Bug 92274

Created attachment 118654 [details]
dmesg from /var/log/dmesg

Created attachment 118655 [details]
VBIOS

Not sure if important, but the vbios is from Gentoo when only one monitor was connected.

Created attachment 118677 [details]
/var/log/dmesg with kernel command line added drm.debug=14 log_buf_len=16M

Having in mind that this could also be a Gennto-specific Bug or could be a badly configured kernel, do you have any suggestions for my kernel config?
I think I followed all instructions to get nouveau to work also due to it
working with only one monitor connected, but there's also the possibility
that I forgot something, so feel free to ask or suggest something that helps solving this.

In your first dmesg, you have:

video=HDMI-0:e video=DVI-I-1:e

Are these required for something? Force-enabling this stuff just leads to trouble down the line.

Interesting....

[    8.872612] nouveau E[    PBUS][0000:01:00.0] MMIO read of 0x00000000 FAULT at 0x690400 [ IBUS ]
[    8.872781] nouveau E[    PBUS][0000:01:00.0] MMIO write of 0xbadf1001 FAULT at 0x690400 [ IBUS ]
[    8.872869] nouveau E[    PBUS][0000:01:00.0] MMIO read of 0x00000000 FAULT at 0x6904c0 [ IBUS ]
[    8.873043] nouveau E[    PBUS][0000:01:00.0] MMIO write of 0xbadf1001 FAULT at 0x6904c0 [ IBUS ]

in drm/nouveau/nvkm/engine/disp/hdmigk104.c:

        /* AVI InfoFrame */
        nvkm_mask(device, 0x690000 + hdmi, 0x00000001, 0x00000000);
        nvkm_wr32(device, 0x690008 + hdmi, 0x000d0282);
        nvkm_wr32(device, 0x69000c + hdmi, 0x0000006f);
        nvkm_wr32(device, 0x690010 + hdmi, 0x00000000);
        nvkm_wr32(device, 0x690014 + hdmi, 0x00000000);
        nvkm_wr32(device, 0x690018 + hdmi, 0x00000000);
        nvkm_mask(device, 0x690000 + hdmi, 0x00000001, 0x00000001);

        /* ??? InfoFrame? */
        nvkm_mask(device, 0x6900c0 + hdmi, 0x00000001, 0x00000000);
        nvkm_wr32(device, 0x6900cc + hdmi, 0x00000010);
        nvkm_mask(device, 0x6900c0 + hdmi, 0x00000001, 0x00000001);

690000 + hdmi / 6900c0 + hdmi each get 2 reads/writes, but you only see one error, both times on the enable. I wonder if this code doesn't need to be more like

u32 foo = nvkm_rd32(device, 0x690000 + hdmi);
nvkm_wr32(device, 0x690000 + hdmi, foo & ~1);
...
nvkm_wr32(device, 0x690000 + hdmi, foo | 1);

and same sort of thing for the second section. Are comfortable enough with code to try this on your own, or do you need patches?

Oh these two video options were for testing and did not show any effect.
I accidently took this kernel log, but the other ones without these options looked the same.
I'm not that into compiling everything on my own (except for some of the easy or
self written parts; or when gentoo does it for me ;)), but I'm willing to learn
and been always looking for a good reason to get myself doing something like this
and not thinking that others will do it or that it does not have enough advantages.

So just that I got it right:
I get the source code from the nouveau git repository,
change the affected code parts, follow the instructions in the readme etc.
and compile nouveau as a module to load it.

Is it enough that I just download a snapshot and do a out-of-tree compilation?
Is it compatible with my kernel?
(because it says it contains the 3.x versions and my version is 4)
Which settings do I need to enable in my kernel that are not trivial?

Sorry for so many questions, but I'd really like to get this done.

