01:38 martsteiner: this multithreaded issue is not something very complicate RSpliet, i had talked about it, neither is doing vulkan sort of thing in opengl's driver, last can be done as linker patch that goes over the code, first can be done in several ways
01:40 martsteiner: just you know in C, you just make sure you understand that program always contains two separate jars of memory in kernel, what are known as stack and heap, first contains all the local/automatic variables, and heap contains all the
01:41 martsteiner: dynamically allocated and static and global variables
01:42 martsteiner: in case of shared context, you wanna share a handle in mesa state somewhere, so the easiest thing to do that, is dynamically mapping that handle to the same memory, in application private stack and heap which is automatical, or in thread memory space
01:43 martsteiner: but i was little suprised why separate context did not work out as planned like in trello it was said, that should refer to some sort of locking bug introduce my nouveaus driver programmer
01:44 martsteiner: *contexts, i.e when there is no shared context by multiple threads, but every cpu thread has it's own context, in such cases, there should not be anything much to be done, everything is handled by hw
01:57 martsteiner: but signing the firmware is slightly more complex issue, as there is a key flashed to the protected memory during a manufacturers initial flash, it is controlled by the falcon secure isa, and memory protection units, so this memory isn't accessible, normally the protection is done with hw fuses
01:58 martsteiner: but since the falcon isa is not entirely familiar to me, i haven't hacked on this stuff, i mean the isa has documents, probably the secure bit's are not that well known, as it is a custom isa, not usual arm
02:06 martsteiner: if anyone knows how phones lock the baseband , then it's analogous to that phenomen, but as arm trustzone evasions are fully comprehended it is basically an easy task to unlock it there there are fuses i.e memory protection registers
02:07 martsteiner: so mwk said there is no key, because it does not allow to read that memory, it comes back as zeros, until the fuse is either reprogrammed or memory protection unit registers are reprogrammed
02:10 martsteiner: in other words if you wanna escalate or evate the falconts trustzone AP analogue, you gotta comprehend the isa and write an exploit to get intouch with rifhts to the protected memory, in other words write an exploit
02:20 martsteiner: exploits idea is to gain access for fetching the key gaining access to that protected memory area and also reversing the algorithimic security, as you understand cryptographically bruteforcing that kind of thing isn't possible
02:22 martsteiner: for me this is slightly too complex work, and i myself will not be dealing with this issue to unlock the hw for recent firmwares with exploits like this, primarly since i do not know the falcon isa very well, getting the docs online could take major time, however for usual arm isas those documents are there in excess over the net, but unfortunently of course not for falco microcontrollers then again
02:33 martsteiner: it is easy on arm, since they have exploits, to be honest writing one, nvidia may end up sueing freely that individual also, allthough apple even hired some exploiters instead, but on qualcomm the fuses can be reprogrammed, one transistor sits in saturation mode, i think it is the same case with nvidia, but of course it does not have to me, and it is easy cause in arm baseband case, the memory is not accessed physically on the hw but
02:33 martsteiner: over the AP's MMU
02:35 martsteiner: it could be that nvidia got those mechanism entirely correct, and this could be situation where this is unhackable overall, as it can be done right too, without mistakes
02:36 martsteiner: for AX processors for apple, there hilarious security mistakes, so probably a child can evade their "paranoid" security measures
02:41 martsteiner: so in hw, somewhere the key is pinned, and if hw accesses it, the memory is given access, so theoretically it is hackable, but practically unhackable, since if done right that kind of memory locations is like finding needle from the haystack
02:57 martsteiner: and generally the arguments about linux desktop not being popular i understand, users amoung being 2% of the market, that has lead me to ignore all microsoft users who are in constant trouble, at least in computing sense, and with all their ungoing troubles cause they ignore the offer that is meant for their own good , as to use proper platform
02:58 martsteiner: they gotta be usually ignored overall in all areas not just computing, in other words this 98% of people tend to be seriously dumb in my opinion
02:59 martsteiner: i am not about to help someone, if their mission in life is to suffer and hence annoy others with that, some even do it deliberately
03:08 martsteiner: i've been user and lately programmer finally for 15years on linux platform, i have an easy go, and i am serious person, evrything works all the time and i am makins sense most of the time, and that is all in contrast to microsoft users, who bother real men with their constant troubles of instability, my impression about such systems i went angry and was about to throw out my computer forced to touch such systems, so the unlocking also
03:08 martsteiner: was managed inclusively under linux as always
03:09 martsteiner: i did some work on apple hw, and all winows tools crashed and system crashed and i mean most of the tools also did not work, used that system over some decade out of interest, and it is entire crap
03:36 martsteiner: and to all musicians, i win every bet when i want to the next task was about to become if someone needed and minor 50eur sum is forwarded to me for 1month work, i demonstrate easily how every VST will run under linux with a mixute of some software, windows blobs but still, without wine, windows kernel module, thin wrapping code and some parts of wine/LLVM, it just proof of concepts, i truely express my opinion that people should not use
