00:04 pmoreau: Freedesktop has a phabricator instance?! :-)
00:04 pmoreau: Since when?
00:04 pq: a couple months?..
00:05 pq: it's still in test use for Wayland and Weston at least.
00:05 pmoreau: Damn... I missed that!
00:05 pq: as in not really officially announced, except for some random links circulating
00:05 pmoreau: Ok
00:05 pq: the official workflow of Wayland and Weston does not include Phab yet
00:06 pmoreau: Is there a plan to move all projects there in some distant future, keep both platforms or have new projects on Phabricator?
00:06 pq: pmoreau, up to each project on their own, I believe
00:06 mupuf: pq: thanks for the link
00:07 mupuf: pmoreau: good to know it is the same in sweden (answer from 10h ago!)
00:07 pmoreau: I've been using Phabricator for my own projects after discovering it when Blender decided to move to it.
00:07 pmoreau: mupuf: ;-)
00:07 pq: handling patch series in Phab is still fairly painful
00:07 pmoreau: mupuf: And finally Swedish courses are going to start! \o/
00:08 mupuf: pmoreau: should be easier than my finnish ones! So I heard
00:09 pmoreau: I should try to use arcanist, but iirc its workflow is mostly one commit for one idea, so no patch series...
00:09 pq: pmoreau, that's exactly the core of the pains
00:09 pmoreau: mupuf: So I heard too! Especially as I learned a lot of German before.
01:17 pmoreau: l1k: I tested your branch this morning, and same results as before: both GPUs get a correctly sized fb, and after some switches I get a "GPU lockup - switching to software fbcon".
01:17 pmoreau: l1k: But I did get it on the second switch to DIS, rather than fourth one with the previous version. https://phabricator.pmoreau.org/P44
01:18 pmoreau: I need to have a look at why, and try to fix it.
02:40 qq[IrcCity]: «nouveau_drv.so» (loaded by Xorg) belongs to xserver-xorg-video-nouveau. but where does in come from?
02:43 qq[IrcCity]: * does it
02:49 qq[IrcCity]: is «nouveau_drv.so» a component of an X server? in short, is it your stuff or other guys are in charge?
02:51 pq: qq[IrcCity], http://nouveau.freedesktop.org - xf86-video-nouveau
02:54 qq[IrcCity]: seemingly not in active development.
02:55 pq: it's matured rather than abandoned.
03:55 karolherbst: ohhhh well
03:56 karolherbst: there is a laptop with intel hd 5500 and nvidia 740M (ddr3, pcie 5.0 max) :/
03:56 karolherbst: how does this even make sense
03:56 karolherbst: why bother
03:57 glennk: apparently there are ones with only intel hd 5500 and single channel memory too
03:58 glennk: maybe its one of those?
03:58 karolherbst: I don't know, I only know that unigine heaven has like 50% mehr perf with the dedicated one
03:59 karolherbst: I would rather save the money for the nvidia gpu and get an intel cpu with even more gpu perf
04:00 karolherbst: cpus with a hd 6000 shouldn't be that more expensive
04:00 glennk: right, but that would imply the laptop hardware designers know anything about the hardware they slap into the devices
04:01 karolherbst: :D
04:01 karolherbst: it is a lenovo one, so they _should_ know
04:01 karolherbst: but I also hear about overheating problems with those
04:01 karolherbst: so I really don't trust them
04:02 karolherbst: and then only 8 lanes for the dedicated gpu?
04:02 karolherbst: uhh
04:02 karolherbst: that is bad for prime
04:02 glennk: i think the lenovo ones tend to be the worst in that aspect
04:06 karolherbst: glennk: wow, the i5-5350U cost like 25$ more and has a hd 6000
04:06 karolherbst: also + 0.2GHz more max speed at same TDP
04:07 karolherbst: ohh higher TDP down though
04:07 karolherbst: and it has 2 memory channels (both)
04:16 karolherbst: ohh wow, there is currently nearly no userspace support for nvenc :D
04:36 kdub: i see this sort of error in dmesg: [88551.624387] nouveau E[ DRM] DDC responded, but no EDID for VGA-1
04:36 kdub: and the monitor's resolution isnt detected (xrandr mode setting can force the right resolution)... any ideas?
04:42 karolherbst: nice
04:46 karolherbst: mupuf: okay, it is nvenc
04:46 karolherbst: ohh wait
04:46 karolherbst: maybe I should disable other video bits first
04:46 karolherbst: okay, done
04:46 karolherbst: nice
05:12 glennk: kdub, http://nouveau.freedesktop.org/wiki/TroubleShooting/#index4h3 ?
