The base class of Registered must provide a virtual destructor to enable
safe deletion with just a base class pointer. This requirement can be
lifted by using Registered_no_delete in places where the deletion
property is not needed.
Fixes#2331
Ldso now does not automatically execute static constructors of the
binary and shared libraries the binary depends on. If static
construction is required (e.g., if a shared library with constructor is
used or a compilation unit contains global statics) the component needs
to execute the constructors explicitly in Component::construct() via
Genode::Env::exec_static_constructors().
In the case of libc components this is done by the libc startup code
(i.e., the Component::construct() implementation in the libc).
The loading of shared objects at runtime is not affected by this change
and constructors of those objects are executed immediately.
Fixes#2332
This patch destructs the environment sessions for the binary and the
dynamic linker along with the other environment sessions to avoid a
warning about reverting quota that occurs when attempting to close
these sessions too late.
The race may happen when element objects get destructed by another thread then
the thread handling the for_each loop. In this case it may happen that the
object is already destructed (left the ~Element destructor) but the thread
handling the loop touches the invalid memory afterwards (the Element lock).
detected during issue #2299Fixes#2320
With the commit "init: session-label rewriting", the stack usage
increased due to the handling of session-label strings as local
variables. The stack overrun occurred in the vmm scenario on
base-hw.
There was a race when the component entrypoint wanted to do
'wait_and_dispatch_one_signal'. In this function it raises a flag for
the signal proxy thread to notice that the entrypoint also wants to
block for signals. When the flag is set and the signal proxy wakes up
with a new signal, it tried to cancel the blocking of the entrypoint.
However, if the entrypoint had not reached the signal blocking at this
point, the cancel blocking failed without a solution. Now, the new
Kernel::cancel_next_signal_blocking call solves the problem by storing a
request to cancel the next signal blocking of a thread immediately
without blocking itself.
Ref #2284
It can happen that when Cpu_free_component is constructed the insertion
of the object through 'manage' succeeds for the EP put not for the pager
EP, which in turn raises an Out_of_meta_data exception. Because we are
within the constructor, the descstructor is not called, leading to a
dangling object pool entry for the EP.
issue #2289
This patch enhances the 'Child' and 'Child_policy' with the ability to
separate the different steps of bootstrapping children. If the
'Child_policy::initiate_env_sessions()' returns false, the child's
environment sessions remain unrouted at construction time. This way,
child objects for many children can be initialized to a state that
allows the children to represent services for other children. Therefore,
session routing can be applied before any child executes.
At this stage, the environment RAM sessions of all children can be
created. Note that this step still has the limitation that RAM sessions
are generally expected to be provided by either the parent or a local
service.
Once all children are equipped with RAM, they can in principle receive
session-quota donations. Hence, all other environment sessions can now
be arbitrarily routed and initiated.
Once the environment of a child is complete, the child's process and
initial thread is created.
This patch improves the accounting for the backing store of
session-state meta data. Originally, the session state used to be
allocated by a child-local heap partition fed from the child's RAM
session. However, whereas this approach was somehow practical from a
runtime's (parent's) point of view, the child component could not count
on the quota in its own RAM session. I.e., if the Child::heap grew at
the parent side, the child's RAM session would magically diminish. This
caused two problems. First, it violates assumptions of components like
init that carefully manage their RAM resources (and giving most of them
away their children). Second, if a child transfers most of its RAM
session quota to another RAM session (like init does), the child's RAM
session may actually not allow the parent's heap to grow, which is a
very difficult error condition to deal with.
In the new version, there is no Child::heap anymore. Instead, session
states are allocated from the runtime's RAM session. In order to let
children pay for these costs, the parent withdraws the local session
costs from the session quota donated from the child when the child
initiates a new session. Hence, in principle, all components on the
route of the session request take a small bite from the session quota to
pay for their local book keeping
Consequently, the session quota that ends up at the server may become
depleted more or less, depending on the route. In the case where the
remaining quota is insufficient for the server, the server responds with
'QUOTA_EXCEEDED'. Since this behavior must generally be expected, this
patch equips the client-side 'Env::session' implementation with the
ability to re-issue session requests with successively growing quota
donations.
