Monday, 4 May 2020

Dual-GPU support: Launch on the discrete GPU automatically

*reality TV show deep voice guy*

In 2016, we added a way to launch apps on the discrete GPU.

*swoosh effects*

In 2019, we added a way for that to work with the NVidia drivers.

*explosions*

In 2020, we're adding a way for applications to launch automatically on the discrete GPU.

*fast cuts of loads of applications being launched and quiet*




Introducing the (badly-named-but-if-you-can-come-up-with-a-better-name-youre-ready-for-computers) “PrefersNonDefaultGPU” desktop entry key.

From the specifications website:
If true, the application prefers to be run on a more powerful discrete GPU if available, which we describe as “a GPU other than the default one” in this spec to avoid the need to define what a discrete GPU is and in which cases it might be considered more powerful than the default GPU. This key is only a hint and support might not be present depending on the implementation. 
And support for that key is coming to GNOME Shell soon.

TL;DR

Add “PrefersNonDefaultGPU=true” to your application's .desktop file if it can benefit from being run on a more powerful GPU.

We've also added a switcherooctl command to recent versions of switcheroo-control so you can launch your apps on the right GPU from your scripts and tweaks.

Wednesday, 1 April 2020

PAM testing using pam_wrapper and dbusmock

On the road to libfprint and fprintd 2.0, we've been fixing some long-standing bugs, including one that required porting our PAM module from dbus-glib to sd-bus, systemd's D-Bus library implementation.

As you can imagine, I have confidence in my ability to write bug-free code at the first attempt, but the foresight to know that this code will be buggy if it's not tested (and to know there's probably a bug in the tests if they run successfully the first time around). So we will have to test that PAM module, thoroughly, before and after the port.

Replacing fprintd

First, to make it easier to run and instrument, we needed to replace fprintd itself. For this, we used dbusmock, which is both a convenience Python library and way to write instrumentable D-Bus services, and wrote a template. There are a number of existing templates for a lot of session and system services, in case you want to test the integration of your code with NetworkManager, low-memory-monitor, or any number of other services.

We then used this to write tests for the command-line utilities, so we can both test our new template and test the command-line utilities themselves.

Replacing gdm

Now that we've got a way to replace fprintd and a physical fingerprint reader, we should write some tests for the (old) PAM module to replace sudo, gdm, or the login authentication services.

Co-workers Andreas Schneier and Jakub Hrozek worked on pam_wrapper, an LD_PRELOAD library to mock the PAM library, and Python helpers to write simple PAM services. This LWN article explains how to test PAM applications, and PAM modules.

After fixing a few bugs in pam_wrapper, and combining with the fprintd dbusmock work above, we could wrap and test the fprintd PAM module like it never was before.

Porting to sd-bus

Finally, porting the PAM module to sd-bus was pretty trivial, a loop of 1) writing tests that work against the old PAM module, 2) porting a section of the code (like the fingerprint reader enumeration, or the timeout support), and 3) testing against the new sd-bus based code. The result was no regressions that we could test for.

Conclusion

Both dbusmock, and pam_wrapper are useful tools in your arsenal to write tests, and given those (fairly) easy to use CIs in GNOME and FreeDesktop.org's GitLabs, it would be a shame not to.

You might also be interested in umockdev, to mock a number of device types, and mocklibc (which combined with dbusmock powers polkit's unattended CI)

Tuesday, 17 December 2019

GMemoryMonitor (low-memory-monitor, 2nd phase)

TL;DR

Use GMemoryMonitor in glib 2.63.3 and newer in your applications to lower overall memory usage, and detect low memory conditions.

low-memory-monitor

To start with, let's come back to low-memory-monitor, announced at the end of August.

It's not really a “low memory monitor”. I know, the name is deceiving, but it actually monitors memory pressure stalls, and how hard it is for the kernel to allocate memory when applications need it. The longer it takes to allocate memory, the longer the kernel takes to allocate it, usually because it needs to move memory around to make room for a big allocation, when an application starts up for example, or prepares an in-memory buffer for saving.

It is not a daemon that will kill programs on low memory. It's not a user-space out-of-memory killer, and does not take those policy decisions. It can however be configured to ask the kernel to do that. The kernel doesn't really know what it's doing though, and user-space isn't helping either, so best disable that for now...

As listed in low-memory-monitor's README (and in the announcement post), there were a number of similar projects around, but none that would offer everything we needed, eg.:
  • Has a D-Bus interface to propagate low memory conditions
  • Requires Linux 5.2's kernel memory pressure stalls information (Android's lowmemorykiller daemon has loads of code to get the same information from the kernel for older versions, and it really is quite a lot of code)
  • Written in a compiled language to save on startup/memory usage costs (around 500 lines of C code, as counted by sloccount)
  • Built-in policy, based upon values used in Android and Endless OS
 GMemoryMonitor

Next up, in our effort to limit memory usage, we'll need some help from applications. That's where GMemoryMonitor comes in. It's simple enough, listen to the low-memory-warning signal and free some image thumbnails, index caches, or dump some data to disk, when you receive a signal.

