AMD R7 4700H: Trouble starting X. No devices connected, no screens found.


i've installed "nvidia-hybrid-graphics-0.5". but it's about 8 degrees hotter than on linux. and in "nvidia-settings" gui, i don't find "on demand" option as its linux version.

and i also found that "X11/nvidia-hybrid-graphics" not work with me ( i used it for first try while i reinstall freebsd to verify my setup script. )

First time I see this line. ( nvidia_xorg_enable="YES" )
it's also here : https://www.freshports.org/x11/nvidia-hybrid-graphics/
it may be used for "nvidia optimus" .
 
Yeah, it uses the accellerated Device: Mesa DRI Intel(R) HD Graphics 4600 (HSW GT2) (0x416) device.

it's because i config to use intel card (card1) to run xfce. with command "nvrun-vgl glxgears", you'll find the difference. i show you two screenshots.
 

Attachments

  • Screenshot_2022-01-06_06-25-59.png
    Screenshot_2022-01-06_06-25-59.png
    203.2 KB · Views: 72
  • Screenshot_2022-01-06_06-30-54.png
    Screenshot_2022-01-06_06-30-54.png
    218.1 KB · Views: 69
i compared "glxinfo" with two configs (compared with meld, not eyes), it's the same. it means that it has auto changed to use intel card.

Section "Device"
Identifier "Card0"
VendorName "NVIDIA Corporation"
Driver "nvidia"
BusID "PCI:1:0:0"
EndSection

Section "Device"
Identifier "card1"
VendorName "Intel Corporation"
Driver "modesetting"
BusID "PCI:0:2:0"
Option "DPMS"
EndSection

Section "Screen"
Identifier "Screen0"
Device "Card1"
EndSection

Section "Screen"
Identifier "Screen1"
Device "Card1"
EndSection

Section "Device"
Identifier "Card0"
VendorName "NVIDIA Corporation"
Driver "modesetting"
BusID "PCI:1:0:0"
EndSection

Section "Device"
Identifier "card1"
VendorName "Intel Corporation"
Driver "scfb"
BusID "PCI:0:2:0"
Option "DPMS"
EndSection

Section "Screen"
Identifier "Screen0"
Device "Card0"
EndSection

Section "Screen"
Identifier "Screen1"
Device "Card0"
EndSection
 
I would recommend avoiding x11/nvidia-hybrid-graphics with modern GPUs. The latest (495) driver supports PRIME render offload out of the box with comparable performance and better functionality.
 
i compared "glxinfo" with two configs (compared with meld, not eyes), it's the same. it means that it has auto changed to use intel card.
No, it just means you botched the test somewhere. Since, apparently, you never read log files, this is fully expected.
 
it's because i config to use intel card (card1) to run xfce. with command "nvrun-vgl glxgears", you'll find the difference. i show you two screenshots.

in fact, it's not for my question when i post in that topic. cause i feel my dual card is working, even if that's not perfect (hotter than on linux).

i just want to say, if someone meet "no screen" matter for a new install,

1) try to install the right driver (e.g. version)
2) do the most simple config let it to work first (e.g. use "scfb" driver)
3) try better driver (e.g. "X11/nvidia-hybrid-graphics" not works with me, but "nvidia-hybrid-graphics" works)

be sure that, to discuss may get right idea.
 
I would recommend avoiding x11/nvidia-hybrid-graphics with modern GPUs. The latest (495) driver supports PRIME render offload out of the box with comparable performance and better functionality.

ver 495 ? in the ports ? i don't find it with "pkg search nvidia"

not here also : https://www.freebsd.org/cgi/ports.cgi?query=nvidia&stype=all

or it may in nvidia's site. i'll visit. ( i found it in new feature branch. i'll come back after try it. thanks. )
 
Don't use the official driver from the NVIDIA site. It's recommended to use the one in ports. (Has many patches etc.)

i don't find it in https://www.freebsd.org/cgi/ports.cgi?query=nvidia&stype=all

and also i don't find it by:

cd /usr/ports/
sudo make fetchindex
make search name=nvidia
echo /usr/ports/*/*nvidia-driver*

i also tried install ver 495 driver which download from the nvidia website. it seems that it works not better than "nvidia-hybrid-graphics". i'll back to that. with ver 495 driver, "nvidia-settings" gui and "nvrun-vgl" not works. i'll paste what i did with ver 495 driver and the problem i meet and solution .
 