For several of core's services (ROM, IO_MEM, IRQ), the default session
quota has now increased by 2 KiB, which should suffice for session
requests to up to 3 hops as is the common case for most run scripts. For
longer routes, the retry mechanism as described above comes into effect.
For the time being, we give a warning whenever the server-side quota
check triggers the retry mechanism. The warning may eventually be
removed at a later stage.
This method is a hook to enable a runtime to respond to state changes.
In particular, in init this hook is used to trigger the generation of a
new state report, if configured.
Furthermore, the patch introduces the 'generate_client_side_info' and
'generate_server_side_info' methods to the 'Session_state', which
generates an XML representation of the session states to appear in
reports produced by init.
Issue #2246
The new return value of 'resolve_session_request' allows the child
policy to define the label used as the policy selector at the server.
Because this patch introduces the distinction of the child-provided
label from the label as presented to the server along with the session
request, the latter is now handled as a dedicated 'Session_state'
argument.
Issue #2248
There existed a race when 'wait_and_dispatch_one_signal' is called form
a RPC context, because the 'signal_proxy' or 'main' will block and the
signal semaphore, when the EP then calls 'wait_and_dispatch_one_signal',
the signal proxy is woken up ands sends an RPC to the EP, leading to a
dead lock if no further signal arrive, because the EP will then remain
blocked in the signal semaphore.
Therefore, for this case, the signal proxy will now perform a semaphore
up operation and does not perform an RPC if the EP is within
'wait_and_dispatch_one_signal'.
First, calls to manage and dissolve signal contexts now check if the
signal receiver was constructed. There is a small window during suspend
where it is destructed before reconstructed again.
Last, we ensure that processing of incoming signal was deblocked by the
suspend signal before entering the suspend operation. This way we ensure
already queued signal are handled.
This commit enables compile-time warnings displayed whenever a deprecated
API header is included, and adjusts the existing #include directives
accordingly.
Issue #1987
If the detach address is not the beginning of the region, one gets:
"virtual void Genode::Allocator_avl_base::free(void*): given
address (0x180e0) is not the block start address (0x18000)"
Instead, print an explicit warning in front of the detach call.
This commit addresses the situation where an environment session
outlives the session-providing service. In this case, the env session
got already invaidated at the destruction time of the server. However,
the underlying session-state structure continues to exist until the
client is destructed. During the eventual destruction of such a dangling
environment session, we have to be careful not to interact with the
no-longer existing service.
Ref #2197
This patch addresses the corner case of destructing a child that
provides an enviroment session to another child. Before this patch,
this situation could result in an infinite loop.
The problem was introduced as a side effect of issue #2197 "base: apply
routing policy to environment sessions".
This patch enables warnings if one of the deprecate functions that rely
in the implicit use of the global Genode::env() accessor are called.
For the time being, some places within the base framework continue
to rely on the global function while omitting the warning by calling
'env_deprecated' instead of 'env'.
Issue #1987
This patch removes the component_entry_point library, which used to
proved a hook for the libc to intercept the call of the
'Component::construct' function. The mechansim has several shortcomings
(see the discussion in the associated issue) and was complex. So we
eventually discarded the approach in favor of the explicit handling of
the startup.
A regular Genode component provides a 'Component::construct' function,
which is determined by the dynamic linker via a symbol lookup.
For the time being, the dynamic linker falls back to looking up a 'main'
function if no 'Component::construct' function could be found.
The libc provides an implementation of 'Component::construct', which
sets up the libc's task handling and finally call the function
'Libc::Component::construct' from the context of the appllication task.
This function is expected to be provided by the libc-using application.
Consequently, Genode components that use the libc have to implement the
'Libc::Component::construct' function.
The new 'posix' library provides an implementation of
'Libc::Component::construct' that calls a main function. Hence, POSIX
programs that merely use the POSIX API merely have to add 'posix' to the
'LIBS' declaration in their 'target.mk' file. Their execution starts at
'main'.
