The requested guest-physical memory range may comprise multiple attached
dataspace regions, which must all be detached. This is not required for
the current vbox5 implementation, but for vbox6 as the current API
suggests these semantics.
This commit can be seen as intermediate fix as a real fix should change
the API to prevent long-running detach loops in core that may lock out
requests by other components.
* Only give managing_system permission when all parent nodes of the
corresponding component agree in doing so.
* Move the physical memory constrains heuristic from sandbox library to core
Fix#4335
This patch changes the 'Allocator' interface to the use of 'Attempt'
return values instead of using exceptions for propagating errors.
To largely uphold compatibility with components using the original
exception-based interface - in particluar use cases where an 'Allocator'
is passed to the 'new' operator - the traditional 'alloc' is still
supported. But it existes merely as a wrapper around the new
'try_alloc'.
Issue #4324
This patch replaces the 'Ram_allocator::alloc' RPC function by a
'try_alloc' function, which reflects errors as 'Attempt' return value
instead of an exception.
Issue #4322
Issue #3612
When rebasing my local branch on top of sculpt-21.10 tag I've noticed
two problems.
The code in new_delete.cc does not include new header file. This works
fine with GCC, but fails with clang because std::align_val_t type is
not defined anywhere according to clang. It looks like GCC pulls this
header indirectly somehow.
The second problem can be seen if one disallows undefined symbols in
executables and shared_libraries. This can be seen with both GCC and
clang by adding --no-undefined to LD_OPT. With such change in place core
fails to link due to:
ld.lld: error: undefined symbol: operator delete(void*, std::align_val_t)
>>> referenced by thread.h:448 (/home/tworaz/devel/genode/repos/base-hw/src/core/kernel/thread.h:448)
>>> thread.o:(Kernel::Core_main_thread::~Core_main_thread()) in archive debug/core-hw-virt_qemu.a
>>> referenced by thread.h:448 (/home/tworaz/devel/genode/repos/base-hw/src/core/kernel/thread.h:448)
>>> thread.o:(non-virtual thunk to Kernel::Core_main_thread::~Core_main_thread()) in archive debug/core-hw-virt_qemu.a
>>> did you mean: operator delete(void*, unsigned long, std::align_val_t)
>>> defined in: debug/core-hw-virt_qemu.a(supc++.o)
If the code would somehow manage call such undefined symbol it'd crash.
Since I generally prefer link time failures to runtime crashes I link
all genode binaries with --no-undefined.
To fix this problem just add a dummy implementation of missing delete
operator.
Fixes#4298
The new 'Env::try_session' method mirrors the existing 'Env::session'
without implicitly handling exceptions of the types 'Out_of_ram',
'Out_of_caps', 'Insufficient_ram_quota', and 'Insufficient_cap_quota'.
It enables runtime environments like init to reflect those exceptions to
their children instead of paying the costs of implicit session-quota
upgrades out of the own pocket.
By changing the 'Parent_service' to use 'try_session', this patch fixes
a resource-exhaustion problem of init in Sculpt OS that occurred when
the GPU multiplexer created a large batch of IO_MEM sessions, with each
session requiring a second attempt with the session quota upgraded by
4 KiB.
Issue #3767
Using 'alignas' in declarations might cause GCC to request for an
implementation of 'operator delete(void*, unsigned long, std::align_val_t)'
although it might actually never be called. This commit adds a dummy
implementation to 'cxx/new_delete.cc' that does nothing more than printing an
error to the log that a proper implementation is missing. This approach is
coherent with our treatment of other global delete operators.
Ref #4217
* Disable trace source and release ownership on subject destruction.
* Note, since the policy module is also destroyed on descruction of the
session component, the traced component must not access the policy
module when acknowledging the disabled state (else: page fault).
Fixesgenodelabs/genode#4247
Introduce two new cache maintainance functions:
* cache_clean_invalidate_data
* cache_invalidate_data
used to flush or invalidate data-cache lines.
Both functions are typically empty, accept for the ARM architecture.
The commit provides implementations for the base-hw kernel, and Fiasco.OC.
Fixes#4207
When loading shared libraries via the 'Shared_object' interface display
all additionaly loaded libraries in case 'ld_verbose' is configured. Up
until now, only the loaded library was displayed. In order to determine
if a dependend library had arlready been loaded prior to loading the
'Shared_object' the reference counter is used.
fixes#4147
This patch adds support for running Genode/Linux on the AARCH64
architecture.
- The kernel-agnostic startup code (crt0) had to be extended to
capture the initial stack pointer, which the Linux kernel uses
to pass the process environment. This is in line with the
existing startup code for x86_32 and x86_64.