(In reply to joseph-thommes from comment #7)
> Oh these two video options were for testing and did not show any effect.
> I accidently took this kernel log, but the other ones without these options
> looked the same.
> I'm not that into compiling everything on my own (except for some of the
> easy or
> self written parts; or when gentoo does it for me ;)), but I'm willing to
> learn
> and been always looking for a good reason to get myself doing something like
> this
> and not thinking that others will do it or that it does not have enough
> advantages.
> 
> So just that I got it right:
> I get the source code from the nouveau git repository,
> change the affected code parts, follow the instructions in the readme etc.
> and compile nouveau as a module to load it.
> 
> Is it enough that I just download a snapshot and do a out-of-tree
> compilation?
> Is it compatible with my kernel?
> (because it says it contains the 3.x versions and my version is 4)
> Which settings do I need to enable in my kernel that are not trivial?
> 
> Sorry for so many questions, but I'd really like to get this done.

Doesn't matter what code you grab, it should basically be the same for the past many kernel versions. Nouveau got a bit of a rewrite in 4.3, so the code might look a bit different if you use an older kernel, but it's the same idea.

If this is your first time compiling a kernel, then it might get a bit tricky. First grab a kernel, compile it, make sure you can boot it, and *then* make the relevant changes.

Oh I misunderstood something about the kernel sources :D.
Thougt this included more afford ;).
I configured my current kernel by myself and it is bootable except for
the multi-monitor problem.
I'm going to try it now.

Ok I'm sorry to ask again, but I've got no idea what the methods nvkm_mask, read, and wr32 are expecting, returning or doing, so I cannot
think by myself what to do...
(By the way in my sources they are just named nv_*.)
So where can I find documentation about it or where are they defined?
I'm familiar with (userspace) C, but not in these dimensions
and I don't know how to locate the definition of a function except for
looking in all included headers recursively...
Did I understand you correctly, that I should make the changes like this:
(again I've got no idea what these functions do, so it might be nonsense)

/* AVI InfoFrame */

        u32 foo = nvkm_rd32(device, 0x690000 + hdmi);
        nvkm_wr32(device, 0x690000 + hdmi, foo & ~1);
        nvkm_wr32(device, 0x69000c + hdmi, 0x0000006f);
        nvkm_wr32(device, 0x690010 + hdmi, 0x00000000);
        nvkm_wr32(device, 0x690014 + hdmi, 0x00000000);
        nvkm_wr32(device, 0x690018 + hdmi, 0x00000000);
        nvkm_wr32(device, 0x690000 + hdmi, foo | 1);

        /* ??? InfoFrame? */
        u32 foo = nvkm_rd32(device, 0x6900c0 + hdmi);        
        nvkm_wr32(device, 0x6900c0 + hdmi, foo & ~1);
        nvkm_wr32(device, 0x6900cc + hdmi, 0x00000010);
        nvkm_wr32(device, 0x6900c0 + hdmi, foo | 1);

And of course I meant effort not afford

(In reply to joseph-thommes from comment #10)
> Ok I'm sorry to ask again, but I've got no idea what the methods nvkm_mask,
> read, and wr32 are expecting, returning or doing, so I cannot
> think by myself what to do...
> (By the way in my sources they are just named nv_*.)
> So where can I find documentation about it or where are they defined?
> I'm familiar with (userspace) C, but not in these dimensions
> and I don't know how to locate the definition of a function except for
> looking in all included headers recursively...
> Did I understand you correctly, that I should make the changes like this:
> (again I've got no idea what these functions do, so it might be nonsense)
> 
> /* AVI InfoFrame */
> 
>         u32 foo = nvkm_rd32(device, 0x690000 + hdmi);
>         nvkm_wr32(device, 0x690000 + hdmi, foo & ~1);
>         nvkm_wr32(device, 0x69000c + hdmi, 0x0000006f);
>         nvkm_wr32(device, 0x690010 + hdmi, 0x00000000);
>         nvkm_wr32(device, 0x690014 + hdmi, 0x00000000);
>         nvkm_wr32(device, 0x690018 + hdmi, 0x00000000);
>         nvkm_wr32(device, 0x690000 + hdmi, foo | 1);
> 
>         /* ??? InfoFrame? */
>         u32 foo = nvkm_rd32(device, 0x6900c0 + hdmi);        
>         nvkm_wr32(device, 0x6900c0 + hdmi, foo & ~1);
>         nvkm_wr32(device, 0x6900cc + hdmi, 0x00000010);
>         nvkm_wr32(device, 0x6900c0 + hdmi, foo | 1);

Seems right to me, except of course it won't compile since you have two variables named "foo" being declared in the same scope. Remove the second 'u32' and you should be good to go.