Issue #2199
These functions are marked as always inline through the 'SELF_RELOC' macro. This
became necessary because on riscv functions calls are performed through the
global offset table, which is not initialized at this point.
Fixes#2203
This patch removes the manually maintained symbol map from the dynamic
linker. This way, the symbol map stays in sync with the ABI and - more
importantly - no longer uses wildcards. So the symbols exported by the
dynamic linker are strictly limited by the ABI.
Issue #2190
This patch changes the child-construction procedure to allow the routing
of environment sessions to arbitrary servers, not only to the parent.
In particular, it restores the ability to route the LOG session of the
child to a LOG service provided by a child of init. In principle, it
becomes possible to also route the immediate child's PD, CPU, and RAM
environment sessions in arbitrary ways, which simplifies scenarios that
intercept those sessions, e.g., the CPU sampler.
Note that the latter ability should be used with great caution because
init needs to interact with these sessions to create/destruct the child.
Normally, the sessions are provided by the parent. So init is safe at
all times. If they are routed to a child however, init will naturally
become dependent on this particular child. For the LOG session, this is
actually not a problem because even though the parent creates the LOG
session as part of the child's environment, it never interacts with the
session directly.
Fixes#2197
This patch removes possible ambiguities with respect to the naming of
kernel-dependent binaries and libraries. It also removes the use of
kernel-specific global side effects from the build system. The reach of
kernel-specific peculiarities has thereby become limited to the actual
users of the respective 'syscall-<kernel>' libraries.
Kernel-specific build artifacts are no longer generated at magic places
within the build directory (like okl4's includes, or the L4 build
directories of L4/Fiasco and Fiasco.OC, or the build directories of
various kernels). Instead, such artifacts have been largely moved to the
libcache. E.g., the former '<build-dir>/l4/' build directory for the L4
build system resides at '<build-dir>/var/libcache/syscall-foc/build/'.
This way, the location is unique to the kernel. Note that various tools
are still generated somewhat arbitrarily under '<build-dir>/tool/' as
there is no proper formalism for building host tools yet.
As the result of this work, it has become possible to use a joint Genode
build directory that is usable with all kernels of a given hardware
platform. E.g., on x86_32, one can now seamlessly switch between linux,
nova, sel4, okl4, fiasco, foc, and pistachio without rebuilding any
components except for core, the kernel, the dynamic linker, and the timer
driver. At the current stage, such a build directory must still be
created manually. A change of the 'create_builddir' tool will follow to
make this feature easily available.
This patch also simplifies various 'run/boot_dir' plugins by removing
the option for an externally hosted kernel. This option remained unused
for many years now.
Issue #2190
The header is foc-specific. It used to shadow the generic one provided
by the base repository, which contradicts with the kernel-agnostic
Genode API. Hence, it had to be moved to a foc-specific location.
The main thread does no longer execute application code. It is solely
responsible for the initialization of the component's entrypoint and for
retrieving asynchronous notifications. Since the stack usage is no
longer dependent on application-specific code, we can significantly
shrink it to reduce the memory footprint of components. In the worst
case - should the stack overrun - we would observe a page fault because
the stack is placed in the stack area, surrounded by guard pages.
This patch replaces the former machine-word-dependent default stack size
by the fixed value of 64 KiB which should suffice for components on both
32 and 64 bit. Previously, the default stack size on 64 bit was 128 KiB,
which is wasteful. If a component needs more stack than 64 KiB, it can
specify a custon stack size by implementing 'Component::stack_size'.
The initial stack is solely used to initialize the Genode environment
along with the application stack located in the stack area. It never
executes application code. Hence, we can make it small. To check that it
is not dimensioned too small, the patch introduces a sanity check right
before switching to the application stack.
This patch unconditionally applies the labeling of sessions and thereby
removes the most common use case of 'Child_policy::filter_session_args'.
Furthermore, the patch removes an ambiguity of the session labels of
sessions created by the parent of behalf of its child, e.g., the PD
session created as part of 'Child' now has the label "<child-name>"
whereas an unlabeled PD-session request originating from the child
has the label "<child-name> -> ". This way, the routing-policy of
'Child_policy::resolve_session_request' can differentiate both cases.
