10:47 karolherbst: mhh, we should try to allign our ioremaps somewhat
10:48 karolherbst: only one 2MB mapping in total :(
10:56 karolherbst: uhhh
10:56 karolherbst: silly tracing bug, everything looks fine, but then again it isn't
10:57 karolherbst: pq: unbind/bind dance (bind hangs): https://gist.github.com/karolherbst/b0c0b7df0955559cac831be55a18b56b
10:57 karolherbst: this looks totally fine
11:00 karolherbst: hang is at arch/x86/entry/entry_64.S:517
11:00 karolherbst: aka: https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/tree/arch/x86/entry/entry_64.S?id=refs/tags/v4.8.8#n517
11:02 karolherbst: last function was nvc0_fbcon_accel_init
11:10 chillfan: which branch/kernel version do I need to use to test the new reclocking code for gtx 780 ti again? Someone mentioned it the other day but I forgot heh
11:11 chillfan: well i say new reclocking code, I think it's just a patch to correct voltages
11:11 karolherbst: chillfan: depends on your kernel
11:11 chillfan: kernel is 4.4.33
11:11 karolherbst: then you need a new kernel anyway
11:11 karolherbst: use the current drm-next kernel
11:12 karolherbst: https://cgit.freedesktop.org/~airlied/linux/log/?h=drm-next
11:12 karolherbst: make sure to use the drm-next branch though
11:13 chillfan: Ok, I'll have a look and I bookmarked it this time, thanks :)
11:14 chillfan: and should go without saying thanks to all who worked on it, I need this :)
11:17 chillfan: will be back later
11:47 antos1: Hello.
14:17 Lekensteyn: yay, got two 33c3 tickets :)
14:20 karolherbst: Lekensteyn: :O
14:20 karolherbst: :(
14:20 karolherbst: I got none
14:20 karolherbst: position 5080 inside the queu
14:22 Lekensteyn: ouch
14:24 Lekensteyn: karolherbst: friend of mine is at 1000 now, will ask if he can try to get another ticket for you
14:25 karolherbst: sold out
14:30 Lekensteyn: next week friday 10am is next chance
14:31 karolherbst: *last
14:33 Lekensteyn: I'll try to get a number next week for you and a friend
14:34 karolherbst: nice
14:34 Lekensteyn: I refreshed after a few seconds (synchronized to IRC time)
14:34 Lekensteyn: what clock were you watching?
14:54 karolherbst: Lekensteyn: no clue, I think I simply missed it somehow
16:14 pq: karolherbst, sorry, I don't think I can help.
16:20 karolherbst: :(
17:02 karolherbst: pq: any clue about the assembler stuff?
17:24 karolherbst: okay, power budget round two
17:30 karolherbst: mupuf: interesting, one power budget entry needs to be higher than one other one
17:30 mupuf: karolherbst: or the driver crashes? :D
17:30 karolherbst: nope
17:30 karolherbst: it just aborts loading
17:30 karolherbst: so it is a check within the driver
17:31 karolherbst: \o/
17:31 karolherbst: round 2 done
17:31 karolherbst: found the header field to select the other entry to cap power
17:31 karolherbst: there are basically two
17:32 karolherbst: uhhh
17:32 karolherbst: is this actually a fip?
17:33 mupuf: a fip?
17:33 karolherbst: yeah
17:33 karolherbst: seems to be a flip to enable power capping
17:33 karolherbst: header+0x6
17:33 mupuf: oh, cool
17:33 karolherbst: entry 0 and 1 are both 40W
17:33 karolherbst: if header+6 is 0, no power capping
17:34 karolherbst: if it is 1, power capping
17:35 karolherbst: if entry 0 is 80W and entry 1 is 40W -> refuse to load
17:35 karolherbst: if I flip header+0x6 to 0, it loads again
17:36 karolherbst: okay, now how do both fields gets their actual meaning
17:38 karolherbst: header+0x5 selects the capping entry
17:38 karolherbst: at least I thought
17:40 mupuf:is looking again at the cracking the formula for the period
17:40 mupuf: if I do not find everything by the end of the month, I will send an email to nvidia to ask what the fuck they were smoking
17:41 karolherbst: I also have this kind of entry: "3: min = 10000 mW, avg = 1000000 mW, peak = 1000000 mW (unkn12 = 0)"
17:41 mupuf: AHAH
17:42 mupuf: well, I guess it is unused, right?
