#milkymist IRC log for Saturday, 2012-02-04

cladamwwpwrak, reading email...01:23
cladamwwill change them based on the feedback you gave, tks. ;-)01:24
wpwrakthanks ! :)01:24
cladamwregards to 'crowbar', i've not gone through all contents about that. so I'll direct add it once i totally understand it. :)01:26
wpwraki don't think we have to work too hard on making the schematics look "perfect", since we'll convert them to kicad anyway, but there are a few nasty bits that can get in the way of reviewing, and it's better to fix these01:26
cladamwsure, whenever you watch it out, all good stuffs we can fix them if just couples minutes works. :)01:28
wolfspraulwpwrak: yes, thank you for the details01:34
wolfspraulthe overall plan seems to be to first finish R4 in Altium, then work on kicad & boom?01:34
wolfsprauland I think the finishing in Altium will take another week or so...01:35
wolfspraulcladamw: you work on a Saturday? or today is not Saturday? :-)01:35
cladamwcladamw, yes, Saturday now, since some details i still haven't absorbed, so ... ;-)01:36
wpwrakyes, i think it's best to finish the altium. then we can do a public call for review. see what feedback we get. after review, adam can talk to the layout guys. and then we can start the conversion.01:37
cladamwalso yes those are small things to go... ;-)01:37
wolfspraulyes sounds right01:37
wpwrakmaybe we'll find issues during the conversion. so it's sort of another review pass.01:37
cladamwsounds good.01:37
wolfspraulI still want to go through the connectors and document how/why/why not software can detect insert/removal action01:37
wolfspraulbut we did that already in chat, so first we just take those old notes, maybe it's just a documentation effort01:38
wolfsprauland maybe we find a few ones to improve, we see01:38
wolfspraulbut after the current burst of activity is settled (j21/j22 etc)01:38
wpwrakheh, good :)01:39
wpwrakwhat do we aim for with DVI ? will there be any design verification for the digital side ?01:40
wpwrakand is DVI settled, as far as the schematics are concerned ?01:40
wpwrakcladamw: oh, i have one more bit of nitpicking for you: the 24 Ohm resistors of USB are all upside-down: R136, ...01:41
wpwrakit's funny. at first, you mentally add an "R" and think you're looking at "R24". then you realize that this can't be right ... ;-)01:41
cladamwwpwrak, i remembered larsc said once that dvi-i pin16 feature in irc, so i think we have to add a part to test this.01:41
cladamwwpwrak, oah..yuh...we few people knew it but else probably can't ! ;-)01:43
wpwrak(dvi pin 16) yes, that's for hotplugging. but i mean in general. is DVI "finished" ?01:43
cladamwwpwrak, i think so, BUT please review those all pins depends to the related... second...01:45
wpwrakah, one BIG task that awaits us when going to kicad is to change all those "bidirectional" signals to something meaningful. things like "AC97_CLK".01:46
cladamwi mostly followed this thread to finish DVI-I circuit, so please also see this important thread too. ;-)01:46
cladamwwpwrak, sure..okay01:46
wolfspraulDVI 'finished', not sure01:48
wolfspraulmaybe we are careful with R4 and first make a handful of boards only01:48
wolfspraulwe have to think about the fastest yet also most economical way forward01:49
wolfspraulcan we do any additional dvi (digital path) design verification now?01:49
wolfspraulI don't think so01:50
wolfspraulpin16 is still todo01:50
wpwrakthe scary bits are probably in signal integrity. we can't usefully verify these without making the real board01:51
wolfspraulbut it's actually good that we start with low resolutions01:51
wpwraki don't know if we could use the extension header to connect DVI-D, or if it would degrade the signal too much to be useful01:52
wolfspraulbecause the little I understand I think the protocol is relatively failure tolerant, that is signal integrity incrementally gets harder as resolution increases01:52
wolfspraulalso I think any digital video-out support depends on first switching to migen01:52
wpwrakhaving DVI-D via the extension header would allow the implementation of some test pattern generator that would then accelerate verification of M1r401:53
wpwrakproper DVI out for sure01:53
wolfspraulthe only person that can implement both the migen switch or digital video-out support in a timely way is Sebastien now, so we have to ask him01:53
