00:05Celmor: how do I blacklist nouveau without editing files?
00:09RSpliet: boot your kernel with nouveau.modeset=0
00:10Celmor: would that parameter go here? https://puu.sh/t1YEL/790d12aac5.png
00:11RSpliet: I think so
00:11RSpliet: I don't boot using efi, but it looks valid
00:39nyef: Jeansf: Defer Hot Plug Detect stuff until post-resume?
00:48Jeansf: i am not sure what it does, i was looking darts, and prolly won't have time tomorrow, it is said in the comment, the yeah basically it queues commands
00:49Jeansf: but it was you who knew what this hpd is suppose to do
00:55Jeansf: those two patches according to hans are needed to fix your issue maybe, but not sure, almost looks like it
00:56nyef: I don't think that this is what I'm looking for, but I'll add it to the queue to check once the current thing I'm trying has played out.
00:58Jeansf: yep, i wish you good luck, and report back maybe in irc here, if you were able to fix this issue, as i have no appriopriate hw i myself gave up doing some hypothesis for time being:)
00:59Jeansf: well he actually kinda says it is, but, diff looks rather small, it just is a sort of a linked_list a structure full of info that is being used
01:02nyef: Ugh. Genkernel added a little something to the filenames to indicate a "dirty" tree, so I was getting old kernels. No wonder nothing was working.
01:03Jeansf: nyef: ok, but... this diff isn't in drm-next anyways
01:04Jeansf: the linux graphics mainter hasn't yet pulled it into that tree
01:05Jeansf: nyef: but do as i advised, well bens tree only touches nouveau directory, and arilieds tree touches only gpu/drm presumably, just build only those dirs
01:07Jeansf: nyef: from nv50_display.c it shows that hans de goede (i was mistaken with the name before) tries to wake up the screen via some sort of hpd or acpi interrupt stuff
01:09Jeansf: someone named ilya just ilya had reported it, and it wasn't fixed in december, same case as for you, probably only chanche is that the named guy patch fixes AFAIK.
01:17Jeansf: drm_helper_hpd_irq_event(drm->dev); sorry for doing my crazy assumptions, maybe worth to try as my last
01:17Jeansf: he says that line should wake up the sink
01:29pmoreau: I might have a working SPIR-V linker before 2017; I wasn’t expecting that!
01:29pmoreau: Still need to perform the actual linking step, but now that all modules have been properly merged into a single one, it shouldn’t be too difficult.
01:29pmoreau: And then, some optional cleanup
01:46mwk: so... anyone here who still remembers old school GL?
01:46mwk: is it legal to enable eg. light 1, without also enabling light 0?
01:57celmor: if `cat /sys/bus/pci/devices/$id/enable` returns 0, does that mean I have to enable it first to be able to manually bind a driver to it? if so, how would I do that (enabling the device)?
02:01nyef: ... I just destroyed the USB device that I was using, accidentally.
02:01nyef: Shoved it a bit to the side while it was plugged in.
02:02nyef: Chance of repair: Unknown.
02:37mooch2: mwk: you
02:37mooch2: you're a fucking WIZARD
02:37mooch2: pls teach me your ways
02:40mwk: mooch2: insomnia
02:41mooch2: i wish i still had access to your hwtesting shit
02:41mooch2: but i forgot the passwords
02:42mwk: well, that can be fixed
02:42mwk: though I'm kind of using it at the moment
02:44mwk: the hwtests have been reogranized a lot lately
02:44mwk: the current form should be much more useful to you, btw
02:45mwk: it mostly separates the testing-related crap (adjust_orig_mthd etc.) and the actual behavior (emulate_mthd, is_valid_val, ...)
02:49mwk: I'm not entirely happy with it yet (the constructor params and class design are quite horrible), but the general idea will stay that way
03:02mwk: well, that's a lot of fucking state
03:03mwk: I can only imagine how hard it'll be to figure out what it actually *does*
03:04mooch2: mwk: weight??? wtf does THAT do
03:05mooch2: is this like, hardware SKELETAL ANIMATION SHIT
03:06mwk: mooch2: https://www.opengl.org/registry/specs/EXT/vertex_weighting.txt
03:06mwk: basically, yes
03:06mwk: it's not as awesome as it sounds, though
03:06mwk: basically you can scale a vertex by a linear interpolation of two matrices
03:07mooch2: well, vertex shaders would obviously be better
03:07mwk: yeah, but we're talking NV10 here :)
03:07mwk: according to my TODO list, I'm halfway through Celsius methods
03:08mwk: and almost all remaining ones deal with state that I don't model yet
03:08mooch2: is there any kelvin stuff on your todo list?
03:09mwk: I'd like to start modeling the state soon
03:09mooch2: where is that, anyway?
