IRC log of #maemo-ssu for Wednesday, 2011-12-07

*** trbs has quit IRC00:08
*** NIN101 has quit IRC00:44
*** scoobertron has joined #maemo-ssu00:44
*** scoobertron has quit IRC01:01
*** scoobertron has joined #maemo-ssu01:02
*** scoobertron has quit IRC01:23
*** Pali has quit IRC01:32
*** mase76 has quit IRC01:40
*** jonwil has joined #maemo-ssu01:46
*** M4rtinK has quit IRC02:34
*** thingie has quit IRC02:37
*** nox- has quit IRC04:46
*** mirandir has joined #maemo-ssu05:44
*** amiconn has quit IRC05:45
*** amiconn_ has joined #maemo-ssu05:45
*** amiconn_ is now known as amiconn05:45
*** regawyte has quit IRC06:23
*** regawyte has joined #maemo-ssu06:24
*** mase76 has joined #maemo-ssu06:36
*** mase_76 has joined #maemo-ssu06:44
*** mase76 has quit IRC06:48
*** mase76 has joined #maemo-ssu06:55
*** mase_76 has quit IRC06:58
*** mase_76 has joined #maemo-ssu07:05
*** mase76 has quit IRC07:08
*** mase_76 has quit IRC07:27
*** m0use has quit IRC07:29
*** DocScrutinizer has quit IRC07:33
*** DocScrutinizer has joined #maemo-ssu07:33
*** mase_76 has joined #maemo-ssu07:40
*** mase76 has joined #maemo-ssu07:44
*** mase_76 has quit IRC07:48
*** mase_76 has joined #maemo-ssu07:53
*** mase76 has quit IRC07:56
*** mase76 has joined #maemo-ssu08:03
*** m0use has joined #maemo-ssu08:06
*** mase_76 has quit IRC08:07
*** mase76 has quit IRC08:17
*** mase76 has joined #maemo-ssu08:30
*** mase_76 has joined #maemo-ssu08:35
*** mase76 has quit IRC08:38
*** mase76 has joined #maemo-ssu08:45
*** mase_76 has quit IRC08:49
*** MSameer has quit IRC08:50
*** mase_76 has joined #maemo-ssu08:55
*** mase76 has quit IRC08:59
*** mase_76 has quit IRC09:42
jonwilMohammadAG: ping09:54
*** scoobertron has joined #maemo-ssu10:03
*** mattlee has joined #maemo-ssu10:11
*** scoobertron has quit IRC10:23
*** mattlee has quit IRC10:23
MohammadAGPong11:02
*** luke-jr has quit IRC11:16
jonwilMohammadAG, what do I do to get operator-name-cbs-widget into CSSU-T?11:26
jonwilIts in the CSSU repos now11:26
jonwilhow do I get it into the next version of CSSU-T?11:26
jonwilguess he left :P11:31
*** xmlich02 has joined #maemo-ssu11:32
*** xmlich02 has quit IRC11:35
*** xmlich02 has joined #maemo-ssu11:35
*** jonwil has quit IRC11:37
*** xmlich02 has joined #maemo-ssu11:37
MohammadAGmerlin1991: Tell him he doesn't have to do anything11:42
freemangordonDocScrutinizer ping12:00
DocScrutinizer51freemangordon: at work12:01
DocScrutinizer51a minute12:02
freemangordonBoth BootROM and x-loader on n900 are thumb compiled. Anywa, ping me when you have time12:02
freemangordonAnyway*12:02
DocScrutinizer51will do12:02
freemangordon10x12:02
*** Milhouse has quit IRC12:20
*** Milhouse has joined #maemo-ssu12:28
*** andre__ has joined #maemo-ssu12:29
*** andre__ has joined #maemo-ssu12:29
*** psycho_oreos has joined #maemo-ssu12:42
*** FireFly has joined #maemo-ssu13:00
*** BCMM has joined #maemo-ssu13:06
*** ekze has quit IRC13:52
*** ekze has joined #maemo-ssu13:53
*** Estel_ has joined #maemo-ssu13:57
*** scoobertron has joined #maemo-ssu13:57
*** jonwil has joined #maemo-ssu14:42
*** mase76 has joined #maemo-ssu16:04
*** M4rtinK has joined #maemo-ssu16:50
*** psycho_oreos has quit IRC16:50
*** jonwil has quit IRC16:51
*** mase_76 has joined #maemo-ssu16:57
*** scoobertron has quit IRC17:04
*** andre__ has quit IRC17:24
*** Estel_ has quit IRC17:25
*** BCMM has quit IRC18:05
*** BCMM has joined #maemo-ssu18:13
*** M4rtinK has quit IRC18:32
*** andre__ has joined #maemo-ssu19:00
*** andre__ has joined #maemo-ssu19:00
*** bsdmaniak has joined #maemo-ssu19:27
*** NIN101 has joined #maemo-ssu19:32
DocScrutinizer10x?19:34
MohammadAGDocScrutinizer: Ten x20:09
MohammadAGTenx20:09
MohammadAGTanzania20:09
MohammadAGTanx20:09
MohammadAGThanx20:09
MohammadAGThanks20:09
MohammadAGGet it?20:09
MohammadAGTanzania was iOS autocorrect20:09
*** scoobertron has joined #maemo-ssu20:14
RaimuDYAC20:17
andre__YMCA20:39
Raimu...that's random, but funny.20:44
*** m0use has quit IRC21:08
*** scoobertron has quit IRC21:09
*** BCMM has quit IRC21:32
*** m0use has joined #maemo-ssu21:34
*** BCMM has joined #maemo-ssu21:37
*** scoobertron has joined #maemo-ssu21:59
DocScrutinizerfreemangordon: ping22:09