17:42 karolherbst: 0x000090a5: 20 0d 22 09 64 00 01 04 06 ff ff e8 03
17:42 karolherbst: last byte
17:42 karolherbst: so maybe?
17:42 karolherbst: no clue?
17:42 karolherbst: it's the header
17:45 karolherbst: mhh, the driver caps to 38.8W
17:49 karolherbst: mhh nvidias dynamic clocking algorithm is crap
17:49 karolherbst: ohh wait, looked at the wrong number
17:49 karolherbst: all good then
18:38 karolherbst: :O http://lkml.iu.edu/hypermail/linux/kernel/1611.2/02473.html
18:38 karolherbst: "We also intend to leverage this inside the open source nouveau driver.
18:44 mupuf: karolherbst: yop :)
18:44 mupuf: cool, ins't it? Jerome will write the support for paging in nouveau :p
18:49 karolherbst: :D
18:49 karolherbst: nice
18:49 karolherbst: we need this anyway
18:49 karolherbst: paging that is
20:34 orbea: two more glupen64 issues/regressions that only seem to occur with nouveau... traces, logs and other info in these issue reports. https://github.com/GLupeN64/GLupeN64/issues/107 https://github.com/GLupeN64/GLupeN64/issues/109 Probably should be fixed in glupen64, but linking it here since it seems related to nouveau.
21:04 bloblo: hello
21:04 bloblo: anyone here ?
21:08 bloblo: I am debian user, with mesa 13.0.1-1, kernel 4.9-rc5, libdrm-nouveau2 2.4.73-1, xserver-xorg-video-nouveau 1:1.0.13-1
21:09 bloblo: I have this: cat /sys/kernel/debug/dri/0/pstate
21:09 bloblo: 03: core 300 MHz shader 600 MHz memory 300 MHz
21:09 bloblo: 0f: core 675 MHz shader 1458 MHz memory 900 MHz
21:09 bloblo: AC: core 399 MHz shader 810 MHz memory 601 MHz
21:09 bloblo: With my old nvidia gts 250
21:11 bloblo: with nvidia driver i can overclock to 740 mhz core and 1000 mhz memory
21:11 bloblo: i want know how i can use personal clock with nouveau ?
21:20 karolherbst: bloblo: well, the source code is out there, but you can also do this with your modified vbios as well (loaded from disc)
21:20 karolherbst: but
21:20 karolherbst: nouveau doesn't check for overheating or power budgets
21:23 bloblo_: thank you
21:24 bloblo_: can tell me more about modified vbios
21:24 bloblo_: is easy to apply ?
21:28 karolherbst: bloblo_: I don'T want to be responsible for anything, if you search for it you should find it. you have to read a lot of sources files anyway to know what to change, or use those crappy windows tools
21:31 bloblo_: I see
21:32 karolherbst: anyway, I wouldn't suggest in doing it, until nouveau is actually able to take actions to protect the hardware
21:37 bloblo_: I understand
21:39 bloblo_: Any way to boost few % performance ?
21:42 karolherbst: bloblo_: switch to the 0f state
21:44 bloblo_: # echo "0f" >> /sys/kernel/debug/dri/0/pstate -su: echo: write error: Function not implemented
21:45 bloblo_: i use kernet param: GRUB_CMDLINE_LINUX_DEFAULT="nouveau.config=NvClkMode=0f"
21:45 karolherbst: uhhh
21:45 karolherbst: wait
21:45 karolherbst: thats a g92
21:45 bloblo_: yes
21:45 karolherbst: with gddr3
21:45 bloblo_: [ 3.483822] nouveau 0000:02:00.0: NVIDIA G92 (092a80a2)
21:45 bloblo_: yes
21:46 karolherbst: kernel?
21:46 bloblo_: 4.9.0-rc5
21:46 karolherbst: mhh
21:46 karolherbst: RSpliet: can't nouveau reclock gddr3 g92 cards?
