03:30mrsinisalu: On Linux and xnu/bsd , there isn't many troubles to write the new OS scheduler , I will do that definitely.
03:30mrsinisalu: however offhand i do not know how to support microsoft windows, which is something that i definitely need to do as well.
03:32mrsinisalu: Generally two most important tables on OS that needs to be cached or stored with an offered compression in their in-kernel memory structures are pid descriptor tables and interrupt descriptor tables.
03:33mrsinisalu: TDT and IDT in short that is.
03:41mrsinisalu: There are lot's of possibilities to tap into new scheduling , basically if microsoft has some ABI/API of the kernel, it also can be done in userspace. Marcel is such approach. Userspace can link into kernel. but not the other way around.
03:41mrsinisalu: I'd like to also turn off paging, and tap into some of the schedulers API and or twiggle with communication syscalls
04:49imirkin: skeggsb: any thoughts on the vbios loading issue in https://bugs.freedesktop.org/show_bug.cgi?id=111371 ?
04:50imirkin: skeggsb: basically the PCIR header is past 0x1000 into pramin
05:10skeggsb: ugh, that's a little annoying
05:11imirkin: yeah. on a NV4 no less :)
08:33mrsinisalu: so general view is a interrupted timeslices around 10-100ms non-preemptive or preemptive, so on context switch regs are saved into memory in kernel and the slices program counter is restored with the rest of the regs for the slice scheduled
08:39mrsinisalu: so definitely one path is not to restore the program counter, and also use special memory variables to restore on context switch in addition to some of the semi-real ones
08:42mrsinisalu: in other words, interrupt descriptor table restores the program counter, but process descriptor tables do not, so it will always restore to the scheduling loop not to new pid, new pid registers ara handled specially
08:43mrsinisalu: but i am not very well present on what people like with kernel and what they do not, it is pretty obvious that i can do implementation which probably will not be accepted first
08:43mrsinisalu: cause i have not followed the kernel debates very carefully
08:52mrsinisalu: i remember from old times that drm memory manager is a flexible kind, it can take also physical address without paging for gart
08:54mrsinisalu: while it seems absolutely secur to do that naturally allready, i also remember vaguely the lines that allowed doing it almost for sure.
08:56mrsinisalu: so to speak driver modification for gpus in queue scheduling is next up after os is fixed, i have lots of work to do, again i was not able to take part of the summer mostly
09:11mrsinisalu: well the pagetables reside itself in memory and branches need special handling, TLB is fine to be implemented, since this is hw structure? but pagetables are not worth to be walked, everything depends on the cache sizes though
09:14mrsinisalu: for a data memory access like textures from gart or gpus mmu, so that things won't be walked
09:14mrsinisalu: my assumption is that memory needs to be physically contiguous
09:14mrsinisalu: and tiling capable hence, i think gem also supported that
09:20mrsinisalu: and in OS terms, i remember that the code of the programs for CPU live in somewhere in the bottom of the stack in code/text section, which is the memory of instruction cache which will be marked as protected
09:21mrsinisalu: i.e don't ever want to access that memory for reading unless specifically told to init the stuff
09:24mrsinisalu: on gpu that is in drm command buffers which belong to instruction stream somewhere, i need to check that, this needs also memory protection toggling on request
09:26mrsinisalu: memory protection quering and marking is provided by the AXI/PCI-E/PCI/AGP and such interconnects AFAIK.
09:28mrsinisalu: on OS if scheduling is done correctly mem protection to bypass ifetches is not needed actually since i allready told only program counter restoring need to be avoided
09:28mrsinisalu: however to bypass instruction fetches on GPU this should be perhaps needed
09:37mrsinisalu: generally this type of GPU scheduling is something that AQP or ACES allready do for AMD cards, but this can be done on any of the shader capable cards, though along with compression the best cards are VLIW cards like kepler and 950GMA, r300-r900 etc.
