When compiling this code, clang complains reference to 'Sandbox' is
ambiguous. It can either mean ::Sandbox namespace from
os/src/lib/sandbox/types.h, or Genode::Sandbox declared in
repos/os/include/os/sandbox.h. Since the code refers to ::Sandbox::Child
and its already in ::Sandbox namespace we can just drop "Sandbox" and
refer to just Child, which makes clang happy.
Issue #3938
This path fixes a void cast used to silence unused return value warning.
Its a common pattern to use void cast to do that. The code uses void *
cast instead. It works for GCC, but clang complains about this.
Issue #3938
Clang is rather picky about this and prints the following warning when
compiling new_delete.cc:
error: function previously declared with an explicit exception
specification redeclared with an implicit exception specification
[-Werror,-Wimplicit-exception-spec-mismatch]
Issue #3938
This was discovered when building the code with clang instead of GCC. In
this setup the run/ping on base-hw/arm_v8a/virt_qemu would crash
on shutdown due to uncaught Deref_unconstructed_object exception thrown
for Genode::Reconstructible<Genode::Account<Genode::Ram_quota>>. The
specific instance throwing this exception was
Pd_session_component::_ram_account. My investigation exposed the
following problem:
1. The Pd_session_component has a _sliced_heap member backed by
_constrained_ram_alloc which in turn uses Pd_session_component itself
as its Ram_allocator.
2. When ~Pd_session_component is called it first destroys _ram_account,
followed by _signal_broker.
3. The signal broker holds a reference to
Pd_session_component::_sliced_heap as Signal_broker::_md_alloc.
4. The base-hw implementation of ~Signal_broker destroys some contexts
and does this by calling Genode::destroy on some slabs using the
_md_alloc (ref to Pd_session_component::_sliced_heap).
5. The Genode::Slab calls the Ram_allocator::free which ends up calling
Pd_session_component::free.
6. The Pd_session_component::free can among other things call replenish
method on Pd_session_component::_ram_account which has already been
freed at this point.
From my POV calling replenish at this point is basically an undefined
behavior. The Genode::Constructible holding the Genode::Account was
already detroyed at this point. GCC builds happen to somehow manage to
go through the -> operator call without raising any alarms, while clang
builds trip on the _check_constructed() call.
This fix moves the _ram_account a bit higher in class declaration to
ensure its destroyed after _sliced_heap. This seems like the simpliest
solution for this problem.
Fixes#3941
This prevents later file-descriptor shortage when opening files on
demand, which can't be reflected to the application in a sane manner.
The real fix is to open socket files not on libc level but on VFS level
only effectively consume one libc file descriptor for one socket.
* Fix GIC model to support priority and cpu target settings correspondingly
* Fix semantic of SGIR register for GICv2
* Minor GIC model IRQ state fix
* Introduce synchronization for VirtIO and GIC models
* Enable multiple CPUs in test run-script for ARMv8
Fix#3926
* Introduce hypervisor-stack per CPU
* Introduce host world context per CPU
* Mark EL2 translation table memory as inner shareable
* The VMID is not bound to a single VCPU, but to the Vm_session as a whole
* Set affinity of the VCPU accordingly
* Add VMPIDR to VM state
Ref #3926
Instead of calling core to run/pause a VCPU, go directly to the kernel.
Apart from the performance win, it would otherwise involve a more complex
protocol, when a VCPU on another core has to be removed from the scheduler.
Core's entrypoint handling those request runs on the boot-cpu only.
Ref #3926
To enable the interaction of a VMM with the kernel directly,
a hidden RPC gets introduced. It allows a kernel-specific
base-library implementation of the Vm_session::Client to request
a kernel-specific capability to address a VCPU, e.g., to
run/stop it.
Ref #3926
Now, the USB connection is established on backend initialization and
terminated on backend exit triggered by high-level libusb code.
Thanks to Peter for the patch.
