ExternalVendorCode/genode
1
0
Fork 0
mirror of https://github.com/genodelabs/genode.git synced 2024-12-23 15:32:25 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
b30bacf86e
genode/doc/release_notes/23-05.txt
Norman Feske 66c3463749 Release notes for version 23.05
2023-05-31 11:10:25 +02:00

862 lines
41 KiB
Plaintext
Raw Blame History

===============================================
Release notes for the Genode OS Framework 23.05
===============================================
Genode Labs
Besides our annual documentation update, our major tool-chain update as
scheduled every two years, and the switch to C++20, version 23.05 puts the
spotlight on the Goa tool, which allows us to leverage existing SDKs like
Lomiri and Rust's cargo for Genode applications. In line with the previous
versions, DDE-Linux is prominently featured as enabler of our cross-platform
Wifi stack and the updated (6.1.20) drivers for Intel graphics and USB.
: <div class="visualClear"><!-- --></div>
: <p>
: <div style="clear: both; float: left; margin-right:20px;">
: <a class="internal-link" href="https://genode.org/documentation/genode-foundations-23-05.pdf">
: <img class="image-inline" src="https://genode.org/documentation/genode-foundations-title.png">
: </a>
: <a class="internal-link" href="https://genode.org/documentation/genode-platforms-23-05.pdf">
: <img class="image-inline" src="https://genode.org/documentation/genode-platforms-title.png">
: </a>
: </div>
: </p>
Before getting to the technical achievements, we'd like to draw your attention
to the books "Genode Foundations" and "Genode Platforms", which have been
updated to reflect the most recent state of the framework. Whereas the
"Foundations" cover Genode's architecture, developer work flows, and reference
material, the "Platforms" document is focused on low-level hardware topics and
provides plenty of practical guidance.
: <div class="visualClear"><!-- --></div>
Every two years, we update Genode's tool chain to the latest stable releases
of GCC and binutils. This time, we took the update as opportunity to switch
Genode's default from C++17 to C++20 so that modern C++ niceties can be used
for regular Genode components. The new tool chain is covered by
Section [New tool chain based on GCC 12.3, C++20 enabled by default].
For application developers, the evolving Goa tool is certainly the most
interesting feature of the current release. As detailed in
Section [Goa tool updated to Sculpt OS 23.04, initial support for Rust],
this tool enables us to reuse existing SDKs to target Genode. In particular,
we enabled the use of the Lomiri mobile UI toolkit (formerly known as Ubuntu
Touch UI toolkit) for targeting the PinePhone, and Rust's cargo.
System integrators may appreciate our continued development of the Linux
device-driver environment, which received an update to Linux 6.1.20
(Section [Device drivers]) and ultimately enabled us to use the same Wifi
stack across PC and ARM platforms
(Section [Uniform Wifi stack across PC and ARM platforms]).
Even though not end-user facing yet, two noteworthy development milestones of
the current release are the new use of our custom base-hw microkernel as x86
hypervisor (Section [Base-HW microkernel]) and the profound work on storage
encryption covered in
Section [Revision of Genode's custom block-encryption infrastructure].
Further topics making an appearance in version 23.05 range from RISC-V, over
WireGuard, VirtualBox, to seL4.
Goa tool updated to Sculpt OS 23.04, initial support for Rust
#############################################################
Last month, we [https://genode.org/news/sculpt-os-release-23.04 - released]
Sculpt OS 23.04 for PC and PinePhone. The new release comes with various
[https://genodians.org/nfeske/2023-05-11-sculpt-os - usability improvements]
such as presets and on-target system updates.
[image mobile_sculpt_23_04]
Interactive software management on the mobile variant of Sculpt OS
In particular, with Sculpt OS 23.04 running on the PinePhone, we have carved
out the base for hosting mobile apps on a Genode-based system. Yet, there are
only very few apps available right now. Since an OS is of no practical use
without apps, this urgently called for an SDK to simplify (mobile) app
development. After careful investigation, we opted for porting the
Ubuntu-Touch-UI toolkit to Genode and integrate it into Goa (Section
[Using Goa for bringing apps based on the Ubuntu-Touch-UI toolkit to Genode]),
our streamlined workflow tool for application development. In addition, we
integrated initial support for Rust's _cargo_ to make Goa palatable to a
broader developer audience (Section [Initial Rust support]).
The growing attention of the Goa tool prompted us to move it under the
[https://genodians.org/nfeske/2023-05-02-goa-genode-labs - umbrella of Genode Labs]
as we are increasing our development and maintenance efforts for the tool.