09:37mrsinisalu: hence on intel i really advise to forget about SIMD and go back to VLIW if possible
09:38mrsinisalu: the patents grant is active status paid for intel VLIW chip technology, so it should be possible
09:39mrsinisalu: i am not sure they had unified VLIW cards though, all where specialized that i am aware about
09:40mrsinisalu: or of
09:41mrsinisalu: as an accelerator SNB is enough of a card, but compared to vliw it is weaker not stronger, but it can be made to work too though
09:47mrsinisalu: and once again to branch in queues to another command CF is not needed for this, so it completetly does not matter that VLIW cards did not have dynamic branches
09:47mrsinisalu: for intel and maybe r300 and such
09:50mrsinisalu: i'd say even branches are a bad design , and totally uneeded waste of resources on GPUs
09:53mrsinisalu: and bitwise emulation on those cards work too very well, but you want to use twos complement to do it faster
10:02mrsinisalu: if I am not mistaken any of the bitwise op can be done almost the same fast as in hw, that includes XOR AND OR and LSH RSH
10:03mrsinisalu: i went through the network on how they do it, and they do not do it correctly i.e performantly, never looked into nir though or mesa IR
10:05mrsinisalu: same should go to double precision emulation, if you want to do stuff fast on shorter pipeline i.e fast pipeline, two's complement is a must in the mix of things
10:12mrsinisalu: so i talked about how to pack booleans using uniformly distributed sums in the variables for the contiguous address space, that one uses the mentioned twos complement too
10:12mrsinisalu: if you were to do it another way it would not work as fast
10:21mrsinisalu: Imo i at least try to do the work properly, however killers do not try anything to be done properly, you are not quite as bad as such, but in my country to get paid to kill people that looks bad.
10:23mrsinisalu: i'd be very satisfied to take off salary from all of them from mental insitution to violate laws like this majorly, of course i am a bit handycapped too being on the meds, which is why i rambled so much
10:24mrsinisalu: i knew professional opinion about this combination i was given 400mg quetiapine XL additive with 15mg aripiprazole was way too big blast to me
10:25mrsinisalu: but at the bottom there is a thing that worked against the death that was alcohol
10:26mrsinisalu: it was such a bad blast or bomb this poisoness combination that 1l vodka did not make me drunk enough just merely made me feel better
10:32mrsinisalu: if you were me, i bet you would want to get rid of such people even more than me, i've been enough civil with scammers like this, who do not have a single hard evidence against me and violate laws
10:37mrsinisalu: i think some of the painkillers and sedaters do not play nicely with doing physical excersizes either, when i did not show any of the agression i was still treated as a threat there and given sedating injection on entry, i am not sure what the fuck happened did it thicken the blood flow, but combined with some
10:38mrsinisalu: sports outside i think i got almost parallized, possibly something went into trombosis kinda path
10:41mrsinisalu: the sides of that injection were untolerably mad so to speak, two hours sort of like on half of the body was dysfunctional, and tongue wanted come out of mouth like it was full of something, highly bad feeling and worthless feeling added
10:42mrsinisalu: i think such people should be banned from doing such things as human tests on innocent people and banned from any work in the future
10:43mrsinisalu: regardless that those are estonian internationals or citizens they are worth it, and it is the easier kind of sentence, if i put them under the same doses, i am sure they are dead
10:45mrsinisalu: what i think you are confused about as this seems to be a friendly fire in between estonians here indeed, but not all estonians are like that, it is mainly about me locally who they complain about though
10:47mrsinisalu: even though on the average those normal good blokes and guys and girls suffer in greater deal too, but there are piles of nice estonians too
10:53mrsinisalu: i mean i do not like the tolerant politics direction where anything unusual or sidedirected stuff is considered with tolerance, like gay relations or some birth deffect to give preemtion or prioritised rights to such, as most of them won't learn and manage to behave and fly in that sense, more important is that the stronger ones develop to take care of them
10:56mrsinisalu: i would consider such groups as humans with some type of rights, but as they attack me daily basis for 20years, i do not think they are human who diserve more rights than me
10:56mrsinisalu: after the war2 everything is messed up, even german most powerful leaders force those kind of insane politics of tolerance
10:59mrsinisalu: so it has been rather the layers who are unsatisfied here in my country who do that, rather than all estonians being bad
11:00glumia: Hi, can someone help me please? I'm working on this bug (https://bugs.freedesktop.org/show_bug.cgi?id=111392) and I would like to find out which registers these addresses (like 022544 or 619444) refer to... I tried to look at envytools but I don't find the infos I need
11:02mrsinisalu: i am going off to drink a bit and then program, a final nice day, i remember the nvapeek command and mmiotraces and some scripts that annotate the regs with names
11:03mrsinisalu: yeah in nouveau code aside from envtyools it's a hacky code where lot of mmio regs are not described and commented on
11:04mrsinisalu: glumia: so before i did not do any research all remember that i was also very confused what those numbers in the codebase do and such.
