[Buildroot] [RFC: 3/4] package/x11r7/xdriver_xf86-video-imx-viv: bump to version rel_imx_5.4.24_2.1.0

[Romain Naour]

Hello Gary,

Thanks for your feedback!

Le 11/01/2021 à 14:12, Gary Bisson a écrit :
> Hi Romain,
> 
> On Sun, Jan 10, 2021 at 06:46:38PM +0100, Romain Naour wrote:
>> Hello,
>>
>> Le 07/01/2021 à 12:04, Romain Naour a écrit :
>>> This package needs to be updated to build with the NXP imx_5.4.x_2.1.0 release.
>>>
>>> Don't override CFLAGS on the command line otherwise we loose some include paths
>>> (ex: -I./vivante_gal).
>>> CFLAGS is already provided by TARGET_CONFIGURE_OPTS in make environment.
>>
>> What's the status of vivante gpu support for X11 ? especially for kernel >= 5.4.
> 
> Don't know. To be honest, i.MX6 will soon be a legacy product for NXP
> releases so I strongly suggest moving to mainline + Etnaviv.

ok, but I used the vivante driver since I started from the nitrogen6x_defconfig
for the sabrelite board that use the linux-imx6 from the Boundary BSP.

Also, I wanted to try the opencl support provided by imx-gpu-viv. I'm not sure
there is an opensource opencl provider for imx6 (mesa Clover only support r600
and RadeonSI).

I'll take a look at nitrogen6x_qt5_gst1_mainline_defconfig:

https://github.com/boundarydevices/buildroot-external-boundary/blob/master/configs/nitrogen6x_qt5_gst1_mainline_defconfig

> 
>> Because it's currently broken with several issues:
>>
>> 0) glxinfo fail to report OpenGL informations
>>
>> # glxinfo -B -display :0
>> name of display: :0
>> Error: glXMakeCurrent failed
>> X Error of failed request:  GLXBadContext
>>   Major opcode of failed request:  148 (GLX)
>>   Minor opcode of failed request:  4 (X_GLXDestroyContext)
>>   Serial number of failed request:  20
>>   Current serial number in output stream:  25
>> Perform final cleanup before process termination!
>>
>> (I'm using a patch that replace glWindowPos2iARB calls with glWindowPos2i:
>> https://github.com/Freescale/meta-freescale/blob/dunfell/recipes-graphics/mesa/mesa-demos/Replace-glWindowPos2iARB-calls-with-glWindowPos2i.patch)
> 
> Have you tried NXP pre-built image from latest release? Although it
> seems they now use Wayland/XWayland. Yet it'd be surprising they'd
> release a broken X11 package.

I haven't tested NXP pre-built image, I'll take a look.

Well, NXP pre-built image use a lot of imx fork. I can expect having some issues
by using official/upstream release when building the image with Buildroot:

https://www.nxp.com/docs/en/release-note/IMX_LINUX_RELEASE_NOTES.pdf

It's more difficult that I expected :)

> 
>> 1) vivante_fbdev is looking for a device with "platform:Vivante GCCore" using
>> drmOpen() [1]
>>
>>  strcpy(pDRIInfo->busIdString,"platform:Vivante GCCore");
>>  fd = drmOpen(NULL,pDRIInfo->busIdString);
>>
>> This string "platform:Vivante GCCore" is present in the linux-imx kernel up to
>> the version imx_4.19.35_1.1.0. The DRM_VIVANTE driver is missing in
>> imx_5.4.70_2.3.0.
>>
>> 2) I was able to workaround this issue by using "130000.gpu" instead of
>> "platform:Vivante GCCore" but the dri driver still fail to load.
>>
>> drmOpenByBusid: Searching for BusID 130000.gpu
>> drmOpenDevice: node name is /dev/dri/card0
>> drmOpenDevice: open result is 12, (OK)
>> drmOpenByBusid: drmOpenMinor returns 12
>> drmOpenByBusid: drmGetBusid reports 130000.gpu
>> (II) [drm] DRM interface version 1.4
>> (II) [drm] Could not create SAREA for DRM lock.
>> (EE) VIVANTE(0): [dri] DRIScreenInit failed.  Disabling DRI
>>
>> 3) I tried to enable X11_DRI3 but xdriver_xf86-video-imx-viv doesn't even build
>> due to recent change [2] (removing pScrPriv variable before using it along with
>> other issues).
>> See patch [3] that try to fix the issue.
>> (Note, X11_DRI3 needs libdrm-vivante fork to build)
>>
>>
>> To compare with a working system, I'm using the Debian (xfce) image provided by
>> Boundarydevices [4] but the kernel is 4.14.98-28-boundary-17b (DRM_VIVANTE is
>> still present). I noticed that the SWRAST DRI driver is used along with the
>> vivante driver:
>>
>> [    21.503] (II) Initializing extension GLX
>> [    21.504] (II) AIGLX: Screen 0 is not DRI2 capable
>> [    22.142] (II) IGLX: Loaded and initialized swrast
>> [    22.143] (II) GLX: Initialized DRISWRAST GL provider for screen 0
> 
> FYI, we haven't updated to 5.4 for i.MX6 and don't plan to.

Ok. I think it would be good to add a comment in the help section of
xf86-video-imx-viv package to recommend using a kernel imx_4.19.35_1.1.0.

> 
>> It means that we really need two LIBGL providers enabled at the same time in the
>> Buildroot configuration.
> 
> OpenGL is another issue though, you can get X11 working without it. But
> yes the implementation on libGL from imx-gpu-viv is messy. Thomas did
> implement it once but never submitted it as the change was pretty
> intrusive & hacky.

I think I'll take into account the comment from Jérôme Pouiller and consider
that imx-gpu-viv is a EGL and GLES provider only.

> 
>> I have the feeling that I need to switch to Yocto to use meta-freescale and all
>> the imx hacks and imx forks (libdrm-imx) to get vivante GPU working with X11
>> (maybe).
> 
> Please let us know how that goes. Once again it feels like it would be
> better to move to Etnaviv at this point.

Well, I'm not sure to try the Yocto way for the moment.

I'll try using Etnaviv.

Best regards,
Romain

> 
> Regards,
> Gary
>