05:14 kdub: glennk, thanks, seems i need to find that edid info then
05:16 mupuf: karolherbst: great!
05:17 karolherbst: mupuf: another thing: https://github.com/envytools/envytools/pull/24/files on gt215 the GR bits are also used, but someone told me, these engine doesn't exist there
05:17 glennk: kdub, i'd probably check the connection to the monitor, fairly common that one of the pins isn't quite making a solid connection
05:18 kdub: glennk, good idea, this has worked before... just stopped working after my latest upgrade
05:42 karolherbst: mupuf: these are the combination used for the specific slots for differentchipsets: https://gist.github.com/karolherbst/1dcb58f2a45b34eed529 maxwell is a bit strange, because across cards some slots are used differently
05:42 karolherbst: last slot seems to be used for total count
05:43 karolherbst: the big pain is, that if an engine doesn't exist, the counter outputs really high numbers, usually higher than total :/
05:44 karolherbst: which means if a wrong bit is set, nouveau would always upclock to max
05:58 mupuf: karolherbst: it just never gets reset
05:58 mupuf: hence why the numbers get so big
05:58 karolherbst: no it get's reset
05:58 karolherbst: because it is always close to the total counter
05:58 mupuf: and I doubt the blob configures the maxwells differently, it is just that you looked at mmiotraces of different blob versions
05:59 karolherbst: could be
05:59 mupuf: sorry, have to go to my finnish class ... it is going to hurt! See you!
05:59 karolherbst: cu
06:07 l1k: pmoreau: sorry I was away for a few hours. is this a regression, i.e. did this issue not occur with earlier versions of my patch set?
07:11 ratherDIfficult: ouh lord what the fuck you've been up tp, chipsies what the hell, all the german blonds are on my side, i.e all the germans and all the russians and stuff
07:11 ratherDIfficult: tell me folks who do you fight with, with an estonian enourmous sports star, and what is our blood dudes or chipsies so to speak?
07:12 ratherDIfficult: its the most beatiful kind of it
07:13 ratherDIfficult: what the fuck chipsies , well hello! take a fucking hike once, when our people have killed many of you
07:16 ratherDIfficult: i was told that biggest portuguese boss is here, i told him and advised son, get the fuck off and run before you die
07:17 ratherDIfficult: its not worth it, i had family conflicts , i had abortion leftovers in my family too
07:18 ratherDIfficult: i've suffered all my life that freaks were cauged out from between womens legs
07:21 ratherDIfficult: yeah you don't know about violations that idiots acquired my family and it was not a fair deal, when you die you will be said that i mean before you die
07:21 ratherDIfficult: world war III launched
07:33 karolherbst: imirkin: as it seems the blod really doesn't to the reclocking calculations on the cpu, but rather somwhere on the gpu. What would be the best way to see where the driver reads information to which clock it has to clock to? I mean the result of the scripts which are running on the gpu
07:34 imirkin: i can't parse that sentence
07:35 karolherbst: I meant in the traces. There needs to be reads the driver does to read out "results" of scripts running on the pmu or falcons or whatever
07:37 ratherDIfficult: i know all possible estonian sport stars, since i was one my own, and when i say christiano ronaldo with his estonian freaks, is mickey then that is the way it is
07:42 ratherDIfficult: the difference being i knew real sport persons, fucktards new fucktards
07:45 imirkin: sorry, not 100% sure how that works
07:45 ratherDIfficult: so i don't know about you, but ill sue them, and with this little money even out sportsmen are better, i'll invest 90% of the pentaly to estonian sports
07:45 imirkin: if i had to guess, i'd say either intr or a shared buffer. probably intr.
07:47 ratherDIfficult: imirkin: i got a computer ...