03:09mwk: but there's a bit of a problem, I've run out of AGP slots
03:09mwk: and I'd have to remove one of my NV1x to make room for an NV2x
03:10mwk: I think I'll buy an AGP motherboard or two...
03:11mwk: but then, I haven't even finished modeling NV10 state yet
03:12mwk: and modeling the PIPE state will be a major pain, I think
03:12mwk: which is why I kept putting it off
03:12mooch2: what's PIPE?
03:12mwk: but now... only two more easy methods left
03:13mwk: mooch2: PIPE knows the gate, PIPE is the gate. PIPE is the key and guardian of the gate. Past, present, future, all are one in PIPE.
03:14mooch2: sounds intense
03:14mwk: it's the doorway to non-obvious state on NV10 and, I think, NV20/NV30 too
03:14mooch2: maybe you should use this opportunity to do some pfifo tests!
03:14mwk: I'm not entirely sure how it works
03:14mooch2: so i can FINALLY get nt4 to work
03:14mwk: but, it's sort of like a command FIFO
03:15mwk: you send a PIPE command, and units along the PIPE pass it down until one of the units processes it
03:15mooch2: OH GOD
03:15mooch2: that sounds fucking awful
03:15mwk: basically: vertex fetching, transform & lighting, color assembly, textures, reg combiners, rop
03:15mwk: these are more or less the units
03:16mwk: and, the weird part, you can also do reads from PIPE
03:16mooch2: mwk: do you think you could do some nv3 pfifo tests, to hammer out what every single register does?
03:16mwk: there's lots of state in the graph engine that's not accessible directly via MMIO
03:16mooch2: oh sweet jesus
03:16mwk: I think... I think reading from the PIPE sends some sort of a readback command through the units
03:16mwk: and, if one of them responds, you get the return value
03:17mwk: but, if you read from an invalid PIPE address, the whole GPU hangs, and your machine hangs with it
03:17mwk: so scanning PIPE state is not exactly easy
03:17mwk: and the things you read back from PIPE don't exactly correspond to data written to the PIPE
03:17mooch2: welp, maybe you should take a break from pgraph then
03:18mwk: eg. if you read vertex position, well, you read the current vertex position
03:18mwk: but if you *write* vertex position... it goes through all the PIPE units on the way to the vertex attribute buffers, including T&L
03:19mwk: so your position will be *transformed* by the T&L unit
03:19mwk: so to perform a proper save/restore, you have to put T&L into bypass mode
03:19mwk: set all scaling factors to 1.0, etc.
03:20mwk: luckily, nvidia introduced a better way to save/restore state on NV20, called RDI
03:21mwk: which is basically direct access to internal RAMs of various units
03:21mwk: so you no longer have to mess with PIPE just to look at the state or manipulate it
03:24mwk: mooch2: well... PFIFO is on the TODO list as well
03:24mooch: mwk: maybe do it sooner rather than later
03:24mwk: you want NV3, right?
03:28mwk: well... I still have 2 celsius methods left to do...
03:29mwk: ugh, the methods taking enums are so ridiculously annoying to test
03:29mooch: i would bet
03:30mwk: I'll probably need some special mode for these
03:30mwk: hammer every value in 0-0xffff range with a clean context and see if it fits
03:31mwk: there, I'm officially out of Celsius methods to RE without modeling PIPE
10:20Jeansf: yeah mwk does pretty magical things, the idea of having docs, is utterly helpful and they are enough good..