21:47 karolherbst: uhh
21:47 karolherbst: it is enabled for g94+
21:47 karolherbst: what is so special about g92?
21:48 bloblo_: i have 3 card, gt 630 (64 bit), gts 250 (256 bit), radeon 7850 (256 bit)
21:48 bloblo_: amd is bad with linux, and my old gts 250 is better than gt 630
21:48 karolherbst: bloblo_: the gt 630 is a gf108?
21:49 karolherbst: or gk107 or gk208?
21:49 bloblo_: i dont know but is 64 bits with 3xx cuda core
21:49 karolherbst: gk208 :O
21:49 karolherbst: you are lucky
21:49 karolherbst: this card shouold be actually faster than the gt 250
21:49 karolherbst: let me check
21:50 karolherbst: 700 GFLOPS theoretically
21:50 karolherbst: the gt 250 has like 470
21:50 karolherbst: bloblo_: 384 cuda cores?
21:50 bloblo_: in glxgears gts 250: 1900 fps, gt 630: 4400 fps, but in game gts 250: 1k7 fps, gt 630: 0.4k
21:51 karolherbst: with nvidia?
21:51 bloblo_: http://www.geforce.com/hardware/desktop-gpus/geforce-gt-630/specifications
21:51 karolherbst: mhh
21:51 bloblo_: is the last colon
21:51 karolherbst: true, yours gt 630 has crappy memory
21:52 karolherbst: oh well
21:52 karolherbst: so some things are faster with the gt250 and some with the gt 630
21:52 karolherbst: but if you use nouveau, than you can actually set higher clocks with the gt 630 one
21:52 bloblo_: yeah :D is bad:D
21:52 bloblo_: hm
21:52 karolherbst: the gt 630 will be better suppoted starting with 4.10
21:53 karolherbst: DDR3 memory is slow though
21:53 bloblo_: but the gts 250 is more powerfull in game
21:53 karolherbst: depends on the game
21:53 karolherbst: but yeah, it will most of the time I guess
21:53 bloblo_: counter-strike :D same game since 10 years
21:53 karolherbst: 256 bit GDDR3 vs 64bit DDR3
21:53 karolherbst: the former always wins
21:55 karolherbst: bloblo_: well, we could try something
21:55 karolherbst: bloblo_: did you compiler your kernel yourself?
21:56 bloblo_: yes
21:56 karolherbst: bloblo_: drivers/gpu/drm/nouveau/nvkm/subdev/clk/g84.c
21:56 karolherbst: (device->chipset >= 0x94)
21:56 karolherbst: turn the 0x94 into a 0x92
21:56 karolherbst: but
21:56 karolherbst: remove GRUB_CMDLINE_LINUX_DEFAULT="nouveau.config=NvClkMode=0f"
21:57 karolherbst: it might not work
21:57 karolherbst: I have no idea
21:57 bloblo_: i can test
21:57 bloblo_: i have 16 thread :D
21:57 bloblo_: finaly is interesting
21:57 bloblo_: ready in 19 minutes
21:58 karolherbst: mhh
21:58 karolherbst: maybe the g92 is more close to a nv50 than a g94 though
21:58 karolherbst: so I expect it to not work at all
21:59 bloblo_: I have found the vbios rom https://www.techpowerup.com/vgabios/?architecture=&manufacturer=Palit&model=GTS+250&interface=PCI-E&memType=GDDR3&memSize=1024&since=
21:59 karolherbst: bloblo_: it's on your gpu anyway
22:00 bloblo_: I can test "on my risk" easy with any kernel "param to new.rom" or i need to flash like with windoose
22:02 karolherbst: first try out reclocking :p
22:21 imirkin: bloblo_: to be clear - did you do reclocking on the GT 630?
22:21 imirkin: bloblo_: it should reclock just fine, even without kernel 4.10
22:22 bloblo_: i done
22:22 bloblo_: i reboot and come
22:23 bloblo_: i dont know how reclock is working
22:23 bloblo_: i can entry manualy clock settings ?
