Bug 86935

Yeah, these are the funky GK208B's (often marketed as GT 720?). You can try adding 0x106 right above case 0x108 in drivers/gpu/drm/nouveau/core/engine/device/nve0.c -- they should be identical or at least similar.

If that doesn't work, please append a dmesg here (after the patch) and send a mmiotrace of the blob + vbios to mmio.dumps@gmail.com -- good guide available at https://wiki.ubuntu.com/X/MMIOTracing

Created attachment 110361 [details] [review]
the patch proposed in the previous comment

Adding 0x106 seems to have worked. I can now boot with nouveau framebuffer and nouveau Xorg driver also works.

Will this patch be included in 3.18 (a bit late I guess) or maybe 3.19?

(In reply to Sven from comment #2)
> Created attachment 110361 [details] [review] [review]
> the patch proposed in the previous comment
> 
> Adding 0x106 seems to have worked. I can now boot with nouveau framebuffer
> and nouveau Xorg driver also works.

Awesome. Please confirm that you're using the nouveau 3d driver (you'll need at least mesa 10.2). Try running some semi-stressing 3d thing too? Not sure if you use gnome-shell or something. [glxgears isn't really enough, although it's a start.]

Also, I think it'd still be nice if you could provide a mmiotrace of the blob, the chip could have different golden init values, and other small differences.

As for the patch, change it so that the 0x106 case gets a full copy of the 0x108 stuff, and call it GK208B in the name string, and send it as a proper patch (with git format-patch) to nouveau@lists.freedesktop.org .

I will try to do the mmiotrace of the blob. Problem is, that the binary nvidia drivers gives me a kernel OOPS. Yay!

Also, while I can make the changes to the kernel source, I'm not really sure what you're asking from me regarding the git format-patch. I'm not really an experienced git user yet.

(In reply to Sven from comment #5)
> I will try to do the mmiotrace of the blob. Problem is, that the binary
> nvidia drivers gives me a kernel OOPS. Yay!

D'oh! Maybe try a newer driver? If it oopses "later" rather than on load/X start, then that should be enough for a mmiotrace.

> 
> Also, while I can make the changes to the kernel source, I'm not really sure
> what you're asking from me regarding the git format-patch. I'm not really an
> experienced git user yet.

Step 1: get a git checkout
Step 2: Make changes
Step 3: git commit -a [pops up editor window, enter in a commit log, subject on first line, blank line, then longer description as necessary, and your sign off]
Step 4: git format-patch HEAD^..
Step 5: review the foo.patch file that's generated
Step 6: git send-email --to 'nouveau@lists.freedesktop.org' foo.patch [or send it as an email in some other way, just make sure that way preserves little details like tabs]

(In reply to Ilia Mirkin from comment #6)
> (In reply to Sven from comment #5)
> > I will try to do the mmiotrace of the blob. Problem is, that the binary
> > nvidia drivers gives me a kernel OOPS. Yay!
> 
> D'oh! Maybe try a newer driver? If it oopses "later" rather than on load/X
> start, then that should be enough for a mmiotrace.

I'm on gentoo. I was using the newest nvidia driver. But I suspect the kernel was too new. Anyhow, on the 3rd try it didn't oops. Trace has been emailed.

(In reply to Ilia Mirkin from comment #4)
> (In reply to Sven from comment #2)
> > Adding 0x106 seems to have worked. I can now boot with nouveau framebuffer
> > and nouveau Xorg driver also works.
> 
> Awesome. Please confirm that you're using the nouveau 3d driver (you'll need
> at least mesa 10.2). Try running some semi-stressing 3d thing too? Not sure
> if you use gnome-shell or something. [glxgears isn't really enough, although
> it's a start.]

I'm using mesa 10.3.4 and glxinfo confirms that I'm using "Gallium 0.4 on NV106" and not software rendering or nvidia binary driver. I can confirm that glxgears, quake3-demo, and 0ad alpha 17 run fine. Everything looks like it should.

Sorry, no gnome-shell here. I'm using the MATE Desktop Environment (good old fast 2D stuff).

Created attachment 110407 [details] [review]
proper patch

There is an error in dmesg with the patch applied:
nouveau E[    PBUS][0000:01:00.0] MMIO read of 0x00000000 FAULT at 0x500c30 [ IBUS ]

It doesn't seem to matter though. Is it something serious?

nouveau 0000:01:00.0: Direct firmware load for nouveau/nv106_fuc084 failed with error -2
nouveau 0000:01:00.0: Direct firmware load for nouveau/nv106_fuc084d failed with error -2
nouveau E[    PBSP][0000:01:00.0] unable to load firmware data
nouveau E[    PBSP][0000:01:00.0] init failed, -2

Does extract_firmware.py need updating?
(I have Gentoo's nvidia-firmware-340.32 installed)

(In reply to Sven from comment #11)
> nouveau 0000:01:00.0: Direct firmware load for nouveau/nv106_fuc084 failed
> with error -2
> nouveau 0000:01:00.0: Direct firmware load for nouveau/nv106_fuc084d failed
> with error -2
> nouveau E[    PBSP][0000:01:00.0] unable to load firmware data
> nouveau E[    PBSP][0000:01:00.0] init failed, -2
> 
> Does extract_firmware.py need updating?
> (I have Gentoo's nvidia-firmware-340.32 installed)

Just add symlinks, same as for nv108, to the bsp/vp/ppp files. I'll update extract_firmware.py shortly.

Any chance we can get this into 3.20?

(In reply to Bjørn Lie from comment #13)
> Any chance we can get this into 3.20?

The current plan is for it to be added into 3.19 I believe.

http://cgit.freedesktop.org/~darktama/nouveau/log/?h=linux-3.19

Running kernel 3.19.0 now, and nouveau driver now recognizes my card :-)

I guess this bug can be closed, but leaving it up to the developer.