*** bsdmaniak has quit IRC22:10
*** MSameer has joined #maemo-ssu22:19
*** trbs has joined #maemo-ssu22:22
*** MSameer has quit IRC22:35
freemangordonDocScrutinizer pong22:37
*** M4rtinK has joined #maemo-ssu22:39
freemangordonDocScrutinizer, as I already told you today both bootrom and xloader (nolo) are thumb compiled. I've been searching alot these days and AIUI two erratas in question regarding thumb are *NOT* mutually exlusive. Workaround for first (430973) is to flush BTB on every context switch. This is done by using special instruction to system control coprocessor. There is a special bit in one of the22:44
freemangordoncontrol registers which enables those instructions otherwise they are treated as NOPs. (TBC)22:44
freemangordonThe second errata has 2 workarounds - either do not enable the bit needed for 430973 errata workaround *OR* make sure that L1 Array debug register is 022:46
DocScrutinizerfreemangordon: I don't think you can know what bootrom code looks like, afaik it's only documented to what it *does*22:46
DocScrutinizerand xloder != NOLO22:46
freemangordonwell, it could be read from linux, if you wish search xda-developers, they have omap3430 disassembled22:47
freemangordonbootrom22:47
DocScrutinizerhmm, ok22:48
freemangordonAnd I disassembed NOLO here, it is thumb compiled22:48
DocScrutinizerwas too bad if they shipped chips with buggy mask programmed firmware ;-P22:48
freemangordonand you were correct, NOLO does not run in secure mode (secure as in TrustZone or whatever it is)22:49
DocScrutinizeranyway, when the ne problem is about flushing pipelibnes/caches on context switch, then xloader and even NOLO are kinda out of the line of fire, as they all ahve no preemptive multitasking22:49
freemangordonIncorrect, Security Manager is in boorom22:50
freemangordonin Ll cache is controlled via calls to it22:50
DocScrutinizereh?22:50
freemangordonsorry, L2 cache22:50
DocScrutinizerI missw22:51
DocScrutinizermissed the link22:51
freemangordonthere are several places in kernel when cache is invalidated, that is achieved via SM call22:51
DocScrutinizerbetween preemptive multitaksing and bootrom22:51
freemangordonand SM call is handled in bootrom22:52
DocScrutinizerand esp the fact that neither bootrom nor xloader or NOLO do have any (preemptive, or whatever) multitasking22:52
freemangordonthe code is called from kernel22:52
freemangordonwhich is22:53
DocScrutinizerso they can all happily ignore all SiERR related to randomly happening context switches, as they are all strictly monothreaded and in that sense 1-dimensional and deterministic22:53
DocScrutinizerin a strictly sequential code there's no need for flushing any cache or pipe22:54
DocScrutinizerexcept for branches, and even those are predefined22:54
freemangordonno, the problem is that it could happen to execute thumb instruction in ARM mode and vice-versa22:54
DocScrutinizeryes, exactly22:55
DocScrutinizeras context switches would switch between the two domains and nothing got implemented on hw level to deal with that22:55
DocScrutinizerthat's my idea of where problems are from22:55
freemangordonwhy domains?22:55
freemangordonAIUI we are talking about virtual(or physical) adresses22:56
freemangordonand I think ROM code is copied in RAM22:56
freemangordonanywa, my point was, that we should never hit second errata, as the register in question is neve modified22:57
freemangordonnever*22:57
freemangordonbootrom does not touch it, and it is too late after switch to unsecure mode22:58
freemangordonmaybe beagleboard, because of some debugging or whatever code in some rom extension modifies it22:59
freemangordonthe errara says that L1 AD register is undefined on *RESET*, not a word about power-on22:59
*** MSameer has joined #maemo-ssu23:00
freemangordonBTW I am not sure if it is not possible to flush BTB by just disabling and re-enabling branch prediction (to wotkaround 430973)23:02
DocScrutinizeryou need to find code that reliably triggers the error, then test what can be done to no longer make it happen, and finally do a theoretical proof why the fix is universally valid and not limited to the set of tests you did23:04
freemangordonwell, that could be problematic, I really have no idea how to trigger that23:05