22:26 karolherbst: bloblo_: you have to echo 0f into the pstate file
22:28 imirkin: bloblo_: there's no automatic reclocking with nouveau
22:28 imirkin: every time you reclock there's a (low) chance of things hanging, so we def don't want to automate that just yet
22:28 imirkin: heh
22:29 karolherbst: so there he goes
22:29 bloblo: here
22:29 karolherbst: bloblo: let me guess, your machine froze?
22:29 Tom^: and i guess the flicker on reclock is a thing worth looking at before you do automate it
22:29 bloblo: no
22:29 bloblo: is same
22:30 karolherbst: Tom^: depends, do you have multiple displays?
22:30 bloblo: small test glxgears give me same result 1900 fps
22:30 Tom^: karolherbst: yup
22:30 bloblo: no just one
22:30 karolherbst: bloblo: did you echo 0f into pstate?
22:30 bloblo: no i have no param kern
22:30 karolherbst: I meant the file in debugfs
22:30 karolherbst: /sys/kernel/debug/dri/0/pstate
22:30 karolherbst: Tom^: disconnect one and no flickering anymore!
22:31 Tom^: hm
22:31 bloblo: 03: core 300 MHz shader 600 MHz memory 300 MHz ---- 0f: core 675 MHz shader 1458 MHz memory 900 MHz ---- AC: core 399 MHz shader 810 MHz memory 601 MHz
22:32 imirkin: bloblo: "echo 0f > /sys/kernel/debug/dri/0/pstate" (as root)
22:32 bloblo: is working
22:32 bloblo: i think
22:32 karolherbst: bloblo: cat it again
22:32 bloblo: is just passed no error say
22:32 karolherbst: what does the AC line say?
22:32 bloblo: no return say
22:32 imirkin: bloblo: cat it again
22:32 bloblo: i check
22:33 bloblo: 0f: core 675 MHz shader 1458 MHz memory 900 MHz AC DC *
22:33 imirkin: what about the AC: line?
22:33 bloblo: AC line : AC: core 675 MHz shader 1458 MHz memory 601 MHz
22:33 karolherbst: hum
22:33 imirkin: boo!
22:33 bloblo: "*" is in end of 0f
22:33 karolherbst: well
22:33 imirkin: memory reclock failed :(
22:33 karolherbst: imirkin: it's g92
22:33 karolherbst: but yeah, it should have reclocked memory
22:33 imirkin: oh?
22:33 karolherbst: maybe it is also disabled there
22:33 imirkin: i thought it was the GT 630
22:34 karolherbst: imirkin: he also has a 630
22:34 imirkin: i think it's disabled by default
22:34 imirkin: NvMemExec=1
22:34 karolherbst: imirkin: but the 630 has like a fourth of memory perf
22:34 imirkin: should flip it on
22:34 imirkin: meh. i dunno. it should make up for it by having correct rendering ;)
22:34 karolherbst: imirkin: 64bit ddr3 vs 256bit gddr3
22:34 karolherbst: let me check
22:34 bloblo: hi good news
22:35 bloblo: glxgears result 2540
22:35 bloblo: before is 1904
22:35 karolherbst: bloblo: sure, but your memory clock is the same
22:35 Calinou: glxgears result is like 30000 here :P
22:35 karolherbst: allthough 600MHz vs 900MHz ain't that bad
22:35 bloblo: i test param ?: NvMemExec=1
22:35 Calinou: (if you want to maximize your score, use a non-compositing window manager)
22:35 bloblo: i have openbox with minimal
22:35 karolherbst: bloblo: nope, kernel, but let me check something before
22:37 karolherbst: imirkin: isn't memory reclocking enabled by default for g94 gpus?
22:37 imirkin: i think so
22:37 imirkin: but what do i know
22:38 karolherbst: bloblo: any error in dmesg?
22:38 bloblo: i check
22:39 bloblo: i think one difference
22:39 bloblo: wait
22:40 bloblo: http://pastebin.com/ykRGHrR7
22:41 bloblo: begin 21 to end is all new
22:41 karolherbst: you can ignore the pci thing
22:41 karolherbst: that means your motherboard/gpu can only do pcie v1
22:41 karolherbst: but the gpu actually reports it can do more
22:41 bloblo: i have x79 board with 2x pcie 3
22:42 karolherbst: :D
22:42 karolherbst: I see
22:42 karolherbst: ohh okay, your gpu refused to go to v2 then
22:42 karolherbst: odd
22:43 karolherbst: well, doesn't matter anyway for you
22:44 karolherbst: the hwsq error is more serious though
22:44 karolherbst: mhhh
22:45 imirkin: karolherbst: tesla is v1 only :)
22:45 karolherbst: imirkin: actually not
22:45 karolherbst: ohh wait
22:45 karolherbst: nope
22:45 imirkin: for G92? i forget.