If I had a compatible NVIDIA card, I'd be tempted to test the new feature branch version i.e. 495.46, but FreeBSD-CURRENT development snapshots are not supported.
if you want to test nvidia drive ver 495, you may prepare a usb key to save your freebsd if it cannot run.
e.g. boot into livecd, mount your freebsd's partition , remove config files in /etc/X11 remove /boot/loader.conf which generated by "make install" the driver, or modify /etc/rc.conf.
i meet a problem after install the driver from nvidia website.
...
efi_check_space: unable to expand staging area
efi_check_space: unable to expand staging area
...
it stop booting. i've to poweroff then use livecd to do what i list steps above.
Jan 6 10:00:47 kernel: nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms 495.46 Wed Oct 27
it's not better than "nvidia-hybrid-graphics", and "nvidia-settings" gui and "nvrun-vgl" cannot work. so i have to remove/install the driver.
 
Which version of FreeBSD for that particular test?

… a problem after install the driver from nvidia website.
... efi_check_space: unable to expand staging area efi_check_space: unable to expand staging area ...
it stop booting. …

Like this?

1641499452995.png
 
Last edited:
Thanks. I apologise: I probably misunderstood what's in the photograph above. I removed some links from the post.

There's no progress beyond EFI buffer information, with which I'm familiar, however the cause is different – not the bug that matches the description.

(I have some cleaning up to do in a topic elsewhere, I should probably leave it until tomorrow.)
 
...
it's not better than "nvidia-hybrid-graphics", and "nvidia-settings" gui and "nvrun-vgl" cannot work. so i have to remove/install the driver.
When using PRIME render offloading, you should not use nvrun-vgl. (I think that there may have some conflict with nvidia-hybrid-graphics)
__NV_PRIME_RENDER_OFFLOAD=1 __GLX_VENDOR_LIBRARY_NAME=nvidia executable

Also try it by using the port tree to build the driver to the feature branch.

You will need an xorg config that will load both driver : nvidia and modesetting
Here a log where the two of them coexist:
Code:
...
[    43.591] (==) ServerLayout "whatever"
[    43.591] (**) |-->Screen "iGPU" (0)
[    43.591] (**) |   |-->Monitor "iGPU-dsp"
[    43.592] (**) |   |-->Device "iGPU-dev"
[    43.592] (**) |-->Screen "dGPU" (1)
[    43.592] (**) |   |-->Monitor "dGPU-dsp"
[    43.592] (**) |   |-->Device "dGPU-dev"
...
[    43.601] (--) PCI:*(0@0:2:0) 8086:191b:1462:115b rev 6, Mem @ 0xdd000000/16777216, 0xb0000000/268435456, I/O @ 0x0000f000/64, BIOS @ 0x????????/65536
[    43.601] (--) PCI: (1@0:0:0) 10de:139b:1462:115b rev 162, Mem @ 0xde000000/16777216, 0xc0000000/268435456, 0xd0000000/33554432, I/O @ 0x0000e000/128
[    43.601] (II) LoadModule: "glx"
[    43.605] (II) Loading /usr/local/lib/xorg/modules/extensions/libglx.so
[    43.628] (II) Module glx: vendor="X.Org Foundation"
[    43.628]    compiled for 1.20.13, module version = 1.0.0
[    43.628]    ABI class: X.Org Server Extension, version 10.0
[    43.628] (II) LoadModule: "modesetting"
[    43.628] (II) Loading /usr/local/lib/xorg/modules/drivers/modesetting_drv.so
[    43.628] (II) Module modesetting: vendor="X.Org Foundation"
[    43.628]    compiled for 1.20.13, module version = 1.20.13
[    43.628]    Module class: X.Org Video Driver
[    43.628]    ABI class: X.Org Video Driver, version 24.1
[    43.628] (II) LoadModule: "nvidia"
[    43.629] (II) Loading /usr/local/lib/xorg/modules/drivers/nvidia_drv.so
[    43.637] (II) Module nvidia: vendor="NVIDIA Corporation"
[    43.637]    compiled for 1.6.99.901, module version = 1.0.0
[    43.637]    Module class: X.Org Video Driver
[    43.639] (II) modesetting: Driver for Modesetting Kernel Drivers: kms
[    43.639] (II) NVIDIA dlloader X Driver  495.46  Wed Oct 27 16:20:16 UTC 2021
[    43.639] (II) NVIDIA Unified Driver for all Supported NVIDIA GPUs
[    43.639] (--) Using syscons driver with X support (version 2.0)
...
 