DocScrutinizerfor your assumed register that never gets touched on N900 the proof seems quite straight and simple, no?23:05
*** MSameer has quit IRC23:05
freemangordonno23:05
freemangordonThere is no way to read that register oudside the secure code23:06
DocScrutinizera) tough the register and cause an exception based on that SiERR, and b) show that actually N900 usually doesn'T use that reg anywhere23:06
DocScrutinizertouch*23:06
freemangordoneven reads are disabled23:06
freemangordonif I could write to it then I will write 0 and be sure second errata is 100% workarounded23:08
freemangordonbut there is just no way23:08
DocScrutinizer:nod:23:08
freemangordonbut I can guess on power-on the content of that register is 023:09
DocScrutinizerso there's another way: a) show the datasheet/whatnot factoids that say "write 0 there and the problem is fixed", rgen b) disassemble bootrom, xloader, and NOLO and show one of them is actually writing 0 to that register23:10
freemangordonthat is what i have done :D23:10
freemangordonoops23:10
freemangordonsorry, misunderstood you23:11
DocScrutinizera2) write to TI and ask what's the POR-default of that register, and if that is sufficient for the assumed-to-fix-it 0 that needs to be written there23:11
freemangordonthat was my idea, bu NOONE is touching that register23:11
Sc0rpiusmy phone is still landscape :(23:11
DocScrutinizerfreemangordon: SPRUF98D is the code of the OMAP3430 datasheet, check what is written there about that register. Might depend on bootmode set by hw pins at low/high on power-on, and whatnot else23:13
freemangordonit is in co-processor, no data in TRM23:14
freemangordonwhatever co-processor c15 means23:14
DocScrutinizeryou might want to talk to jacekowski in #maemo, he also knows a bit about that early boot and coproz et al stuff23:18
*** MSameer has joined #maemo-ssu23:19
*** m0use has quit IRC23:19
DocScrutinizeras long as we can't find an official word about what that reg does or looks like on boot, and we don't know bootrom/xloader touching it, and we have indications the thumb is flawed by a SiERR related to it, we can't consider thumb safe by any metrics23:20
*** m0use has joined #maemo-ssu23:20
freemangordonagree23:20
freemangordonI will try to workaround first errata without enabling IBE bit23:21
DocScrutinizerunless of course we could show the situation that triggers this particular SiERR can never happen on N900 under fremantle23:21
freemangordonwell, it seems to be the case, but lets wait till the end of a week23:22
DocScrutinizerwe'd then live with an open barrel of gunpowder but nobody ever enters the room with any light  or cigarette23:22
freemangordonas my /var/lib/dsme/stats is still empty23:23
DocScrutinizerand it's not enough state "this place isn't that croded, how likely is it somebody cmes by with a burning cigarette" - no we have to make *sure* nobody ever does. I.E. publish a howto advisory what code sequence never must get used, etc23:24
DocScrutinizerthere might be assembler hackers out there that code shit your gcc never would produce23:24
freemangordonwell, there is fix in binutils for another errata23:25
DocScrutinizerfrom the cursory read of MLs I think indications were they "tested" it with heavy load, means: many context switches23:26
freemangordonand I assume it working, as same (or similar) fixes exists in another arm compilers/linkers23:26
freemangordonpaid ones23:26
DocScrutinizersure, still that fix might work just golden in thumb-only environment23:27
freemangordonAIUI, it artificially swithes to arm and then again to thumb, or something like that. Should always work23:28
DocScrutinizeran OS with mixed-mode operation may still behave quite different to a WRT router compiled ALL for thumb23:28
DocScrutinizeron an all-thumb system there can't be arbitrary switches between thumb and arm mode23:29
freemangordonwell, we know wor sure gcc version in scratchbox produses buggy code from time to time, so what, shall we stop using it?23:29
freemangordonfor plain arm23:29
DocScrutinizero.O23:29
freemangordonyes, remember Qt raster engine?23:30
DocScrutinizerbbl23:30
freemangordonok23:30
*** andre__ has quit IRC23:35
*** BCMM has quit IRC23:48
*** mirandir has quit IRC23:49
*** MSameer has quit IRC23:56

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