10:24Jeansf: back days i allready knew roughly what the fixes are, but i would had been lost in the code, basically purely based of those good third party docs, so if i come across having time again, there are also possible fixes to be implemented by me on nouveau also
10:29Jeansf: mwk: it's my belief that it is possible to run games on memory isolated threads the way it is too, without doing atomics i.e locks and mutexes and the bits of pfifo or pgraph to do it i remember was nicely documented
10:33Jeansf: the lockless way is lot more easier and predictable behavior, however usings locks may have unpredited behavior still in some cases..if the accesses are not almost fully serialized, also lockless performs better but needs more memory
10:38Jeansf: i can not remember offhand but i allready had a plan how to modify the engine for parallel lockless algos, it needed a little hack
10:39Jeansf: i think it was pfifo that needed a minor modification the one with subchannels , but pgraph may have been needed minor work too, it's easy to recap based of the docs, once i rearrive there
10:46Jeansf: i read about this hpd stuff, it's yeah hotplug detection capability, and probably hans was right there, according to spec it should open the aux channel for some devices indeed
10:47Jeansf: hopefully has potential to fix nyef issues, with this single hack, of course ben's work blows be off too, that is what the real programmer life is
10:57Jeansf: *me, but i am in parallel planning two patchets one for AMD performance improvements and other for nouveau cpu multithreading, i have some more graphics related work queued, but those are the most important at the moment
11:02Jeansf: one is doing a native hw blob compiler based of wine's direct3d, it is sort of a precompiler, it seems wine has gained full dx11 support in 2016 perhaps, not sure how stable it is, but it supposedly supports tessellation and compute shaders
11:11Jeansf: i did a lot of research on kernel and compiler based multithreading, we remember the dfixer, lots of llvm tools like mcfi, also austrians had https://github.com/thorstent basically liss and canrepair, so there are many automated and manual repair approaches
11:18Jeansf: this is in my opinion in programming the most active research on the web, it really had piles of hits from google, i remember reading hundreds of different pdf's some of them were crap, but lots of relevant material still
11:21Jeansf: the relevant code that would want to get concurrency bugs found is bit more complex as you'd think, for nouveau it's coded in c, but kernel is also involved
11:24Jeansf: i dunno what about calim's glsl codegen, is that thread-safe, but z3 could handle c++ too though
11:24Jeansf: it should convert the stuff to goto-c and back to original representation after mods, and generate a diff
11:28Jeansf: so this could be so that we end up fixing a bit of pfifo and relevant engines with minor patch, then the rest could almost be done ideally automatically if in time hurry
11:32Jeansf: in any event terror is responded with terror as violations with violations, and by set of local people, i feel no sorry that they have begun to face it and face it to their rest of their lives nowdays in responce
11:38Jeansf: don't think we should be bothered by that, as i am not actively involved in the punishment, like human crap is regardless wiped off the business by communities that support also me, and are pretty much annoyed by such activity by those local beefers, have done it couple of decades, and now they are charged responsible for such scams
11:44Jeansf: i applied for a job instead, and i won't blink my eye, if someone comes at me having humiliated me for 20years in a row, and his head is cracked my the community
11:45Jeansf: so for AMD, things work pretty much as i posted, but that is irrelevant here in most part, though scheduling can work for nv50 and to pascal in similar strategy as well
11:46Jeansf: i just changed the DS_CMP2ST against DS_MIN/MAX as it was cheaper ternary on shared memory, 12cycles vs 4cycles
11:50paulk-aldrin: mupuf_ and others who might be into ISA reverse engineering: does that binary ring a bell to anyone: https://chromium.googlesource.com/chromiumos/third_party/arm-trusted-firmware/+/firmware-oak-8438.B/plat/mediatek/mt8173/drivers/spm/spm_mcdi.c#62 ?
11:50paulk-aldrin: I'm trying to figure out its ISA
11:50paulk-aldrin: sofar it looks like 0xf0000000 is return
11:50paulk-aldrin: and I'd bet it's 32-bit fixed-width instructions
11:51Jeansf: as you may know estonian police is paranoid about me as being a betraydor in my country, soul reason for it, has been some retarded rappers thinking that they are someone, and violating my rights, id o not care when they are punished, is it a russian or whoever taht boots such off, cause they fully being also tried to kill me all the time diserve i
11:51paulk-aldrin: 0x17c07c1f seems to be padding
12:01Jeansf: paulk-aldrin: it feels like arm assmebly, i did not open the link, yes on cpu world they do padding to get better cacheline accessing
12:01Jeansf: it's normally done with bunch of nop's
12:02paulk-aldrin: Jeansf, is 0xf0000000 return in ARM assy?
12:02Jeansf: let's check, i myself can not remember, i don't think i ever knew arm assembly, hold on
12:05Jeansf: i have not yet found, mnemonics page is needed for ARM or opcode so to speak
12:06paulk-aldrin: looks like it's usually "mov pc, lr" to return in arm
12:07Jeansf: ok, but why do you discuss this in here, is it some tegra firmware involved in graphics world or some microcontroller in arm isa or what?
12:07paulk-aldrin: well I know that people here worked on the falcon ISA
12:07Jeansf: yes yes, it falcon stuff?
12:08Jeansf: it/is it
12:08paulk-aldrin: so I'm assuming people around here are skilled when it comes to figuring out ISA
12:08paulk-aldrin: I guess not, it's not nvidia
12:08paulk-aldrin: (I realize this is off-topic)
12:08Jeansf: well ok, how about #asm though, there are really skilled assembly guys?