When using PRIME render offloading, you should not use nvrun-vgl. (I think that there may have some conflict with nvidia-hybrid-graphics)

the problem is i cannot run the gui nvidia-settings, and it still feels hot.

in the /var/log/messages ( it's the log that i saved before reinstall driver. )

Jan 6 10:00:47 kernel: nvidia0: <Unknown> on vgapci0
Jan 6 10:00:47 kernel: vgapci0: child nvidia0 requested pci_enable_io
Jan 6 10:00:47 syslogd: last message repeated 1 times
Jan 6 10:00:47 kernel: nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms 495.46 Wed Oct 27 16:15:12 UTC 2021
Jan 6 10:00:47 kernel: drmn1: <drmn> on vgapci1
Jan 6 10:00:47 kernel: vgapci1: child drmn1 requested pci_enable_io
Jan 6 10:00:47 syslogd: last message repeated 1 times
Jan 6 10:00:47 kernel: [drm] Unable to create a private tmpfs mount, hugepage support will be disabled(-19).
Jan 6 10:00:47 kernel: VT: Replacing driver "efifb" with new "dummy".
Jan 6 10:00:47 kernel: Failed to add WC MTRR for [0xd0000000-0xdfffffff]: -22; performance may suffer
Jan 6 10:00:47 kernel: [drm] Got stolen memory base 0xafe00000, size 0x20000000
Jan 6 10:00:47 kernel: [drm] Supports vblank timestamp caching Rev 2 (21.10.2013).
Jan 6 10:00:47 kernel: [drm] Driver supports precise vblank timestamp query.
Jan 6 10:00:47 kernel: [drm] Connector VGA-1: get mode from tunables:
Jan 6 10:00:47 kernel: [drm] - kern.vt.fb.modes.VGA-1
Jan 6 10:00:47 kernel: [drm] - kern.vt.fb.default_mode
Jan 6 10:00:47 kernel: [drm] Connector eDP-1: get mode from tunables:
Jan 6 10:00:47 kernel: [drm] - kern.vt.fb.modes.eDP-1
Jan 6 10:00:47 kernel: [drm] - kern.vt.fb.default_mode
Jan 6 10:00:47 kernel: [drm] Connector HDMI-A-1: get mode from tunables:
Jan 6 10:00:47 kernel: [drm] - kern.vt.fb.modes.HDMI-A-1
Jan 6 10:00:47 kernel: [drm] - kern.vt.fb.default_mode

i didn't find the ver 495 driver in the ports. maybe it's in the stable branch not the release branch ?
 
Feature branch will not be available as a package see PR 254668
Alexey Dokuchaev said:
Sorry, but no. We won't be serving short-lived/beta driver versions as a port or package, this is too large of a commitment to make. Current matrix -- that is, combination of the main driver, three legacy drivers, and the same for Linux libraries, is already hard enough to test and maintain. We do, however, aim to support building those ports against arbitrary DISTVERSIONs, because naturally, at some point beta version would turn stable and those hooks would stay.

That said, enthusiastic users who wish to play with 465.xx should be able to do that by typing ``make DISTVERSION=465.19.01 -DNO_CHECKSUM ...'' which should DTRT now.

so normaly, make DISTVERSION=495.44 -DNO_CHECKSUM would do the trick.
But beware, some patching could be needed.

The nvidia settings gui will not work if you keep nvidia-hybrid-graphics installed after installing the 495.x version.
It will run a separate script which will do something like: nvidia-settings -c :8 (too lazy to put the libexec path), which will return an error since there will be no server on display :8, only one on display :0 .

For the heating issue, unless you instruct it with the proper command to shutdown the gpu it will run since the start of the computer.
 
The nvidia settings gui will not work if you keep nvidia-hybrid-graphics installed after installing the 495.x version.

i've removed "nvidia-hybrid-graphics" when i tried ver 495 driver.

i think it's better to wait for its release version. do not use it for now. it's still in the feature branch in official website.
 
Back
Top