11:04glumia: Unfortunately I get "WARN: Can't probe 0000:01:00.0\nPCI init failure!" when I try to run nvapeek or nvalist (yes, I'm running the command as root)
11:06mrsinisalu: no clue about this one, not even a real thought present, so you need to wait for others to comment anyways
11:06glumia: Okay, thanks anyway!
11:07mrsinisalu: what card is that, and is the cards pci even functional, what works and what does not? 2d 3d works?
11:08mrsinisalu: gm107 right?
11:08glumia: Nvidia GTX960M (gm107), everything works fine but I get those error messages on dmesg
11:08mrsinisalu: it should be kepler mid-level card perhaps, best supported card imo, shocking.
11:09mrsinisalu: ahah, yeah yeah, this card should work well
11:09mrsinisalu: allthough it could be maxwell have to check
11:12mrsinisalu: glumia: yeah this is the last very good card from nvidia
11:13mrsinisalu: i mean generation of cards it is low power, and performant
11:17mrsinisalu: so let's try to hypothesize why an mmio read which is reading from chip into the memory should fault?
11:17mrsinisalu: one obvious thing is that such number is reserved or does not exist and code has a bug right
11:21glumia: Yeah, it's what I'm thinking, that's why I would like to know to what registers those addresses refers to
11:21mrsinisalu: glumia: but about that card, i have one question to you! has that one also freezed due to mesa multithreading bugs
11:22mrsinisalu: have you faced a freezing when using consitently app like google earth or what was the war game that did it on my kepler.
11:24glumia: hmmm if I remember correctly some time ago it freezed on vlc
11:24mrsinisalu: it shows up as program crash and so to speak command processor scheduler should reset it
11:28glumia: Sorry, had to reboot
11:31karolherbst: glumia: iomem=relaxed
11:31karolherbst: as a kernel param
11:31karolherbst: you can't use the nva tools when a driver is using the device
11:32karolherbst: or at least that's the case with the nvidia driver
11:32RSpliet: glumia: also, the tool from envytools you were looking for is "lookup". e.g. lookup -a 117 0x22544
11:32RSpliet: where -a 117 specifies it's an NV117 (a.k.a GM107)
11:33RSpliet: Not sure how much useful information you'll get out of it, I'm not sure how well it's kept up to date with anything newer than Maxwell, but that's how to navigate the docs we do have
11:35karolherbst: I think this is the display engine
11:35karolherbst: (in the bug)
11:35RSpliet: By the way, for what it's worth, read faults are bad and should be fixed (and should be relatively easy to fix), but do not necessarily manifest into actual problems.
11:35karolherbst: should be trivial to fix actually
11:35karolherbst: or might be already fixed
11:35karolherbst: ohh, new kernel
11:35karolherbst: so not
11:36karolherbst: RSpliet: I think in this case (111392) it's just a laptop GPU with no display functionality
11:36karolherbst: but reported as a VGA device
11:37karolherbst: glumia: lspci -s 0000:01:00.0
11:37karolherbst: does it say VGA or 3D?
11:37RSpliet: karolherbst: Thanks for looking into the bug in greater detail. My remarks were very high level ;-)
11:37karolherbst: I think I actually sent out a patch for some of that already...
11:37karolherbst: it's just sad that this VGA vs 3D pci stuff is super meaningless
11:38glumia: lspci output: 01:00.0 3D controller: NVIDIA Corporation GM107M [GeForce GTX 960M] (rev a2)
11:38karolherbst: ohh, interesting
11:39karolherbst: we access that reg inside gm107_gr_init_bios_2 it seems
11:39glumia: It uses the integrated intel as vga controller.
11:39karolherbst: glumia: yeah.. but sometimes external ports are wired to the nvidia one
11:39karolherbst: and on some laptops.. they aren't
11:40karolherbst: glumia: are you into compiling your kernel?