xiangfuwpwrak, Hi. I just test the SET_PROTOCOL.02:13
xiangfuI guess most usb keyboard support the SET_PROTOCOL. I send the SET_PROTOCOL to combo device's mouse interface. then the mouse totally not working02:14
xiangfuwpwrak, so my plan is only check the REPORT_ID from REPORT_DESCRIPTOR02:16
wpwrakxiangfu: did the SET_PROTOCOL work with a normal mouse ?02:18
xiangfuwpwrak, yes. normal mouse works fine.02:19
xiangfuwpwrak, I only have one normal mouse.02:19
wpwrak(normal mouse ok, but combo not) ah, pity then. i was afraid that something like this would happen :-(02:19
wpwrakone is enough to test for success ;-)02:20
xiangfuwpwrak, I would guess those combo device only make keyboard support boot protocol but mouse02:21
xiangfuwpwrak, since for BIOS boot system. it only check keyboard02:21
xiangfuwpwrak, I have tested the code of check REPORT_ID. it's make both combo device and normal mouse works fine.02:22
wpwrakkewl. so it's the report protocol then02:23
wpwrak... and we've learned something more about things going wrong with usb :)02:23
xiangfuwpwrak, another thing is when I issue the GET_HID_REPORT to normal mouse. it always give RX timeout error, SETUP not ACKed02:26
xiangfuprint_hex the return of control_transfer is 0xFF.02:26
xiangfuI don't know what wrong with this GET_HID_REPORT. so far only ignore this error.02:27
wpwrakhmm, is the interface number correct ? (in the case of the normal mouse)02:30
wpwrakah wait, you determine this at run time. so it should be fine02:30
xiangfuthis is the first patch . pass ep_status to process: http://pastebin.com/PUKCx0Jy02:38
xiangfuthis is the second patch : http://pastebin.com/sbmy8DTK02:38
xiangfutry to check if there is REPORT_ID in the report data.02:38
xiangfuline 78, 79 still hard code.02:39
xiangfuworking on that now.02:40
xiangfuwpwrak, can you please paster the Rii RF mini-keyboard  hiddump somewhere. just want compare with my mini-keyboard. :)02:42
wpwrakyou need to check for ep == NULL at line 10802:42
wpwrakand you can simply line 112 and so on by replacing  case X: if (Y) { ... } break;  with  case X: if (!Y) break; ...; break;02:44
wpwraksaves one indentation level :)02:44
xiangfuline 112. yeah.02:50
xiangfuI like save indentation level.02:51
xiangfunow much better: http://pastebin.com/eJzV42w402:52
xiangfuand fix a NULL bug.02:52
wpwrakyup :)02:54
wpwrakif you want to emphasize the NULL check, you could begin both USB_DT_HID and USB_DT_ENDPOINT with  if(!ep) break;02:55
wpwrakbut what you have now looks good to me, too02:55
xiangfuyou mean like this one: http://pastebin.com/bgyG1FbH02:58
xiangfusorry. I mean this one: http://pastebin.com/eRbkj3Yy02:58
xiangfuI got a mini wireless keyboard  too.03:00
wpwrakwith built-in mouse ? or just keyboard ?03:01
xiangfusame as yours. but it's a ShanZhai device.03:01
xiangfuwith a build-in mouse03:02
wpwrakkewl. and i'm sure it was a lot cheaper ;-))03:02
wpwrakwolfgang is looking for such a device to source for M1r4. so it's good that with both have one, and can make sure it works well.03:03
xiangfuyes. ~200RMB. I forget the price. Yi bought it under taobao.com.03:05
xiangfusince it's a ShanZhai device. not work out of box. :(03:06
xiangfuwith these two patches works fine now. :)03:06
xiangfubut I need change the hardcode better :)03:06
wolfspraullet's be accurate. that's not 'shanzhai'.03:07
wolfspraulshanzhai I'd say are really only mtk-driven small outfits. the keyboard you got comes from unisen, anterton or some other keyboard maker.03:07
xiangfuoh. another problem you mentioned before. about those keys. like PrtSc = Fn + F2.03:07
wolfspraulI know the real shanzhai and they work from the trunks of their cars...03:07
xiangfuand screenshot not working. since we need press. Ctrl + Fn + F2.03:07
wolfspraulour ambition must be: if the keyboard or kbd+mouse combo works with Linux/notebook, it should work with m1 :-)03:09
wpwrak(keys) yes ! ;-) from that i know that sebastien can see the future. and he's vicious :) he accurately predicted all the key combinations one couldn't produce with the kind of keyboard we have, and he used almost exclusively these in FN. about the only thing that works is Esc.03:10
xiangfuwolfspraul, yes. shanzhai = mtk-driven  or if the device that same with foreign device and people call then shanzhai also. :)03:10