As a consequence, the stricter labeling must now be considered wherever
a precise label was specified as a key for a session route or a server-
side policy selection. The simplest way to adapt those cases is to use a
'label_prefix' instead of the 'label' attribute. Alternatively, the
'label' attribute may used by appending " -> " (note the whitespace).
Fixes#2171
The heap typically first tries to allocate larger chunks than necessary, and
if it fails the actual minimal one. The first attempt already triggers warnings
which are not critical at all. If the second (critical) allocation fails,
then there are additionally checks and warnings already in place.
Issue #1039
The code in Core's Cpu_session_component::create_thread might throw a
capability refernce-count overflow if one creates many threads, which would
kill core if not handled.
Ref #2120
This is a redesign of the root and parent interfaces to eliminate
blocking RPC calls.
- New session representation at the parent (base/session_state.h)
- base-internal root proxy mechanism as migration path
- Redesign of base/service.h
- Removes ancient 'Connection::KEEP_OPEN' feature
- Interface change of 'Child', 'Child_policy', 'Slave', 'Slave_policy'
- New 'Slave::Connection'
- Changed child-construction procedure to be compatible with the
non-blocking parent interface and to be easier to use
- The child's initial LOG session, its binary ROM session, and the
linker ROM session have become part of the child's envirenment.
- Session upgrading must now be performed via 'env.upgrade' instead
of performing a sole RPC call the parent. To make RAM upgrades
easier, the 'Connection' provides a new 'upgrade_ram' method.
Issue #2120
This data structure is meant as a safe alternative for a list wherever
the list is solely used to remember objects and iterate through them in
an unspecified order. One use case is the 'Service_registry'.
We preserve lower RAM for device drivers with physical memory
constraints. If no physical RAM constraint exists, the allocations above
3G (32-bit) or 4G (64-bit) are preferred.
This fixes a regression on Ubuntu 16.04 (resp. Linux systems with recent
kernel versions) and address-space randomization originating from an
uninitialized relocation base of 0.
Replace 'dump()' debug utilities within Allocator_avl with Output::print
equivalents, and use the new Avl_tree::for_each utility to simplify
the implementation.
Ref #2159
This patch fixes a race condition triggered by the thread test running
on Linux inside VirtualBox. The 'test_stack_alloc' sporadically produced
one of two errors: A segfault in the 'Thread::deinit_platform_thread' on
the attempt to access the 'native_thread' of the to-be-destructed thread
(this data structure is located on the thread's stack). Or, an error
message about a region conflict within the stack area.
The problem was that two instances of 'Region_map_mmap' issued a
sequence of munmap and mmap each. Even though each instance locked the
attach/detach operations, the lock was held per instance. In a situation
where two instances performed attach/detach operations in parallel, the
syscall sequences could interfere with each other.
In the test scenario, the two region-map instances are the test's
address space and the stack area. When creating a thread, the thread's
trace-control dataspace is attached at an arbitrary place (picked by
the Linux kernel) within the address space whereas the stack is attached
at the stack area. The problem is the following sequence:
Thread A wants to destruct a thread:
1. Remove stack from stack area
(issue unmap syscall)
2. Preserve virtual address range that was occupied from the stack
so that Linux won't use it
(issue mmap syscall)
Thread B wants to construct a thread:
1. Request trace-control dataspace from CPU session
2. Attach trace-control dataspace to address space at a location
picked by the Linux kernel
(issue mmap syscall)
The problem occurs when thread B's second step is executed in between
the steps 1 and 2 of thread A and the Linux kernel picks the
just-unmapped address as the location for the new trace-control mapping.
Now, the trace control dataspace is mapped at the virtual address that
was designated for the stack of the to-be-created thread, and the
attempt to map the real stack fails.
The patch fixes the problem by replacing the former region-map-local
locks by a component-global lock.
Furthermore, it cleans up core's implementation of the support function
for the region-map-mmap implementation, eliminating the temporary
unlocking of the region-map lock during RPC.
