The gpu-manager should allow RandR offloading without bbswitch
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
nvidia-settings (Ubuntu) |
Triaged
|
Medium
|
Alberto Milone | ||
Trusty |
Triaged
|
Medium
|
Alberto Milone | ||
ubuntu-drivers-common (Ubuntu) |
Fix Released
|
Medium
|
Alberto Milone | ||
Trusty |
Fix Released
|
Medium
|
Alberto Milone |
Bug Description
SRU Request
While bbswitch may fail on some hybrid systems (e.g. on the desktop or when the bbswitch module fails to build), this is not a good reason to prevent the system from offloading rendering to the NVIDIA GPU.
[Impact]
* This problem makes it impossible to use the NVIDIA GPU on hybrid systems with Intel+NVIDIA GPUs when bbswitch is not available (i.e. when it fails to load).
[Test Case]
* Make sure to be using a hybrid system with Intel+NVIDIA GPUs.
* Check that bbswitch is not available. You can do so by posting output of the following command:
test -e /proc/acpi/bbswitch && echo true || echo false
* Make sure that the gpu-manager is not disabled (only necessary if you disabled it manually).
* Install ubuntu-
* Remove all fglrx and nvidia drivers (keep the nvidia-common and the nvidia-prime packages):
sudo apt-get --purge remove nvidia-331
sudo apt-get --purge remove nvidia-331
sudo apt-get --purge remove fglrx
sudo apt-get --purge remove fglrx-updates
* Restart the system.
* Install the nvidia binary driver, reboot, and check that the binary driver is enabled (attach your /var/log/
- Expected: the NVIDIA discrete GPU is enabled.
- Bad behavior: the system switches back to the intel driver and the discrete GPU is not used, despite the fact that the system was configured properly.
[Regression Potential]
* Low. Systems that currently work will keep working as usual, the ones that currently fail should finally work.
[Other Info]
* N/A
-----------
The gpu-manager should allow RandR offloading without bbswitch. This would allow desktop systems to make use of offloading even when bbswitch is not supported.
Also, nvidia-settings should not complain in such scenarios. See https:/
Changed in ubuntu-drivers-common (Ubuntu): | |
status: | New → Triaged |
importance: | Undecided → Medium |
assignee: | nobody → Alberto Milone (albertomilone) |
Changed in nvidia-settings (Ubuntu): | |
importance: | Undecided → Medium |
status: | New → Triaged |
assignee: | nobody → Alberto Milone (albertomilone) |
description: | updated |
description: | updated |
Changed in ubuntu-drivers-common (Ubuntu Trusty): | |
status: | New → In Progress |
importance: | Undecided → Medium |
assignee: | nobody → Alberto Milone (albertomilone) |
Changed in nvidia-settings (Ubuntu Trusty): | |
status: | New → In Progress |
status: | In Progress → Triaged |
importance: | Undecided → Medium |
assignee: | nobody → Alberto Milone (albertomilone) |
description: | updated |
This bug was fixed in the package ubuntu- drivers- common - 1:0.2.98
--------------- drivers- common (1:0.2.98) utopic; urgency=medium
ubuntu-
* gpu-manager.c, gpu-manager.py:
- Add support for testing core alternatives. This will be needed
by fglrx.
- Add the --backup-log parameter. This is only used for debugging
and disabled by default.
- Fix a regression that caused the gpu-manager to switch to mesa
after enabling fglrx on hybrid systems (LP: #1310489). Also
add a test case so that we don't regress in the future.
- Check that either fglrx or nvidia is set in xorg.conf when
xorg.conf is needed.
- Check if kernel modules are blacklisted before choosing the
driver (LP: #1310489).
- Enable only GPUs with open drivers that have connected outputs.
Attaching to GPUs that have no connected outputs would result
in a black screen.
- Allow RandR offloading even without bbswitch (LP: #1349282).
- Fall back to mesa when failing to enable prime.
- Do not try to enable prime if the nvidia driver is older than
331.
- Do not abort if the settings for prime cannot be found.
Try creating the file before failing.
- Avoid false positives of unloaded modules.
- Use the cleanup attribute.
- Always use snprintf instead of sprintf.
-- Alberto Milone <email address hidden> Wed, 27 Aug 2014 10:39:17 +0200