End of Week Cleanup

Friday, May 19, 2017

Trying to tie some loose ends and close some gaps.

Machine updates

Shuttle rebooted during a file copy again. It looks like the problem is really related to networking activity, since most other activies, including disk-heavy VM runs, do not cause reboots. But a machine without a network is not that useful

Creating a media image for Fedora26 failed the first time. It turns out that Linux over-optimizes file transfer: 

[root@shuttle ddd]# dd if=/data/systems/ISO/Fedora-Workstation-Live-x86_64-26_Alpha/Fedora-Workstation-Live-x86_64-26_Alpha-1.7.iso of=/dev/sdc1 bs=1M
1373+0 records in
1373+0 records out
1439694848 bytes (1.4 GB, 1.3 GiB) copied, 38.5871 s, 37.3 MB/s
[root@shuttle ddd]# dd if=/data/systems/ISO/Fedora-Workstation-Live-x86_64-26_Alpha/Fedora-Workstation-Live-x86_64-26_Alpha-1.7.iso of=/dev/sdc1 bs=1M
1373+0 records in
1373+0 records out
1439694848 bytes (1.4 GB, 1.3 GiB) copied, 0.503722 s, 2.9 GB/s
[root@shuttle ddd]# sync

The problem is this: 

1439694848 bytes (1.4 GB, 1.3 GiB) copied, 0.503722 s, 2.9 GB/s

I don't think that my flash drive suddenly became capable of being written to at 2.9GB/s !!! So the most likely explanation is that the file cache somehow applies to /dev/sdc1, which was very surprising to me. But the fact that sync makes the device LED blink, and takes quite a while, gives credence to that theory.

Now, macOS gives /dev/disk and /dev/rdisk devices, where one is cached and the other is raw. HP-UX and I believe other historical Unixen have the same difference. Linux does it differently. A quick Google search led me to this page, which seems to demonstrate that on Linux, you don't have raw devices by default, you must create them.

Lesson learned. That will only bite me once, I guess

By the way, the Fedora installer geo-locates me, and offers French as the first language choice. But what I positively love is that it offers Brezhoneg (Breton) and Catalan as alternative options One day, I'll create a virtual machine with Breizhoneg language.

Enabling streaming VM on Turbo

Still working on this one. It took a few reboots, but blacklisting nouveau seems to do the trick. The problem is to get ahead of the system while it's booting. You have to type the spicy command line (spicy -h turbo -p 5901 in my case, but not run it (since the guest is not up, so not listening). Then start the guest with virsh start f25-turbo. And very quickly, switch to the terminal with the spicy command (which must be a different terminal, since it's a different machine), launch spicy, hope that it connects quickly enogh, and very quickly type e to stop the boot process and edit the command line.

Took me 5 tries to get it Now I can edit my grub configuration to blacklist nouveau, the guest becomes responsive, and I can edit the grub configuration with more refinement from the guest.

But then, I'm not sure how I'm supposed to add custom menu entries in GRUB. Apparently, /etc/grub2.cfg is all automatically generated. For my current specific case (blacklisting nouveau), that's all dandy. I can simply edit /etc/sysconfig/grub. But what if I want to add multiple menu entries for the same kernel with different options (which is exactly what I want for Turbo, the host)?

And by the way, disabling nouveau means lspci no longer hangs. To be added to Uri's documentation.

Installed the drivers. Rebooted. Does not seem to work too well: 