07:48 ratherDIfficult: imirkin: is a russian , a quite intelligent one as most of them are
07:49 ratherDIfficult: i do not care about portuguese palaces, i care about russian sports, and they have always supported estonians to become stars
07:52 ratherDIfficult: it's kinda like hitler did not like retards, so do not russians like them
07:52 ratherDIfficult: but when you are someone russians tend to understand right away
07:53 ratherDIfficult: so about this syria conflict, i don't have anything against if usa and russia and germany will work together
07:53 ratherDIfficult: its the most logical path
08:01 ratherDIfficult: you gotta understand that what those fucktards speak has hardly any meaning
08:02 ratherDIfficult: kinda a like a biggest losers who compensated the stuff exponating their fucking deals
08:04 ratherDIfficult: so you kinda blacked out joss193 for punch of retards really, who from the moment of birth hardly had any clue what they do anyways, pretending to be adults cause of their fucks
08:09 pmoreau: l1k: This time I was the one being away. :D It is a regression compared to earlier versions in the sense that it *crashes* as soon as the second switch, compared to 8 switches. And it is a regression in the sense that it doesn't happen without your patches.
08:12 ratherDIfficult: imirkin: when you do not work with me, then good luck listening to those retards, and i am dissapointed to be honest, but as always i can get over it
08:16 ratherDIfficult: ill tomorrow start working regardless though, today i am too drunk, mother got me an usb stick, so i can boot my new machine
08:16 ratherDIfficult: an/a
08:24 ratherDIfficult: someone named stirling formed SAS, maybe david or someone, we can proudly accept english to the allies, in nato sense all but russia is missing
08:24 ratherDIfficult: SAS and united states army protected me when i had worse times
08:29 ratherDIfficult: definitely ban portuguease, italians , france, and accept russia
08:30 ratherDIfficult: also spaneints , i mean all the chipsies should be banned
08:37 ratherDIfficult: you end up like either you run away, or you get fucking burned, i have like every fucking sane person with me, and ill hope youll join
08:37 ratherDIfficult: c ya
10:04 karlmag: Meh... order cancelled... Out of stock (forever) also at supplier.
10:05 karolherbst: karlmag: you are really unlucky
10:07 karlmag: karolherbst: well... not always, but sometimes.
10:08 karlmag: *looking at new workstation-to-be* Not always.. :-)
10:10 karlmag: but yeah, that was a bit of a bummer
10:11 jefferym: I am running kernel 4.3.0-rc3 with an nv4e device. I noticed a WARNING in the kernel log inside of drm_vblank_count_and_time which looks like it is coming from nouveau_finish_page_flip
10:14 karlmag: but.. I'm still a nice, new graphics card out
10:14 jefferym: It looks to me that nouveau_finish_page_flip always passes a -1 as the pipe argument to the drm_send_vblank_event function.
10:15 imirkin_: karlmag: what's an online retailer that delivers to your area?
10:16 imirkin_: jefferym: is -1 invalid?
10:16 karlmag: imirkin_: that was one of them.. other major one doesn't have any
10:17 imirkin_: karlmag: can you provide a link?
10:17 karlmag: two big ones are multicom.no and komplett.no
10:17 jefferym: this code seems to be setting it to -1 by default inside of nouveau_finish_page_flip and chaning it in the device.info.family >= ..TESLA
10:17 karlmag: in norwegian (mostly) obviously
10:18 imirkin_: jefferym: yeah, i'm guessing we don't know which head it's coming from pre-tesla
10:18 jefferym: this card, I believe to be less than TESLA, thefore -1 is always sent.
10:18 imirkin_: jefferym: yeah, nv40 is pre-tesla
10:18 imirkin_: curie, i think, is the family name
10:18 karlmag: imirkin_: and look for "skjermkort"
10:19 imirkin_: karlmag: chrome has built-in translation support
10:19 karlmag: ah, ok
10:19 jefferym: Does the page flipping not occur frequently, I would assume to see this warning more frequently on this sytem
10:20 imirkin_: impressive. only has a handful of low-end (probably fermi) cards, the rest are all maxwell
10:20 imirkin_: jefferym: 60x per second, if you have a 60hz display
10:20 imirkin_: jefferym: perhaps it's a WARN_ONCE?
10:22 imirkin_: karlmag: this is most likely a kepler GK208... kinda low end though. https://www.komplett.no/product/817919/datautstyr/skjermkort/pci-express/msi-geforce-gt-740-2gb-physx-cuda
10:24 jefferym: it's WARN_ON. however, it is only called if dev->num_crtcs > 0, perhaps that's what is differnet in my case
10:25 imirkin_: no, you have crtc's :)
10:25 imirkin_: crtc = scanout engine
10:25 imirkin_: [at least in modern parlance]
10:26 jefferym: ok, I wasn't sure if it could go to 0
10:26 RandomRead: so the portuguease when they come will be buried very quick, any sort of other questions you chipsies may have?
