Ubuntu 17.10 with RADV on GCN 1.1 (290) help

[foxx1337]

bridgman, Kano and debianxfce: you helped me fix this!

I had previously wrongly considered that my 290 is an "si" card, hence I had the "radeon.si_support=0 amdgpu.si_support=1" params passed to the kernel. Apparently it's a "cik" card.

- leave radeon on default state (aka not blacklist it)
- add "radeon.si_support=0 amdgpu.si_support=1 radeon.cik_support=0 amdgpu.cik_support=1" to the kernel args

$ dmesg | grep -P "(modeset|radeon|amdgpu)"
[    0.000000] Command line: BOOT_IMAGE=/boot/vmlinuz-4.13.0-16-generic root=UUID=f6e04b5c-6951-4bc6-9cc5-01e3168f060b ro quiet splash radeon.cik_support=0 amdgpu.cik_support=1 radeon.si_support=0 amdgpu.si_support=1 vt.handoff=7
[    0.000000] Kernel command line: BOOT_IMAGE=/boot/vmlinuz-4.13.0-16-generic root=UUID=f6e04b5c-6951-4bc6-9cc5-01e3168f060b ro quiet splash radeon.cik_support=0 amdgpu.cik_support=1 radeon.si_support=0 amdgpu.si_support=1 vt.handoff=7
[    0.981653] [drm] radeon kernel modesetting enabled.
[    0.985255] fb: switching to radeondrmfb from VESA VGA
[    0.985435] radeon 0000:01:00.0: CIK support disabled by module param
[    1.032905] [drm] amdgpu kernel modesetting enabled.
[    1.033066] [drm] initializing kernel modesetting (HAWAII 0x1002:0x67B1 0x174B:0xE285 0x00).
[    1.056573] amdgpu 0000:01:00.0: VRAM: 4096M 0x0000000000000000 - 0x00000000FFFFFFFF (4096M used)
[    1.056574] amdgpu 0000:01:00.0: GTT: 4096M 0x0000000100000000 - 0x00000001FFFFFFFF
[    1.056661] [drm] amdgpu: 4096M of VRAM memory ready
[    1.056662] [drm] amdgpu: 4096M of GTT memory ready.
[    1.217077] amdgpu 0000:01:00.0: amdgpu: using MSI.
[    1.217091] [drm] amdgpu: irq initialized.
[    1.222803] [drm] amdgpu: dpm initialized
[    1.223276] amdgpu 0000:01:00.0: fence driver on ring 0 use gpu addr 0x0000000100000010, cpu addr 0xffff9c73017a5010
[    1.223372] amdgpu 0000:01:00.0: fence driver on ring 1 use gpu addr 0x0000000100000020, cpu addr 0xffff9c73017a5020
[    1.223463] amdgpu 0000:01:00.0: fence driver on ring 2 use gpu addr 0x0000000100000030, cpu addr 0xffff9c73017a5030
[    1.223521] amdgpu 0000:01:00.0: fence driver on ring 3 use gpu addr 0x0000000100000040, cpu addr 0xffff9c73017a5040
[    1.223581] amdgpu 0000:01:00.0: fence driver on ring 4 use gpu addr 0x0000000100000050, cpu addr 0xffff9c73017a5050
[    1.223642] amdgpu 0000:01:00.0: fence driver on ring 5 use gpu addr 0x0000000100000060, cpu addr 0xffff9c73017a5060
[    1.223665] amdgpu 0000:01:00.0: fence driver on ring 6 use gpu addr 0x0000000100000070, cpu addr 0xffff9c73017a5070
[    1.223689] amdgpu 0000:01:00.0: fence driver on ring 7 use gpu addr 0x0000000100000080, cpu addr 0xffff9c73017a5080
[    1.223712] amdgpu 0000:01:00.0: fence driver on ring 8 use gpu addr 0x0000000100000090, cpu addr 0xffff9c73017a5090
[    1.223760] amdgpu 0000:01:00.0: fence driver on ring 9 use gpu addr 0x00000001000000a0, cpu addr 0xffff9c73017a50a0
[    1.223786] amdgpu 0000:01:00.0: fence driver on ring 10 use gpu addr 0x00000001000000b0, cpu addr 0xffff9c73017a50b0
[    1.224075] amdgpu 0000:01:00.0: fence driver on ring 11 use gpu addr 0x000000000088bd30, cpu addr 0xffffa83ac3238d30
[    1.224150] amdgpu 0000:01:00.0: fence driver on ring 12 use gpu addr 0x00000001000000d0, cpu addr 0xffff9c73017a50d0
[    1.224168] amdgpu 0000:01:00.0: fence driver on ring 13 use gpu addr 0x00000001000000e0, cpu addr 0xffff9c73017a50e0
[    1.615136] fbcon: amdgpudrmfb (fb0) is primary device
[    1.615205] amdgpu 0000:01:00.0: fb0: amdgpudrmfb frame buffer device
[    1.825786] [drm] Initialized amdgpu 3.18.0 20150101 for 0000:01:00.0 on minor 0
[   21.449357] [drm:amdgpu_connector_dvi_detect [amdgpu]] *ERROR* HDMI-A-1: probed a monitor but no|invalid EDID

I really appreciate you caring and giving me the guidance through this process; hope someone else finds this thread useful at a later point.

[bridgman]

That looks much better. Glad you got it working.

Now you can start trying to get radv going

[foxx1337]

I finally got home and had the time to play around with vulkan-smoketest and the SaschaWillems/Vulkan (on github) [1] demos. At first I was experiencing core dumps, then I found a mailinglist listing [2] suggesting to replace the Ubuntu 17.10 provided libvulkan_radeon.so with one from Padoka or Oibaf.

I just grabbed "libgl1-mesa-glx_17.3" from Padoka [3] and replaced my /usr/lib/x86_64-linux-gnu/libvulkan_radeon.so with it. Dota 2 works, all of Sascha's demos I built also work (and run more than ok).

[1] https://github.com/SaschaWillems/Vulkan
[2] https://www.mail-archive.com/desktop...msg503442.html
[3] https://launchpad.net/~paulo-miguel-...mesa/+packages and click there on mesa - 1:17.3~git<something>~padoka0, then grab from there the included libgl1-mesa-glx_17.3~git<something>~padoka0_amd64.deb, open it with an archive manager and find inside libvulkan_radeon.so

[airlied]

Just FYI we discovered today, that Ubuntu patch the vulkan drivers with MIR support and broke radv while doing so.

[DarkFoss]

After my initial attempt at that bug report I waited for Ubuntu to package the mesa 17.2.2 upgrade I made sure apport had all the extra components needed. I immediately installed all the vulkan related ddbug then ran the vulkan-smoketest and let apport do it's thing auto-reporting it. Hopefully that helped discover this since it did a far better job than I could have.

Afterwards I tried unsuccessfully to rebuild the Ubuntu source I saw the quilt patch that only seemed to enable the Intel side. I then assumed that radv was either disabled/broken by design and would be turned on when they deemed fit. Because of that false assumption I never updated the initial bug report about the patch since I thought it was deliberate. It was also the reason I never filed upstream I didn't want to waste dev's time for what was to me a packaging choice not a bug with radv.

Linux has changed a lot since I ran it outside of a vm, longest stint ever with Ubuntu/Debian much more comfortable with Magia Urpmi/rpm. I've got much to unlearn, re-learn as well as learn. I'll get there eventually.

foxx1337 While I'm glad that hack helped you out good to see it's won't be needed much longer. Don't forget to revert it once the updated packages land and before you install them.