IRC log of #maemo-ssu for Saturday, 2011-12-10

*** int_ua has quit IRC00:16
*** int_ua has joined #maemo-ssu00:16
*** int_ua has joined #maemo-ssu00:18
*** int_ua has quit IRC00:19
*** int_ua has joined #maemo-ssu00:19
*** mase_76 has joined #maemo-ssu00:36
*** mirandir has left #maemo-ssu00:38
*** mase76 has quit IRC00:40
*** int_ua has quit IRC00:46
*** mase_76 has quit IRC01:03
*** int_ua has joined #maemo-ssu01:03
*** int_ua has quit IRC01:04
*** scoobertron has joined #maemo-ssu01:16
*** Sc0rpius has quit IRC01:26
*** scoobertron has quit IRC02:21
*** M4rtinK has quit IRC02:29
*** NIN101 has quit IRC02:35
*** trbs has quit IRC02:56
*** Sc0rpius has joined #maemo-ssu03:40
*** jonwil has joined #maemo-ssu04:29
*** nox- has quit IRC04:34
*** amiconn has quit IRC05:41
*** amiconn_ has joined #maemo-ssu05:41
*** amiconn_ is now known as amiconn05:42
*** DocScrutinizer has quit IRC07:35
*** DocScrutinizer has joined #maemo-ssu07:35
*** psycho_oreos has quit IRC08:18
*** rd has joined #maemo-ssu10:32
*** M4rtinK has joined #maemo-ssu11:12
*** jonwil has quit IRC11:33
*** rd has quit IRC11:39
*** mirandir has joined #maemo-ssu12:02
*** scoobertron has joined #maemo-ssu12:10
*** rd has joined #maemo-ssu12:26
*** NIN101 has joined #maemo-ssu12:32
*** rd has quit IRC12:45
*** NIN102 has joined #maemo-ssu12:51
*** NIN101 has quit IRC12:53
*** l32606 has joined #maemo-ssu13:06
*** scoobertron has quit IRC13:06
*** barzam has joined #maemo-ssu13:38
*** barzam has left #maemo-ssu13:38
*** Pali has joined #maemo-ssu13:55
*** l32606 has quit IRC14:09
merlin1991Sicelo: I'll have a look at that14:49
merlin1991Sicelo: I can't reproduce it on testing or stable14:54
merlin1991MohammadAG: new testing this weekend? :)14:55
freemangordonmerlin1991 could you please pull git@gitorious.org:fremantle-720p-support/osso-camera-ui-l10n.git in CSUU repo14:58
freemangordon*CSSU14:58
Palimerlin1991, can you also push into CSSU patched osso-calculator-engine?15:06
freemangordonPali, any plans to move KP in CSSU?15:08
Paliwhat?15:08
Paliwhy?15:09
freemangordonBecause omap1 is dated15:10
freemangordonwell, not exactly KP, but KP with stripped capabilities15:11
freemangordonoverclocking stuff and such15:11
Paliso only bugfix to CSSU?15:11
freemangordonwell, new functionality too, new modules, etc.15:11
freemangordonthose that don't have known issues15:12
Paliok, but ask merlin1991 and MohammadAG15:12
freemangordonsee, I am not talking about moving KP in CSSU today. For the fiture.15:12
freemangordon*future15:13
freemangordonof course I will :D15:13
freemangordonactually it should be done in some meeting15:13
Palithere will be problem with upgrading via HAM15:13
Paliusers with KP will get update too15:13
merlin1991and users with other kernel too15:13
freemangordonsee, after all thumb seems rock stable, new DSP driver is in KP, all that stuff15:14
freemangordonboth?15:14
freemangordonsorry, misread, scratch that15:14
freemangordonPali, what if users of KP get the update?15:14
Paliif they get CSSU kernel update, it will flash and replace cssu kernel15:15
freemangordonwell, that is some maintaner work to do, what if they flash newer KP over CSSU one?15:16
freemangordonit is the same now15:16
freemangordonbut you have omap1 instead of cssu115:16
*** obironbo has joined #maemo-ssu15:16
freemangordonbbl15:17
MohammadAGI already suggested we check in postinst if a different kernel flavor exists15:24
MohammadAGif so prompt the user on what to do15:25
*** freemangordon has quit IRC16:28
*** freemangordon has joined #maemo-ssu16:28
*** Pali has quit IRC16:31
*** l32606 has joined #maemo-ssu16:31
*** l32606 has quit IRC16:39
*** Aranel has joined #maemo-ssu17:11
*** Aranel has quit IRC17:11
*** Aranel has joined #maemo-ssu17:11
*** mirandir has left #maemo-ssu17:29
*** jonwil has joined #maemo-ssu17:34
*** fw190 has joined #maemo-ssu17:48
*** mase76 has joined #maemo-ssu17:57
*** jonwil has quit IRC18:00
fw190hello18:11
fw190any news about uboot for kp49 or later?18:11
merlin1991fw190: it's only a matter of doing it18:14
merlin1991I had the requirements in place at some point, but don't have it anymore18:15
fw190hmmm would be greatfull- it's needed for testing nemombile. can't leave kp for it so im in fead end ;)18:18
fw190dead18:19
merlin1991though that question wasn't exactly cssu related :D18:21
fw190it isn't but dont know where to ask...18:21