Instead of solving the problem to deliver ROM modules to core while booting
differently for the several kernels (multi-boot, elfweaver, core re-linking),
this commit unifies the approaches. It always builds core as a library, and
after all binaries are built from a run-script, the run-tool will link an
ELF image out of the core-library and all boot modules. Thereby, core can
access its ROM modules directly.
This approach now works for all kernels except Linux.
With this solution, there is no [build_dir]/bin/core binary available anymore.
For debugging purposes you will find a core binary without boot modules, but
with debug symbols under [run_dir].core.
Fix#2095
base generic code:
* Remove unused verbosity code from mmio framework
* Remove escape sequence end heuristic from LOG
* replace Core_console with Core_log (no format specifiers)
* move test/printf to test/log
* remove `printf()` tests from the log test
* check for exact match of the log test output
base-fiasco:
* remove unused Fiasco::print_l4_threadid function
base-nova:
* remove unused hexdump utility from core
base-hw:
* remove unused Kernel::Thread::_print_* debug utilities
* always print resource summary of core during startup
* remove Kernel::Ipc_node::pd_label (not used anymore)
base*:
* Turn `printf`,`PWRN`, etc. calls into their log equivalents
Ref #1987Fix#2119
This patch adds the missing exception handling for depleted RM session
quotas. If core runs out of session quota while creating a new region
map, it now reflects this condition as a Region_map::Out_of_metadata
exception to the RM client. Thanks to Denis Huber for reporting the
issue!
This patch revives our ds_ownership test from 2012, which just revealed
a regression in core where the dataspace-free operation of the RAM
service would unconditionally destroy dataspace objects from foreign
sessions. The patch fixes the bug and adds an updated version of the
test to the autopilot.
Fixes#2065
Besides adapting the components to the use of base/log.h, the patch
cleans up a few base headers, i.e., it removes unused includes from
root/component.h, specifically base/heap.h and
ram_session/ram_session.h. Hence, components that relied on the implicit
inclusion of those headers have to manually include those headers now.
While adjusting the log messages, I repeatedly stumbled over the problem
that printing char * arguments is ambiguous. It is unclear whether to
print the argument as pointer or null-terminated string. To overcome
this problem, the patch introduces a new type 'Cstring' that allows the
caller to express that the argument should be handled as null-terminated
string. As a nice side effect, with this type in place, the optional len
argument of the 'String' class could be removed. Instead of supplying a
pair of (char const *, size_t), the constructor accepts a 'Cstring'.
This, in turn, clears the way let the 'String' constructor use the new
output mechanism to assemble a string from multiple arguments (and
thereby getting rid of snprintf within Genode in the near future).
To enforce the explicit resolution of the char * ambiguity, the 'char *'
overload of the 'print' function is marked as deleted.
Issue #1987
and show some message about. On sel4 sometimes exceptions are thrown and no
message is shown, it just seems to hang. With this patch at least you get
an idea that something bad happened.
Issue #2044
Those headers implement a platform-specific mechanism. They are never
used by components directly.
This patch also cleans up a few other remaining platform-specific
artifact such as the Fiasco.OC-specific assert.h.
Issue #1993
Conveying the ROM filename as the final label element simplifies
routing policy and session construction.
Annotations by nfeske:
This commit also changes the ROM session to use base/log.h instead of
base/printf.h, which produced build error of VirtualBox because the
vbox headers have a '#define Log', which collides with the content of
base/log.h. Hence, this commit has to take precautions to resolve this
conflict.
The commit alse refines the previous session-label change by adding a
new 'Session_label::prefix' method and removing the use of 'char const *'
from this part of the API.
Fixes#1787
Session_label constructor now takes a bare string rather than a
serialized argument buffer.
Replace all instances of previous constructor with 'label_from_args'
function.
Issue #1787
This patch establishes the sole use of generic headers across all
kernels. The common 'native_capability.h' is based on the version of
base-sel4. All traditional L4 kernels and Linux use the same
implementation of the capability-lifetime management. On base-hw, NOVA,
Fiasco.OC, and seL4, custom implementations (based on their original
mechanisms) are used, with the potential to unify them further in the
future.