Aligned with the Sculpt release, the Goa tool has been updated with the
corresponding depot archive versions. With this Genode release, we put a
cherry on top and added bash completion to improve the user experience even
further. Having Goa installed, bash completion is enabled by the following
commands:
! goa update-goa master
! GOA_DIR=$(realpath $(which goa) | sed s#bin/goa##)
! echo "source ${GOA_DIR}/share/bash-completion/goa" >> ~/.bashrc
:Goa tool:
[https://github.com/genodelabs/goa/]
Using Goa for bringing apps based on the Ubuntu-Touch-UI toolkit to Genode
==========================================================================
While writing mobile apps might be fun, it is outside our core expertise.
Therefore, we have looked into ways of supporting established open-source SDKs
for app development on Genode. We investigated two possible options in depth,
namely Ubuntu Touch's UI toolkit now called [https://lomiri.com - Lomiri] and
the [https://docs.sailfishos.org/Tools/Sailfish_SDK - Sailfish SDK]. We have
tried to port applications for both stacks and after many iterations settled
with the Ubuntu UI toolkit. The full story can be read
[https://genodians.org/ssumpf/2023-05-06-ubunutu_ui - here]. Therefore, a port
of the Ubuntu UI toolkit is available on Genode right now and support for it
has been added to the Goa tool.
The workflow for crafting an app for the PinePhone using the Goa tool is a
fairly streamlined experience now:
# Since the UI toolkit depends on Qt5, add "genodelabs/api/qt5" to your
[https://genodians.org/nfeske/2019-11-25-goa - _used_apis_ file]
# Add "ssumpf/pkg/ubuntu_ui_toolkit" to your _archives_
[https://genodians.org/nfeske/2019-11-25-goa - file] to have the UI toolkit
available within your package
# In order to have your QML code within your packet installed, add
"<packet-name>.tar: install/" to your
[https://genodians.org/nfeske/2019-11-25-goa - _artifacts_ file]
# Configure your
[https://genodians.org/nfeske/2019-12-19-goa-unix-terminal - _runtime_ file]
# Execute your scenario on Linux for development
! goa run
# Build for the PinePhone
! goa build --arch arm_v8a
# [https://genodians.org/nfeske/2020-01-16-goa-publish - Publish] your package
! goa publish --depot-user john --depot-overwrite
Examples using QML, Qt5, and C++ can be found
[https://github.com/ssumpf/goa-projects - here]
Initial Rust support
====================
The Rust programming language has grown in popularity in the recent years.
The Genode OS Framework had support for the Rust programming language
before, contributed to Genode release 16.05 by Waylon Cude. However, as an
on-off contribution it never got traction and the support was eventually
removed with release 20.05.
While the original support focused on some low-level runtime libraries and
integration into the Genode build system, our new attempt has a somewhat
different objective, which is to facilitate the use of the existing Rust
ecosystem on the Genode OS Framework. The removal note already envisioned a
possible comeback using the Goa tool and Rust's cargo build system, for which
we have added initial support with this release.
Our objective led to the following guidelines for Rust integration:
# Make use of the native build system, cargo, to make the existing ecosystem
accessible.
# Aim for a seamless integration into the Genode OS Framework using the Goa
build tool.
# Instead of introducing our own Genode
[https://doc.rust-lang.org/nightly/rustc/platform-support.html - target triples],
leverage Genode's FreeBSD-based C library interface to use existing
supported standard library targets like 'x86_64-unknown-freebsd'.
# Strive to use the upstream tool chain, or at least stay as close to upstream
as possible.
While we largely succeeded in following these guidelines, our initial
proof-of-concept implementation relies on a marginally adapted tool chain to
work around missing support for versioned library symbols in our linker.
We are exploring avenues to overcome these limitations and expand the support
to cover more complex use cases in the next release.
To learn more about our Rust support, head over to the
[https://genodians.org/atopia/2023-05-30-bringing-rust-back-to-genode - article on Genodians.org].
Uniform Wifi stack across PC and ARM platforms
##############################################
Support for wireless LAN was mostly focused on the
[https://genode.org/documentation/release-notes/14.11#Intel_wireless_stack - PC platform]
as it was the platform predominately used for using Genode and, in extension,
Sculpt on a daily basis. In the last couple of years, however, we started to
embrace ARM-based platforms like the MNT Reform 2 and the PinePhone as well,
longing for thorough support of Sculpt OS on such systems. Thanks to our Linux
device-driver environment, we have now taken the opportunity to reuse the
existing wireless stack on vastly different platforms.
