Bug 18517

I had no issues with the display during the Live Gnome install.

Mageia-6-sta1-LiveDVD-GNOME-x86_64-DVD.iso via usb efi

Vga  GeForce 560GTX
Monitor Acer AL2216W
Resolution 1680x1050
Driver in use nouveau

The display was fine throughout but the install was a failure.
The booloader would not install giving the error
"Could not install grub2-efi package"

Not willing to fail at this point I rebooted to a boot.iso and ran an upgrade install.
It did not work on 1st try, no ESP partition found.
Booted a PM live dvd and used gparted to change the flags on the fat32 partition
from "msftdata" to "boot esp".
Rebooted again using the boot.iso and the upgrade ran and completed without issue.
Am now easily able to boot EFI to the installed Live Gnome.

CC: (none) => cae

Tried this whole procedure with the Plasma Live DVD iso as well and found exactly the same problem on both machines.  Since this is essentially the same bug I am not opening another bug report.

Another test on the BIOS boot laptop:
Specified run-level 3 using the F6 configuration option and also specified x11driver=vesa.  Same result.

Using isodumper 0.47 to write to the USB stick, a third Transcend pendrive.

Attaching a snapshot of the directory system on the USB key.

Created attachment 7842 [details]
tree directory of usb key

Tried the latest 64-bit Plasma Live iso - 6sta1 2016-06-12.  Exactly the same problem.  No possibility of obtaining debug information without a monitor.

This happens on an old Dell laptop with nvidia and a workstation with nvidia GTX 970.  In fact it kills the machine altogether.  It has to be switched off completely and the monitor cable removed and replaced after a wait before the monitor is accessible again.

Is a level 3 boot possible with live ISOs?

I tried a level 3 boot with the GNOME live iso but ended up in the same place, almost immediately.

How did I do that?  Forget the level 3 business.

Tried both Plasma and GNOME live isos on an Aorus X5 laptop with twin nvidia cards and again failed to access the monitor.  The screen went black and stayed black.  No consoles available and CtrlAltDel caused the screen to flash continuously with a multicoloured noise pattern, probably the monitor test mode kicking in again.  The only way out of that is to switch off and restart.

In all these cases the system fails to establish a connection to the monitor.  One BIOS system and two UEFI.

Is it the graphics driver or something to do with EDID?  I have no idea.  No way to obtain diagnostics.

x86_64  CSM boot.

The suggestion by Charles Edwards to use modeset provides a way in.  After booting Plasma5 Live the live filesystem was accessible.  Appending the journal file and Xorg.0.log for the latest session.  I had run drakx11 but failed to connect to the X server or to start sddm.service.

There seems to be some chat about raid in the journal - can't interpret it.  None of my systems have RAID arrays.  Note also the remark about probing the monitor and failing to obtain the EDID.  "Xorg may do better".

Created attachment 8030 [details]
Journal file for Plasma5 Live

journalctl -ab > file

Created attachment 8031 [details]
Log file for X after running drakx11

drakx11 -> nvidia series 8100 and later
resolution automatic 24 bpp (native 1920x1200)

The modifier was nomodeset of course.

These are extracts from reports by tarazed on the riseup pad:

Latest x86_64 Classic installed on a Dell laptop with nvidia GeForce 8700M using BIOS boot and a UEFI workstation with nvidia GTX970.

Installations:
Dell laptop, CSM boot.  Install ran smoothly with custom partitioning.  Proprietary nvidia driver installed.  Skipped wifi configuration.

         UEFI workstation.  Whistled through the install (custom partitioning), enabled ethernet connection and installed media sources.  The proprietary
         nvidia driver was set up.  No apparent problem there.

After reboot:
Same old problem; failed to connect to the monitor, presumably because the nvidia driver was not working.  dkms seemed to be building the module
         but then the monitor test pattern kicked in.  This is for the BIOS boot.
         Rebooted with nomodeset kernel parameter to revert to text mode.  The "can't launch graphical display" message came up with the advice to login and
         run drakx11.  Did that and rebooted and reached the Plasma desktop.

         UEFI system: the monitor went dead soon after boot.  Rebooted and edited the grub2 boot to use nomodeset which provided the opportunity to configure
         the X display driver.  Rebooted with nomodeset and watched dkms start to rebuild the driver module (kmod?) but it exited with an error 10 fault - 
         failure to build module.  No point in continuing beyond that but a good place to obtain diagnostics.  See attachments.

Addition to comment 11: the installs were for Plasma5 only.

Created attachment 8040 [details]
Journal file for the UEFI system reboot.

Reboot followed Plasma5 install from June 20 Classic iso.

Created attachment 8041 [details]
Xorg.0.log file from UEFI install

Tried a multi-desktop UEFI install from the classic iso and configured the nvidia driver at the end of the installation before rebooting.  Graphics failed on reboot - monitor switched to power-saving mode.  CtrlAltDel rebooted to the grub2 screen.
Pressed Esc as soon as Plymouth appeared and saw the dkms messages about building kmod but was unable to capture them; about three lines then something about starting dkms-auto-installer on every boot...." or something like that, then back to monitor power saving mode.