- unlink shared memory files
- lower maximum number of socket pool sockets to reduce chance of file
descriptor exhaustion
- fix a build dependency which caused sporadic parallel build errors
Fixes#3910
With this commit, the alignment of anonymous 'mmap()' allocations can be
configured like this:
<config>
<libc>
<mmap align_log2="21"/>
</libc>
</config>
Fixes#3907
This plugin gives access to the Audio_out session by roughly
implementing a OSS pseudo-device. It merely wrapps the session and does
not provide any resampling or re-coding.
Fixes#3891.
In the same vein as the terminal and block I/O controls, the sound
controls are implemented via poperty files and match the OSS
API ([1] features a nice overview while [2] is v3 and [3] gives
in-depth information on the current v4.x API we eventually might want
to implement).
[1] https://wiki.freebsd.org/RyanBeasley/ioctlref/
[2] http://www.opensound.com/pguide/oss.pdf
[3] http://manuals.opensound.com/developer/
The controls currently implemented are the ones used by the cmus OSS
output plugin, which was the driving factor behind the implementation.
It uses the obsolete (v3) API and does not check if the requested
parameter was actually set, which should be done according to the
official OSS documentation.
At the moment it is not possible to set or rather change any
parameters. In case the requested setting differs from the parameters
of the underlying Audio_out session - in contrast to the suggestion in
the OSS manual - we do not silently adjust the parameters returned
to the callee but outright fail the I/O control operation.
The following list contains all currently handled I/O controls.
* SNDCTL_DSP_CHANNELS sets the number of channels. We return the
available channels here and return ENOTSUP if it differs from
the requested number of channels.
* SNDCTL_DSP_GETOSPACE returns amount of playback data that can
be written without blocking. For now it amounts the space left
in the Audio_out packet-stream.
* SNDCTL_DSP_POST forces playback to start. We do nothing and return
success.
* SNDCTL_DSP_RESET is supposed to reset the device when it is
active before any parameters are changed. We do nothing and return
success.
* SNDCTL_DSP_SAMPLESIZE sets the sample size. We return the
sample size of the underlying Audio_out session and return ENOTSUP
if it differs from the requested number of channels.
* SNDCTL_DSP_SETFRAGMENT sets the buffer size hint. We ignore the
hint and return success.
* SNDCTL_DSP_SPEED sets the samplerate. For now, we always return
the rate of the underlying Audio_out session and return ENOTSUP
if it differs from the requested one.
This commit serves as a starting point for further implementing the
OSS API by exploring more users, e.g. as VirtualBox/Qt5/SDL2 audio
backend or a more sophisticated progam like sndiod.
Issue #3891.
At least on some PIT-based platforms (x86_32 + pistachio/okl4/sel4), we run
into trouble with the reworked timeout framework that now proccesses all
pending timeouts before calling their handlers. This order change leads to a
higher rate of handling of short periodic timeouts in the timer driver which
can cause lower prioritized components to starve. Especially, if submitting
signals (from timer to client) isn't cheap (as is the case on qemu + pistachio
for example).
Issue #3884
The driver is faily simple and does not support fancy features like
TCP checksum offloading or vlan filtering, but it is fully capable of
running every Genode network based scenario I've tried. Its currently
known to work on virt_qemu arm platforms and x86_64.
Fix#3825
To simplify writing native VirtIO drivers for Genode add helper classes
representing VirtIO device and queue. The queue implementation should
be platform independant. The device abstraction however is closely tied
to the VirtIO transport being used (PCI/MMIO). Both PCI and MMIO
implementations expose the same public API so the actual driver logic
should be the same regardless of which transport is used.
Its also important to note that the PCI version of Virtio::Device
currently does not support MSI-X interrupts. Unfortunately my kowledge
about PCI bus is very limited and my main area of interest was to get
VirtIO drivers working on virt_qemu ARM/Aarch64 platform. As such all
the VirtIO drivers I plan to submit will work with PCI bus, but might
not use some extended capabilities.