Making the wireless stack globally accessible
---------------------------------------------
The
[https://genode.org/documentation/release-notes/23.02#Realtek_Wifi - previous release]
already featured additional support for a different wireless LAN device driver -
the rtlwifi driver that supports Realtek-based devices - giving us a good
intuition on how easy it has become to extend even a complex Linux-based
driver component stack such as our wifi-driver component ('wifi_drv').
The first step was making it less x86-centric. We started by making the various
ingredients of the driver available on the ARM platforms.
On the one hand, that includes the WPA supplicant and its dependencies like
the 'nl80211' driver that in turn depends on 'libnl'. Enabling them was
straight-forward because they are already pretty platform independent and
the platform-dependent portions, e.g. libcrypto, are readily available for ARM.
On the other hand, the wireless stack was slightly more complicated because
the hardware integration of wireless networking devices on ARM platforms
varies from platform to platform. In case of the MNT Reform 2 and PC, the
integrated wireless devices are normally connected via PCIe. In contrast, the
PinePhone relies on SDIO. We separated the code to allow for a "mix-and-match"
way of selecting the necessary compilation units as the used Linux
configuration might differ between each target and could result in compilation
issues otherwise.
The next step was to make the wireless stack globally accessible by moving it
from the _pc_ to the _dde_linux_ repository. This move was motivated by the
fact that the _dde_linux_ repository is already available in all platform or
rather board-specific repositories while the _pc_ repository is not. It is
in itself a board-specific repository and therefore having it appear as
dependency for other such repositories feels unnatural.
So the bulk of the driver code now lives in the _dde_linux_ repository from
where it can be referenced by other repositories.
While moving the code, we noticed that in contrast to all other Linux-based
drivers the 'wifi_drv' is special. Since the binary itself is a libc component,
care was taken to isolate the application code, the 'wpa_supplicant', from
the driver code, the library containing the Linux wireless stack and drivers.
On all platforms, the binary stays the same while the driver library contains
all the platform-specific code. For this reason, the 'wifi_drv' binary is now
delegated to be a generic harness that includes all configuration and
management functionality shared by all wireless device driver components,
e.g., the WPA supplicant. The code of the device driver emulation environment
is located in _repos/dde_linux/src/lib/wifi_. It is referenced by the
platform-specific driver library that resides in the corresponding platform
repository. The runtime configuration needs to point the driver to a proper
driver library.
The platform-specific library is in charge of orchestrating the 3rd-party
sources utilized by the driver as well as providing the _source.list_ and
_dep.list_ files. It must include the generic library snippet
_repos/dde_linux/lib/wifi.inc_ that deals with managing the emulation
environment code. The amount of code added by the platform-specific libraries
is unimposing as it mostly consists of the dummy implementations needed by
the Linux configuration.
[image wifi_drv_architecture]
Composition of the wireless LAN driver component
All recipes for the depot archives are prefixed to the specific driver, for
example 'pkg/pc_wifi' contains a reference to 'src/pc_wifi_drv' as well as to
'raw/pc_wifi_firmware'.
Thanks to the steps outlined above, we now have three different wireless LAN
drivers, one for the PinePhone, one for the MNT Reform 2, and one for the PC
that nicely follow the same approach.
New firmware loading mechanism
------------------------------
Additionally to making it easier to enable and use the driver for new
platforms, we also refined how the driver loads its firmware images. In the
past, the driver contained a list of well-known working firmware images that
needed to be updated every now and then when new devices where enabled or the
firmware version changed due to a Linux update. In particular using the driver
with new devices was cumbersome as the driver itself already supported the
device most of the time, but it solely missed the corresponding entry in the
firmware list and adding that required recompiling the driver.
[image wifi_firmware_loading]
Firmware image loading sequence
So instead, the driver now loads the firmware images via its local VFS rather
than requesting a predetermined ROM module. Since the platform-specific driver
library has no direct access to the VFS - after all both worlds are
intentionally isolated from each other - a request/response interface was
added. The library submits a request to the _wifi_drv_ binary and will suspend
its execution waiting for the completion of the request. The binary will
acquire the firmware image and notify the driver library in return.
Streamlining the firmware acquisition in such a manner allows for using the
original probing mechanism available in Linux. Rather than following the
firmware list the actual driver code is now free to probe as it sees fit,
exactly pointing to the required uAPI revision in case the firmware is
missing.