11:41glumia: Never done, but I can give it a try
11:41karolherbst: that's the function I think causes that
11:42karolherbst: at least the 0x619444 one
11:43karolherbst: but it might be tricky to figure out what's up with that one
11:46glumia: Hmm I'll take a look and see what I can do
11:47glumia: RSpliet: thanks for the lookup tip
12:56mrsinisalu: that jtag stuff seems very real but quite complex, at least under meds :D cortex a7 has itm and etm/etb mode of the jtag, this cpu is not very popular but is used in iphone SX clones, LOL!
12:57mrsinisalu: realistically it should be possible to make it work some thousound and million times faster without even dealing yet with gpu which add another boost mali-400 on the clone :D
12:59mrsinisalu: because the performance boost on cpu is a curve from all the timeslices amount of instruction during some deeper amount of time
12:59mrsinisalu: it will be huge since only with one instruction when bypassing paging and ifetch and decode the boost is about 3or more fold
13:00mrsinisalu: if you think correctly on this problem then yeah the boost is around million fold even on cortex a7
13:01mrsinisalu: but if we talked to maybe a whitehat a security guy, than i would not know how to make debug pipeline very secure yet
13:03mrsinisalu: microsoft has a joint program with mediatek on their platform for drivers and also a sales with vouchers for mediatek mt3620
13:04mrsinisalu: this is azure sphere, and that hw is a freak and complex to use, but largely a lot more powerful then users would expect
13:12mrsinisalu: cortex a9, that iphone 4 had, has even two methods to do the fast computation, it came with two cores, and has both debug pipeline and speculative out-of-oder 60entry rob per core
13:13mrsinisalu: i loved iphone 4 and 5SE very nice phones, 5 is manufactured again even with modern hw by apple
13:15mrsinisalu: so yeah sw is under propriatary drivers and microcode you try to follow and disasm and decompile also miles away from real hw capabilities
13:16mrsinisalu: i am not entirely sure if this is intentional or agreed upon in sw to lag behind or other complication
13:17mrsinisalu: so if you were or are a big brain without much troubles in life (I have many my own) youd end up pimping up anything in your tree cheaply
13:18mrsinisalu: and would also understand finally that life is very beautiful even with physical injuries until your brain works!
13:20mrsinisalu: I also thought that there are no subjects as sharp to lock me up, by outsiders like this, but appareantly the born cripple therapeut wanted to kill me cause of envy, and by definition on such road it was also certain that his brain did not work well to understand other paths in life.
13:21mrsinisalu: to sum up, they initiated a conflicting scam, cause no physical strengths neither thinking present.
13:21mrsinisalu: when one of them was present it would be almost still ok
13:23mrsinisalu: lots of born cripples think they are the most intelligent people in the world, but it's not a reality!
13:24mrsinisalu: I am daily basis said that i do not make any work, when doing all the garden work and most the heavilifting and scoring, they only think they are geniuses, actually they only crap.
13:26mrsinisalu: I could not live with the fact, that i killed an innocent person, but leftover can, cause he faced injustice from the beginning.
13:32mrsinisalu: it is just different levels of evolution, those who are in the same level with me, never violate either basically, and it is right away spotted too
13:32mrsinisalu: but those who have lagged behind much and are in a level of stoneage with their brain structure and such, those tend to be vicous and problematic to me
13:40mrsinisalu: i remember a slight time ago, some woman wanted me to pay her bills, and daily basis try to tease me with humiliations that i am not good enough partner for her, was some zombie fucktard who tried to cause me to make suicide that way until that slut was again sent to pick strawberries with her finnish and other partners
13:40mrsinisalu: she just overestimated her skills and influence on my princibles and understandings
13:41wrl: the hell going on in here
13:43mrsinisalu: now she has allies among some muslims who this undereducated idiot deals with, i have more then one such sources of sexual humiliating scammers
13:44mrsinisalu: estonians have a term of such deluded scammers, lõpetas eluülikooli eriauhinnaga "cum laubal"
13:45mrsinisalu: that rhimes in estonia, meaning is life school was graduated with award cum on the forehead
13:49mrsinisalu: i have not commented on those things much, but it has been done to me quite a lot, those freak fucktards friends are mostly screaming outside to me a wanker wanker wanker stuff.
13:51mrsinisalu: morans of such a degree, that like a wind plays with a leave they do not have any difference if they are alive or dead, it is not even worth to punish freaks like this.