12:09Jeansf: probably here are some folks too, but as you might have seen only active dude here at this point is me
12:10mwk: paulk-aldrin: no way it could be ARM
12:11mwk: I could try a few picks, but it could always just be something custom
12:15Jeansf: but then again he is right, this is a return in arm assmeblu, mnemonics are correct, where is the opcode i.e machine code equivalent, don't want to assemle here
12:17mwk: the binary is real short, too
12:19Jeansf: paulk-aldrin, correct lr is 0xf000000 indeed
12:20Jeansf: http://www.pjrc.com/teensy/beta/DDI0403D_arm_architecture_v7m_reference_manual.pdf page 46
12:22Jeansf: paulk-aldrin: maybe mwk knows what isa falcon uses, i can just google it, i doubt it is arm though, allthough your mediatek firmware seems to be
12:27Jeansf: will read about how generally firmwares are laid out in memory, later on, i have to go prolly, it seems like they have no containers right, pure machine code
12:28Jeansf: they are likely put into one pointer, where they will be executed and always stored in memory
12:31paulk-aldrin: Jeansf, falcon has its own ISA
12:31Jeansf: also up for a debate if that is microcontroller blob, is it that processors MMU is used to seek and translate the virtual address space, probably
12:31paulk-aldrin: Jeansf, before switching to risc-v anyway
12:31paulk-aldrin: well given the size, I doubt there's a mmu
12:34Jeansf: so you think those are raw values in main processor memory, that are transferred to microcontroller specifically, well that makes sense too
12:34paulk-aldrin: mwk, yeah I'm a bit worried it might be a custom ISA
12:34paulk-aldrin: mwk, maybe andes ISA
12:34paulk-aldrin: apparently mediatek used that alreayd
12:35RSpliet: paulk-aldrin: my ASM RE-ing skills are rather limited, but the last line seems to hint at 32-bit immediate support in the ISA (and this variable length instructions) or at code/data section interleaving (0xaa55aa55)
12:36paulk-aldrin: RSpliet, yes you're right 0xaa55aa55 is indeed a constant
12:36paulk-aldrin: you can spot deaddead and beefbeef too
12:36RSpliet: variable length instructions? it doesn't sound like an off-the-shelf ISA to me yet
12:37paulk-aldrin: RSpliet, or that's rodata
12:38Jeansf: yeah no clue either, never got llvm disassembler to work, but gcc's one could be tried i belive
12:38Jeansf: if it disassembles the the raw binary or something
12:39Jeansf: err GNU that was
12:46Jeansf: non-of my tools, including llvm-objdump , gnu objdump or llvm-mc are able to understand that blob
12:53Jeansf: arm-linux-gnueabi-objdump -D --target binary -marm mis.bin
12:54Jeansf: this one did it, but some undefined lines are actual
12:56Jeansf: apt-get install binutils-arm-linux-gnueabi
13:04Jeansf: let's remove commas, with minor sed script though
13:17Jeansf: pretty terrible as i386 disassembler on that blob does a cleaner job...:D
13:23Jeansf: mwk: is right, it is not arm machine code
13:23Jeansf: neither it is x86 of course
13:26Jeansf: hmm, yeah well let's try to to convert it to bin , before disasming it though
13:38Jeansf: it does not still look like to be an arm machine code, i dunno how they managed to reverse engineer such blobs, but this is entirely terrible
13:45Jeansf: aah this is that chinease mtk processor company, you can try their arm standarts
13:49Jeansf: taiwanease though, pardon, seems they have a57 and more powerful arm processors also, not only budget ones
13:55Jeansf: The systems are powered by a MediaTek MT8173 quad-core processor with two ARM Cortex-A72 CPU cores, two ARM Cortex-A53 CPU cores
13:56Jeansf: it is some very powerful arm processor solution, so it is ARM instruction set, but 64-bit of one of those
13:59mupuf_: paulk-aldrin: you need to speak with mwk
13:59mupuf_: he has a new tool that would help you too
14:00paulk-aldrin: mupuf_, yay! thanks
14:03Jeansf: paulk-aldrin: anyways it appears that mediatek is soc manufacturer, using arm processors, so this firmware can be purely disassembled
14:06Jeansf: paulk-aldrin: because it is executed on the host processor there is no dma upload of scpecial kind probably, it is executed by the processors stil using it's mmu
14:06Jeansf: i.e you can't execute the command so, because the kernel knows what is the real physical aaddress where they are stored, those commands
14:12Jeansf: so obviously if that is a userspace execution it goes that treats this memory blob, it's cortexes memory management unit in hw, which knows itself where the physical addresses are
14:13Jeansf: err where the data is in physical addresses
14:20Jeansf: in graphics land there is no concept of accessing device without communicating with kernel though, the concept of userspace execution of commands is bit misleading, it either has to be done with mmap or ioctl, which are both system calls, but luckily open source cudes fairly well knew that ioctl is faster
14:30Jeansf: just now got another set of news, that estonian court extended 5years period for me to be under the custody
14:30Jeansf: so i do not really have any time to deal with those programming attempts at the moment
14:31Jeansf: i have to go with another set of trials in europes court, and hopefully be going to overseas to find a job
14:32Jeansf: so i come back when this stuff is sort of resolved here, i did apply to markham ontaria canada, and boxborough united states, to work for AMD, i prolly do not get those jobs
14:35Jeansf: well the planned work at AMD is very easy kind.. i only applied for the job, cause noone was willing to do that, and i just do not have time to do it as volunteer, but same goes with nvidia remaining work