This change achieves binary compatibility of dynamically linked programs
across all kernels.
Furthermore, the patch introduces a Native_capability::print method,
which allows the easy output of the kernel-specific capability
representation using the base/log.h API.
Issue #1993
This patch alleviates the need for a Native_capability::Dst at the API
level. The former use case of this type as argument to
Deprecated_env::reinit uses the opaque Native_capability::Raw type
instead. The 'Raw' type contains the portion of the capability that is
transferred as-is when delegating the capability (i.e., when installing
the parent capability into a new component, or when installing a new
parent capability into a new forked Noux process). This information can
be retrieved via the new Native_capability::raw method.
Furthermore, this patch moves the functions for retriving the parent
capability to base/internal/parent_cap.h, which is meant to be
implemented in platform-specific ways. It replaces the former set of
startup/internal/_main_parent_cap.h headers.
Issue #1993
The static 'Thread::mystack()' function returns the stack boundaries of
the calling thread. It is useful when a thread uses a diffent stack than
the primary one.
Fixes#2037
This patch introduces the Genode::raw function that prints output
directly via a low-level kernel mechanism, if available.
On base-linux, it replaces the former 'raw_write_str' function.
On base-hw, it replaces the former kernel/log.h interface.
Fixes#2012
- add a new function 'binary_ready_hook_for_gdb()' in ldso. GDB can set a
breakpoint at this function to know when ldso has loaded the binary
into memory.
- get the thread state from the NOVA kernel immediately on 'pause()'
Fixes#1968
This patch moves the thread operations from the 'Cpu_session'
to the 'Cpu_thread' interface.
A noteworthy semantic change is the meaning of the former
'exception_handler' function, which used to define both, the default
exception handler or a thread-specific signal handler. Now, the
'Cpu_session::exception_sigh' function defines the CPU-session-wide
default handler whereas the 'Cpu_thread::exception_sigh' function
defines the thread-specific one.
To retain the ability to create 'Child' objects without invoking a
capability, the child's initial thread must be created outside the
'Child::Process'. It is now represented by the 'Child::Initial_thread',
which is passed as argument to the 'Child' constructor.
Fixes#1939
It turns out that the name function does not have much use in practice
except for naming the thread of the component's initial entrypoint. For
dynamically linked components, this thread is created by the dynamic
linker. It is named "ep" in these cases. Considering that we will
eventually turn all regular components into dynamically linked
executables, the additional information provided by the
Component::name() function remains unused. So it is better to not bother
the component developers with adding boilerplate code.
Now rlibs are actually linked to programs. Target files have been
modified to not generate code that requires compiler-rt. Added a target
for libstd-rust, but it's very broken right now. Moved alloc_system to
the libports folder because either a memory allocator needs to be
written in rust or posix_memalign needs to be implemented. Changed
liblibc to use freebsd as the OS instead of netbsd. Added a library with
unwind dummy functions.
Rust relies on atomic builtins, which are not implemented in libgcc for
ARM. One was implemented in rust, which was sufficient to get the
current rust test to run. Rust libs were added into the group of libs
for the linker so order no longer matters. The raspberry pi now uses an
armv6 target.
This patch cleans up the thread API and comes with the following
noteworthy changes:
- Introduced Cpu_session::Weight type that replaces a formerly used
plain integer value to prevent the accidental mix-up of
arguments.
- The enum definition of Cpu_session::DEFAULT_WEIGHT moved to
Cpu_session::Weight::DEFAULT_WEIGHT
- New Thread constructor that takes a 'Env &' as first argument.
The original constructors are now marked as deprecated. For the
common use case where the default 'Weight' and 'Affinity' are
used, a shortcut is provided. In the long term, those two
constructors should be the only ones to remain.
- The former 'Thread<>' class template has been renamed to
'Thread_deprecated'.
- The former 'Thread_base' class is now called 'Thread'.
