#qi-hardware IRC log for Friday, 2014-07-18

nicksydneywpwrak: http://www.instructables.com/id/How-Can-StikNPeel-TM-Rework-Stencil/01:13
hozerhey at least with the linux kernel we get to see all the bad code. No such chance will all the damn firmware for various hardware devices02:23
BlackGyver_Good morning everyone07:03
BlackGyver_I'm a bit of a newb when it comes to hardware07:03
BlackGyver_I was looking to know what kind of performance increases (and in which kind of applications) I would get by having my RAM's clock speed higher than what my CPU asks for07:03
BlackGyver_ie. grabbing 2133Mhz RAM on a processor asking for 1333/160007:04
BlackGyver_oh I don't think I'm actually on a good channel for that kind of questions, am I07:05
BlackGyver_Well, feel free to answer anyway if you have an idea07:06
DocScrutinizer05BlackGyver: for the channel it's maybe a tad off topic, but anyway faster RAM will do nothing when your CPU doesn't use faster speed.07:21
DocScrutinizer05clock speed of RAM bus even07:21
DocScrutinizer05it's like with your car engine: it doesn't get more horsepower when your fuel has higher octane07:22
DocScrutinizer05fuel needs suppicient octane for the compression of your engine, and RAM needs to be fast enough for the RAM bus clock speed of your CPU. Using more than "sufficient" doesn't bring any benefit07:23
ysionneauwell ram is not only about bus speed07:24
ysionneauit's also about switching from one bank to another speed07:24
ysionneauswitching from read to write speed07:24
ysionneauso "speed" of RAM matters (for DRAM I mean)07:25
DocScrutinizer05of course there are car engines that dynamically optimize the ignition point, and there are CPU memory controllers that "dynamically" adapt memory bus clock speed07:25
DocScrutinizer05ysionneau: well, yes. There are more parameters than only mere clock speed. There's stuff like RAS timing etc. But all that are parameters of the memory controller and unless the memory controller adjusts them, no increased speed of the RAM you use will change the system performance07:27
BlackGyverAlright! Thanks a bunch07:27
BlackGyverSorry for the confusion about the channel07:27
BlackGyverCheers! Have a good day!07:28
ysionneauthese are parameter of the RAM, but indeed the memory controller needs to "know" them and use them07:28
DocScrutinizer05for PC architecture (x86) it's usual that CPU reads the parameters out of a config ROM in RAM and adjusts parameters of mem controller07:28
DocScrutinizer05but often you need to set them manually in BIOS07:28
ysionneaubut then, if your memory controller is clever and know how to adapt, the limiting factor is the speed of your RAM07:28
ysionneauif your memory controller is stupid then indeed you don't care about RAM parameters07:29
ysionneauyou just access it slowly, in a non pipelined way etc07:29
DocScrutinizer05there's no such thing like stupid memory controllers nowadays in all but the most stupid CPUs which only come with embedded RAM07:30
DocScrutinizer05usually RAM sticks have a small serially attached (EP)ROM with the parameters of the RAM on that stick, and the BIOS of your PC reads that out and sets memory controller parameters accordingly07:32
DocScrutinizer05some RAM sticks come with bogus or broken info in that ROM, and then you need to config RAM parameters in BIOS setup manually07:33
DocScrutinizer05there's no mem controller that can really autodetect speed of RAM07:34
DocScrutinizer05since there's no physical parameter on the signal lines that tells mem controller if RAM can cope with the speed07:35
DocScrutinizer05and RAM access is not async, i.e. there's no data-ready or other ACK feedback from RAM to controller07:36
DocScrutinizer05and re >><BlackGyver_> ie. grabbing 2133Mhz RAM on a processor asking for 1333/1600<< I doubt there are CPUs out there that "ask for" 1333/1600 but could go as fast as 213307:40
DocScrutinizer05usually the SPI data has not even parameters for RAS/CAS etc on other than max clock speed of RAM07:41
DocScrutinizer05so when the RAM is saying 2133;3CAS;5RAS then a CPU running the RAM at 1333 nevertheless will use 5 CAS clock cycles and 3 RAS. So no speed increase at all, unless you calculate CAS/RAS etc manually and config it in BIOS setup manually. --- disclaimer AFAIK07:43
DocScrutinizer05and even that isn't worth it, the speed increase is rather marginal07:44
apeleteHi larsc07:45
apeleteyesterday I was testing my mmc async requests patches against 3.16-rc5, and I hit a dma-jz4740 related bug: http://paste.debian.net/110350/07:45
apeletethis is the one solved by these two patches:07:46
apeleteI think this is the first time I hit it the bug on the upsteam mainline, found it funny :)07:46
apeletelarsc: anyway, mmc async requests is now working, by mapping only dma area in pre_request(), like you said07:47
apeleteand it was also explicitely documented in dmaengine.txt, must have missed the point when I first read it07:48
larscgood :)07:51
DocScrutinizer05larsc: do you know anything (existence for example) of crtouchb10 touchscreen controller drivers in upstream or elsewhere?07:54
apeletelarsc: is there anything substantial left to do on qi-kernel ?07:57
larscDocScrutinizer05: no07:58