xiangfuwolfspraul, yes. just check. there are company name. tel and address on manual. so it's no shanzhai.03:10
wpwrakwolfspraul: the problem is that FN requires exotic key combinations that some of the tiny keyboards can't generate03:11
xiangfushanzhai device don't have tel. addr. company name. etc. :) sorry.03:11
xiangfuwpwrak, it's made by NOVICE. model N950mini.03:11
wpwrakwhatever :)03:11
wolfspraulwe need to move all default keys to such keys that are easily accessible on the most basic keyboards03:14
wolfsprauland we have plenty of keyboards now, a good time to do that...03:14
wolfspraul(some more samples are on the way)03:14
wpwrak(move) agreed. shouldn't be too hard to find better mappings. the mouse replacement may be a bit tricky. maybe with shift it'll be alright.03:17
wolfspraulI've found some mini keyboards that advertise 'multi-touch' on the pad03:24
wolfspraulnot sure what that means - sample on the way03:24
wolfspraul(and I'm not sure it's relevant to us, but can never hurt to look at what it actually is)03:24
wpwrakthat sounds weird ;-)03:25
wolfspraulmaybe they export 2 mice interfaces? who knows...03:25
wolfspraulor maybe just a marketing gimmick. we shall find out soon.03:26
wpwrakyeah. "welches maeuserl haetten's denn gern ?" (for an obscure old german reference :)03:26
wpwrakmy bet would be on marketing03:26
wpwrakmaybe mouse + scroll strip03:27
wolfspraulbut in any case xiangfu has to get the one he has now working, since it easily works on a Linux notebook, and our codes cannot be so fragile that they only work with a rare select list of models03:27
wolfspraulyes, could be with scroll strip, because some models also seem to say something about 'scroll' feature03:28
wpwrak(rate models) sure. we have a number of well-known compatibility issues. not considering report formats is but one of them :)03:29
wolfspraulwpwrak: I'm leaning towards an upright one like this one right now http://www.ipazzport.com/05E.html03:32
wolfspraulbut selection/sampling process is ongoing, and it will depend on quality, feel, whether we can get it to actually work on m1, etc.03:32
wolfspraulthat 'E' one includes multi-touch & scroll03:33
wolfspraulthe same without E, 1-2 USD cheaper http://www.ipazzport.com/05.html03:33
wpwrak“jest like your TV remote control”03:35
wpwrakthere's a comma missing ;-)03:36
wpwrakboth look suitable03:37
wolfspraulthis one is narrower, for 'single hand operation' http://www.ipazzport.com/08.html03:37
wolfspraulnot sure what's better with m1 - larger touchpad or narrower / single-hand03:38
wolfspraulwow, the -08 weighs 43 gram :-)03:39
wpwraklooks cramped03:39
wolfspraulI'm also not 100% sure yet about batteries, of course I would prefer standard batteries03:39
wolfspraulbut I think those ones have nokia-style batteries, maybe even the same as on Ben NanoNote03:40
wpwrakas if you had a choice :) li-ion it is03:40
wpwrakah, "standard" li-ion03:41
wolfspraulsure, but a CR25 is easier to replace x years later than a nokia bl-4c03:41
wpwrakwell, you can peek inside. most likely, it won't be anything you'll find convenient to source outside of china03:41
wolfspraulah, increasingly what is convenient to source inside China is the same that is convenient to source worldwide...03:42
wolfspraulbut still, I would prefer standard, we see03:42
wolfspraulI'm learning03:42
wolfspraulyou think the -08 looks cramped?03:42
wpwrakwhat's cr25 ?03:43
wolfspraulthose round button batteries03:43
wolfspraulsorry maybe it's cr2016, 2025, 204303:44
wolfspraul16/25/32 is the thickness03:44
wpwrak-08 looks like my keyboard but with the keys oriented vertically instead of horizontally. so each button would be half the size.03:44
wpwrakalready mine is small.03:44
wolfspraulyeah, so maybe 05 or 05E is better03:44
wolfspraulI'm leaning towards those now, but I need to sample and compare more03:45
wpwrakah :) all those keyboards seem to use rechargeable batteries, for better or worse03:45
wpwrak("mine is small") and i like things to be tiny. so i'm already at the tail of the curve :)03:46
wpwrakchecked the size. it's indeed just as wide as mine is tall03:47
wpwrakso that's a size reduction of about 2:3. very small.03:49
wpwrakbut try it. if it works, even cooler ;-)03:49
wolfspraulall moving03:50