The following snippet illustrates the configuration of the driver on the
PinePhone (omitting any integration-related routes for the config ROM as well
as state and scan reports):
!<start name="wifi_drv" caps="250" priority="-1">
! <resource name="RAM" quantum="32M"/>
! <config ld_verbose="yes">
! <report mac_address="true"/>
! <libc stdout="/dev/log" stderr="/dev/log"
! rtc="/dev/rtc" rng="/dev/urandom"/>
! <vfs>
! <dir name="dev"> <log/> <null/> <rtc/>
! <jitterentropy name="random"/>
! <jitterentropy name="urandom"/>
! <wifi/>
! </dir>
! <dir name="firmware">
! <tar name="wifi_firmware.tar"/>
! </dir>
! </vfs>
! </config>
! <route>
! <service name="ROM" label="wifi.lib.so">
! <parent label="a64_wifi.lib.so"/>
! </service>
! <service name="ROM" label="wifi_firmware.tar">
! <parent label="a64_wifi_firmware.tar"/>
! </service>
! <service name="ROM" label="dtb">
! <parent label="wifi-pinephone.dtb"/>
! </service>
! […]
! <any-services> <parent/> <any->child/> </service>
! </route>
!</start>
In this configuration, the firmware images are provided as a _.tar_ archive
that itself is requested via a ROM connection. The driver will always look
into the _/firmware_ directory to access any firmware related files. How the
directory is populated is up to the integrator of the driver.
As a further simplification step, we removed the need for the firmware library
used to contain firmware images. It is superseded by the use of a plain data
depot archive, e.g., _raw/pc_wifi_firmware_.
Additional device support and updates
-------------------------------------
We updated the firmware images to the most recent ones supported by
Linux version 6.1.20.
We enabled the ath9k PCIe driver that can be used on the MNT Reform 2 and the
PC. As the ath9k device (168c:0034) used to test the driver on the PC exhibited
problems when using MSIs, we disable their usage in the 'pci_decoder'. Similar
treatment might be necessary if other ath9k-based devices are used.
The device support in the 'rtlwifi' driver got extended by additionally
enabling support for RTL8192CE devices.
Furthermore, we updated the WPA supplicant to its latest v2.10 release and
introduce preliminary support for joining networks secured by WPA3.
Base framework and OS-level infrastructure
##########################################
New tool chain based on GCC 12.3, C++20 enabled by default
==========================================================
Following a regular cycle of two years, we updated our tool chain to recent
versions again, this time in particular to GCC 12.3.0, binutils 2.40, and GDB
13.1 while taking the opportunity to enable C++20 by default.
A noticeable change with GCC 12 is that auto-vectorization with the
'-ftree-vectorize' option is now enabled by default when building with the
'-O2' optimization level. This has the effect that more SIMD instructions are
generated, which required adaptations throughout our code base, for example by
making sure that memory allocations in ported Linux drivers adhere a suitable
address alignment and by saving and restoring ARMv8 FPU registers in the
dynamic linker.
In addition to that, GCC 12 reports new warnings and errors, which we had to
rectify at various places, the most common ones being:
* Deprecated arithmetics between different enumeration types,
* Deprecated use of '++' and '--' operators with volatile variables, and
* Undefined references to 'strlen' inside custom implementations
of 'strlen'-like functions, related to the
'-ftree-loop-distribute-patterns' option.
As an extra feature, we added Genode's library name patterns to the linker so
that the '-l' option has become able to find the corresponding libraries.
This is useful while porting 3rd-party software based on Autoconf, whenever a
'configure' script checks for a library dependency by linking a test program
with this option. This change thereby removes the need for dummy libraries
that were formerly used to satisfy the probing.
API changes
===========
As part of Genode's
[https://genode.org/documentation/release-notes/16.08#Cultivation_of_the_new_text-output_API - great API revision]
in 2016, we largely *abolished* the use of *format strings* throughout the
framework. This is desirable because a code base without format strings cannot
have format-string vulnerabilities. Still, a few occurrences, specifically the
interface for passing session-construction arguments, remained untouched since
then. With version 23.05, we finally attained our initial goal by wrapping up
the transition.
In particular, we revised 'Genode::Connection', which now accepts the session
label, affinity, and session-specific parameters as constructor arguments,
whereas the parameters are passed as a 'Genode::String'. This eliminates the
need for rendering a format string. Given this new interface, we were able to
remove format strings from all connection types, updated all components that
still happened to rely on format strings, and ultimately removed format
strings from Genode's base API.
Format strings still play a role to accommodate 3rd-party code ported
to Genode. Whenever the 3rd-party code targets the C runtime, format
strings are readily available via the libc. For free-standing ports that
avoid the dependency from the full C runtime, e.g., ported device drivers,
a new 'format' library based on Genode's former _base/snprintf.h_ and
_base/console.h_ provides rudimentary format-string support. The library
is hosted in the libports repository.