Ref #3825
The VirtIO device configuration on Qemu is dynamic. The
order and presence of different command line switches affects
base address and interrupt assignment of each device. One could
probably hard-code the necessary switches and resulting XML ARM
platform driver configuration in each run script, but this seems
like troublesome and hard to maintain solution.
This patch explores an alternative approach to the problem.
It implements a ROM driver which probes the address space region
Qemu virt machines assign to VirtIO MMIO devices and exposes the
result as XML via a ROM session. This XML output can be fed directly
as config to the generic ARM platform driver.
Ref #3825
Right now the same code dealing with nic setup on qemu is duplicated
in many different run scripts. It makes it unnecesarily complex to
change the existing config or add support for new nic types. Lets move
all this common code to qemu.inc.
Ref #3825
- make GPIO server more robust on imx by not throwing exceptions for
unknown pins, use '_with_gpio' instead
- use 'Gpio::Pin' data type instead of POD 'unsigned'
issue #3900
The patch handles the case, that the memory for the MSI-X table is part
of one of the Pci::Resource Memory BARs, which got allocated beforehand already.
With this commit, the platform driver will not fall back to use legacy IRQs or MSI, whereby MSI-X is available actually. Additionally, this patch avoids a lot of red
messages about non available IO-MEM printed by the roottask.
Fixes#3904
The deadlock occured with three concurrently running threads: two
waiters calling pthread_cond_timedwait() and one signaller calling
pthread_cond_signal().
If waiter W1 hits its timeout, the signaller may have called
pthread_cond_signal(), detected this waiter and posted the internal
'signal_sem' concurrently. Then, the signaller waits for 'handshake_sem'
to ensure the waiter got woken up.
Waiter W1 can't consume the 'signal_sem' post by
'sem_wait(&c->signal_sem)' because another waiter W2 may have consumed
the post already above in sem_wait/timedwait(). Waiting for a post on
'signal_sem' would block the waiter W1 in perfect deadlock with
signaller on 'handshake_sem'. As W1 also owns 'counter_mutex' in this
situation, waiter W2 would block when trying to aquire 'counter_mutex'
and can't resolve the situation.
So, W1 does nothing in this case and we accept the spurious wakeup on
next pthread_cond_wait/timedwait().
* get rid of alarm abstraction
* get rid of Timeout::Time type
* get rid of pointer arguments
* get rid of _discard_timeout indirection
* get rid of 65th bit in stored time values
* get rid of Timeout_scheduler interface
* get rid of uninitialized deadlines
* get rid of default arguments
* get rid of Timeout::_periodic
* get rid of Timeout::Raw
* use list abstraction
* only one interface for timeout handlers
* rework locking scheme to be smp safe
* move all method definitions to CC file
* name mutexes more accurate
* fix when & how to set time-source timeout
* fix deadlocks
Fixes#3884
In case of contexts blocked in select() the monitor updates the
file-descriptor status, but if the entrypoint is just blocked for the
select handler, the status must be updated explicitly on
dispatch_select().
This patch fixes the corner case where the keyboard focus is defined
independently from user interactivity, e.g., the activation of a
screensaver or lock screen.
In this case, nitpicker would update its internal focus state not before
the next input event is handled. Should this input event be a press
event, this event would wrongly be delivered to the prior focused
session. Another problematic situation is the initial state before the
very first input event occurs. Since the focus remains undefined until
the first input event is handled, an initial key press event would not
be delivered.
This is a regression caused be the transition to the event-session
interface and the removal of the nitpicker's periodic way of operation.
The patch fixes the problem by applying pending focus changes not only
at the input processing but also on the code path that responds to focus
changes (e.g., focus-rom update).
Issue #3812
Adjust the base-* platforms to acknowledge new thread location solely if
migration is supported and succeeded. Otherwise the wrong thread
locations are observed via the trace session and utilization time calculation
get wrong.