         Starting Fallback to nouveau if nvidia did not load...
[   54.820898] wmi: Mapper loaded
[   55.976167] nouveau 0000:00:0a.0: NVIDIA GM206 (126300a1)
[   56.056475] nouveau 0000:00:0a.0: bios: version 84.06.62.00.01
[   56.059658] nouveau 0000:00:0a.0: disp: dcb 15 type 8 unknown
[     *] (2 of 2) A start job is running for...ia did not load (45s / no limit)[   57.970568] nouveau 0000:00:0a.0: fb: 4096 MiB GDDR5
[   57.974435] nouveau 0000:00:0a.0: bus: MMIO write of 800000fe FAULT at 10eb14 [ IBUS ]
[   57.981248] nouveau 0000:00:0a.0: DRM: VRAM: 4096 MiB
[   57.981721] nouveau 0000:00:0a.0: DRM: GART: 1048576 MiB
[   57.982208] nouveau 0000:00:0a.0: DRM: TMDS table version 2.0
[   57.982717] nouveau 0000:00:0a.0: DRM: DCB version 4.1
[   57.983199] nouveau 0000:00:0a.0: DRM: DCB outp 00: 04800f96 04400020
[   57.983770] nouveau 0000:00:0a.0: DRM: DCB outp 01: 04000f92 00020020
[   57.984359] nouveau 0000:00:0a.0: DRM: DCB outp 02: 04811f86 04400010
[   57.984933] nouveau 0000:00:0a.0: DRM: DCB outp 03: 04011f82 00020010
[   57.985521] nouveau 0000:00:0a.0: DRM: DCB outp 04: 02822f66 04400010
[   57.986100] nouveau 0000:00:0a.0: DRM: DCB outp 05: 02022f62 00020010
[   57.986681] nouveau 0000:00:0a.0: DRM: DCB outp 06: 02833f76 04400020
[   57.987378] nouveau 0000:00:0a.0: DRM: DCB outp 07: 02033f72 00020020
[   57.988099] nouveau 0000:00:0a.0: DRM: DCB outp 15: 01df4ff8 00000000
[   57.988678] nouveau 0000:00:0a.0: DRM: DCB conn 00: 02000046
[   57.989191] nouveau 0000:00:0a.0: DRM: DCB conn 01: 01000146
[   57.989710] nouveau 0000:00:0a.0: DRM: DCB conn 02: 00010246
[   57.990380] nouveau 0000:00:0a.0: DRM: DCB conn 03: 00020346
[   57.991058] nouveau 0000:00:0a.0: DRM: DCB conn 04: 00000470
[   57.992125] nouveau 0000:00:0a.0: DRM: Pointer to flat panel table invalid

Next boot gives me the same thing: 

[  OK  ] Started Builds and install new kernel modules through DKMS.
         Starting Fallback to nouveau if nvidia did not load...
[   45.417949] wmi: Mapper loaded
[   46.027370] nouveau 0000:00:0a.0: NVIDIA GM206 (126300a1)
[   46.102917] nouveau 0000:00:0a.0: bios: version 84.06.62.00.01
[   46.103804] nouveau 0000:00:0a.0: disp: dcb 15 type 8 unknown
[   48.216073] nouveau 0000:00:0a.0: fb: 4096 MiB GDDR5
[   48.216572] nouveau 0000:00:0a.0: bus: MMIO write of 80000115 FAULT at 10eb14 [ IBUS ]
[   48.227391] nouveau 0000:00:0a.0: DRM: VRAM: 4096 MiB
[   48.227870] nouveau 0000:00:0a.0: DRM: GART: 1048576 MiB
[   48.228364] nouveau 0000:00:0a.0: DRM: TMDS table version 2.0
[   48.228881] nouveau 0000:00:0a.0: DRM: DCB version 4.1
[   48.229420] nouveau 0000:00:0a.0: DRM: DCB outp 00: 04800f96 04400020
[   48.230200] nouveau 0000:00:0a.0: DRM: DCB outp 01: 04000f92 00020020
[   48.231170] nouveau 0000:00:0a.0: DRM: DCB outp 02: 04811f86 04400010
[   48.231841] nouveau 0000:00:0a.0: DRM: DCB outp 03: 04011f82 00020010
[   48.232532] nouveau 0000:00:0a.0: DRM: DCB outp 04: 02822f66 04400010
[   48.233254] nouveau 0000:00:0a.0: DRM: DCB outp 05: 02022f62 00020010
[   48.238024] nouveau 0000:00:0a.0: DRM: DCB outp 06: 02833f76 04400020
[   48.238700] nouveau 0000:00:0a.0: DRM: DCB outp 07: 02033f72 00020020
[   48.239371] nouveau 0000:00:0a.0: DRM: DCB outp 15: 01df4ff8 00000000
[   48.240042] nouveau 0000:00:0a.0: DRM: DCB conn 00: 02000046
[   48.240640] nouveau 0000:00:0a.0: DRM: DCB conn 01: 01000146
[   48.241221] nouveau 0000:00:0a.0: DRM: DCB conn 02: 00010246
[   48.241807] nouveau 0000:00:0a.0: DRM: DCB conn 03: 00020346
[   48.242394] nouveau 0000:00:0a.0: DRM: DCB conn 04: 00000470
[   48.242985] nouveau 0000:00:0a.0: DRM: Pointer to flat panel table invalid
[   48.523212] nouveau 0000:00:0a.0: DRM: unknown connector type 70