10:26 RandomRead: taken to the woods and buried like old times
10:27 RandomRead: what is the question you are asking me?
10:27 imirkin_: jefferym: yeah, it can go to 0, e.g. if you have a gpu without any display capabilities
10:28 RandomRead: eaxactly what is that slut trying to acheive?
10:29 RandomRead: this one is a braindead person, she does not know notrmal estonian people
10:31 RandomRead: what is the question again, when slut invites ronado to attack me, yeah he will be killed
10:31 jefferym: ok, thanks for the info, I am not seeing it now and am not sure what I may have done to produce it previously
10:32 imirkin_: jefferym: probably some funny race, maybe with dpms
10:32 jefferym: I'll keep an eye out in case I see it again. Should I assume that since it has display capabilities right now it shouldn't change as the system runs?
10:32 jefferym: ok
10:32 imirkin_: yeah, i think num_crtcs is fixed at startup
10:33 imirkin_: it'll be 2 for you
10:33 jefferym: well I have been seeing other issues, namely an occasional flicker and an issue where it hangs
10:34 imirkin_: =/
10:34 jefferym: I cannot seem to reprodce the flicker consistently
10:34 jefferym: I have not seen it with the nvidia driver but I use the nvidia driver far less than nouveau on the system
10:35 jefferym: it happened, I think, more frequently with my 4.0.5-4.0.7 kernels I tried, and less frequently on the 4.3.0 or 4.2.0 or 4.1.7 kernels I have tried
10:35 RandomRead: what you mean you attack estonian beaty and you survive?
10:35 imirkin_: at one point or another i disabled MSI on all the nv4x IGP's, that could have increased stability
10:35 RandomRead: what the hell is that what makes you think so?
10:35 imirkin_: also a bunch of mesa fixes made it out at various points, potentially improving stability
10:36 imirkin_: hmmm... commit 4761703bd0 went into 3.19
10:37 jefferym: ok, I am testing recently with the mesa/nouveau that I built yesterday, previously I had used the rpms with fedora 22
10:37 karlmag: imirkin_: ah, yeah... I want 4 outputs too
10:37 imirkin_: karlmag: oh right. forgot.
10:37 imirkin_: jefferym: cool. i just pushed a small fix that should help things
10:38 imirkin_: jefferym: make sure you have http://cgit.freedesktop.org/mesa/mesa/commit/?id=47d11990b2ca3eb666b8ac81fee7f7eb5019eba1
10:38 imirkin_: i'm not 100% sure why i was seeing it on nv30 and not on the other models
10:38 imirkin_: nor am i sure why i wasn't seeing it on nv30 before but all of a sudden started seeing it now
10:39 jefferym: ok, ill go confirm that now
10:41 RandomRead: its my duty to advise you that you just get the fuck off from my areas, really that slust is amazingly big slut
10:41 imirkin_: could be that nv4x is somehow less sensitive to it than nv3x (this was happening on a nv34) but... won't hurt :)
10:41 RandomRead: get the fuck off from here
10:41 RandomRead: or youll be sorry
10:41 jefferym: I didn't have that fix, i was a bit behind it
10:41 jefferym: I'll rebuild and retest.
10:42 pmoreau: Argh... I shouldn't have updated drm-next... Nouveau doesn't want to build anymore
10:42 imirkin_: jefferym: in my case, this was causing assert failures (and segfaults in opt builds)
10:42 imirkin_: jefferym: not flickering
10:42 jefferym: the hangs I see give me output such as EBUSY for failed pushbufs
10:43 imirkin_: hmmmm
10:43 imirkin_: that indicates gpu hang
10:43 jefferym: err, -16 which I am hopefully getting right as EBUSY, but ill try your fix first and get back with more
10:43 imirkin_: or gpu commands being submitted a lot faster than the gpu can process them
10:44 jefferym: is it up to the program (chrome in this case as far as I can tell) to keep from sending too many commands?
10:44 imirkin_: no
10:44 imirkin_: but nouveau has no real throttling mechanism
10:45 imirkin_: the hope is that this would happen naturally since presumably you make decisions based on the results you compute
10:47 qq[IrcCity]: in what can a DoS attack on the nouveau driver result?
10:48 imirkin_: system hang
10:49 jefferym: if it is useful, the messages usually start with "error fencing pushbuf: -16"
10:49 imirkin_: huh
10:49 imirkin_: error *fencing* pushbuf? that's odd.