Issue #3842
Instead of using the old 'ioctl' Vfs::File_io_services API implement
the I/O control functionality in a buch of files. This is similar to
the terminal-VFS plugin.
Fixes#3889.
Like already done for terminal I/O controls use collect the information
by reading property files instead of using the old VFS ioctl interface.
Fixes#3888.
There is a type mismatch as in the FreeBSD contrib code the type of the
request is 'unsigned long'. So far, only I/O controls where the request
falls into the signed range where used and this was not a problem.
Some of the SNDCTL requests, however, have the bit set.
Fixes#3887.
This patch is a follow-up for the commit "libc: use monitor for fork".
It removes the use of the monitor mechanism from the
'Local_clone_service::close' RPC function because the fork_ep must stay
responsive for the destruction and creation of 'Child' objects.
Issue #3874
The current version of the driver leaves the screen in a black state
after a mode change, e.g., when connecting an external display.
With this patch, the framebuffer content is fetched for the entire
screen after a mode change. Thanks to Sebastian Sumpf for investigating.
Issue #3878
- Show "..." for resource-assignment menu entry because it leads
to a sub menu.
- Avoid dialog changes when clicking on "Add component" while a
routing or resource dialog is unfolded.
The initial implementation of the affinity configuration in
"depot_deploy: support affinity configuration" added the affinity
location attributes to the <start> node of the deploy config. This patch
moves the information into a dedicated <affinity> sub node as done by
the init configuration. So the context of the attributes 'xpos', 'ypos',
'width' and 'height' becomes clear.
It also fixes a usability issue in Sculpt that occurred during testing:
When configuring multiple components with custom affinities, the
resources dialog of later components would wrongly display the state of
earlier components instead of displaying the fresh (default) state. The
resulting configuration would then not match the displayed information.
This is fixed by resetting the dialog state.
As another minor cosmetic change, the patch adds a line break in front
of copied <config> or <heartbeat> nodes.
Issue #3597
The combination of Net::Mac_address and
Genode::ascii_to(Net::Mac_address) required shaky quirks in several
places because GCC is not able to resolve the ascii_to overload if
base/xml_node.h was included to early. The current solution moves the
several ascii_to overloads "closer" to the Net types by putting them
into the Net namespace, where GCC reliably picks them up.
Hence, co-locating the ascii_to() utility with the overload type in the
same scope/namespace is good practice.
This patch removes the now obsolete <nic/xml_node.h> header file.
With the move of the storage-management dialogs to the graph in Sculpt
version 20.02, the ability to unselect the buttons for destructive
storage operations such as format or expand disappeared. This patch
restores the original behavior.
* On this platform there is no need to zero out the framebuffer
dataspace, which is already blanked by core
* But it might happen that the GUI server is sending a capture
event before the zero out happens. Thereby screen content can get
lost
Fix#3878
* Fixes faulty algorithm in fractional pll rate calculation
* Enables clocks that are set as reference clock to prevent system freeze
* Enables/disables root clocks of gates implicitely
Fix#3876
By default, bash brings its own version of 'getenv', named exactly like
the libc function. This becomes a problem in fork/execve scenarios if
the dynamic linker resolves 'getenv' to the bash binary instead of the
libc.
This patch fixes the generate step of the genodians.org scenario.
Issue #3882
This patch enables the menu view's new ability to respond to font-size
changes in Sculpt so that the menu view instanced no longer need to be
restarted whenever the screen resolution or font size is changed.
Fixes#3875
This patch allows the use of the VFS watch mechanism for the glyph file
of the TTF VFS plugin so that clients become able to dynamically respond
to font reconfigurations.
Issue #3875
This is needed to enable VFS plugins to notify VFS clients about file
changes that depend on the plugin configuration, E.g., whenever the vfs_ttf
plugin responds to a font-size change, it generates a watch notification
for the glyphs file. Since the change is independent from I/O, we need
to manually call 'handle_io_progress'.