Apparenlty, the "fallback" bypasses the blacklisting entirely. Not good. Will attempt to completely uninstall nouveau.

Uninstalled nouveau. Screen is still entirely black. Starting X manually gives me an error: 

[root@f25-turbo ddd]# startx
xauth:  file /root/.serverauth.5187 does not exist
xauth:  file /root/.Xauthority does not exist
xauth:  file /root/.Xauthority does not exist




X.Org X Server 1.19.3
Release Date: 2017-03-15
X Protocol Version 11, Revision 0
Build Operating System:  4.9.3-200.fc25.x86_64
Current Operating System: Linux f25-turbo.dinechin.lan 4.10.8-200.fc25.x86_64 #1 SMP Fri Mar 31 13:20:22 UTC 2017 x86_64
Kernel command line: BOOT_IMAGE=/vmlinuz-4.10.8-200.fc25.x86_64 root=UUID=7b7d6716-597f-488e-9e4e-53f44b56550b ro rootflags=subvol=root console=ttyS0 modprobe.blacklist=nouveau rd.driver.blacklist=nouveau
Build Date: 15 March 2017  06:37:12PM
Build ID: xorg-x11-server 1.19.3-1.fc25
Current version of pixman: 0.34.0
	Before reporting problems, check http://wiki.x.org
	to make sure that you have the latest version.
Markers: (--) probed, (**) from config file, (==) default setting,
	(++) from command line, (!!) notice, (II) informational,
	(WW) warning, (EE) error, (NI) not implemented, (??) unknown.
(==) Log file: "/var/log/Xorg.0.log", Time: Fri May 19 16:09:12 2017
(==) Using config file: "/etc/X11/xorg.conf"
(==) Using config directory: "/etc/X11/xorg.conf.d"
(==) Using system config directory "/usr/share/X11/xorg.conf.d"
(EE)
Fatal server error:
(EE) no screens found(EE)
(EE)
Please consult the Fedora Project support
	 at http://wiki.x.org
 for help.
(EE) Please also check the log file at "/var/log/Xorg.0.log" for additional information.
(EE)
(EE) Server terminated with error (1). Closing log file.
xinit: giving up
xinit: unable to connect to X server: Connection refused
xinit: server error

The /var/log/Xorg.0.log complains that no screens were found: 

Fatal server error:
[   175.993] (EE) no screens found(EE)
[   175.993] (EE)

The card is clearly detected, and the 381.22 driver does load: 

