be nice17:12
kristianpaulJFDI :-D17:12
Action: wpwrak in iDiot mode: lekernel: (picocomputing) but it's cheaper than the M1 ! it has an FPGA, so it's the same ! ;-)17:49
lekernelwpwrak, the picocomputing thing is $39917:54
lekernelit's more expensive than the m1 :)17:55
wpwraklekernel: isn't M1 USD 500 ?17:55
lekernelthat'll be the package, the board alone (which we have now) is cheaper17:55
Action: kristianpaul try to remenber how much paid for his m1 board17:57
wpwrakaah, i see :) bah, such a big board without a nice case. that's so /technical/. and where's the app store ? ;-)17:58
kristianpaulyou mean patch library?17:58
kristianpaulso you dont need bother for apps, all work out of the boxes, and you can get cool patches from free ;-)17:59
kristianpaulalso there are special edition patches, that you can easilly buy18:00
wpwrakkristianpaul: an app store ! where i can buy apps blessed by Steve Holy Jobs :)18:03
kristianpaulSr, the only app you run in milkymist is the VJing stuff called FLickernoise,18:04
wpwrakkristianpaul: you techies talk funny. "patches". that's why this stuff sucks !18:04
kristianpaulhow i should call it then?18:05
wpwrakkristianpaul: 1 app ?!? we have a hundred millions in the app store !! aah, go away, geeks !18:05
kristianpaulwpwrak: what if i tell you dont need to be a VJ to use MM1, you can just turn in on, press a button and your friends will get amazed with the coolest effects18:10
kristianpaulif you want further, you can be social and use twitter to display messages on the go18:11
kristianpauland thats not all, from you iphone you can control the performance too, is as easy as always :-)18:12
wpwrakhah. the iPhone is in control ! so you morlocks did at least something right. keep on trying. maybe in 10-20 years, you'll have something that is worth our money. until then, we shall shower you with disdain and nonsensical feature sugg... no, demands !18:15
wpwrakheh, a bit more exercise and i could become a slashdot troll ;-)18:15
kristianpauljeje ;-)18:16
lekernelwpwrak, the sad thing is they probably buy iphones18:16
wpwraklekernel: it probably feels like a basic necessity to them. like wearing clothes. they would feel naked and shunned by their peers if they didn't have an iSomething too.18:19
kristianpaulwear is very correct, and iphone an apple product have nice design, so many people surelly just like wear/show it :-)18:29
kristianpaullike all those geeks wearing t-shirts with fork bombs written in the back, or that funny penguino :-)18:30
GitHub104[milkymist] sbourdeauducq pushed 1 new commit to master: http://bit.ly/jTvmCe18:54
GitHub104[milkymist/master] rc5: adapt timing to the t003 remote - Sebastien Bourdeauducq18:54
carlobarhi, problem with HPDMC was solved.. it was an error in memory addressing19:12
lekernelZPU redesign: http://alvie.com/zpuino/images/core_comparison.png20:53
kristianpaulfor god and lekernel happiness? :D20:59
lekernelwell, given the ZPU core I tried last year was some 10-20x slower than that, Alvie seems to have done a good job21:00
wpwrakfractional cycles ? interesting ;) addsp must do some interesting magic21:12
wpwrakand "store" is marginally slower than 2 cycles21:13
mwallelekernel: instead of patching the lm32 interrupt controller, we could add our own interrupt core and hook it up in the cpu interrupt line, thats imho the cleanest solution21:25
lekernelwhat cpu interrupt line? the lm32 has 3221:26
mwalleall 32 are ored to one, iirc21:26
kristianpaul+1 to the own interrupt core21:26
mwalleso we basically patch out the lm32 interrupt controller21:27
lekernelso 1) remove most of the LM32 interrupt logic and turn it into a single line IRQ input 2) implement an off-core interrupt controller?21:27
mwallejust keep IE21:27
lekernelsounds ok, as long as the extra bus cycles involved do not impact performance21:27
mwallewe could add some vector logic, so we wont have to walk over the IP register21:29
mwallehow many cycles does a csr access need?21:30
lekernelat least 321:31
lekernelwhat is the advantage of an external interrupt controller over making the LM32 interrupt lines level-sensitive and IP read-only?21:34
kristianpaulrun linux? :-)21:35
kristianpaulah sorry, miss read21:36
mwalleeasier to modify, eg xilinx has some nice set/clear bit pseudo registers for IE, so you wont have to RMW21:36
mwalleand i hoped that it would be easier to patch, but i think this isnt the case, just looking at the lm32_interrupt.v21:37
lekernelyou can make the register that assert the IRQ in each core clear the IRQ bit on write, so you won't have to RMW either21:38
mwalleyoure right, still thinking in the 'old' per interrupt IE bits terms21:39
mwallecool devices trees are basically working22:17
GitHub0[llvm-lm32] jpbonn pushed 70 new commits to master: http://bit.ly/lDpsa723:13
GitHub0[llvm-lm32/master] SI, DI, BP, and SP don't have 8-bit sub-registers in x86 mode. - Jakob Stoklund Olesen23:13
GitHub0[llvm-lm32/master] Zap the last reference to allocation_order_begin(). - Jakob Stoklund Olesen23:13
GitHub0[llvm-lm32/master] Only call TRI::getRawAllocationOrder to resolve a target-dependent hint. - Jakob Stoklund Olesen23:13
--- Tue Jun 21 201100:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!