14:35Jeansf: it's quite sort of very doable, where the heavylifting has been allready done
14:39Jeansf: and today i decided, i won't participate on the estonian side in the war, when it will be initiated, i laugh when the nuke is sent here
14:53Jeansf: intel work is also very elementary and easy for scheduling they have their own ways
14:59Jeansf: i myself have delt with a company in estonia i wanted to start a business, but i have to do that ovearseas instead, i am mostly interested in low-power DSP processors, that run graphics apis and cpu code on the same die
15:03Jeansf: it's fairly doable i can get it done all the circuit with a year or so, but the biggest issue is with display controllers, i originally planned to work with altera and xilinx fpga's then found that they are pretty pricey for being a commoditiy hw to write software to
15:03Jeansf: but they had like display controller circuits allready available, but when doing all from scratch this needs to be done in addition
15:07Jeansf: as we noted with nyef: it is no gimme those displayport and such specifications
15:12Jeansf: for electronic design i have all open source software for circuit design, things like berkeley sw and austrian programmers additions to it they are governed by software licenses and free to be used by all verndors commercially
15:14Jeansf: i belive also new technologies in the lithography world, would allow me to show my strenghts of understandings without bigger risks taken financially
15:17Jeansf: this humans acheivement capablity is mostly intellectual level those properties as of nowdays, it's important to be able have a capability to think and resist to pressure
15:19Jeansf: mostly when the techonlogical side of the machinery and theory nowdays is nearing to mature work, it's only important for people to be able to understand how to work with those machines how to control them
15:26Jeansf: yeah estonian country is all corrupted , i barely had a conflict with some "important" politicians, they had one against me though, i have been charged six times in a row, vastly retatarded self composed protocols, desrcribing non of what i had been saying
15:28Jeansf: so i just need to fill out digitally signed information to european court, hope all those get really what their sentence cause of not being able to think
15:31Jeansf: allthough i know most of the estonian beaties , like brian adams was saying it was a nation that had very beatiful women, i know that aussies had it times more, and in higher density
15:31Jeansf: so basically screw them
15:38Jeansf: canadian and united states science is ranked nr.1 in my book, so obviously i will be turned down there, some fellow kidding here but..probably some job i am able to get overseas
15:39Jeansf: it's just lovely that they have such a statement on their web that they treat people with mental diagnosis equally, defintely a first pick of mine to work for such company
15:47Jeansf: so yeah eddisson or tesla, i have alsa read about those men, both were kinda famous , one was cruel and rich, but tesla had ideas that were later used and nowdays direct current is very common
15:48Jeansf: tesla was lived very poor life, but you know they start to use charging stations in the europe for electrical cars , eu program
15:48Jeansf: it's all direct current
15:52Jeansf: it's new technolgoeis nowdays, as i understand even armenians did the calculations very intelligently with symphony 7A prototype
15:53Jeansf: bob from united states, also saw that for hydrogen electrolysis, the answer is bit of both and a lot more modulations then only dc and ac
15:54Jeansf: bob was a speedboat racer, discovering his elementary ..what we call perpetum mobile
15:55NanoSector: the fuck
15:55Jeansf: allegedly he was positioned with verichip
16:05Jeansf: all the friggin estonian retards know this not seeing any of their own thoughts to ever function , but it's a country and i have been tracked with one of such too, through the corrupted illegal estonian police acitivity
16:10Jeansf: during the foodstamps votes, people in us, voted it off, for obvious reasons, cause it is prone to very big violations inside communities
16:12Jeansf: and people who have never been in mental institutution we have angry persons there that officer JÜRI PIHL can inspect everything which happens in our assholes
16:12Jeansf: and quite frankly it is true
16:16Jeansf: they are mofo retarded persons never knowing any of the technology their own, and they use it to violate their own highest stars, by lying to interpol network
16:18Jeansf: i hate to say this, but your biggest fastrack and skype or kazaa hero got stuff done by finnish cause their brain was so little, that they could not even program
16:18Jeansf: wanting to kill me cause i can
16:20Jeansf: fasttrack, it's worlwide network, it's nodes have been shut down long ago, but it was the network that had highest amount of users
16:23NanoSector: how do you diagnose sleep/resume issues with nouveau? I'm suspecting it's nouveau causing the hang on resume but I'm not sure
16:25Jeansf: NanoSector: well to be honest i've been under huge pressure, but we talked about this try hans's patch , and if that ain't gonna work, it's gonna end up to be quite complex
16:25Jeansf: perhaps bskeggs should at least come and talk about it
16:26Jeansf: but i really hope that red hat guys got it right
16:26NanoSector:adds Jeansf to ignore list
16:27Jeansf: nyef: how were you'r results with suspend resume with goede patch?