- The link order of the host libraries linked to lx_hybrid
programs had to be adjusted such that libgcc appears at last
because the other libraries depend on symbols provided by
libgcc.
- When using AARCH64 Linux as host, one can execute run scripts
via 'make run/<script> KERNEL=linux BOARD=linux' now.
Issue #4136
When we allowed symbol resolution during exceptions, we used the shared
object lock to protect ELF object list manipulation (e.g., dlopen,
dclose) when executing exception unwinding code in the linker.
Unfortunately, sometimes libraries that are loaded by 'dlopen' may raise
exceptions in the process, leading to a deadlock within the unwind code.
In order to resolve this, we now protect the object list operations
(i.e., enqueue, removal, iteration) by a separate mutex. This allows
the shared object interface to throw exceptions.
issue #4071
This patch unifies the core-internal 'Mapping' type across all base
platforms.
As one minor downside on seL4, the diagnostic error messages when
observing faults other than page faults no longer print the faulting
thread and PD names.
Issue #2243
When the own cap quota of a client does not suffice for a cap upgrade of
an existing session to a server, the client must issue a cap-resource
request to the parent. This logic was already in place for RAM quota but
was missing for cap quota.
Issue #4072
When callback functions of `dl_iterate_phdr` required further jump slot
relocations this lead to a deadlock. Therefore, we allow the resolution
of further symbols from callback functions, but protect the ELF object
list during the iteration, which blocks any dynamic loading (e.g.,
dlopen/dlcose) of shared object by other threads while in program header
iteration.
fixes#4071
This patch changes the 'alloc_aligned' interface as follows:
- The former 'from' and 'to' arguments are replaced by a single
'range' argument.
- The distinction of the use cases of regular allocations vs.
address-constrained allocations is now overed by a dedicated
overload instead of relying on a default argument.
- The 'align' argument has been changed from 'int' to 'unsigned'
to be better compatible with 'addr_t' and 'size_t'.
Fixes#4067
The 'Timer::Session::trigger_periodic' RPC function used to accept 0 as
a way to de-schedule the periodic processing. Several components such as
nitpicker relied on this special case. In "timeout: rework timeout
framework", the value of zero was silently clamped to 1, which has the
opposite effect: triggering signals at the maximum rate. This results in
a visible effect in Sculpt where the leitzentrale-nitpicker instance
produces a constant load of 2% CPU time.
This patch restores the original timer semantics by
- Documenting it in timer_session.h,
- Handling the case explicitly in the timer implementation, and
- Replacing the silent clamping of the unexpected value 0 passed
to the timeout framework by a diagnostic error message.
Issue #3884
This patch fixes a corner case where a child is destructed while a
asynchronous close request to a sibling server is still pending.
The child immediately discarded the session ID as the end of the
close-session processing, assuming that this ID is never to be needed
again. The session-state continues to exist to handle asynchrous close
protocol with the server.
However, if the child is destructed at this point (before the server
responded to the session request), the destruction of the child would
not cover the discharging of the session state because the session state
was no longer be part of the client's ID space. So once the asynchronous
close response from the server came in, the session state contained
stale information, in particular a stale closed_callback pointer.
The patch fixes the problem by deferring the discarding of the client ID
to the point where the session state is actually destructed. So the
session of a pending close response is covered by the child destructor.
Thanks to Pirmin Duss for reporting this issue along with a test
scenario for reproducing it!
Fixes#4039
Do not link base and core libraries into on large relocatable .o file,
which is linked later to core - causing long link times. Create an
independent library archive out of the base and core libraries that can
be linked faster.
issue #4027
By first removing unused ranges, implicitly meta data allocations are freed
up. This leads to more unused slab blocks and freed up meta data allocations
in the avl tree.
Issue #4014
The code in base-hw/src/bootstrap/platform.cc uses segment flags for
identification purposes. Based on this information the code decides
what to do with each segment. Unfortunately the linker script does
not actually ensure the flags for a specific named segment match
expectations. The code relies on implicit linker behaviour.
This implicit behaviour can vary between linkers. This breaks
arm_v7a base-hw builds linked with LLVM's lld linker. The segment
named "ro" ends up having writeable flag set when using LLD.
This patch ensures that all ELF segments in genode.ld have their
required perimssion flags set explicitly.
Fixes#3988
* Remove SPEC declarations from mk/spec
* Remove all board-specific REQUIRE declaratiions left
* Replace [have_spec <board>] run-script declarations with have_board where necessary
* Remove addition of BOARD variable to SPECS in toplevel Makefile
* Move board-specific directories in base-hw out of specs
This commit restores the diag feature for selecting diagnostic output of
services provided by core. This feature became unavailable with commit
"base: remove dependency from deprecated APIs", which hard-wired the
diag flag for core services to false.