As another matter of housekeeping, we removed the _util/avl_string.h_ utility.
The use case of organizing objects by using strings as keys is covered by the
_util/dictionary.h_ now.
Towards kernel-agnostic DMA protection
======================================
As sketched in our [https://genode.org/about/road-map - road map], we plan
having a feature-complete PC version of Sculpt OS based on base-hw by the end
of this year. One of the reasons why we are still sticking to base-nova for
the PC version is the fact that we are relying on NOVA's IOMMU support. One
necessary step to decouple Sculpt OS from base-nova is to integrate the IOMMU
handling into the platform driver.
Motivated by our
[https://genode.org/documentation/release-notes/23.02#Custom_IP_block_for_DMA_protection_on_AMD_Xilinx_Zynq - custom IP block for DMA protection on AMD/Xilinx Zynq],
we integrated the notion of IOMMU-like devices into the platform driver with
this release as a preparatory step. The platform driver automatically acquires
known IOMMU-like devices for itself by looking at the device types. Other
devices can then reference these devices by using '<io_mmu>' nodes. This is
best illustrated by looking at the devices ROM for the Zynq's dma_guard IP
block:
! <devices>
!
! <device type="dma_guard" name="dma_guard_0">
! <!-- [...] -->
! </device>
!
! <device type="axi_dma" name="axi_dma_0">
! <io_mmu name="dma_guard_0"/>
! <!-- [...] -->
! </device>
!
! </devices>
This tells the platform driver that, whenever a DMA buffer is allocated/freed
for the session owning the 'axi_dma_0' device, the 'dma_guard_0' must be
configured accordingly in order to allow/deny access to the corresponding
memory ranges. With the structural changes to the platform driver, the support
for dma_guard devices is simply added by implementing specific 'Io_mmu' and
'Io_mmu_factory' objects. You can find the code in the _dma_guard.h_ within
the
[https://github.com/genodelabs/genode-zynq/blob/master/src/drivers/platform/zynq/dma_guard.h - genode-zynq repo].
For the PC version of the platform driver, we implemented a _kernel_iommu_
device that still uses device PDs to pass IOMMU configuration to the NOVA
kernel. The _kernel_iommu_ is automatically instantiated and used as a default
for each device until we replaced this by a kernel-agnostic implementation in
a future release.
With these preparations, we paved the way for implementing configuration logic
for arbitrary IOMMU-like devices within the platform driver. In particular,
the platform driver has been made capable of managing multiple IOMMU-like
devices at the same time. However, there is one limitation that comes from the
fact that DMA buffers are not device-specific but allocated per session: All
IOMMU-like devices must either operate as MMU (virtual addressing) or as MPU
(physical addressing).
Revision of Genode's custom block-encryption infrastructure
===========================================================
Tresor library
~~~~~~~~~~~~~~
For about two years, our Ada/SPARK-based CBE block encryption and its GUI
front-end, the file vault, served us well with rather manageable workloads
such as configuration and credential files in Sculpt OS on the PC. However,
with the rise of mobile Sculpt on the PinePhone, the CBE ecosystem was
suddenly confronted with new challenges and requirements.
First, mobile platforms are usually less forgiving when it comes to
performance and the CBE still exhibited a lot of potential for optimization.
Second, we envision encrypted storage to become an integral part of the base
system - the "appliance role" of mobile Sculpt OS - which shifts the role of
the component from an optional feature to a foundational mechanism. With this
role shift, however, maintainability becomes increasingly important. Third,
now that we decided to settle on this block-encryption approach and to
increasingly expose it to real workloads, we can expect new requirements to
pop up more frequently and with higher priority. Last but not least, our
Ada/SPARK runtime, so far, lacks ARM support.
This prospect forced us to carefully reconsider our relation to the existing
CBE approach, and especially to the fact that its core logic and crypto
back-end were entirely written in Ada/SPARK. When we started developing the
CBE in Ada/SPARK, we were positive that the language might become popular
among the core developers of Genode and that, eventually, other, especially
critical parts of the framework could benefit from it as well. But this idea
didn't come to fruition. Only a few of us came in touch with the new language
and, of those, even fewer acquired profound experience with it. We ultimately
realized that the friction caused by the added language boundary that emerged
with the CBE approach became a bottleneck, inhibiting the further evolution of
our block-encryption stack with a strong sense of collective code ownership.