[   175.948] (II) systemd-logind: logind integration requires -keeptty and -keeptty was not provided, disabling logind integration
[   175.948] (II) xfree86: Adding drm device (/dev/dri/card0)
[   175.948] (II) xfree86: Adding drm device (/dev/dri/card1)
[   175.954] (--) PCI:*(0:0:2:0) 1af4:1050:1af4:1100 rev 1, Mem @ 0xf2000000/8388608, 0xf2800000/8388608, 0xfe078000/4096, BIOS @ 0x????????/131072
[   175.954] (--) PCI: (0:0:10:0) 10de:1430:10de:1190 rev 161, Mem @ 0xfd000000/16777216, 0xe0000000/268435456, 0xf0000000/33554432, I/O @ 0x0000c000/128, BIOS @ 0x????????/131072
[   175.954] (II) LoadModule: "glx"
[   175.955] (II) Loading /usr/lib64/xorg/modules/extensions/libglx.so
[   175.958] (II) Module glx: vendor="NVIDIA Corporation"
[   175.958] 	compiled for 4.0.2, module version = 1.0.0
[   175.958] 	Module class: X.Org Server Extension
[   175.958] (II) NVIDIA GLX Module  381.22  Thu May  4 00:17:15 PDT 2017
[   175.958] (II) LoadModule: "nvidia"
[   175.958] (II) Loading /usr/lib64/xorg/modules/drivers/nvidia_drv.so
[   175.958] (II) Module nvidia: vendor="NVIDIA Corporation"
[   175.958] 	compiled for 4.0.2, module version = 1.0.0
[   175.959] 	Module class: X.Org Video Driver
[   175.959] (II) NVIDIA dlloader X Driver  381.22  Wed May  3 23:53:41 PDT 2017
[   175.959] (II) NVIDIA Unified Driver for all Supported NVIDIA GPUs
[   175.993] (EE) No devices detected.
[   175.993] (EE)
Fatal server error:
[   175.993] (EE) no screens found(EE)
[   175.993] (EE)
Please consult the Fedora Project support
	 at http://wiki.x.org
 for help.
[   175.993] (EE) Please also check the log file at "/var/log/Xorg.0.log" for additional information.
[   175.993] (EE)
[   176.031] (EE) Server terminated with error (1). Closing log file.
1

Here are the things I tried:

  • Add modprobe.blacklist=nouveau and rd.driver.blacklist=nouveau to the Linux command line.
  • Add blacklist nouveau to /etc/modprobe.d/blacklist.conf
  • Add nomodeset to command line, as it is said to make the NVIDIA driver fail.
  • Uninstall nouveau with dnf remove xorg-x11-drv-nouveau-1:1.0.13-1.fc25.x86_64
  • Rebuild the GRUB configuration with grub2-mkconfig -o /etc/grub2.conf
  • Rebuild the initramfs with dracut -f --kver `uname -r`
  • Reboot in 4.8 kernel
  • Reinstall NVIDIA driver 381.22 on top of 4.8 kernel

That last step apparently did the trick. FINALLY!!!!

Published projects

Tried to update both GitHub and GitLab with the projects I made some contributions to recently (BTRFS, Mesa, autoconf). It looks like there is absolutely no reaction to my autoconf patch (GNU Bug 25740).

vGPU video

The vGPU demo that was given at Red Hat Summit is here:

I learned several things by watching this video. One of them was the existence of the command watch.

Star Trek: Discovery Trailer?

Seen on Slashdot, but does not work for me? Not accessible from France, or taken down?

Found one that works, though:

Red Skin Cat

Daily rants and notes about my work on Spice at Red Hat

May 2017
Sun Mon Tue Wed Thu Fri Sat
1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28 29 30 31
Apr  Jun
RSS

Links

This is the work blog of Christophe de Dinechin for Red Hat. Think of it as a public lab book.

This should not be particularly interesting except to myself. It's really more intended as a set of personal notes, so there is no formatting whatsoever. I type tings as I go along during the day. All you'll find here are things like stack traces, Linux commands I want to remember or bugzilla links I want to be able to return to.

If you want to read stuff about my work at Red Hat that I believe is more interesting to the general public, you will find it on A Spice Odyssey. If you want to read my rants about everything else, go to Grenouille Bouillie. This is about as much advertisement as you will find on this site.

2016-2017 - Christophe de Dinechin

All views expressed on this blog are strictly personal. This blog does not represent the views of my employer, nor mine for that matter. This blog is provided as-is, without warrrany of any kind, express or implied, in particular regarding accuracy or interest of the content. The contents of this blog is distributed under a Creative Commons BY-SA 4.0 license .