13:51mrsinisalu: it is just annoying to be annoyed daily basis though
13:54mrsinisalu: so i have those humiliaters all around the world, but all of them are orginally orginating from some relation the the girl who did see hallucinations and her male fuckers in a fucker fuckee scandal intrigue
13:57mrsinisalu: where as i hate scandals tention and intrigues, there are certain types of humans who can not live without them.
13:57mrsinisalu: scamming and flooding their intrigues and scandals onto my identity
13:58mrsinisalu: the reason for this under the hood, is the lack of any type of resillience all in all
13:58mrsinisalu: and internal strength and such.
14:00mrsinisalu: it it understanable for the woman, they tend to hallucinate while pussy bleeds and are easily manipulative and dominated by retarded men, who are stronger
14:01mrsinisalu: but intruguant men is something i do not understand
14:05mrsinisalu: you want to give some fame to your estonian palls, all attacks on me has been very cowardish ones, no real men have attacked me for ages since they are all at my side
14:06mrsinisalu: thos who did those scams are all chromosome faulted lunatics
14:10mrsinisalu: hence you find coming to attack me at my home, i have always interferring sources to this matter, as you have noticed
14:10mrsinisalu: i.e out of the blue my people come to defend me
14:12mrsinisalu: it is clear to all, that i got massively injured without none having tried to help me , and it has changed for now
14:13mrsinisalu: i always have defence now when i can not clear up the conflicts my own, most real people are really embarrassed cause of this
14:20mrsinisalu: no it isn't that fradulent person will actupon feeling the guilt, the innocent ones are feeling this that could not help me on time!
14:34tomeu: karolherbst, pmoreau: btw, I was able to get plaidml to do some matrix calculation on a jetson
14:34tomeu: running an actual model got into an infinite loop in the nir_lower_goto_ifs pass, though
14:35karolherbst: tomeu: yeah... I guess that can happen depending on the shader
14:35tomeu: had to disable the spirv validator, though
14:36karolherbst: there are still some bugs in the validator
14:37karolherbst: I think a validator from master works fine
14:37karolherbst: tomeu: it's the vloadn and generic address space issue you ran into, no?
14:38tomeu: I'm actually using spirv-tools from master
14:39tomeu: hmm, no, it was always complaining about undefined variables
14:41karolherbst: ohh, I see
14:41karolherbst: might be a different issue
14:42tomeu: have pushed my pile of hacks to https://gitlab.freedesktop.org/tomeu/mesa/commits/clover_support_nir_target
14:42karolherbst: tomeu: we already have vec16 stuff
14:43karolherbst: there is even an MR
14:43karolherbst: tomeu: https://gitlab.freedesktop.org/mesa/mesa/merge_requests/477
14:43karolherbst: you have that on your branch
14:44karolherbst: that callback stuff
14:44karolherbst: that I've fixed
14:45tomeu:is happy to maintain a consolidated branch in the near futur
14:46karolherbst: tomeu: https://github.com/karolherbst/mesa/commit/d7a8e2efec34243f51d76ce51bc4bd108a4fa2ed
14:46karolherbst: but I bet it doesn't apply anymore
14:47tomeu: np, will apply tomorrow
14:47karolherbst: tomeu: my plan is to get the basic stuff merged first and then reapplying everything I have on a wip branch...
14:47karolherbst: otherwise I always end up solving merge conflicts
14:47tomeu: yeah, mainlining stuff is always better :)
14:49karolherbst: tomeu: do you think you have time and motivation to take care of the goto_if lowering?
14:49karolherbst: shouldn't be too much work... just requires like a week of dedication or so
16:42mrsinisalu: actually karolherbst: unstructured to structured control flow pass, not only that it existed in the glsl compiler and many other compilers, is not very needed feature, can be of course done but real performant code does not find much use for this
16:43mrsinisalu: acutally on correct in queue computation where you map FUs to queue entries, everything is a branch
16:44mrsinisalu: every instruction in the stream of cont address space is a branch somewhere on the wfid line and or row
16:45mrsinisalu: err column
16:45mrsinisalu: those branches are calculated highly fast though at runtime
16:48mrsinisalu: what i am saying there are lots of wise people in sweden and probably the spir-v and khronos stuff at the compiler level is ok, but it is not entirely correct for performance
16:57mrsinisalu: I also recommend you to skip the opencl, since we need to and are about to do lots more performant things ;)
16:58mrsinisalu: 60percent of the inventions in the spec actually should be quite quickly denied to implement
16:58mrsinisalu: cause they are not performant
16:59mrsinisalu: nvidia or amd would want to make illusion that they are, but karolherbst and some many other folks work at red hat so:
16:59mrsinisalu: they can ignore this shit.