This observation in mind, and the above-mentioned challenges in sight, we
decided to drop the CBE library and create a new implementation strongly
inspired by the CBE design but in C++, our "mother tongue". The new library is
called tresor, brings the same feature set as the CBE and is compatible with
containers created with the CBE. The file vault has been adapted to run with
the tresor library. So file-vault users can continue using their containers as
usual without further ado. The entire tresor-based ecosystem is
architecture-agnostic, which lifts the former restriction to x86.
File Vault
~~~~~~~~~~
Some new features have been added to the file vault. For instance, the
component can now be driven with one of two available user interfaces: The
usual graphical front-end or the new non-interactive interface that is driven
by a textual configuration and provides feedback through a report. This allows
for the integration of the file vault with automated controls respectively
lower or higher-level UIs. The interactive interface remains the default, but
one can replace it with the text-based variant using the new "user_interface"
configuration attribute. An example of operating the text-based interface is
provided by the new _file_vault_config_report.run_ script.
As another rather small but handy feature, a file vault can now be locked and
unlocked without having to restart the component. In the locked state, all key
material is removed from the cryptographic back end and the block-encryption
driver is shut down. The user is then prompted to provide the correct
credentials in order to re-establish access to the container.
Custom virtual machine monitor on ARM
=====================================
The
[https://genode.org/documentation/release-notes/23.02#Interactive_graphical_VMs_on_ARM - previous release], introduced interactive graphical VMs on ARM systems.
Genode's custom virtual machine monitor was enhanced by VirtIO device models
for input events and GPU. However, dynamic changes of the virtual GPU's
framebuffer resolution weren't yet handled by the initial version. With the
current release, these restrictions got removed. Now, the user is able to
resize the window of a virtual machine as expected.
NetBSD rump kernel on RISC-V
============================
We have added RISC-V to our port of the
[https://wiki.netbsd.org/rumpkernel - rump kernel].
This enables Genode to access commodity file-systems on RISC-V based devices.
Strengthened fault tolerance of on-target package management
============================================================
Genode's way of safely installing and deploying packages on-target - as
introduced in
[https://genode.org/documentation/release-notes/18.02#On-target_package_installation_and_deployment - version 18.02] -
is a corner stone of Sculpt OS. The recent move of Sculpt OS to mobile
devices, however, revealed a couple of limitations that we address with the
current release.
First, in contrast to the PC version of Sculpt OS that allows for the
straight-forward management and editing of files using a regular command-line
interface, a touch-based user interface as present on the phone is far more
constrained. Problems that can be solved by manual intervention on the PC
without second thought can become insurmountable showstoppers on the phone.
The most prominent problem is recovery from the situation where package
dependencies remain incomplete due to an interruption of the installation
process or due to packaging mistakes. On the PC, such a situation can be
resolved by simply clearing the depot using a single terminal command,
followed by a reinstall of the package. On the phone, the user was left out in
the cold with the message "package installed but incomplete" but with no
obvious or non-obvious way of recovery. The new version gracefully handles
this failure state by offering the retry of the package installation.
Second, network connectivity is far more fluctuating on mobile devices, which
increases the likelihood for download errors. The previous version that
regarded download errors as rare and sporadic issues, responded to such errors
by repeated and silent retries. We found that a mobile phone demands a more
graceful way to reflect such failure situations to the user, and to limit the
rate of futile download attempts. The new version preserves information about
download failures for user inspection and re-issues new downloads only if not
already flagged as unavailable.
Finally, we encountered the manual addition of software providers to the
system as a hurdle on the phone. On the PC, a new software provider can be
added by manually placing the provider's _download_ and _pubkey_ files in a
local depot directory, which is straight-forward when using a shell. However,
on a touch-screen device, there is no obvious and simple way to supplement the
system with such information. To still accommodate the user's desire to
download and install software from arbitrary providers, we added the option to
explicitly skip the signature verification for downloads. This is useful in
scenarios where the lack of integrity of downloaded content does not pose a
risk, e.g., for untrusted applications that are rigidly sandboxed, or during
development.
Whenever the depot-download subsystem encounters the attribute 'verify="no"'
for an '<installation>' item, it accepts the installation even if no key is
available. It still applies verification for dependencies whenever possible.
E.g., if a package of the provider "john" gets installed via 'verify="no"' and
the package depends on an archive by "genodelabs", for which the public key is
known, the integrity of the content originating from "genodelabs" is verified.
Libraries and applications
##########################
Qt5 reorganization
==================
When the Goa tool is used to build an application, all libraries of the used
API packages get linked to the application and the single Qt5 API package with
big libraries like QtWebEngine was a bit too much for simple Qt applications.