To control this feature, three possible policies can be expressed in a
routing target of init's configuration:
* Forcing silence by specifying 'diag="no"'
* Enabling diagnostics by specifying 'diag="yes"'
* Forwarding the preference of the client by omitting the 'diag'
attribute
Fixes#3962
Clang correctly asserts this private member variable is not used
anywhere in the code. I'm not sure what the intention of the code is,
might be this is a part of some unfinished feature. This patch just does
the minimum amount of work to allow the code to build with clang. If
required I can also drop the parent_levels constructor argument and
clean up the call sites.
Issue #3950
The control area is constructed during session creation and the caller can
handle the Out_of_* exception by increasing the quota by the next attempt.
Fixes#3917
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
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
* 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
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.
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
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
This remove the call to Io_progress_handler::handle_io_progress() from
wait_and_dispatch_one_io_signal() to prevent unexpected nesting
I/O-progress handling in case of custom dispatch loops (e.g., in libc).
The original intention of Io_progress_handler was to inform the
entrypoint just before blocking in the entrypoint loop.
Issue #2635
- base/cancelable_lock.h becomes base/lock.h
- all members become private within base/lock.h
- solely Mutex and Blockade are friends to use base/lock.h
Fixes#3819
_timestamp() returns CPU local values which may not be in sync with _ts
taken from another CPU. Be robust and don't produce wraparound/negative
timeout values.
Issue #3657
- Since Genode::strncpy is not 100% compatible with the POSIX
strncpy function, better use a distinct name.
- Remove bogus return value from the function, easing the potential
enforcement of mandatory return-value checks later.
Fixes#3752
The former ldso-startup static library (now called ldso_so_support) is
used to spice each shared object/library with local support code for the
dynamic linker (execution of static constructors and ARM-EABI).
Therefore, the library must be statically linked to each dynamic
library.
As a result recipes for dynamic libraries must always depend on the "so"
API, which makes ldso_so_support.mk and so_support.c available
independent of "base". Additionally, ldso_so_support is also provided in
the libc API to cut the dependency early for libc/posix libraries.
Issue #3720
This patch fixes the handling of the corner case where the allocation of
a trace buffer throws 'Out_of_caps' or 'Out_of_ram'. Under this
circumstance, the '_buffer' would still be flagged with the 'size',
which prevented any subsequent allocation attempt. This patch fixes the
problem by initializing the 'size' after the potentially throwing
allocation.
The problem triggered with the test-trace_logger after the accounting of
core's TRACE service (replacing the 'Allocator_guard' by
'Constrained_ram_allocator') became more accurate.
Related to issue #3750
The 'WHITESPACE' case of the _calc_len method wrongly accessed the
character before checking upper bound of the token. The problem is fixed
by switching the order of both conditions.
Fixes#3756
This patch removes old 'Allocator_guard' utility and replaces its use
with the modern 'Constrained_ram_allocator'.
The adjustment of core in this respect has the side effect of a more
accurate capability accounting in core's CPU, TRACE, and RM services.
In particular, the dataspace capabilities needed for core-internal
allocations via the 'Sliced_heap' are accounted to the client now.
The same goes for nitpicker and nic_dump as other former users of the
allocator guard. Hence, the patch also touches code at the client and
server sides related to these services.
The only remaining user of the 'Allocator_guard' is the Intel GPU
driver. As the adaptation of this component would be too invasive
without testing, this patch leaves this component unchanged by keeping a
copy of the 'allocator_guard.h' locally at the component.
Fixes#3750
This patch largely reverts the commit "base: lay groundwork for
base-linux caps change" because the use of 'epoll' instead of 'select'
alleviated the need to allocate large FD sets, which motivated the
introduction of the 'Native_context' hook.
Related to issue #3581
If trace is enabled for component than an attempt to put message into
trace buffer is performed using log_output policy. If it succeeds than
message is not put to logs using log service.
Fixes#3714
The former scheme left open a race window between
_process_incoming_signals() and wait_and_dispatch_one_io_signal()
resulting in both threads calling block_for_signal() and blocking
forever with one unprocessed signal.
Fixes#3704
This is needed to execute the tool-chain scenario on base-nova.
Otherwise, the fork mechanism stumbles upon a region conflict
between ldso allocations and the application heap.
Fixes#3706
The applicant_to_wake_up() member must be initialized to 0 to keep the
same semantic as before the change by
"base: add mutex as derivate of lock"
Issue #3662
The mutex class is more restrictive in usage compared to
Genode::Lock.