17:06mrsinisalu: even though, well what opencl does is conditional warp id based execution on the arbiter, once you run many iteration back to back it can run many copies of the program
17:06mrsinisalu: but once you don't it won't work out well
17:10mrsinisalu: there are so called blocks of intra-CTA in nouveau terms, iner-CTA and software managed iterations
17:11mrsinisalu: which just repeat all the same stuff again, once you have an iteration to do, opencl is good
17:11mrsinisalu: but when you do not , you need another hack...
17:16mrsinisalu: it can be rather checked maybe i confuse the terms, but the idea remains the same
17:16mrsinisalu: in CUDA terms there are dims grids and blocks
17:31mrsinisalu: actually i am aware of this long time ago, so you have threadindex which is inter-CTA, you have block index which is intra-CTA
17:32mrsinisalu: you also have griddim which is software abstraction and you have also blockdim which is also software abstraction
17:33mrsinisalu: griddim should be inter-cta threads aligned to dimensions
17:33mrsinisalu: but to make iterations on intra-cta several of them you need blockdims
17:40dirbaio: Hello! I'm trying to debug https://bugs.freedesktop.org/show_bug.cgi?id=100228
17:41dirbaio: tldr is 1050ti mobile (GP107M) fails in hybrid graphics mode but works perfectly in discrete graphics mode
17:41dirbaio: and nvidia blob works OK in both (but only with X and with tons of other issues)
17:42dirbaio: any pointers on what I can do to debug? I'm willing to put in the work and send patches if I manage to get it working
17:44dirbaio: mmiotrace maybe?
17:53mrsinisalu: yeah please help the user yourself, i am not very uptodate with bugs. However with cuda and opencl upon meeting a branch, you can write so that if threadid.x+blockid.x=somevalue you branch somewhere
17:54karolherbst: dirbaio: uff, that's the issue where booting the secured firmware is failing
17:54karolherbst: it's an annoying issue and skeggsb is already looking into it
17:55dirbaio: hm.. but what could cause it to fail only in hybrid graphics mode?
17:56karolherbst: random things
17:56karolherbst: we don't know
17:58dirbaio: well, I got an affected machine, let me know if there's anything I can test to help
17:58karolherbst: dirbaio: mhh, I know of some stuff, but I also know it works... it's still nothing we could push, because it's quite hacky
17:58karolherbst: we are still trying to figure out the root causes
17:58karolherbst: I have an affected machine as well
17:59dirbaio: is the code somewhere available?
18:04mrsinisalu: someone may second, but i have not found a way to go back to queue entry which was allready executed using opencl abstractions, in other words, the sequence is like this: if the threadid and blockid or something is some value or modulo of or whatver criteria, it will fetch the instruction and execute as many time as needed fast
18:53mrsinisalu: in my opinion you have no control with opencl at all where in the queues stuff will end up. So you do some start arithmetic that you run the counter 4096 times, nested somehow or straight incremeting by one i mean whatever
18:53mrsinisalu: but the chip decides where it places it
18:57mrsinisalu: in other words absolutely no chanche to reuse the onchip queue entry
19:04mrsinisalu: what i think is that opencl counter and fetch/decode and simd arbiter stuff is completly decoupled
19:06mrsinisalu: you do like 4*5 is 20 if idx=1 do this if idx=2 do that , the best way you can replace the slot with a fluke somehow, but after the opencl counter is mathing per instruction , it only then does the real arbitration
19:11mrsinisalu: the distribution in the queues will end up entirely random hence, opencl neither cuda is not useful hence , it is only useful when multiple concurrent copies run once per single instruction
19:14mrsinisalu: opencl lets user to decide how ever he wants to schedule stuff, but only on full pipeline where only several iterations can work in parallel and back to back from queues
19:15mrsinisalu: cause programmer arbitrarily controls scheduling things are utterly messed up right away
19:16mrsinisalu: it seems that some illuminaties or someone wants to make you belive that this is the limits what hw can do, but it is not so of course
19:24mrsinisalu: no it isn't secure either, you still get the instructions extracted purely, but in a slow mode you see...