16:28mlankhorst: hm no imirkin here
16:28Jeansf: i just can not test, cause i have no edp MST hw
16:28Jeansf: probably killed off
16:29NanoSector: nope :(
16:32Jeansf: to be honest, i have had a lots of conspiracy against me, and most of them are cripples today
16:32Jeansf: never felt sorry about it
16:33nyef: Jeansf: I broke part of my test setup last night, so I'm basically dead in the water until either I get a replacement (arriving sometime next week, apparently) or figure out some temporary measure.
16:33Jeansf: oooh damn, i have lots of usb sticks, i could send you one for free...
16:33Jeansf: sending is going to cost more only..:)
16:34NanoSector: are they a genuine person?
16:36Jeansf: yeah, cause in my country lots were growing with issues, it's not delveoped country and they hated me with passion arranging constantly wrong things, and they were taken care off
16:37Jeansf: that is why the decisions against me, cause random people have beat the shit out of this crap
16:39Jeansf: patritotic decisions against me, because other folks saw visions that they are someone and humiliated me because of their vast illusions, later getting beaten up by those false flashes
16:43Jeansf: when you are bluffing that you are someone no matter how i try this thing is going to end up recognized and revealed
16:49Jeansf: they thought they'd bluff and someone gonna give them some goods for that, generally more serious persons said, you have humiliated and annoyd a person that is with 100times stronger class then you are, hence they instead got beaten up
16:52Jeansf: imagine when you bluff that you are so friggin marvelous guy and one dude sits in mental insitution and it ends up that everyone recognized that domestic people were full of shit
16:53Jeansf: themselves, yeah they gonna get beaten up by russians they gonna get beaten up by estonians illegaly, and i am being responsible for this
16:57Jeansf: i gonna just take a hike from estonia, and after five years when they are recongized to be empty spots again compared to me, they gonna come again charging me because of that
17:19Jeansf: if the results would be shown i am not saying it was the most complicated draw back days to get into national team, i am willing to listen about such acheivements regardless, allthough knowingly it really was tough during my times and i did it
17:22Jeansf: i had 10event on horizon where i allready was rewarded at the age of 14, you know but i had only suffered ever since, i am not hampered by this and i am willing to having a 10year black hole in my life, try to patehtically advance from 20year old things ahead
17:22Jeansf: still do not have time unless you show me the results
17:37Jeansf: but i see yeah that this suspend and resume has had some complicationd during the history, i am willing to try to solve it, but i do not do pointless slave activity, first try hans de goede patch
17:38Jeansf: if he did his work properly then this is going to work
17:42Jeansf: and my the looks of it nyef was pretty much more on course then me in this subject, bit hampered by troubles
17:46Jeansf: during the gentoo days everyone spammed, so it's ok, conversations produce outcomes pretty oftenly
17:49nyef: Discovered: USB<->spinning-rust adaptors for PATA and SATA, and the battery for my test system (I thought that was in another state entirely). Not discovered: spare spinning rust that is suitable for scratch storage.
17:54Jeansf: mensa club, when i reconed one finnish guy being intelligent this weekend, he said he was in mensa club, it's a club full of people with hi IQ levels, i really spotted this right away
17:55Jeansf: i think against all odds, sharon stone had something like 160 from 180 or so
17:55nyef: I don't know if I want to go with the "personsa" joke or the "monthly meetings" joke.
17:57Jeansf: hehee, nyef: so you are running pata or sata adapther for usb
17:58Jeansf: i think sata was around 3gb/s which is around 6times usb 3.0 speed
17:58nyef: I had been running an SD-card adaptor.
17:58nyef: The SD-card is fine, but I managed to tear the USB connector from its moorings.
17:58Jeansf: whatever storage element you have, ouh, so you did not not have internal hdd for the cpu?
17:59Jeansf: or you are missing an internet connection on this laptop, what is the real trouble there?
18:00nyef: I have two of these systems. I'm using one, and the other one is the one that I used to use before it developed problems that I didn't want to deal with on a regular basis. So I swapped all of the bulk storage over to the new machine.
18:01nyef: So I do my kernel work on the working machine, deploy to storage over USB, and boot the test machine from USB.
18:01Jeansf: hmm, ok, but how to you communicate with the other machine?
18:02Jeansf: aah ok...
18:02Jeansf: what about sharing a newtork tftp boot?
18:02Jeansf: or iscsi nas , or nfs or whatever
18:03nyef: PXE boot? I might try that.