- The new 'name()' accessor returns the thread's name as 'Name'
object as centrally defined via 'Cpu_session::Name'. It is meant to
replace the old-fashioned 'name' method that takes a buffer and size
as arguments.
- Adaptation of the thread test to the new API
Issue #1954
This patch moves the base library from src/base to src/lib/base,
flattens the library-internal directory structure, and moves the common
parts of the library-description files to base/lib/mk/base.inc and
base/lib/mk/base-common.inc.
Furthermore, the patch fixes a few cosmetic issues (whitespace and
comments only) that I encountered while browsing the result.
Fixes#1952
Most slab allocators in core use a sliced heap as backing store. Since
sliced-heap allocations are performed at page-granularity, it is
sensible to dimension the slab blocks to fill whole pages.
This patch cleans up the implementation of the sliced heap, adds a
constructor that takes references instead of pointers, and adds the
function 'meta_data_size' to determine the meta-data overhead per block.
The latter can be used to dimension slab allocators such that slab
blocks use whole pages.
The original 'Env' interface as returned by 'Genode::env()' has been
renamed to 'Env_deprecated' and moved to deprecated/env.h. The new version
of base/env.h contains the interface passed to modern components that
use the component API via base/component.h.
Issue #1832
This patch makes the former 'Process' class private to the 'Child'
class and changes the constructor of the 'Child' in a way that
principally enables the implementation of single-threaded runtime
environments that virtualize the CPU, PD, and RAM services. The
new interfaces has become free from side effects. I.e., instead
of implicitly using Genode::env()->rm_session(), it takes the reference
to the local region map as argument. Also, the handling of the dynamic
linker via global variables is gone. Now, the linker binary must be
provided as constructor argument.
Fixes#1949
This patch replaces the former 'Pd_session::bind_thread' function by a
PD-capability argument of the 'Cpu_session::create_thread' function, and
removes the ancient thread-start protocol via 'Rm_session::add_client' and
'Cpu_session::set_pager'. Threads are now bound to PDs at their creation
time and implicitly paged according to the address space of the PD.
Note the API change:
This patch changes the signature of the 'Child' and 'Process' constructors.
There is a new 'address_space' argument, which represents the region map
representing the child's address space. It is supplied separately to the
PD session capability (which principally can be invoked to obtain the
PD's address space) to allow the population of the address space
without relying on an 'Pd_session::address_space' RPC call.
Furthermore, a new (optional) env_pd argument allows the explicit
overriding of the PD capability handed out to the child as part of its
environment. It can be used to intercept the interaction of the child
with its PD session at core. This is used by Noux.
Issue #1938
This patch integrates three region maps into each PD session to
reduce the session overhead and to simplify the PD creation procedure.
Please refer to the issue cited below for an elaborative discussion.
Note the API change:
With this patch, the semantics of core's RM service have changed. Now,
the service is merely a tool for creating and destroying managed
dataspaces, which are rarely needed. Regular components no longer need a
RM session. For this reason, the corresponding argument for the
'Process' and 'Child' constructors has been removed.
The former interface of the 'Rm_session' is not named 'Region_map'. As a
minor refinement, the 'Fault_type' enum values are now part of the
'Region_map::State' struct.
Issue #1938
The return code of assign_parent remained unused. So this patch
removes it.
The bind_thread function fails only due to platform-specific limitations
such as the exhaustion of ID name spaces, which cannot be sensibly
handled by the PD-session client. If occurred, such conditions used to
be reflected by integer return codes that were used for diagnostic
messages only. The patch removes the return codes and leaves the
diagnostic output to core.
Fixes#1842
We don't want Genode environment objects that register their destructor
for program exit as it is mostly unnecessary and easily produces
dangling pointers. Thus, use unmanaged_singleton instead of the static
keyword.
Fixes#1941
When using the Allocator interface, one can't tell which alignment
resulting allocations fulfill. However, at least on ARM, given the
architectural alignment requirements of ARM memory accesses, one wants
memory allocations (what allocators are for in most cases) to be word
aligned automatically. Previously, at least the AVL allocator simply
called alloc_aligned without defining align in its alloc implementation.
