00:40 imirkin: Lyude: unfortunately that test is hardcoded for 160x160 =/
00:40 imirkin: anyways, wtvr
00:40 imirkin: i think a lot of tests are buggy like that
04:00 vpelletier: imirkin: if you are around, I just booted my ppc64 on the same 4.11.8 kernel, this time with nouveau.config=NvMSI=0 nouveau.vram_pushbuf=1 (as I was not intending to debug the MSI issue) and I caught this line during boot:
04:01 vpelletier: nouveau 0001:06:00.0: bus: MMIO write of 00000000 FAULT at 6813c8
04:02 vpelletier: the card is still in the "wrong" slot (PCIe 4x under a bridge under hypertransport bus, and not on the dedicated PCIe 16x (native ?) bus)
04:05 imirkin: did it have one of the patches i had you try with endianness?
04:05 vpelletier: (but this boot seems to have other issues too, windfarm kernel thread is stuck and cpu0 was soft-locked when I tried to reboot)
04:08 vpelletier: I do not think, checking
04:10 vpelletier: there is no "+" suffix to the uname, so the working copy was clean AFAIK. and I do not see any local commit I would have done with these patches
04:11 vpelletier: OTOH, after a forced shutdown and another boot, I do not see this error anymore :(
04:17 vpelletier: another issue I noticed was that the card is not considered as supporting opengl 2.1 and I remember you said this comes from endianness and GL_* color formats
04:18 mooch2: mwk: pls bby work on nv3
04:18 mooch2: :D
04:18 vpelletier: if I understand correctly, this would be a mesa-level fix (likely nouveau-specific)
04:19 imirkin: correct
04:20 mooch2: imirkin, i added the dumb fucking pci device id configuration to my nvidia emulation
04:20 mooch2: why oh why did they have to do that
04:40 vpelletier: imirkin: ...and I cannot identify where mesa is probing driver for supported extensions (as nouveau does not reference sRGB I guess it instead exposes some lower-level capabilities that mesa uses to decide which GL-level extensions are available). Any hint ?
04:41 vpelletier: fwiw I'm looking in src/mesa as of 9ac1432a5714f2c946d005dcdaa90dc5f738a6d8
04:41 vpelletier: from git://anongit.freedesktop.org/git/mesa/mesa
04:42 imirkin: src/mesa/state_tracker/st_extensions.c
04:42 imirkin: and nv30_format.c
04:42 imirkin: the issue could also be a lack of a byte-swapped srgb format
08:28 ccaione: hey guys, any idea on https://bugs.freedesktop.org/show_bug.cgi?id=101782 ?
15:30 docmax: boah geht mir das auf den keks
15:30 docmax: etwas total simples und es schreibt keiner eine vernünftige anleitung im netz
15:30 docmax: /dev/sda als NBD exportieren
15:30 docmax: wie geht das?
15:30 docmax: oh wrong channel
15:31 mupuf: indeed :)
15:32 RSpliet: poor biscuits...
16:07 karolherbst: docmax: very wrong :p
22:38 tobijk: mhh does the atomic modeset thingy really handle the complete modeset enable/disable for nouveau already? :o
22:40 imirkin: for nv50+ yea
22:41 tobijk: so thats ith then: https://hastebin.com/qizekuhepi.md
22:41 tobijk: pretty unspectactular
22:42 tobijk: whoops wrong one
22:43 tobijk: this one: https://hastebin.com/uzetimemes.md
22:43 imirkin: the if might go elsewhere. it's also questionable whether pre-nv50 needs it. i think skeggsb is just removing it.
22:44 tobijk: yeah the if goes outside the loop for the final one
22:44 tobijk: not sure if it is needed fpr pre nv50, but maybe keep it for now?!
22:45 imirkin: ben's taking care of it :)
22:45 tobijk: meh
22:45 imirkin: there are some additional vblank-related issues he's trying to address while he's at it
22:46 tobijk: mhk
22:51 skeggsb: just removing it leads to hitting other WARNs
22:52 imirkin: of course it does.
22:52 skeggsb: unfortunately.. the interactions of off/on/get/put and a combination of how the hw works + how we use it makes it non-trivial to implement all the rest "properly"
22:52 tobijk: so encapsulate the for_each with !drm_drv_uses_atomic_modeset(dev)
22:53 tobijk: does the trick for me
22:53 skeggsb: tobijk: the warns happen *on* the atomic path
22:53 skeggsb: unload the module, for example..
22:53 tobijk: with the crtc attached?
22:54 skeggsb: i don't think we need it for the non-atomic path either (we shutdown all crtcs, and the dispnv04 code calls off when a crtc is disabled already), but i hvaen't explicitly tested that yet
22:58 tobijk: mh i'm no expert in unloading :D (dont get it done)
22:58 tobijk: does it happen when you put the system to hibernate?
23:01 tobijk: never mind, got the warning