The signal also gives you a “warning level”, with 255 being when low-memory-monitor would trigger the kernel's OOM killer, and lower values different levels of “try to be a good citizen”.

The more astute amongst you will have noticed that low-memory-monitor runs as root, on the system bus, and wonder how those new fangled (5 years old today!) sandboxed applications would receive those signals. Fear not! Support for a portal version of GMemoryMonitor landed in xdg-desktop-portal on the same day as in glib. Everything tied together with installed tests that use the real xdg-desktop-portal to test the portal and unsandboxed versions.

How about an OOM killer?

By using memory pressure stall information, we receive information about the state of the kernel before getting into swapping that'd cause the machine to become unusable. This also means that, as our threshold for keeping everything ticking is low, if we were to kill high memory consumers, we'd get a butter smooth desktop, but, based on my personal experience, your browser and your mail client would take it in turns disappearing from your desktop in a way that you wouldn't even notice.

We'll definitely need to think about our next step in application state management, and changing our running applications paradigm.

Distributions should definitely disable the OOM killer for now, and possibly try their hands at upstream some systemd OOMPolicy and OOMScoreAdjust options for system daemons.

Conclusion

Creating low-memory-monitor was easy enough, getting everything else in place was decidedly more complicated. In addition to requiring changes to glib, xdg-desktop-portal and python-dbusmock, it also required a lot of work on the glib CI to save me from having to write integration tests in C that would have required a lot of scaffolding. So thanks to all involved in particular Philip Withnall for his patience reviewing my changes.

Friday, 13 December 2019

Dual-GPU support follow-up: NVIDIA driver support

If you remember, back in 2016, I did the work to get a “Launch on Discrete GPU” menu item added to application in gnome-shell.

This cycle I worked on adding support for the NVIDIA proprietary driver, so that the menu item shows up, and the right environment variables are used to launch applications on that device.

Tested with another unsupported device...


Behind the scenes

There were a number of problems with the old detection code in switcheroo-control:
- it required the graphics card to use vga_switcheroo in the kernel, which the NVIDIA driver didn't do
- it could support more than 2 GPUs
- and it didn't really actually know which GPU was going to be the “main” one

And, on top of all that, gnome-shell expected the Mesa OpenGL stack to be used, so it only knew the right environment variables to do that, and only for one secondary GPU.

So we've extended switcheroo-control and its API to do all this.

(As a side note, commenters asked me about the KDE support, and how it would integrate, and it turns out that KDE's code just checks for the presence of a file in /sys, which is only present when vga_switcheroo is used. So I would encourage KDE to adopt the switcheroo-control D-Bus API for this)

Closing

All this will be available in Fedora 32, using GNOME 3.36 and switcheroo-control 2.0. We might backport this to Fedora 31 after it's been tested, and if there is enough interest.

Wednesday, 21 August 2019

low-memory-monitor: new project announcement

I'll soon be flying to Greece for GUADEC but wanted to mention one of the things I worked on the past couple of weeks: the low-memory-monitor project is off the ground, though not production-ready.

low-memory-monitor, as its name implies, monitors the amount of free physical memory on the system and will shoot off signals to interested user-space applications, usually session managers, or sandboxing helpers, when that memory runs low, making it possible for applications to shrink their memory footprints before it's too late either to recover a usable system, or avoid taking a performance hit.

It's similar to Android's lowmemorykiller daemon, Facebook's oomd, Endless' psi-monitor, amongst others

Finally a GLib helper and a Flatpak portal are planned to make it easier for applications to use, with an API similar to iOS' or Android's.

Combined with work in Fedora to use zswap and remove the use of disk-backed swap, this should make most workstation uses more responsive and enjoyable.

Thursday, 8 August 2019

libfprint 1.0 (and fprintd 0.9.0)

After more than a year of work libfprint 1.0 has just been released!

It contains a lot of bug fixes for a number of different drivers, which would make it better for any stable or unstable release of your OS.

There was a small ABI break between versions 0.8.1 and 0.8.2, which means that any dependency (really just fprintd) will need to be recompiled. And it's good seeing as we also have a new fprintd release which also fixes a number of bugs.

Benjamin Berg will take over maintenance and development of libfprint with the goal of having a version 2 in the coming months that supports more types of fingerprint readers that cannot be supported with the current API.

From my side, the next step will be some much needed modernisation for fprintd, both in terms of code as well as in the way it interacts with users.

Saturday, 6 April 2019

Developer tool for i18n: “Pseudolocale”

While browsing for some internationalisation/localisation features, I found an interesting piece of functionality in Android's developer documentation. I'll quote it here:
A pseudolocale is a locale that is designed to simulate characteristics of languages that cause UI, layout, and other translation-related problems when an app is translated.
I've now implemented this for applications and libraries that use gettext, as an LD_PRELOAD library, available from this repository.


The current implementation can highlight a number of potential problems (paraphrasing the Android documentation again):
- String concatenation, which displays as one message split across 2 or more brackets.
- Hard-coded strings, which cannot be sent to translation, display as unaccented text in the pseudolocale to make them easy to notice.
- Right-to-left (RTL) problems such as elements not being mirrored.

Our old friend, Office Runner 


Testing brought some unexpected results :)