22:45 karolherbst: g94 can do it for sure
22:45 imirkin: G80 was definitely v1
22:45 karolherbst: or is g94 v2?
22:45 karolherbst: well
22:46 karolherbst: g9x can do v2
22:46 karolherbst: g8x can't
22:46 bloblo: is palit e-green gts 250, 256 bit
22:46 karolherbst: imirkin: is there a reason why g92 is basically treated like g82?
22:47 karolherbst: bloblo: drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
22:47 karolherbst: bloblo: you have there a nv92_chipset = {
22:47 karolherbst: .bus = nv50_bus_new,
22:48 karolherbst: make it .bus = g94_bus_new,
22:48 karolherbst: or... wait
22:48 karolherbst: I have no idea
22:49 karolherbst: mupuf: do you know something?
22:50 karolherbst: bloblo: can you prove lspci -vv -s 02:00.0
22:50 karolherbst: *provide
22:50 bloblo: wait
22:51 karolherbst: uhh lspci was pimped
22:51 bloblo: http://pastebin.com/XwAA4gVX
22:51 karolherbst: v1 alright...
22:51 karolherbst: bloblo: mind installing envytools?
22:52 bloblo: i have not "envytools"
22:52 bloblo: debian
22:52 karolherbst: uhh I see
22:52 bloblo: http://www.pcgameshardware.de/Grafikkarten-Grafikkarte-97980/News/9800-GT-and-9800-GTX-with-new-names-on-Cebit-675268/
22:52 karolherbst: bloblo: https://github.com/envytools/envytools
22:55 bloblo: have cloned git envytools
22:55 karolherbst: bloblo: inside it: mkdir build; cd build; cmake ..; make
22:57 bloblo: cmake empty build dir
22:58 karolherbst: bloblo: ohh right, you need to install the dev packages needed
22:58 karolherbst: mainly libpciaccess-dev
23:01 bloblo: i have installed dependency, mkdir build cd build, cmake, but cmake want params
23:01 bloblo: ah is two point
23:01 bloblo: not work
23:01 karolherbst: well
23:01 karolherbst: you create new builds dir over and over
23:01 bloblo: ah no wrong dir
23:01 bloblo: i need root perms or i can with user ?
23:02 karolherbst: you don't need to install
23:02 bloblo: ok istarted
23:02 bloblo: error bison
23:02 bloblo: i check bison-dev
23:02 karolherbst: the package is bison
23:03 karolherbst: uhhh
23:03 karolherbst: it seems like the cpie stuff is my fault indeed, let me check
23:04 karolherbst: ohh
23:04 karolherbst: imirkin: guess what, we have no g92 mmiotraces....
23:04 bloblo: http://pastebin.com/xt5tvDB7
23:04 bloblo: libbison-dev installed
23:04 imirkin: karolherbst: inconceivable
23:04 karolherbst: bloblo: looks good
23:05 imirkin: karolherbst: should be some on mmio.dumps worse-case
23:05 karolherbst: imirkin: that's why I enabled the pcie stuff just for g94+
23:05 karolherbst: imirkin: already did add all tesla+ traces
23:05 karolherbst: ...