DocScrutinizer05thanks! :-)07:58
DocScrutinizer05though actually :-S07:58
larscapelete: migrating to the common clock framework07:58
DocScrutinizer05larsc: wpwrak: could you have a look at http://www.freescale.com/webapp/sps/download/license.jsp?colCode=CRTOUCH_DEMO_LINUX_BSP_IMX53QSB&location=null&fpsp=1 and see if that's any useful?07:59
apeletelarsc: \o/ (<-- infinite hacker joy)08:00
apeletewas thinking you would say no :)08:00
larscapelete: I started with that some time ago, but never finished it08:00
apeletelarsc: what is that about, any more details ?08:01
Action: apelete just found a lwn article about common clock framework08:02
DocScrutinizer05wpwrak: larsc: http://privatepaste.com/6f58478d6f08:02
DocScrutinizer05excerpt from the .pdf to be found in above URL08:02
larscapelete: we implemnt the clock API with a custom implementation right now08:09
larscswitching over to the common clock framework means we'll use the the generic implementation08:09
apeletelarsc: does "custom implementation" mean "not upstream" ?08:11
larsccustom means, we implement the API functions in arch/mips/jz4740/clk.h08:11
larsccustom means, we implement the API functions in arch/mips/jz4740/clk.c08:11
apeleteok, will take a look at that08:12
larscDocScrutinizer05: do you know where the touchscreen_crtouch.patch is?08:15
DocScrutinizer05only click accept08:15
DocScrutinizer05all the way down at bottom08:16
larscwell, thats a /dev/input/eventX driver, but not upstream material08:16
DocScrutinizer05nfc what I accepted ;-P08:16
DocScrutinizer05yup, guessed that much. Does it look useful though?08:17
DocScrutinizer05IOW could some kernel hacker easily build an upstream compliant driver from it?08:17
larscit doesn't look to complex08:18
DocScrutinizer05I'm asking you since I have no idea about "easily"08:18
DocScrutinizer05fine \o/ :-D08:18
DocScrutinizer05thanks a megaton!08:18
Action: apelete was thinking about hacking on JZ4770 kernel support for GCW0 soon, maybe he'll have to manage workload on both qi-kernel and gcw-kernel08:19
apeletenights are going to get ever shorter...08:19
DocScrutinizer05longer ;-)08:20
DocScrutinizer05and the days are endless then08:20
larscI have a at least somewhat working gcw board now08:20
apeletehaha, good point08:20
apeletelarsc: oh, you're working on gcw0 already ?08:21
larscbut I have the board08:22
larscI could work on it if I had time08:22
larscI think they were looking on switching to the jz4740 mmc driver08:23
apeleteyes, I'll talk with pcercueil and mth about that08:51
DocScrutinizer05I don't like vox too much, but those sites sound not too much biased and unfair09:22
rjeffrieswpwrak: suggest you add "anelok password device" to list of what channel covers16:19
kyakwe can't discuss it without adding it to the list? :)16:22
wpwrakDocScrutinizer05: (license) hmm, i guess it depends on what you want to do16:50
wpwrake.g., "to use, only as part of, or integrated within, Authorized Systems" and ""Authorized System" means the hardware system(s) or software program(s) marketed by you which contains an i.MX"16:51
wpwrakalso, it only allows you to distribute object code, no source16:53
wpwrakso i'd heed nancy reagan's wide advice and "just say no" :)16:54
wpwrak"The included patches were created based on" sounds like derivative work. if the work it was derived from in under that license, then it may even be in violation to that license.16:56
wpwrakgeneral rule: if some piece of hardware doesn't come with sufficient documentation to let you write your own complete firmware, but instead requires something provided by the vendor, then this almost always means that you can't use it for Free Software16:57
DocScrutinizer05sorry, I can't follow. We have a .c fule with a GPLv2 license and we have a datasheet that for sure has all needed info to write whatever you want17:13
DocScrutinizer05I don't give a **** which way that .c file got created17:14
DocScrutinizer05the chip manufacturer provided c code with a GPL and in my book that GPL can't be overridden by any random general license that obviously is a standard disclaimer used for all code that manuf provides17:15
wpwrakah, so freescale released source under GPL ? and added a non-GPL license ? nice organization :)17:16
DocScrutinizer05btw this is kinda ECHAN17:16
wpwraki was just commenting on the license, since you asked for opinions on it17:16
DocScrutinizer05I honestly wonder what you refer to when you haven't seen the original17:17
wpwrak<DocScrutinizer05> larsc: wpwrak: could you have a look at http://www.freescale.com/webapp/sps/download/license.jsp?colCode=CRTOUCH_DEMO_LINUX_BSP_IMX53QSB&location=null&fpsp=1 and see if that's any useful?17:17
DocScrutinizer05I obviously didn't mean that nonsensical biolerplate "licence"17:18
DocScrutinizer05how could *that* be useful?17:18
wpwrakoh, there are buttons at the bottom ;)17:18
wpwrakwell, "useful" as in "can we live with that"17:18
DocScrutinizer05please reread full chanlog17:19
DocScrutinizer05just two lines later I told that there are buttons17:19
DocScrutinizer05and the convo with larsc wasn't about licence issues17:19
--- Sat Jul 19 201400:00

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