10:49 jefferym: then it may say something such as "cal_space: -16"
10:49 imirkin_: yeah, the cal_space thing is expected
10:49 jefferym: ok
10:49 imirkin_: [for that situation]
10:50 qq[IrcCity]: imirkin_, so a system running nouveau is virtualy defenceless against any application having access to the video system? DoS on the video card can be engineered even with a tty.
10:51 imirkin_: qq[IrcCity]: i'm not sure what your point is... an application that can submit commands to the accel engines can just hang the whole box in the first place
10:51 imirkin_: qq[IrcCity]: nouveau has basically no real recovery from gpu hangs
10:51 imirkin_: and it's pretty easy to engineer a hang
10:55 qq[IrcCity]: imirkin_, I talk about (user-mode) applications that can talk to the driver, possibly indirectly.
10:55 imirkin_: qq[IrcCity]: yeah, that's what i'm talking about
10:56 imirkin_: qq[IrcCity]: nouveau exposes the ability to feed commands in directly
11:17 karlmag: tried two more webshops; conclusion; noone really have any of those anymore
11:17 imirkin_: really weird
11:17 jefferym: imrikin: I am unable to test that out right now but I will test out the new mesa build with nouveau in the morning
11:17 imirkin_: tons of them on newegg, not that it's much consolation
11:17 karlmag: but! There was one demo-used of these; http://uk.gigabyte.com/products/product-page.aspx?pid=4511#ov
11:18 jefferym: thanks for the help, I'll be back with more
11:18 imirkin_: karlmag: looks perfect for you
11:18 karlmag: I guess most people in Norway want the latest and greatest, so everyone are geared towards that.
11:19 karlmag: imirkin_: yeah.. and the price?
11:19 karlmag: about $55 including shipping
11:19 imirkin_: seems fair :)
11:19 karlmag: shipping and handling being about 1/3 of that :-P
11:19 imirkin_: usually is.
11:20 karlmag: Hope it's ok and all that.
11:20 karlmag: Well, should be, but you know.
11:20 karlmag: got expensive anyway. I bought some server ram from the same place too :-P
11:22 karlmag: Now.. TLW(tm)...
11:23 karlmag: Hmm.. up to about 12 days.. Fun.. oh well
13:32 hakzsam: mwk, are you sure the compute class for GF110 is 0x91c0 ? because I just traced the blob and it's 0x90c0 as for the other Fermi chipsets
13:41 mwk: hakzsam: it's 91c0, but it supports the older class too
13:42 imirkin: hakzsam: GF110 and GF108 have their own iirc
13:48 hakzsam: mwk, well, creating a compute object with 0x91c0 fails in mesa
13:48 hakzsam: imirkin, GF108 can use 0x90c0 also
13:48 imirkin: oh wait. compute.
13:49 imirkin: i thought graph
13:49 imirkin: er, 3d
13:49 hakzsam: ok :)
13:49 imirkin: hakzsam: check nvd7/nvd9
13:49 imirkin: perhaps mwk was off-by-one
13:50 hakzsam: nvd9 uses 0x90c0
13:50 hakzsam: I have on,
13:50 hakzsam: *one
13:50 imirkin: hmmm
13:50 imirkin: and you checked GF110 as well?
13:50 hakzsam: I did few minutes ago but never tried before
13:52 hakzsam: I'll try by replacing 0x91c0 by 0x90c0
13:53 hakzsam: mwk, what is different between 0x91c0 and 0x90c0 classes?
13:54 imirkin: hakzsam: looks like a typo in nouveau
13:54 imirkin: { -1, -1, FERMI_COMPUTE_A },
13:54 imirkin: should probably be FERMI_COMPUTE_B
13:54 hakzsam: mmh
13:54 hakzsam: yeah, probably
13:55 imirkin: it _probably_ doesn't matter
13:56 hakzsam: seems like there are only minor changes between COMPUTE_A and COMPUTE_B
13:56 hakzsam: I'll try with _B
13:56 imirkin: [in gf110/117/119.c
13:57 hakzsam: gf117 and gf119 use _A as expected
13:57 hakzsam: mmh maybe not
14:02 hakzsam: imirkin, http://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nv_object.xml.h#n200
14:02 hakzsam: ...
14:02 imirkin: hehe
14:02 imirkin: oops :)
14:03 imirkin: pretty sure 91c0 is right
14:04 hakzsam: imirkin, it is, if I update nouveau as well
14:05 hakzsam: imirkin, to fix the compute class in mesa, I need to re-generate that xml file, right?