By default, Vim renames a written file to a backup file suffixed with
"~" before writing the current buffer to a new file. Consequently, there
exists an intermediate state when no file exists. Should a client watch
such a file to obtain dynamic configuration info, it observes the empty
state.
Some components have builtin heuristics for such a situation. In
particular, the window layouter falls back to a predefined default
'rules' if no rules are provided as a file. So when interactively
editing window-layouter rules using Vim, it can happen that the manually
maintained rules get overwritten by the default rules.
By setting 'set nowritebackup', we can sidestep this issue by preventing
Vim from producing the bad intermediate state.
With the change of nitpicker to the event session interface, the
formerly periodic hovering updates moved to the - now sporadic - input
processing. This has the unfortunate side effect that hovering changes
caused by non-user-input, in particular view-stack changes issued by the
GUI clients that change the view under the current pointer position,
would no longer be reported immediately but only after receiving the
next incoming input event.
This patch reworks the hover handling such that potential hovering
changes due to view-stack operations are evaluated immediately by
those operations, covering the update of the hover report and the
generation of artificial enter/motion events.
Issue #3812
This patch solves a corner case where one long-active job (e.g.,
read-ready request) stays at the beginning of the '_active_jobs' queue
without an ack. In this case, the '_try_acknowledge_jobs' method would
wrongly stop processing the subsequent acknowledgements. In practice,
this can lead to a delayed sending of acknowledgements until new I/O or
client requests occur. In particular, Vim in Sculpt's inspect window
sometimes did not immediately respond to key presses during tab
completion. Here, the read-ready request of the terminal prevented the
acknowledgement for read of directory entry from being delivered until
the next key was pressed.
Fixes#3873
The quota for the argument buffer is already accounted by using the
Attached_ram_dataspace _argument_buffer, which uses the Constraint_ram_allocator
_ram, which uses the Ram_quota_guard from the Session_object. Running on
Sculpt with more than 1000 Subject_info objects/trace IDs the memory
waste become noticeable.
Follow-up commit to fix the old usb_drv. Under rpi one and the
same irq can be requested several times. Therefore, we've to track
the Irq_connection objects.
Ref #3865
The Lx_kit::Irq abstraction of DDE Linux was always using the very first
interrupt of a Platform::Device stored in it. Instead of handing over the
platform device it is much more flexible and sane to hand over the actual
interrupt capability.
Ref #3865
By now, the enumeration of peripheral interrupts on Raspberry Pi 1 was
different in between base-hw kernel and Fiasco.OC. Therefore, hacks were
needed in every driver to request the correct interrupt number dependent
on the kernel. Before reproducing the same in the platform driver for rpi,
we can more easily use the same enumeration with base-hw.
Ref #3864
This is a temporary workaround to not harm drivers, which aren't
converted yet to request all device resources including power and
clocks from this new platform driver.
Ref #3863
To access the ARM Trusted Firmware from the platform driver
fill the new `managing_system` call of the PD session with life resp.
do a SMC call on behalf of the client.
Fix#3816
Introduce the managing_system privilege for components like the
platform_driver to allow it to call system management functionality
that is reserved by kernel or special firmware, e.g., ARM Trusted Firmware.
The former RAM resource configuration attribute `constrain_phys`,
which enabled to constrain the region of physical RAM to be used,
gets replaced by the new, broader managing_system configuration
attribute of a `start` node. It gets enforced by the sandbox library.
Ref #3816
Normally CLOCK_REALTIME is used. However libraries, like glib, want to
use CLOCK_MONOTONIC. To make those users happy add setting the clock.
Note, the pthread_cond implementation uses the POSIX semaphore API
internally that does not have means to set the clock. For this reason,
the private 'sem_set_clock' function is introduced.
Fixes#3846.
When the 'raw' attribute is set to 'yes' the terminal VFS plugin will
ignore control characters. This in necessary for terminal connections
that transport data that contain such characters as part of message.
Fixes#3860.
The driver wrongly rejected a block request for the very last block of
the device, which prevented part_block from successfully parsing the
partition table (when attempting to access the GPT backup).
