Switch to kernel branch, that determines the available system memory during
boot time. The overall kernel memory is still static, but during boot time
dynamically the amount can be chosen. Following 3 config option exists:
CONFIG_MEMORY_BOOT is the amount of kernel memory allocated in the BSS
statically - effectively chosen during link time - see linker script.
CONFIG_MEMORY_DYN_MIN && CONFIG_MEMORY_DYN_PER_MILL configures the dynamic
part of the kernel memory allocation applied during early kernel boot time.
CONFIG_MEMORY_DYN_MIN is the amount of memory which should be allocated at
least. CONFIG_MEMORY_DYN_PER_MILL defines the amount of the system memory in
per mill which should be allocated at most. The overall maximum kernel memory
is restricted to ~1G (64bit), due to the chosen internal virtual memory layout.
Fixes#2985
On some machines, both the AHCI and NVMe drivers are started within the
[drivers -> dynamic] subsystem. This ultimately exhausted the RAM quota
of this subsystem.
Issue #2987
This patch introduces the distinction of the manually managed
config/deploy from the managed config/managed/deploy. The latter
incorporates interactive changes of the system by the user. There are
two user interactions supported.
First, by clicking on the '+' button at the top-left of the runtime
view, the user can select a component to launch. All launchers at
config/launcher/ are listed in the popup menu. Each launcher can be
lauched only once. While running, is not available in the popup
menu.
Second, when selecting a node that corresponds to a start node in
config/deploy or that was interactively launched, the detailed view
shows a 'remove' button, which can be used to exclude the component
from the deployment.
The result of the interactive manipulation is always available at
config/managed/deploy. Hence, the current situation can be made
persistent by using it as config/deploy.
Fixes#2986
The tasklet_schedule implementation was missing the unblock call, which
was no problem in the past because the correspending task would get un-
blocked by other work items. Now that every workqueue has its own task,
there is nobody left to unblock the task and thereby execute the
tasklet. As a tasklet is the only way to transmit any pending frames
after transmission was temporarily suspended - because the device/fw
was not able to process the TX request - this rendered 6xxx devices
unusable whenever this situation occurred.
Fixes#2973.
The Arora main thread sometimes blocks on a pthread condition variable,
which prevents Genode signal processing with the current implementation.
This is especially a problem when the thread who could unblock the main
thread calls 'Libc::suspend()'.
As a workaround until the pthread locking mechanisms get adapted to the
Genode libc execution model, the Arora main function can be called from
a dedicated thread.
Fixes#2978
Close the stdin, stdout, and stderr I/O channels at child exit. This
serves to flush buffers at the I/O resources which might not be written
otherwise.
Ref #2919
The VFS LwIP plugin is page-faulting on connect error because the LwIP
library frees a failed TCP protocol control block before calling the
error callback, and then the VFS plugin dereferences the PCB to free it
a second time. This problem was caused by a failure to follow
documentation during a transition from a C callback to a C++ method.
Fix#2972
1) The loop for determining the line length read from a character offset
before checking whether the offset is smaller than the given string
length. This could have caused access outside the string buffer.
2) The routine for determining the line length first seeked for the
offset of the last real character of the line and than added one for
getting the length but only if the following character was '\n'. This
has to be done for any other line-terminating character too. The only
case where you don't want to do this is when the end of the whole
string is reached.
Issue #2967
When 'notify_read_ready()' is called during 'select()' and fails,
suspending can cause a deadlock when the libc IO response handler becomes
active and calls 'select_notify()', which tries to acquire the
'select callback list lock', which is already acquired by the suspended
'select()' call.
It seems possible to ignore a failed 'notify_read_ready()' call instead of
suspending. When the VFS plugin calls the IO handler later when the
notification request can be processed, the 'select_notify()' call of the
libc IO response handler will eventually call 'notify_read_ready()' again.
Fixes#2970
Implement the passwd database subroutines with single-user database.
This database is populated with a <passwd/> sub-node of the libc
configuration node. All fields of the "passwd" struct may be specified
with reasonable defaults provided for a "root" user. This allows a
libc-based component to spoof user information for the sake of porting
existing Unix software.
A test is provided at run/libc_getpwent.
Fix#2919
GNU tar interprets the passed time value in local time by default, which
lead to unstable hashes again. So, now provide a complete date/time
value incl. time zone.
Related to #2842
The VFS server does not support file-system with one single-file plugin
providing the root. This is because the "leaf_path" is not universally
implemented to handle the path "/". This fix is simply to skip a
"leaf_path" check when opening the path "/".
Ref #2919