18:06Jeansf: nyef: does it need a wire connection , i.e a cross-nat cable stuff , or nowdays can it be done wirelessly too?
18:08Jeansf: efi and such possibilities, they have this edk2 might be recognizing wireless upfront too, right
18:09nyef: I have a crossover cable, but I'm thinking that it might be easier for me to "just" hook up to my existing NFS/tftp/bootp arrangement, which is on a normal-ish 10/100 ethernet setup.
18:12Jeansf: ok, what about wifi though, can this edk2 based bios recognize wifi as well, as this friggin clock domain works
18:12Jeansf: there is new standart coded, that bios fetches the clock over network , or os or i can not remember
18:13nyef: (One reason that it would be easier is that neither network manager nor some DHCP server that I don't have good control over would get in the way.)
18:15Jeansf: what you mean? hmm, not sure either, prolly it starts from ramdisk, and suddenly parses the dhcp
18:17mlankhorst: inglor: was about to say :)
18:17mlankhorst: erm imirkin_ *
18:17imirkin_: do you not have ops?
18:18imirkin_: hrm, apparently not
18:19imirkin_: and i'm not cool enough to grant access bits
18:19mlankhorst: I don't touch nouveau much any more, so no big deal
18:21imirkin_: nyef: i saw in the logs you had various issues, but with all the spam from joss it's a bit hard to tell... did you get your issue resolved, and if not, what was it?
18:30whompy_: NanoSector: Are you on 4.10rc1?
18:31whompy_: My machine has also stopped resuming. I haven't dug into it yet to see why it's silently dying.
18:32NanoSector: whompy_: yes, but it stopped with 4.8 too
18:32NanoSector: with any kernel really
18:34whompy_: Interesting. I had a bit of nouveau spam on boot. I'll need to drop it in here when I turn it back on.
18:39nyef: imirkin_: No, I did not get things resolved. The issue is that the display panel on my laptop doesn't resume properly from suspend, possibly also not from DPMS blanking (not yet tested). An LVDS panel on this hardware works fine, but this is an eDP panel of some sort and it does not work.
18:40imirkin_: nyef: out of curiousity, how are you swapping eDP and LVDS? those tend to be pretty hard-wired?
18:40imirkin_: or are they just two separate boards with slightly different configurations?
18:41nyef: The mainboard has an LVDS and an eDP connector right next to each other, alongside a connector for some group of accessories (camera + logo backlight) and a connector for the 3D Vision IR emitter.
18:42imirkin_: huh ok. and you also mentioned that it was a MXM gpu?
18:42nyef: It's a fairly basic, if tedious, operation to remove the entire lid assembly and put a different one in.
18:42nyef: Yeah, it's an MXM.
18:42imirkin_: the thing about eDP is that it's extremely finicky =/
18:43nyef: Yeah, I got that.
18:43imirkin_: i think skeggsb asked you for some logs, were you able to get them?
18:43nyef: Some of them. Do you need URLs for the ones that I have available?
18:43nyef: Basically, post-suspend, whenever the AUX channel is accessed, it can't find the sink.
18:44nyef: Which means no link training.
18:47nyef: With the drm-next kernel (airlied's tree, I believe), I no longer get the backlight-destroying blank-white-to-black flickering that I had been getting, and there's about a second of garbage on the panel at resume, so I'm thinking that it's getting a partial wake-up and then going back to sleep.
18:49imirkin_: when you log in remotely, does everything look ok?
18:49imirkin_: i mean, does /sys/class/drm/card*-*/status show "connected", and modes shows the list of modes?
18:50nyef: I don't know. I'll have to get the system working again.
18:51imirkin_: iirc you had a message link "sink not found"?
18:53nyef: At 44.713041 is the first hit to the eDP port post-suspend.
18:54nyef: At 44.716487 is a message about not being able to read the link configuration.
18:54nyef: And by 44.718431 the jig is basically completely up.
18:54imirkin_: so it basically looks like the AUX channel isn't up by the time we resume
18:56nyef: The DP spec is quoted as saying that the link has 1 ms to come up to ready from the first "differential signal" on the AUX channel, and that it doesn't have to respond to that transaction, but I'm not sure that said signal is actually happening... And the DP spec ALSO says that for "embedded" configurations the time is actually 20 ms.
19:05nyef: Basically, it looks as though something tries to wake the panel, but doesn't do the things that it needs to do afterwards, so the panel goes back to sleep. And IIRC, it's not the EC that does it, because I don't remember the panel trying to wake if nouveau isn't loaded.