- At initialiation time it is ever unlocked.
- No thread is permitted to lock twice. Warn about it
in case it happens.
- Only the lock onwer is permitted to unlock the mutex.
Warn about it and don't unlock the mutex in case it happens.
Issue #3612
If the root child requests a LOG service with the label "unlabeled" then
return a LOG session that logs unprefixed messages. This allows a external
test controller to recognize log messages produced by a blessed component.
This patch changes the code of '_alloc_two_blocks_metadata' to not
leak the result of a partial allocation in the out parameters of
the method. This eases the reasoning about the absence of a
use-atfer-free problem (there was none).
Even though the call of map.metadata is known to always return a valid
pointer (because the meta data is assigned in the code just above),
better add an explicit nullptr check.
This patch removes the global variable 'blocker', which was expected to
be constructed via the global ctors. This mechanism, however, is not
used for the base library, which resulted in the use of an unconstructed
object. Specifically, the spinlock of the 'Lock' of the 'Registry'
defaulted to the LOCKED state (value 0), which eventually would lead to
a deadlock in the contention case of the cxa guard.
I could observe this deadlock once on during the component startup on
base-linux during the construction of the 'startup_lock'.
This patch fixes the problem by explicitly initializing the registry
of blockers via an init function.
Issue #2299
Issue #3578
Make sure that at least entry is present in the .dynamic section, so the
PT_DYNAMIC segment points to something valid in case there are no
dynamic symbols.
issue #3537
This patch removes ldso's builtin policy of removing any path elements
prepending the ROM module name. Instead, the ROM name is used as is.
This clears the way to access different ROM modules that share the same
name but are stored at different directories behind an fs_rom (e.g.,
/bin/bash vs. /usr/local/bin/bash).
Issue #3500
The allocation of regions within the linker area is normally left to the
best-fit 'Allocator_avl', which happens to populate the linker area
starting with the binary followed by all loaded libraried with no gaps
in between.
When replacing the binary during execve, however, we need to ensure that
the new binary does not conflict with any library that stays resident
during execve. This patch tweaks the linker's region allocation scheme
such that these libraries are placed at the end of the linker area.
Issue #3481
This patch extends the interface of the dynamic linker with the ability
to replace the running binary executable by another one. It is
designated for the implementation of execve. The interface consists of
two new functions.
'Dynamic_linker::keep' marks the specified shared object as unloadable.
This can be used to pin a set of libraries (i.e., the libc) within the
local address space while replacing the binary and other higher-level
libraries.
'Dynamic_linker::respawn' unloads the current binary, loads the one
specifed as first argument, and looks up the entry point symbol of the
new binary, which would be "main" for POSIX programs.
In addition to implementing the new interface, the patch adjusts the
linker at various places that previously assumed the binary to be
constant over runtime.
Issue #3481
This patch enables the fork.run script to run on base-linux. It should
be regarded as an interim solution, however, because the randomization
performed by the Linux kernel may still - by chance - produce a
situation where one of the libc's malloc heap regions intersects with
another dataspace dynamically attached to the child.
The better solution would be to make the 'Region_map_mmap'
implementation not depend on the kernel's allocation policy by using a
locally implemented allocator.
Issue #3478
By specifying the config attribute 'check_ctors="no"', the dynamic
linker won't abort the program on a missing call of
'Env::exec_static_constructors'. This is the case for forked programs
where the ctors were already executed by the forking program prior the
fork operation.
Issue #3478
This patch enhances the 'base/shared_object.h' interface of the dynamic
linker with the function 'for_each_loaded_object', which allows the
caller to obtain information about the currently loaded binary and
shared libraries.
The new interface is a base mechanism needed for implementing 'fork' in
the libc.
Issue #3478
This patch handles the case where a PD's cap quota becomes exhausted
on the attempt to transfer caps via 'Env::pd().transfer_quota'. The
solution mirrors the existing code for RAM quota.
Prior this patch, the 'transfer_quota' operation would trigger an
'Out_of_caps' exception.
The warning falsely detected cases where shared objects where loaded before
exec_static_constructors() was called as unneeded even in cases were the
binary itself contained static globals.
The commit also removes one redundant call to exec_static_constructors()
from the block tester.
The lazy-timer test depends on the faster-timer handler to be executed
before the fast timeout occurs, which was pretty hard to achieve on Qemu
and a busy host machine. Therefore, I increased the fast-to-faster
timeout ratio from 50/25 ms to 200/25 ms and set the test runtime to
4000 ms.
* added global offset table relative loading for global symbols
* removed 'initial_sp' and 'initial_x0' because they are currently not
used on this platform. If required they are easy to resurrect.
issue #3260