03:36 martsteiner: microsoft though still overall
03:36 martsteiner: i for instance don't need those VST's i use native software but, it is an easy task
04:09 martsteiner: hw wise i was in and endless loop deciding what to purchase what hw, bought used AMD hw, it was like an insult word for others, i ignored that, i am entirely impressed by it's performance and architecture/design, and it transformed my way of thinking that those days, if companies invest into doing hw and sw, they likely still entirely succeed, this hw has not been exception, and i think likely tegra from nvidia allthough being arm64 is
04:09 martsteiner: good too
04:17 martsteiner: with preliminary/initial inspection about the laptops design to meet the requirements, like maybe ventilation openings on the side not only in the bottom, then otherwise by hw and sw, it is not that big of a worry, those days most laptops do very well, and it should had not been such a long process to pick one
04:18 martsteiner: i account if my equipment is stolen those days, i get new one easily, nice technlogy out, so aight, i head off to sleep again
06:56 gnarface: well, that was quite a rant
06:56 gnarface: but i can't say it all sounded crazy
06:56 gnarface: actually alot of it sounded really sane
06:57 gnarface: especially the stuff about the unaccounted-for damage microsoft has done to society
06:59 gnarface: i thought it was overall harmless, or at least a sum-of-zero sort of balanced right&wrong which i don't totally agree with but still shrugged off as "the way the world works" until i realized they'd get away with using it on gradeschool children
13:40 martsteiner: hei gnarface, thanks for encouraging comments, i am not sure what was the confusion about baseband, i point out couple of misconceptions, i first parsed the documentation incorrectly myself too, nck is not randomly generated per device, it is unique fixed per device, but random among different devices, it blown into one certificate fuse, and is concatenated of two other fuse blown id's
13:42 martsteiner: upon restart of the device the gsm connects to auth server , and it sends out the nckey, and it is compared with one in the pk hash on the device, known as unlock code
13:42 martsteiner: if the server says device is locked it sends locked nck and it would not match
13:44 martsteiner: so a little misconception is on the net, that nck is not known, but it can be read from the device, no need to bruteforce it
13:46 martsteiner: but yeah i am kinda of a saint and i refuse to go along with the masses, i refuse to work with the such systems what i previously ranted about, representing the minorities of computing here, i am lonely and rejected, but i still have interest in dealing in with stuff in the wrong way
13:55 martsteiner: this kind of lock is nontheless childlock for experts, but i bet nvidia does it better, in addition i fixed some tools, and qualcomm binary container easily revealed all the nvitems physical and virtual addresses in it's , they came as zeros without privilege escalation for the values, but from there, minor hack can be used to evade the security and overwrite them
14:00 martsteiner: i do not think it's much help in this case, but arm case winocm's tzboot and some german named kirk vertigo are two basic ways how to escalate and invate into trustzone, they are called scm and svc methods, last one just rewrites the pointer of the syscall, and taps into security zone, where everything is open
14:00 martsteiner: in other words, it hijacks the pointers to system calls
14:04 martsteiner: in addition to that hack, i also forced ios downgrade on with a hack, and downgraded iphone 4 to ios 4.0.2 with older baseband firmware, i issue was i was delivered a lost phone that was entirely locked with all elements on,as the owner was foreigner and unable to contacted to and never cared about his lost phone, i put that phone into use again, it is stylish phone in its appareance
14:17 martsteiner: but overall agreed, we had lots of chanches to rebuild all the infrastrcuture, and pisses little off as it turns out instead of migration to proper platforms in a long run, what people instead did was bully and simply locked their minds into improper methods and platforms, it affects everyone in the country, and it is simply dead wrong and pathetic, it seems migration is hard due to different people not being used to think their own
14:19 martsteiner: i early in migrated and practiced all the time, to raise my knowledge in a proper, but, i am one of the few here, dissapointing overall so to speak, one mans knowledge or 100 can't be used to change things, it should had been in thousounds or more, so it will take too much time
14:23 martsteiner: so when unix system experience from the groundup of starting point currently for many is zero level, it will take too big effort to migrate them into correct ways
14:23 martsteiner: their hunches are underdeveloped etc. compared to what i allready have 15years of experience
14:32 martsteiner: i have talked with my friends about it for ages, now maybe final time i offer, that i make ome more intelligent ones a gift at some stage hw like proper laptop, but only prequisite or requirement to collect and use it in the future, is to use proper system on it
14:42 martsteiner: ok cheers, i head off, and i have some things to do, few know it is possible to e-mail me too personally, i respond to all e-mails, and none the less most technology is given out for free, if i see sense on the way or troubles that people have
14:42 martsteiner: bye
19:21 orbea: After updating mesa to the latest git master I found that mplayer segfaults when starting a video file now. Here is a backtrace. https://pastebin.com/yZqN7kpW
19:21 karolherbst: \o/
19:21 karolherbst: dolphin with ubershaders: 18% -> 28% speed
19:21 orbea: cool, as of when?