19:13imirkin_: basically if a spec says something, you can be pretty sure that eDP breaks it
19:14imirkin_: it's probably a race between the panel claiming it's there and it not actually being there
19:14imirkin_: you could try sticking sleep's somewhere and see if it helps? dunno
19:15imirkin_: ideally skeggsb will have some better ideas when he returns, but i'm not sure what his vacation schedule is
19:18nyef: I've already tried to add a section to try to wake the panel with retries at one point. I need to double-check to see if it's the right point, and to bump the number of retries a bit to see if that helps.
19:19nyef: Otherwise, possibly a few other things. And I still need to check to see if DPMS causes similar results.
19:20imirkin_: if you can trigger an hpd signal somehow, that could kick it back into action
19:20imirkin_: (hpd = hotplug detect)
19:21nyef: Mmm. Might try to find some way to fake up an HPD IRQ or similar from /sys or /proc.
19:21imirkin_: iirc just running "xrandr" might do it.
19:21nyef: Mmm. Except that I don't have X installed in my test environment. (-:
19:22imirkin_: well, there's some scan thing it triggers i think.
19:22imirkin_: [obviously xrandr doesn't do anything directly, but it causes ioctl's to get sent down from the X server]
19:24nyef: Right. I can hunt for those IOCTLs if necessary, and write a short test program.
19:24imirkin_: might be able to even do it with 'modetest', part of libdrm
19:29nyef: Ugh. Typical. This machine has an SD-card slot, but it's not working for some reason. /-:
19:29imirkin_: i've yet to have a SD card work for me, in any device
19:30nyef: They tend to work fine for me if I use a USB adaptor for them.
19:30imirkin_: (it's not like i've tried far and wide...)
19:30nyef: But my USB SD card adaptor is the thing that I broke last night.
19:43nyef: No wonder I couldn't fix this thing last night. At least three of the pads for mounting the USB connector have lifted from the board.
21:55haimegen: i do not have fucking time for you guys you do not tend to know first thing about linux, as for an example that there is no bios for android phones.... there is only an eeprom fs image of os
21:56imirkin_: wrong channel?
21:58haimegen: imirkin_: you hould be well aware of the fact, that fpga's also have no bioses, but somehow some do not understand the fact, like leio coreboot dude from estonia
21:59haimegen: being so fucking dumb gets me nervous all the time
21:59imirkin_: apparently not.
22:11imirkin_: mwk: modeling the whole hidden pipeline seems crazy to test... i wonder if it'd make more sense to just have a bunch of "integration" tests, i.e. run the pipeline and look at all the various results
22:12imirkin_: instead of trying to model (and test!) every stage of it
22:41mooch: i'd prefer that second one tbh. it helps me in case i ever have to emulate nv10
23:00mwk: imirkin_: I'd guess modeling the stages one at a time is much better than dealing with all stages at once
23:00imirkin_: mwk: unless you spend all the time figuring out how to test just that one stage. obviously you're in a better position to figure it out, but just an opinion from the sidelines :)
23:01mwk: also, modeling the state correctly is needed for context switching, which is kind of important
23:02mwk: quite frankly, the current nouveau nv10 ctxsw code looks way too simple to be correct...
23:04imirkin_: that seems eminently likely
23:04imirkin_: the question is what the implications are of getting it slightly wrong
23:07karolherbst: imirkin_: maybe the same situation with the kepler cards, it usually works, but sometimes it doesn't and nobody knows why
23:07imirkin_: karolherbst: those context switches are a little different
23:07mwk: that sounds like a common result
23:07mwk: I'd say they're not different enough...
23:08imirkin_: mwk: actually i think the nv50 state can get messed up - i think that's the cause of those occasional CACHE_ERROR things that happen
23:08imirkin_: seems like the RAMFC just disappears sometimes
23:08imirkin_: or whatever that thing is called
23:09imirkin_: it's been a persistent issue since the dawn of time
23:09imirkin_: across all tesla chips
23:09mwk: could be good old memory corruption
23:09mwk: or... a hw context swithing bug in PFIFO
23:09imirkin_: search for "406040" and "400040" in bug reports
23:10mwk: in which case you probably just need to flip some random DEBUG bit somewhere and it'll magically get fixed
23:11karolherbst: and by the way: next time we so have to make a nouveau assembly at the c3....
23:39janeismo: anyways that is the beauty of open source, that is why third party vendors is not involed like abit or say what, you burn some sort of initramfs on andoid phone mobo's eeprom, and done with it. billions sold
23:50skeggsb: nyef: i don't believe adding retries or any such thing will help you, the aux channel check is for whether the hpd signal is asserted at all (ie. whether there's anything on the other end)
23:50skeggsb: that should be asserted regardless of what state the sink is in
23:50skeggsb: i have no idea *why* it's not for you though
23:59imirkin_: skeggsb: perhaps we're not waiting for something to happen?
23:59imirkin_: like the panel coming up properly