Fixes#3861
With this patch, sculpt uses init's heartbeat-monitoring mechanism to
detect the failure of part_block instances during storage disovery.
If part_block gets stuck, the device is released and can thereby
be accessed at the whole-device level.
Issue #3861
The patches disable VBOX_IGNORE_FLUSH and a sanity check in the VMDK
backend. This enables passing an explicit flush request by the guest
down to the VFS.
Fixes#3743.
The check prevents the Ttf_font from violating the bounding box in the
presence of very small scale values. This can happen during the startup
of Sculpt. Before the framebuffer driver is up, Sculpt bases its dynamic
font-size setting on a screen resolution of 1x1.
Issue #3812
* add libsparkcrypto source-recipe
* provide ALI files through a new repository by mstein and add the repo
download to the libsparcrypto port-file
* remove dependencies to non-existent contrib ADB files from the library make
files
Fixes: #3852
The sculpt manager used to defer the initialization of the GUI until
nitpicker's first display report became ready. This way, Sculpt was able
to run headlessly even if the framebuffer driver failed to start up.
Thanks to #3827, nitpicker no longer depends on a working framebuffer
driver. So the sculpt manager is safe to rely on nitpicker in any case,
simplifying the code.
Issue #3827
This is a follow-up commit to "nitpicker: make framebuffer and input
optional". It restores the dynamic mode-change support when using
'request_framebuffer="yes"' as needed in scenarios where multiple
nitpicker instances are used in a cascaded way. E.g., Sculpt's
Leitzentrale. The previous version missed to reconstruct the
'_fb_screen' on mode changes.
Issue #3812
This patch untangles the dependency of VFS operations that need RTC
information from the 'clock_gettime' libc function that must never be
called from the libc kernel context.
- The 'Rtc' class uses the VFS directly for reading the rtc file instead
of relying on libc functions.
- The 'Rtc' instance has become part of the 'Kernel' instead of
being construced as a side effect of the first call of
'clock_gettime'.
- Changed 'Rtc::read' to return a timespec value, which has a higher
precision than the formerly used time_t value.
- The 'Rtc::read' returns a value with the relative 'current_time'
already applied. The former handling of subsequent rtc-value
updates has been rewritten to become more logical.
- The 'Vfs_plugin' no longer calls 'clock_gettime' but the new
kernel-level 'Current_real_time' interface.
Issue #2635
This patch untangles the interplay of the base library and the libc
during the exit handling.
- The CXA ABI for the atexit handling is now provided by the libc.
For plain Genode components without libc dependency, __cxa_atexit
is a no-op, which is consistent with Genode's notion of components.
- The 'abort' implementation of the base library no longer calls
'genode_exit' but merely 'sleep_forever'. This way, the cxx library
no longer depends on a 'genode_exit' implementation.
- The libc provides 'atexit' support by storing metadata on the
libc kernel's heap now, thereby eliminating the former bounded
maximum number of atexit handlers.
- Shared-library dtors are no longer called via the atexit mechanism
by explicitly by the dynamic linker. This slightly changes the
call order of destructors (adjustment of the ldso test). Functions
marked as destructors are called after the atexit handlers now.
- The libc executes atexit handlers in the application context,
which supports the I/O operations in those handles, in particular
the closing of file descriptors.
Fixes#3851
For the time being, we put all spec and body files into the recipe although
there might be a (so far to us unknown) way to have only the API relevant files
in it.
Fixes#3849
* switch to a libsparkcrypto fork of m-stein
* switch to a libsparkcrypto state that makes some units pure that are used by
the consistent block encrypter, in order that the latter can become
completely pure
Ref #3849
* switch to fork of the Ada runtime provided by m-stein
* switch to a Ada-runtime state that provides the Exp_Int package (exponential
function on integers)
* adapt spark lib, symbols, and recipes to incorporate the Exp_Int package
Fixes#3848