19:26mrsinisalu: this is rather a business strategy it very distant and orthoganal thing about real physics/electrotechnics and hw capabalities
19:27mrsinisalu: you see those are totally separate things science and business
19:28mrsinisalu: the strategy in trolling is done so people would purchase cards with gradual improvements done at very slow pace for the next 100or200years
19:34mrsinisalu: similarly using bent science to troll people to pay money, back days when albert einstein was living a german jew, he had a grim in his face with dissapointment, when his ideas to do science that people would feel good were used to do stuff to kill people
19:43mrsinisalu: i don't think the evilness will totally be wiped off the planet, but there is a chanche of some smaller amounts present, when the goods are offered, I am on that line my own
19:43mrsinisalu: i think i will be violated until the general satsifaction and welfare reaches to towards better
19:44mrsinisalu: most strategicians also offered that way, education is general way to stop the overpopulation naturally
19:44Lyude: mupuf: ^
19:45mrsinisalu: and also stop overly too big gaps of wealth differences
19:58mrsinisalu: when you want to feel what i felt, then try one dose...of 400mg quetipine aka seroquel brand name combined with 15mg aripiprazole named abilify and show your spirit and intelligence and patients like i have shown
19:58mrsinisalu: you'll never make it like i did
20:00mrsinisalu: patience heck
20:02mrsinisalu: i would like to see Lyude what kind of face you show up with then!
20:14mrsinisalu: i'd lay bets you do not make past the first months, because the tough i am , i had severa problems with such a dose.
20:14mrsinisalu: and did it all the way max 2months
20:18mrsinisalu: all the relatives who made those scams are looking over the shoulder for the rest of their lives with embarrassement with the unsuccsessful kill of atempt gone soar as illegal lawless conspiring event
20:19mrsinisalu: their relatives will be recognized cause of the anchestors failed attempts etc.
20:35Lyude: don't like that
21:02glumia: Hi, can someone help me please on https://bugs.freedesktop.org/show_bug.cgi?id=111392 ? I'm on a dead point
21:33dirbaio: karolherbst: took a look at your github :)
21:33dirbaio: this commit makes it work -> https://github.com/karolherbst/nouveau/commit/7e24d8aba04726dde912434420ba2f8a962a6dab
21:33karolherbst: ahh, yeah
21:33karolherbst: but it's still not something we can push out sadly
21:34dirbaio: the others ones on secboot_fixes don't seem to make a difference
21:34dirbaio: yeah I understand
21:34dirbaio: "turning it off and on" is not a proper solution, hahaha
21:34karolherbst: yeah.. the other things are runpm related
21:35dirbaio: runpm_fixes? should I check thouse out too?
21:35dirbaio: oh they're the same commits
21:35dirbaio: I see
21:37dirbaio: oh well
21:37dirbaio: thanks a lot for all your hard work
21:38dirbaio: hopefully you guys will figure it out soon
21:39dirbaio: is your affected machine also optimus?
21:39karolherbst: I have enough workaround to make it work.. but it's not something I wish users to have to do
21:39dirbaio: and if you switch it on bios to discrete graphics does it work without the hack? mine works
21:40karolherbst: ohh, I can't switch it on this one
21:40dirbaio: thinkpad x1 extreme
21:40dirbaio: ohh.. some can't
21:40dirbaio: my previous laptop also was optimus and it couldn't.. this one can
21:43dirbaio: maybe the fact that the issue is optimus-specific is a clue to find the root cause... dunno
22:15Lyude: Anyone seen this issue with nouveau on drm-tip? https://paste.fedoraproject.org/paste/A8KSiJuz27bwetWrh3AEmw
22:16Lyude: seems to happen just from loading gnome-shell
22:40Lyude: skeggsb: btw, didn't see your response to my SOR/PIOR/drm_encoder question until just now. I looked into just bumping the encoder limit to 64 and it ended up actually being pretty easy, think it's OK if we go with that solution instead
23:45Lyude: well this is weird. now that I've got the thinkpad P71's discrete GPU working, every single DP port connected to the nvidia GPU reports being unplugged for about 3 seconds when there's nothing being displayed on whatever's copnnected to it, then reports being reconnected a few moments later and repeats