23:05 karolherbst: took me quite some time
23:05 bloblo: 85% make error
23:06 karolherbst: well good enough
23:06 karolherbst: bloblo: you should have build/nva/nvapeek
23:06 bloblo: yes i have
23:06 karolherbst: bloblo: as root
23:07 karolherbst: build/nva/nvapeek 0x88460
23:07 bloblo: # build/nva/nvapeek 0x88460 00088460: b0602220
23:08 karolherbst: PPCI.CONFIG_LINK => { UNK2 = 0 | TARGET_SPEED = 5_0GT | CARD_SPEED = 5_0GT | SYSTEM_SPEED = 0 | UNK16 = 0xb06 | UNK28 = 0 }
23:08 karolherbst: uhm
23:08 karolherbst: k
23:08 karolherbst: ohh
23:08 karolherbst: forgot a 2
23:08 karolherbst: PPCI.CONFIG_LINK => { UNK2 = 0 | TARGET_SPEED = 5_0GT | CARD_SPEED = 5_0GT | SYSTEM_SPEED = 5_0GT | UNK16 = 0x60 | UNK28 = 0x3 | UNK31 }
23:08 karolherbst: right, this issue
23:09 karolherbst: bloblo: build/nva/nvapoke b0602220
23:09 karolherbst: bloblo: build/nva/nvapoke b0602221
23:09 karolherbst: bloblo: then lspci -vv -s 02:00.0 again
23:09 bloblo: No value specified.No value specified.
23:09 karolherbst: ohhh right
23:09 karolherbst: build/nva/nvapoke 0x88460 b0602220
23:09 karolherbst: build/nva/nvapoke 0x88460 b0602221
23:09 karolherbst: both as root again
23:10 bloblo: return nothing both
23:10 karolherbst: bloblo: that's why the lspci
23:11 bloblo: http://pastebin.com/FWVe3hc5
23:11 karolherbst: imirkin: there are g92 dumps :O
23:11 karolherbst: ugh
23:11 karolherbst: wait a second
23:11 karolherbst: ahh true
23:11 karolherbst: I forgot something
23:11 karolherbst: bloblo: nvapeek 0x00154c
23:12 bloblo: 0000154c: 0000007c
23:12 karolherbst: k
23:12 karolherbst: nvapoke 0x00154c 7d
23:13 karolherbst: then lspci again
23:14 bloblo: http://pastebin.com/Rg5ssXZj
23:14 karolherbst: nvapoke
23:14 karolherbst: not peek
23:15 bloblo: ah, return nothing
23:16 bloblo: lscpi after poke: http://pastebin.com/GxxkKwsn
23:16 karolherbst: nice
23:16 karolherbst: so it should work
23:16 karolherbst: next thing
23:17 karolherbst: nvapoke 0x00154c fd
23:17 karolherbst: then LnkCap should print 5.0 GT/s
23:18 karolherbst: then build/nva/nvapoke b0602221
23:18 karolherbst: and LnkSta should print 5,.0 GT/s as well
23:18 bloblo: http://pastebin.com/wJwNVWgL
23:18 bloblo: i test the last
23:18 bloblo: now*
23:19 bloblo: build/nva/nvapoke b0602221 say no value specified
23:19 karolherbst: build/nva/nvapoke 0x88460 b060222
23:19 bloblo: is done
23:19 bloblo: 5gt
23:19 karolherbst: in lnkCap and lnkSta?
23:20 bloblo: LnkCap: Port #0, Speed 5GT/s, Width x16, ASPM L0s L1, Exit Latency L0s <512ns, L1 <1us
23:20 bloblo: http://pastebin.com/ydvQi99b
23:20 karolherbst: silly lnksta
23:20 karolherbst: uhhh
23:20 karolherbst: build/nva/nvapoke 0x88460 b0602221
23:21 karolherbst: forgot the last digit :O
23:21 bloblo: both 5gt
23:21 bloblo: now
23:22 karolherbst: thanks for confirming :)
23:22 bloblo: http://pastebin.com/SDwE5t0C
23:27 bloblo: still need me ?
23:27 karolherbst: well, you could try out to enable memory reclocking, but I have no clue what would be safe to do
23:28 bloblo: is not problem i have 2 other cards
23:28 bloblo: how enable ?
23:29 bloblo: with the kernel param ?
23:30 karolherbst: it won't help
23:32 bloblo: i go to test with my fps game :D
23:32 bloblo: thank you
23:32 bloblo: ++
23:46 karolherbst: imirkin: actually the g8x GPUs have pcie v2 stuff implemented, but it is totally broken
23:47 karolherbst: they have all the regs and so on, but the PCIe controller gets upsets and throws the GPU out