wolfspraulbut we need to make the ones we have work first as well, our sampling effort is gated on several sides :-)03:51
wpwrakoh, and the -08 uses a lot of Fn+key for special characters03:54
wpwraknothing that looks too offensive, though03:55
wpwrakand so does the -05. i'm discovering the secrets of the upright form factor :)03:57
wpwrakwell, try them. edit a few patches. then you'll see what feels right.03:58
lekernel_if I understand correctly, this "bank group" thing is about selecting two banks (one in each group) on read/write operations, each providing half of the data bus?11:28
wpwrakhmm, why does it scare me if they state that adding CRC increases performance ?11:45
lekernel_yeah I don't get that either11:50
lekernel_and it seems that those CRCs cause actual data bursts of length 9, which sounds like more scheduling problems11:50
wpwrakto me the bank groups sound more like an optimization of (burst) accesses in the same area of the chip, with penalty cycles when crossing a boundary11:52
wpwrak(crc) it could mean that they intend to run at speeds at which errors are expected to occur11:53
lekernel_nah, it has do to with data prefetch (aka data bus serialization to a multiple of the DRAM core clock frequency)11:54
wpwraksimilar to NAND ECC not just being an extra safety feature11:54
lekernel_eg DDR1 has a DRAM core running at half the data rate, but which is 2n bit wide for n data pins11:55
lekernel_DDR3 is 8n11:55
wpwraki think my fairly broad definition would include such cases :)11:56
lekernel_and it seems DDR4, instead of having 16n, is 8n + 8n in two selectable banks11:56
wpwrakit doesn't say that you can select them separately11:56
lekernel_I would guess that they simply have to belong to different groups11:57
wpwrakmaybe it's just 8n bits from bank X at tCCD_S, then it switches internally to bank X+1 with the memory controller selecting tCCD_L for that access11:57
lekernel_nah, the numbers have to add up - you have to provide 16 bits per DRAM cycle per data pin11:59
lekernel_tCCD is the minimum delay between two column commands (ie read/write) from the memory controller11:59
lekernel_which can be larger than the burst length starting with DDR312:00
lekernel_I wonder how much this costs: http://www.lecroy.com/protocolanalyzer/protocoloverview.aspx?seriesid=39812:06
wpwrak(tCCD) still sounds compatible like my theory :) maybe tCCD_S is without data read or a somehow reduced data read12:07
lekernel_"the Kibra 480 system allows SoC designers to leverage their R&D budgets and increase their test coverage of next generation DDR memory technology."12:08
wpwrak(kibra) if you have to ask you can't afford it ;-)12:11
mumptaimhmm, crc? fec-codes would12:12
wpwrakmumptai: CRC+ARQ12:16
mumptaiurks, but than you get unpredictive timing ...12:17
wpwraka small price to pay for more overall speed ;-)12:19
wpwrakand maybe the clock speed can be "trained" as well. cheap motherboard or cheap RAM modules = DDR4 runs a bit slower. nothing new, actually.12:20
mumptaiand temperature dependant refreshes made timing already a bit unpredictable12:21
wpwrakDDR6 will require the system to boot at a low-speed setting, then consult an astrology server over the internet to determine the exact operational parameters for that day12:28
lekernel_wpwrak: (clock speed training) I've read XBox consoles do this sort of thing... during production, they basically throw in whatever cheap DRAMs they find on the market that day, and their ASIC is supposed to cope with them12:33
lekernel_mumptai: I've never heard of temperature dependent refreshes (except for systems that are supposed to operate at more than 80C or so)12:34
lekernel_but then you just stay in the "high refresh rate" mode12:34
mumptaior if refresh causes a significant impact on your power budget (and you application runs rather cold)12:38
lekerneland is this spec'd by the DRAM vendors, or do you have to do the testing and characterization yourself?12:39
lekernelI think you can actually do that to a large extent... just for the heck of it, I've tried killing refresh for several seconds on a M1, without much data loss12:41
mumptaidozens of minutes, if cooled (about -20°C)12:43
lekerneldata loss is much faster when the DRAM power is cut, though12:43
kristianpaulmorning :)13:39
--- Sun Feb 5 201200:00

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