GPU firmware naming and problems with loading

[Alexey Dokuchaev]

On Thu, Apr 23, 2020 at 10:42:22PM +0200, Niclas Zeising wrote:
> On 2020-04-21 15:09, Alexey Dokuchaev wrote:
> > ...
> >Why is it trying to load TAHITI modules is another question.
> 
> I don't know why it tries to load TAHITI, I'll try to figure more
> out, but in general, the firmware selection code is from the original
> source, so it should be the same elsewhere.  Does TAHITI load with
> drm-legacy-kmod?

Nope.  This is for legacy (kldstat | grep radeon_):
41    1 0xffffffff81b31000     2385 radeon_ARUBA_pfp_bin.ko
42    1 0xffffffff81b34000     2383 radeon_ARUBA_me_bin.ko
43    1 0xffffffff81b37000     1985 radeon_ARUBA_rlc_bin.ko

This is for drm-{current,devel}-kmod (surprisingly, couple of my later
attempts to "kldload radeonkms" did not hang the laptop, but the very
latest still did):

48    1 0xffffffff819d3000     2385 radeon_ARUBA_pfp_bin.ko
49    1 0xffffffff819d6000     2383 radeon_ARUBA_me_bin.ko
50    1 0xffffffff819d9000     1985 radeon_ARUBA_rlc_bin.ko
51    1 0xffffffff819db000    35ba7 radeon_TAHITI_uvd_bin.ko
52    1 0xffffffff81a11000     beb7 radeon_TAHITI_vce_bin.ko

Both drm-fbsd{11.2,12.0}-kmod still lock up the laptop, but the logs show
they do load TAHITI modules as well.

> If you have the opportunity, can you check which firmwares are loaded on
> a Linux system with the same hardware?

I've booted Ubuntu 18.04, lsmod(8) output is too cluttered on Linux so
perhaps we should look elsewhere, but per dmesg(8) it seems that only
ARUBA firmware is loaded.  Is there some special command on GNU/Linux
that shows exactly which firmware files were loaded?

> Does it work if you remove the TAHITI modules before loading the
> graphics ko?  Looks like it loads TAHITI ok though.

It does not seem to make any noticeable difference.  I'm starting
to think that it might not be related to the firmware.  I'll continue
my investigations.

Another thing is, that even if "kldload radeonkms" succeeds, glxinfo -B
still reports software mode.  In the /var/log/Xorg.0.log, I see these
lines, does it ring any bell?

    (II) AIGLX: Screen 0 is not DRI2 capable
    (EE) AIGLX: reverting to software rendering

./danfe