merlin1991#maemo? :D18:22
fw190ok you got me ;)18:23
fw190so now for something cssu related18:23
fw190maybe we could change te meeting schedule?18:23
merlin1991to what?18:24
fw190maybe this could be a daily chit chat if something interesting would pop uout?18:24
merlin1991we do have a 24/7 chit chat since this is irc :D18:24
merlin1991the meeting schedule was meant for people who don't end up on irc usually18:25
fw190ah now I understand18:26
fw190but decisions are made on the meetings18:26
fw190but sometimes we are missing some key members and there are no meetings or no decisions can be made18:26
*** Aranel has quit IRC18:40
*** rd has joined #maemo-ssu18:58
*** Aranel_ has joined #maemo-ssu19:25
*** Aranel_ is now known as Aranel19:27
*** Aranel has joined #maemo-ssu19:27
*** Pali has joined #maemo-ssu19:44
*** Pali has quit IRC19:44
*** Pali has joined #maemo-ssu19:44
*** Kaptenen has quit IRC19:45
*** Pali has quit IRC19:56
*** rd has quit IRC20:14
*** rd has joined #maemo-ssu20:14
*** trbs has joined #maemo-ssu20:16
*** rd has quit IRC20:25
*** rd has joined #maemo-ssu20:25
*** nox- has joined #maemo-ssu20:30
*** rd has quit IRC20:41
*** rd has joined #maemo-ssu20:41
*** rd has quit IRC20:55
*** rd has joined #maemo-ssu21:00
freemangordonDocScrutinizer, ping21:11
DocScrutinizer51pong21:12
freemangordonPM-ed something, worth reading21:12
DocScrutinizer51pm as in?21:12
freemangordonpersonal messgage21:13
freemangordonmessage*21:13
freemangordonon TMO21:13
DocScrutinizer51to?21:13
DocScrutinizer51aah21:13
freemangordonjoerg_rw21:13
DocScrutinizer51omg21:13
freemangordon?21:13
* DocScrutinizer51 hates this friggin tmo pm21:13
freemangordonyep, but there is a reason.21:14
DocScrutinizer51hmm21:14
DocScrutinizer51couldn't figure any21:14
DocScrutinizer51I'll see when I eventually come across it21:14
*** mase76 has quit IRC21:23
DocScrutinizerfreemangordon: I tend to support your notion, slowly but a bit more every day21:29
freemangordon:D good21:31
DocScrutinizerstill reluctant to dive into all thumb head-first, without decent benchmarks. Also a proper forced context switching benchmark and stresstest would tell us a lot more than random tests on a whole system. But... yes, your argumentation is straight and sound21:31
freemangordonDocScrutinizer, BTB flush workaround is enabled in KP for an year or so, too late for benchmarks.21:33
freemangordonbbl21:33
DocScrutinizerwe got no thumb binaries - basically - though21:33
freemangordonit affects all binaries, not only thumb. It invalidates BTB on every context switch no matter if code is thumb or not21:34
*** rd has quit IRC21:34
DocScrutinizer1) I need to read and understand that SiERR and what this BTB thing actually does. 2) I need to know if it'S a thumb-only thing or has impact on system performance in all situations. 3) we still need some *real* numbers so we can do an educated guess rather than stating "nobody complained so far" as that'S actually not correct: users complain all the time about a zillion of things, and we don't even know exactly what particular side21:36
DocScrutinizereffects flushing BTB might show on our fremantle system21:36
DocScrutinizerwell, you answered a few of my points as I needed so long to type21:37
DocScrutinizer:-D still we need better stories about all this21:37
DocScrutinizermaybe the scheduler flushes BTB on every context switch, but is arm code actually using BTB? - for example21:38
DocScrutinizerI need to see a in-depth theoretical evaluation of the impacts this btb-flush fix has, under some selected situations21:39
DocScrutinizerto specify what exactly I need to see, I first have to understand what the fix does, which in turn needs me to understand what btb does21:40
DocScrutinizerand how it's been handled in context switches before that patch, how it's handled now with this patch, and how all this fits into the SiERR thing21:41
DocScrutinizerthen when we've done all this correctly, we can publish an official "this bug got fixed" announcement that is applicable to all similar hw21:42
DocScrutinizernot only to N90021:42
DocScrutinizeractually you'll do that, and you're free to mention my help when doing so. Then we basically "import the bugfix" from upstream21:43