14:05 imirkin: that way since 6d1cdec3 :)
14:05 imirkin: errrrr
14:05 imirkin: i wouldn't
14:05 imirkin: just modify it directly
14:06 imirkin: you'll run into major headaches regenerating it
14:06 hakzsam: since 2012... :)
14:06 hakzsam: okay
14:07 hakzsam: imirkin, and what about nouveau? because if we change the classes in nouveau we will break mesa except if we check the drm_version or something
14:08 imirkin: hakzsam: except mesa doesn't care
14:08 imirkin: coz it doesn't do compute
14:08 imirkin: and it clearly wasn't working before
14:09 hakzsam: it does if you use MP counters :)
14:09 hakzsam: imirkin, so, we need two patches, one for mesa and one for nouveau, right?
14:11 imirkin: hakzsam: but MP counters are disabled by default
14:11 imirkin: hakzsam: but check what class blob uses...
14:11 imirkin: if blob uses the A class, then just use tha teverywhere and move on
14:11 hakzsam: imirkin, they will enabled by default once I have finished my series
14:12 hakzsam: imirkin, it uses 90c0 everywhere I can see
14:12 imirkin: then you should also always use 90c0
14:12 hakzsam: yeah, good
14:12 imirkin: but... some traces "proving" that would be nice
14:12 imirkin: mmt traces
14:13 hakzsam: I have a trace if you want to double check, but I'm 100% sure it is 90c0
14:15 hakzsam: in my opinion, 91c0 just contains some improvements/extra suff regarding 90c0
14:15 hakzsam: and mwk said that 90c0 is also supported on gf110
14:15 hakzsam: so, make use of 90c0 seems reasonable
14:27 mwk: hakzsam: there's like 1 or 2 new methods, and we have no idea what any of them do.
14:28 mwk: kernel should definitely be changed to allow 91c0 too, as for which one to use in mesa, meh
14:28 hakzsam: mmh
15:32 imirkin: hakzsam: are you going to send the patch?
15:40 imirkin: hakzsam: i've sent one
15:44 skeggsb_: imirkin: merged
15:44 imirkin: skeggsb_: thanks!
15:54 imirkin: skeggsb_: btw, someone on a very different powermac than mine confirmed that the OF fix worked for him too
15:54 imirkin: well... not THAT different, but it's a NV47
15:54 imirkin: vs my NV34
15:55 skeggsb_: cool, not unexpected :)
15:56 imirkin: no, not unexpected. but still nice to know.
15:56 airlied:has my ppc f22 upgrade waiting for a reboot, but I'm not game to reboot it remotely
16:16 imirkin: airlied: might as well just shut it down remotely :)
16:16 imirkin: airlied: it'll save the anticipation of whether it's just slow at coming up or not
20:31 k-man: can i run triple monitors with the nouveau driver?
20:32 imirkin: k-man: if your gpu has 3 crtc's, sure
20:32 imirkin: k-man: pre-kepler, only 2. kepler+ gained 4.
20:33 k-man: err... what is a crtc?
20:34 k-man: what is kepler?
20:34 k-man: and will this work for gnome?
20:34 imirkin: kepler is a gpu generation
20:34 imirkin: a crtc is a piece of hardware that reads memory and feeds it into an encoder
20:34 k-man: its a Quadro NVS 420 -
20:35 imirkin: k-man: lspci -nn -d 10de:
20:36 k-man: http://sprunge.us/SJgO
20:36 imirkin: you have 2 GPUs in there, each can do 2 monitors
20:37 imirkin: it's probably that single-card dual-gpu board
20:37 k-man: yep
20:37 k-man: yeah
20:37 k-man: it is
20:37 k-man: with a special connector that breaks out the 4 DVI connectors
20:37 imirkin: should basically work
20:37 k-man: ok, currently i'm using the nvidia driver for some reason
20:37 k-man: i'll switch
20:38 k-man: might take a wihle
20:38 k-man: bbiab
20:59 k-man: do i need to do anything to get the 3rd monitor to be detected?
20:59 k-man: xrandr only reports 2
21:01 imirkin: http://nouveau.freedesktop.org/wiki/Optimus/
21:01 imirkin: look for the bit about "Using outputs on discrete GPU"
21:02 imirkin: xrandr --setprovideroutputsource 1 0
21:02 imirkin: should make the secondary gpu's outputs appear
21:03 k-man: yeh yeah
21:03 k-man: thats working!