19:21 karolherbst: some opts I try to do
19:21 orbea: :)
19:21 karolherbst: we have a bug inside the compiler doing useless stuff when spilling
19:22 karolherbst: "total instructions in shared programs : 161535 -> 142203 (-11.97%)"
19:22 orbea: they are already fast enough for some games to be full speed with exclusive and reclocking (Mario Kart Wii)
19:22 karolherbst: I am trying Zelda Wind Waker
19:22 karolherbst: with my patches ~28% in the intro menu
19:23 karolherbst: native res
19:24 karolherbst: orbea: can you check if this patch helps perf a lot for you as well without breaking things? https://github.com/karolherbst/mesa/commit/ee62f048d5d1e42df664e86db84a7f9b6f57b5b8
19:25 karolherbst: it should get you a roughly 50% speedup
19:25 karolherbst: ohhh wait
19:25 orbea: i'll rebuild mesa...oh?
19:25 karolherbst: orbea: did you had a gk110?
19:25 orbea: gk110b
19:25 karolherbst: meh
19:25 karolherbst: it won't do anything there
19:25 orbea: oh well
19:26 karolherbst: gk110b has 255 registers per thread, prior keplers only 63
19:26 karolherbst: so on kepler1 codegen spills a lot
19:35 orbea: ww exclusive is about half speed here while hybrid is full speed
19:36 karolherbst: mhh
19:36 karolherbst: only half?
19:36 karolherbst: I thought you had like a 780ti or so
19:36 orbea: i do, i guess its doing it very slowly...
19:37 karolherbst: odd
19:37 karolherbst: native res?
19:37 orbea: yes
19:37 karolherbst: codegen messes up so hard...
19:37 karolherbst: 53: mov u32 $r5 $r4 (8) + 54: st u32 # l[0xa0] $r5 (8) and to be funny: 55: mov u32 $r5 $r6 (8)
19:38 karolherbst: there is something terribly wrong with codegen
19:39 orbea: :\
19:40 karolherbst: okay, codegen is just terrible regarding branched code
20:47 karolherbst: ohh, I think I've figured it out where those silly st/ld pairs came from
21:32 mangix: karolherbst: https://pastebin.com/DNi6jpvJ
21:32 mangix: branch: maxwell2_playground
21:32 mangix: the first MMIO error is unrelated
21:37 karolherbst: ohh
21:37 karolherbst: you can ignore those
21:38 karolherbst: is it new with the reclocking code?
21:39 karolherbst: mangix: I assume everything works as expected though?
21:39 mangix: i don't notice any issues
21:39 mangix: and yeah, the clk stuff
21:40 karolherbst: have to see where we read those
21:40 karolherbst: but from dmesg it looks clike clk init then
21:40 karolherbst: weird
21:42 mangix: nv_clk_src_hubk01 and nv_clk_src_pmu
21:42 mangix: :)
21:42 karolherbst: ohhhhh
21:43 karolherbst: yeah okay, good point
21:43 mangix:has wasted a few hours looking at the code before realizing those patches introduced the errors
21:43 karolherbst: I should check if we respect all the clock domains properly on maxwell
22:21 karolherbst: oops while doing something in steam: https://gist.github.com/karolherbst/bcbbdfbb52e85b504cf6ed93bbc400c3
22:23 karolherbst: vpgd->obj->addr is broken
22:24 karolherbst: skeggsb: knowing anything about race conditions related to gf100_vm_flush?
23:28 mardinator: i did not elaborate how vulkan kind of thing would work inside opengl driver, you see vulkan is all about banging the ringbuffer with multiple CPU threads, but opengl multithreading has structure that is easily patchable
23:30 mardinator: shared context would have fences or glflush or glfinish in the stream as well seperate context, you'd end up doing a patch which precalculates the offsets in the first case with shared context, naturally it is not handled, cause ringbuffer methods in hw, pushbufs is how you called it for nvidia hw are parallel for separate hw rings naturally
23:31 mardinator: in case of shared context you go over the commands in the linker for instance, and parallelize the command buffers