> Date: Wed, 7 Aug 2019 20:17:26 +0200
> From: Sebastien Marie <sema...@online.fr>
>
> On Wed, Aug 07, 2019 at 03:16:56PM +0200, Mark Kettenis wrote:
> > >
> > > I built a kernel with DRMDEBUG and put 0xffff in drm_debug variable.
> > >
> > > initializing kernel modesetting (RV610 0x1002:0x94C3 0x1028:0x0402 0x00).
> > > [drm] COMBIOS detected
> > > drm:pid0:r600_init *ERROR* Expecting atombios for R600 GPU
> > > drm:pid0:radeondrm_attachhook *ERROR* Fatal error during GPU init
> > > [drm] radeon: finishing device.
> > > [TTM] Memory type 2 has not been initialized
> > > [drm]
> > > drm0 detached
> > > radeondrm0 detached
> > > vga1 at pci1 dev 0 function 0 "ATI Radeon HD 2400 Pro" rev 0x00
> > >
> > >
> > > I tried to follow a bit the initialisation path from radeon_get_bios():
> > > - radeon_get_bios()
> > > - radeon_read_disabled_bios()
> > > - r600_read_disabled_bios()
> > > - init registers
> > > - radeon_read_bios()
> > > - restore registers
> >
> > Can you try to figure out where exactly reading the BIOS from the card
> > goes wrong?
>
> radeon_read_bios (called by r600_read_disabled_bios) returns something
> good enough to pass some tests (bios signature check) and finally
> returns true. the fail is later (in r600_init function, where it checks
> for ATOM bios. it isn't so its abort init).
>
> so I tried something stupid: commenting the call to
> radeon_read_disabled_bios, in order to force radeon_get_bios to try
> another way to get bios.
>
> it called radeon_read_platform_bios() (the next method) and succeed. it
> returned an ATOM bios (so r600_init is happy), radeondrm(4) finished its
> initialization, and xenodm started correctly.
>
> $ dmesg
> ...
> initializing kernel modesetting (RV610 0x1002:0x94C3 0x1028:0x0402
> 0x00).
> radeondrm0: 1680x1050, 32bpp
> wsdisplay0 at radeondrm0 mux 1: console (std, vt100 emulation), using
> wskbd0
> ...
>
> $ xrandr
> <
> Screen 0: minimum 320 x 200, current 1680 x 1050, maximum 8192 x 8192
> DIN disconnected primary (normal left inverted right x axis y axis)
> DVI-0 connected 1680x1050+0+0 (normal left inverted right x axis y
> axis) 430mm x 270mm
> 1680x1050 59.88*+
> 1280x1024 75.02 60.02
> 1152x864 75.00
> 1024x768 75.03 60.00
> 800x600 75.00 60.32
> 640x480 75.00 59.94
> 720x400 70.08
>
> Below the patch for what I did (it could be more evident than my explain)
>
> Index: radeon_bios.c
> ===================================================================
> RCS file: /cvs/src/sys/dev/pci/drm/radeon/radeon_bios.c,v
> retrieving revision 1.15
> diff -u -p -r1.15 radeon_bios.c
> --- radeon_bios.c 14 Apr 2019 10:14:54 -0000 1.15
> +++ radeon_bios.c 7 Aug 2019 17:41:42 -0000
> @@ -815,8 +815,10 @@ bool radeon_get_bios(struct radeon_devic
> r = igp_read_bios_from_vram(rdev);
> if (r == false)
> r = radeon_read_bios(rdev);
> +#if 0
> if (r == false)
> r = radeon_read_disabled_bios(rdev);
> +#endif
> if (r == false)
> r = radeon_read_platform_bios(rdev);
> if (r == false || rdev->bios == NULL) {
>
>
> Now, I know it is only a workaround on this specific machine.
>
> I still dunno if:
>
> - radeon_read_disabled_bios() should return false (because it
> isn't the right way for this specific card), and so the use of
> radeon_read_platform_bios() is the right thing to do.
>
> - radeon_read_disabled_bios() should return true and an ATOM bios,
> and calling radeon_read_platform_bios() is a workaround.
Could you try booting an i386 kernel. I think you can just boot the
kernel without reinstalling. Obviously you'll crash as soon as it
tries to run init(8) but at that point the bios reading should already
have failed.