Note that this sequence can be traced once only.  On the next attempt to reboot the Plymouth screen does not even show up before monitor connection is lost, which implies that the state of the system has changed to some extent.

Performed a cold boot and chose nomodeset.  After the Plymouth splash screen back to text mode and the nvidia build messages.  These were as before.  It looked like the build succeeded - three rows of dots then "searching for new hardware" and then the graphics failure message and "press any key".  It is a loop.

xorg.conf indicates that the nouveau driver was used.  The nouveau driver has never worked on any of my systems.

As a last resort I installed the vesa driver and rebooted to the Mate desktop with a screen resolution of 1024x768.  Unusable apart from examining system files.  What this all seems to point to is the failure of dkms to build a viable kmod.

Attaching diagnostic files.

Created attachment 8045 [details]
Journal file from boot of 64-bit multiple desktop

Created attachment 8046 [details]
xorg.conf after failure to build nvidia module

Created attachment 8047 [details]
Xorg log for the multiple desktop install

Plasma Live iso for x86_64: 2016-06-23

Dell CSM laptop with GeForce 8700M

Running a straight boot from USB drive fails immediately on the graphics - monitor test pattern kicks in.  Dead in the water.

Eventually succeeded in getting to the Plasma desktop under the nouveau driver.

The prescription was to modify the kernel parameters; remove the vga option and add nomodeset, which meant console mode to start with.  The Plymouth splash screen appeared at some point with the cauldron bubbles adding slowly then after a long interval with a mouse pointer on a black screen the desktop arrived.  The setting up stage was missed so the default keyboard was configured instead of UK.  The screen resolution was correct at 1920x1200 and the laptop suspended when the lid was closed.

Please advise if journal and Xorg.0.log would be useful.

That trick cannot be applied on a UEFI system - the grub2 entry looks quite different.

(In reply to Len Lawrence from comment #20)
> That trick cannot be applied on a UEFI system - the grub2 entry looks quite
> different.

You it can.

Click E on the boot entry you wish to edit.
The last entry listed should be 'initrdefi /boot/initrd-------------

Just above it will be 'linuxefi /boot/lmlinuz------------- ro'
This is the line to which you wish to add the append.
Use arrow keys and add nomodeset to the right of the ro 
If you are currently using appends they will also appear in that line.

example:
linuxefi /boot/lmlinuz------------- ro  splash=verbose nomodeset resume=-----

When finished the edit use the F10 key to boot it.

Created attachment 8062 [details]
UEFI install high end nvidia card

This install succeeded after choice of vesa X11 driver via drakx11.

Attachment 7842 is obsolete: 0 => 1
Attachment 8040 is obsolete: 0 => 1
Attachment 8041 is obsolete: 0 => 1

There was an install.log1 file in /root/drakx which probably refers to the earlier install attempt on the GTX970 GTX machine.  Attaching the last 40 lines from that log because it contains an error message regarding missing kernel devel files.

Created attachment 8063 [details]
tail -40 for install.log1

Tail end of /root/drakx/install.log1 for UEFI install for GTX970 system.

This is getting weird.  Reinstalled from the classic iso, choosing multiple desktops as before.  At the configuration stage I agreed to setting up online media sources and updated the system.  Running on autopilot I "accidentally" reinstalled the nvidia driver. Rebooted using nomodeset kernel parameter and waited for the dkms failure.  No show.  Straight to the sddm login and the Mate desktop.  

4.6.2-desktop-3.mga6 and nvidia 367.27.

Logged out and rebooted without using the nomodeset kernel parameter and everything came up normally.

Another successful x86_64 classic iso install, this time on a Dell laptop, CSM boot and nvidia GeForce 8700M.  Chose Mate custom install and set up media sources and updated the system before reboot.  The system booted to an sddm login and then the Mate desktop.  Noted that Plasma was offered at login although I had not asked for it.  It must have been pulled in because of something chosen in the customized package installation and that could explain why sddm was installed.

The wired connection allowed for system updates before the reboot and this seems to be the common factor distinguishing my successful installs from the failed ones.  I would guess that something is being provided in the final updates which brings the kernel and nvidia into line.  An earlier install log for a failed install contained an error which seemed to indicate a mismatch somewhere;

Error! your kernel devel files for kernel 4.6.2-desktop-1.mga6 cannot be found at /lib/modules/4.6.2-desktop-1.mga6/build or /lib/modules/4.6.2-desktop-1.mga6/source.

The kernel version in the installed system ends up at 4.6.2-desktop-3.mga6.

Is this bug still valid for newest 6sta2?

Keywords: (none) => 6sta1
Assignee: tmb => isobuild
Summary: 6sta1 monitor access failure in stage 1 install of Live iso => monitor access failure in stage 1 install of Live iso

All the recent Live isos boot to the desktop provided the kernel parameter nomodeset is provided.  I would like time to redo some of the tests before finally retiring this bug.

This bug is no longer evident in any recent installation tests so it is time to close it, as FIXED?

Resolution: (none) => FIXED
Status: NEW => RESOLVED