Ok I changed the file, it compiled fine (with a little C90 standard warning).
On bootup I still get the message at the HDMI Monitor: No input. The DVI Monitor just stays black, not sure if this is specific to this monitor, but I think it usually shows such a message, too. So the symtpoms are the same.
In the last post I accidentally removed one line too much, which I did not in the source code I used to compile it.
The kernel log has changed a bit.

Created attachment 118683 [details]
/var/log/dmesg after changes in ./drm/.../hdmigk104.c

I changed to the nouveau driver in Manjaro and the same issue came up.
Unfortunately I'm not able to obtain a kernel log in a chrooted environment in Manjaro, so there is none attached. If someone knows how to do this, please just answer. (There's no file like dmesg in other distributions in /var/log/).
Anyway, the symptoms are the same, so it is not a gentooish bug, but really a bug in nouveau...

Just saw this (https://bugs.freedesktop.org/show_bug.cgi?id=91705) thread which is about the same problem on another card so its not just this card. I don't know if this helps, but I hope so. (You probably know this already, but just in case.)

Just compiled the new gentoo-4.0.9 sources and the problem persists.

When I changed it in Manjaro it was using 4.1.11 I think.

Created attachment 121229 [details]
kernel log after upgrade to gentoo-4.1.15-r1

The monitors still turn off and I think it's the exact same thing as before.
Just wanted to keep this up-to-date.

(In reply to Joseph Thommes from comment #19)
> Created attachment 121229 [details]
> kernel log after upgrade to gentoo-4.1.15-r1
> 
> The monitors still turn off and I think it's the exact same thing as before.
> Just wanted to keep this up-to-date.

The latest release would be 4.4, which includes the big rewrite from 4.3, whereas 4.1.15 will only include fixes that were back ported (and Nouveau doesn't send many fixes to stable version).

Sorry I can't help more than that: I don't have the correct setup and have no knowledge about that part of Nouveau.

I meant the latest kernel release in sys-kernel/gentoo-sources, but I think, I'll try some more recent sources to find out whether the problem is fixed or not and I'll keep this thread up-to-date.

Created attachment 121284 [details]
dmesg of the 4.4 kernel from the gentoo sys-kernel/git-sources

I upgraded to the 4.4 release of the kernel today, but still the same error happens. I think the syntax of the output was slightly changed, but all in all it's still the same thing.
If you need any other information, tell me.

Hi,
I now have both monitors attached via DVI and now the system boots up. However the monitors now display both the same picture with the same frequency and resolution - they are mirrored. I haven't really tried to solve this by now and I am not sure whether this is a X or a nouveau problem, but my X configuration is apparently not read at all. On the other hand, this mirroring is also the case while booting. So I need to figure out what the problem is and let you know what it was.
This all being the case and me not knowing how the driver is constructed, I think you can isolate the problem to somewhere, but I don't really know where.
I hope this somehow helps.

Ok, the same problem occurs just in the consoles, too.
So it's not a X problem.

AFAIK, the mirrored behaviour in console mode is expected. What happens if you run `xrandr --output ID_OF_YOUR_RIGHT_SCREEN --right-of ID_OF_YOUR_LEFT_SCREEN`, with the id being of the form DVI-I-1?

Can you link your Xorg config file here, and tell us where it is placed on your system?

Ok the fact that the mirroring was also present in console mode made me not even check for the X configuration. Thanks for this tip. I can calibrate and configure the monitors now. Sometimes the simplest things can confuse one :D
But the bug concerning the HDMI and DVI combination was still not fixed before I bought the DVI-cable, so I'm leaving this thread open. :)

I switched back to nouveau (kernel 4.10.1) after a time using the proprietary nvidia driver and now X and tty work even with three monitors attached. But now the switching between X and tty does not work properly, so I'll open a new thread for that.