For this reason, we split the Qt5 API into smaller packages according to the
corresponding Qt modules.
As preparation for the release of a binary version of the Qt5 host tools, we
also reduced the external dependencies of these tools for improved
compatibility with different host systems and changed their install location
to the location of the other Genode host tools.
And finally, we added a 'ubuntu-ui-toolkit' meta package in the genode-world
repository which pulls in all dependencies for the Ubuntu UI toolkit,
including a runtime with the required ROMs.
WireGuard improvements
======================
There are two smaller changes related to Genode's port of WireGuard. First,
peers can now be removed from WireGuard at runtime by removing the
corresponding '<peer>' tags from the component's configuration. This operation
enforces the same assurances as removing a peer from a native WireGuard driver
in Linux.
The second change has to do with the nature of the port. The WireGuard port is
one of the rare examples where we use our Linux device driver environment
(dde_linux) for porting software that is not exactly a driver. The component
does not depend on a specific hardware configuration and therefore, the
emulated Linux kernel can be platform-agnostic. Consequently, while porting,
we created such a variant of the Linux emulation specifically for WireGuard.
However, we realized that this variant can come in handy for ports of other
hardware-agnostic kernel parts (for instance, lxip) as well. Therefore, we now
cut it out of the WireGuard port in order to make it a self-contained version
of the 'lx_emul' library. The new library is called 'virt_lx_emul' and is
accompanied by the 'virt_linux' target that can be used to build the
corresponding Linux kernel and run it in Qemu.
Updated or removed 3rd-party software
=====================================
VirtualBox updated to version 6.1.44
------------------------------------
Our port of VirtualBox underwent some maintenance work published in this
release. With the tool chain updated to GCC 12, it became necessary to update
VirtualBox to version 6.1.44 to keep up with the tool-chain changes and fix
many upstream bugs alongside. Also, we improved several aspects of the port to
improve robustness of networking, USB, multi-threading, and VM reboot. After
thorough testing in every-day scenarios, we finally adopted the handling of
the x86 time-stamp counter from version 5 and disabled the VM exit for the
RDTSC instruction, which improves the performance of selected scenarios
significantly. For Windows guests, it has become crucial to configure the
paravirtualization provider like follows in the _machine.vbox6_ file.
Otherwise, the guest's TSC calibration fails resulting in a bogus CPU
frequency assumption.
! <Paravirt provider="HyperV"/>
Removed ports of pcre16 and icu libraries
-----------------------------------------
The pcre16 and icu libraries had been used by Qt5 in the past but were not
used anymore since the last Qt updates. So we removed them from the _libports_
repository.
Device drivers
##############
Linux device driver environment updated to Linux 6.1.20
=======================================================
According to [https://genode.org/about/road-map - our roadmap], the update of
Genode's Linux device driver environment (DDE) to a more recent 6.x Linux
version was planned for release 23.08. Now, we decided to tackle this update
with this version already.
Besides the Wireguard port to Genode, the following ported drivers use the
latest Linux kernel 6.1.20 version now:
* Zynq SD-card driver
* PCI Wifi driver for i.MX 8MQ
* all PC drivers (USB host, Wifi, Intel display)
Note that a few drivers are not listed above. The existing drivers for the
Allwinner and i.MX 8MQ SoC still use older 5.x Linux kernel versions as base.
However, the Linux device driver environment has been tweaked carefully to
support a range of Linux kernel versions from 5.11 till 6.1.20.
While doing the update work, we investigated a more sustainable link between
the Linux kernel drivers for USB and display drivers (DRM/KMS) on the one
hand, and the Genode API on the other. The outcome is explained in the next
two sections.
Intel display driver
====================
During the update of DDE Linux to the Linux 6.1.20 version, the dependency on
internal structures of the Intel framebuffer driver (intel_fbdev) became a
hassle. Although the update was successful finally, we decided to remove the
direct usage of intel_fbdev in our ported Intel display driver, in order to
ease future updates. Nevertheless, the functionality of intel_fbdev is
required to manage the framebuffer memory to provide a working Genode GUI
interface by the driver. For that, we investigated the use of the
[https://www.kernel.org/doc/html/v5.0/gpu/drm-kms.html - Linux DRM/KMS]
interface, specifically to allocate and manage so called
[https://www.kernel.org/doc/html/v5.0/gpu/drm-kms.html#dumb-buffer-objects - dumb buffer objects].