*** xnt14 has quit IRC21:51
*** MohammadAG has quit IRC21:51
freemangordonDocScrutinizer, BTB is a kind of cache memory used by branch predictor, i.e. when a predicted (or not) branch is taken the logic searches BTB if VA address of the branch target is there, if it is there, the branch is taken using PA(physical address) from BTB. The effect is that we skip VA->PA translation which has very good performance impact. But I would trade even 10% of performance for22:01
freemangordon40MB of free RAM on n90022:01
freemangordon40MB *MORE* free ram22:02
*** rd has joined #maemo-ssu22:27
*** bsdmaniak has joined #maemo-ssu22:36
*** rd has quit IRC22:40
DocScrutinizerI bet it's no 10% we will have to suffer from, for the mere BTB flush. But we need a proper analysis if the one SiERR is evidently fixed by this. And we need a rough estimation about the actual impact on performance, in percent_increased_contextswitching_time plus a rough picture how many context switches usually and under worst case can happen per time.22:40
DocScrutinizerooh, of course plus impact on code execution *after* context switch. Along the line "BTB depth: 16 branches. Time to fill one BTB position vs time to execute a branch based on a valid BTB position: NN clock cycles. Ergo: 16*NN = max penalty for each taskswitching. Plus penalty for the flushing during taskswitching: NN cycles. Estimated avrg taskswitches / second: 50. Total avrg penalty: clockcycles/s / (50 * sigma(penalties per switch))"22:47
freemangordonAnd how we are going to evaluate penalty for BTB flush against speed gain because of more free RAM, better cache usage (because of the shorter code), less swapping, etc. Not sure it is possible to do such evaluation.22:52
*** rd has joined #maemo-ssu22:54
DocScrutinizerall is possible22:56
freemangordonhmm, trust me on that(tm) approach seems more feasible to me than some indepth academical research, I am not going to do it.22:57
DocScrutinizerwe however are not even really interested in that, as that's our private pet thing, not sth about fixing a SiERR properly. In the end it's up to the end user to decide if she prefers more free RAM or faster code, all we can and have to do is deliver proper fix and evaluation of the impact it brings22:57
freemangordonyou mean non-n900 users?22:58
DocScrutinizerand more important than impact analysis is evaluation of general validity of the fix, performance impact estimation comes "for free" with that22:59
DocScrutinizeractually that'S sth the original author of the fix should have done, I'm wondering why he didn't23:01
freemangordonI think I have done enough analysis these days, just need to gather all the pieces. And send a mail to maemo-developers or such.23:01
freemangordonit is done23:01
freemangordonby the original author23:02
DocScrutinizerso why don't you just share a URL so we all could read it up by ourselves? :-)23:02
DocScrutinizerwill save us lots of analysis and evaluation work23:03
freemangordon+bool "ARM errata: Stale prediction on replaced interworking branch"23:03
freemangordon+depends on CPU_V723:03
freemangordon+help23:03
freemangordon+  This option enables the workaround for the 430973 Cortex-A823:03
freemangordon+  (r1p0..r1p2) erratum. If a code sequence containing an ARM/Thumb23:03
freemangordon+  interworking branch is replaced with another code sequence at the23:03
freemangordon+  same virtual address, whether due to self-modifying code or virtual23:03
freemangordon+  to physical address re-mapping, Cortex-A8 does not recover from the23:03
freemangordon+  stale interworking branch prediction. This results in Cortex-A823:03
freemangordon+  executing the new code sequence in the incorrect ARM or Thumb state.23:03
freemangordon+  The workaround enables the BTB/BTAC operations by setting ACTLR.IBE23:03
freemangordon+  and also flushes the branch target cache at every context switch.23:03
freemangordon+  Note that setting specific bits in the ACTLR register may not be23:03
freemangordon+  available in non-secure mode.23:03
RST38hplease stop23:03
freemangordonsorry23:04
RST38hpastebin23:04
freemangordonhttp://pastebin.com/FLCVKqfb23:05
freemangordonDocScrutinizer ^^^23:06
*** StyXman has joined #maemo-ssu23:14
*** StyXman has left #maemo-ssu23:27
*** chainsawbike has quit IRC23:35
*** chainsawbike has joined #maemo-ssu23:36
*** fw190 has quit IRC23:51
*** freemangordon has left #maemo-ssu23:54

Generated by irclog2html.py 2.15.1 by Marius Gedminas - find it at mg.pov.lt!