21:03 k-man: awesomeness!
21:04 k-man: thank you so much
21:04 imirkin: enjoy
21:05 k-man: gah i spent a lot of time on the nvidia driver and couldnt get it to work
21:06 k-man: so its great to see the open source driver does a better job
21:07 imirkin: don't worry, you'll get to know nouveau's little foibles as well
21:07 k-man: oh, i remember why i switched to the proprietary driver now. i seem to get these momentary hangs in the nouveau driver where everything stops responding for 2 seconds or so
21:07 k-man: yeah, just remembering one now
21:08 imirkin: still happens?
21:08 k-man: yeah, its happening now
21:09 imirkin: never heard of any such thing. but i have heard of weird stuff with xrandr offloading
21:11 k-man: hmm.. it makes it very unusable in fact
21:11 k-man: very disconcerting
21:12 imirkin: try booting with nouveau.config=NvMSI=0
21:12 imirkin: oh, and try turning off sync-to-vblank
21:12 imirkin: Option "GLXVblank" "off" iirc
21:12 k-man: where do i set that?
21:12 k-man: in xorg.conf
21:13 k-man: not quite sure how to even google for momentary hangs. i seemto only find results with hard lockups
21:19 imirkin: could be that the G98's are so lowly clocked that they can't keep up with the 3 screens
21:43 AndrewR: RSpliet, hello.
21:46 AndrewR: I think I have mmiotrace from my nv43 during reclocking up (from 300/1000 -> 500/1000 - max) and back, not counting initial bootup upclock.
21:46 AndrewR: anyone wish to look (13 mb gz compressed)
21:50 AndrewR: http://s000.tinyupload.com/index.php?file_id=07309909598025436542 _ hope it will be useful, but while I'm on it - I can reboot and trace for something else, like FSAA
21:55 AndrewR: also, what I found interesting - blob can read temperature sensor (seems to rise according to load - from 72c to 76c) and also thinks fan has variable speed, while I don't know way how to slow it down, for me it always stay at 100% ...
21:55 imirkin: AndrewR: and nouveau doesn't?
21:59 AndrewR: imirkin, no, as far as I can see.... it can't find sensor (more like some info in vbios? I looked at it some time ago) and then probably disables also fun-controlling logic. But then, I hope(d) it will be clean how blob raises and lowers voltage ..?
22:00 imirkin: run 'sensors'
22:01 AndrewR: imirkin, "No sensors found!"
22:01 imirkin: =/
22:01 imirkin: probably related to your voltage fail too
22:01 imirkin: can't read the gpio's
22:01 imirkin: is this the case on linux 4.2 as well?
22:02 AndrewR: imirkin, ohm, sorry, I'm still on blob ...guess it doesn't expose itself this way
22:02 imirkin: oh
22:02 imirkin: yeah, blob doesn't
22:03 AndrewR: imirkin, I can reboot into 4.2 with nouveau.debug=debug (or I better add another string as param for debug?)
22:04 imirkin: no special debug params required
22:06 AndrewR: imirkin, I was hoping for slightly more informative dmesg (I also tried 4.2 with this param - it doesn't talk too much about voltage , only one line saying 'reset' ...unlike nouveau module from 4.3 tree)
22:07 AndrewR: imirkin, so may be before it really worked because none touched this part of reclocking
22:07 AndrewR: imirkin, I also hope your BE adventure goes well
22:18 AndrewR: imirkin, http://www.fpaste.org/276217/ - 4.2.0 dmesg
22:18 AndrewR: imirkin, no sensors too
22:26 imirkin: hmmmm
22:26 imirkin: temp sensors not detected
22:26 imirkin: but you do get fan control via pwm
22:27 imirkin: it might be a therm sensor type we don't recognize
23:04 AndrewR: imirkin, I just grep'ed demmio output, it definitely shows a lot of lines with PBUS.DEBUG_1 . Can it be sensor read (I had nvidia-setings running, for watching if GPU actually upclocks)
23:09 AndrewR: also, [0] 600.221334 MMIO32 R 0x0015b8 0x14800000 PBUS.THERM.CFG1 => { UNK0 = 0 | UNK9 = 0 | CONNECT_SENSOR | 0x14000000 } at the very end looks promising ....
23:22 hakzsam: imirkin, thanks for the patch, I'll make the mesa one and add it to my series