As described in the linked article, the dumb buffers are a standardized and
streamlined way to make early boot graphics possible driven by user-land
tools. We adjusted our port along the ioctl's of the dumb buffer functionality
to manage the framebuffer in our ported display driver.
USB
===
Connecting different USB clients to a USB host controller driver is a delicate
task. When using a port of a Linux kernel driver, it can quickly become
brittle because the USB driver API in the Linux kernel is complex and contains
some semantic dependencies, for instance regarding synchronization, which are
not always obvious. However, the Linux kernel offers a USB device I/O API to
the user-land that is used for instance by libusb. This API has to guard the
USB subsystem against wrong usage, and implements the necessary semantics
regarding synchronization and dynamic changes of clients and devices. In the
past, we repeatedly encountered corner-case issues, if clients or devices
vanished and appeared at a high rate. For the sake of robustness, we decided
to redesign our internal linking in between the Genode USB API and Linux to
use the user-level device I/O API of the latter. Moreover, we extended the
capacity of USB packets in-flight that can be handled by the controller in
parallel to 32, to enhance the throughput for some USB devices.
NVMe storage
============
Our custom NVMe driver received the following improvements. First we added
'host-memory-buffer' (HMB) support to the driver, which is a performance
optimization for NVMe devices that do not make use of a DRAM cache for its
operational data.
The amount of memory used for the HMB can be set by adding the 'max_hmb_size'
attribute in the '<config>' node of the driver. This value is checked against
the constraints imposed by the device. Should the value be less than the
minimal required amount of memory, it will not be used and a warning is
issued. On the other hand, if the specified value is larger than the preferred
amount of memory as favored by the device, it will be capped to the useful
amount instead.
Naturally, when using the HMB, the required RAM quota of the driver component
increases by that amount.
Second, we fixed a problem detecting the block size (LBA format) of a given
namespace. The lower 4 bits of the 'FLABS' register indicate which of the (up
to) 16 supported LBA formats is used by the namespace. However, instead of
only making use of those bits, the driver looked at the whole register that
also includes other information. This led to using the wrong index for reading
the LBA format and, on certain devices, rendered the driver unusable as the
assumed block size was detected wrong.
Audio-driver update
===================
We updated the audio driver for HDA devices ported from OpenBSD to version 7.3.
The functional changes are minimal, but the new version supports more recent
PC platforms and recognizes more codecs.
Platforms
#########
Base-HW microkernel
===================
Principle x86 virtualization support (on Qemu)
----------------------------------------------
This release brings limited support for AMD's Secure Virtual Machine (SVM)
vCPUs to Genode's custom base-hw microkernel. Supporting SVM is meant as an
intermediate step towards enabling advanced virtualization workloads using
VirtualBox on Intel VMX later this year. The approach allows us to craft the
kernel's virtualization infrastructure using Qemu - which is able to emulate
SVM in software - and cross-test our implementation against other hypervisors
in a tightly controlled setting. For reference, we used the time-tested Qemu
version 4.2 for this line of work.
Implementing principle vCPU support revealed a few points of friction between
base-hw's kernel interface, which had been designed for the needs of our
custom ARM VMM, and our kernel-agnostic VM interface on x86 that has been
carefully crafted to support a range of 3rd party hypervisors, but relies on
more logic in the kernel-specific VMM library to manage the vCPU state.
The current implementation is able to run several test VM workloads like the
artificial 'vmm_x86' test, our seoul VMM run scripts with Linux, and - of
course - Genode VMs on one vCPU. It has thereby reached an important stepping
stone towards our actual goal of hosting VirtualBox on Intel hardware.
Having shown that base-hw can support the generic x86 VM interface, we will
mature our implementation and may adapt our interface to make it a better fit
to base-hw's vCPU execution model in the future.
Boot-time RAM detection on the PinePhone
----------------------------------------
For the PinePhone, we implemented dynamic detection of the system RAM size by
parsing the values of the DRAM controller as programmed by U-Boot. This way, 2
and 3 GB models of the PinePhone are supported by Genode.
Updated seL4 microkernel
========================
With this release, we updated the support of the seL4 kernel from 9.0.1 to
12.1.0 for i.MX6 Sabrelite board and x86_64 PC. The support for 32-bit PC got
removed since it is unused, and the i.MX7 Sabrelite support got removed since
it is not supported by the new seL4 kernel anymore.
The updated seL4 kernel requires additional host tools installed, namely
CMake, Ninja and additional Python3 modules, jinja2, jschonschema, and pyfdt.
Depending on the distribution, the modules are available as distribution
package or need to be installed with the python pip3 tool.
Reference in New Issue View Git Blame Copy Permalink