This led to unaligned access faults (the default was 0) when using the
AVL allocator as Allocator (as done in the metadata management of a SLAB
of an AVL that uses the AVL as backing store). To avoid such pitfalls
in the future, we force users of alloc_aligned to always specify align
(why use alloc_aligned without align anyway).
Ref #1941
Besides unifying the Msgbuf_base classes across all platforms, this
patch merges the Ipc_marshaller functionality into Msgbuf_base, which
leads to several further simplifications. For example, this patch
eventually moves the Native_connection_state and removes all state
from the former Ipc_server to the actual server loop, which not only
makes the flow of control and information much more obvious, but is
also more flexible. I.e., on NOVA, we don't even have the notion of
reply-and-wait. Now, we are no longer forced to pretend otherwise.
Issue #1832
This patch unifies the CPU session interface across all platforms. The
former differences are moved to respective "native-CPU" interfaces.
NOVA is not covered by the patch and still relies on a custom version of
the core-internal 'cpu_session_component.h'. However, this will soon be
removed once the ongoing rework of pause/single-step on NOVA is
completed.
Fixes#1922
This patch changes the organization of the slab blocks within the slab
allocator. Originally, blocks were kept in a list sorted by the number
of free entries. However, it turned out that the maintenance of this
invariant involves a lot of overhead in the presence of a large number
of blocks. The new implementation manages blocks within a ring in no
particular order and maintains a pointer to the block where the next
allocation is attempted. This alleviates the need for sorting blocks
when allocating and deallocating.
Fixes#1908
This patch ensures that the 'Allocator_avl' releases all memory obtained
from the meta-data allocator at destruction time. If allocations are
still dangling, it produces a warning, hinting at possible memory leaks.
Finally, it properly reverts all 'add_range' operations.
This patch fixes a use-after-free problem raised by the recent ability of
the slab allocator to dynamically release empty slab blocks. The
Rm_session_component::detach function used to rely on the assumption
that the region metadata co-located with the allocator metadata of the
'_map' would stay intact even after a 'free' if the region.
This patch makes sure that the dataspace pool is flushed before
destructing the heap-local allocator-avl instance. With the original
destruction order, the allocator would still contain dangling
allocations on the account of the dataspace pool when destructed. In
practice, this caused no problem because the underlying backing store is
eventually freed on the destruction of the pool. But it triggers a
runtime warning of the allocator since it has become more strict with
regard to dangling allocations.
This commit introduces the new `Component` interface in the form of the
headers base/component.h and base/entrypoint.h. The os/server.h API
has become merely a compatibilty wrapper and will eventually be removed.
The same holds true for os/signal_rpc_dispatcher.h. The mechanism has
moved to base/signal.h and is now called 'Signal_handler'.
Since the patch shuffles headers around, please do a 'make clean' in the
build directory.
Issue #1832
This commit replaces the stateful 'Ipc_client' type with the plain
function 'ipc_call' that takes all the needed state as arguments.
The stateful 'Ipc_server' class is retained but it moved from the public
API to the internal ipc_server.h header. The kernel-specific
implementations were cleaned up and simplified. E.g., the 'wait'
function does no longer exist. The badge and exception code are no
longer carried in the message buffers but are handled in kernel-specific
ways.
Issue #610
Issue #1832
Rust relies on atomic builtins, which are not implemented in libgcc for
ARM. One is implemented in rust, which is sufficient to get the
current rust test to run.
Issue #1899
Check if the binary pointer is valid before attempting to lookup the
symbol. Shared objects with unresolved symbols and missing depencies,
e.g a library that references 'errno' but is not linked against libc,
will now produce an error message when they are loaded by the dynamic
linker.
Fixes#1904.
This patch moves details about the stack allocation and organization
the base-internal headers. Thereby, I replaced the notion of "thread
contexts" by "stacks" as this term is much more intuitive. The fact that
we place thread-specific information at the bottom of the stack is not
worth introducing new terminology.
Issue #1832
By moving the stub implementation to rm_session_client.cc, we can use
the generic base/include/rm_session/client.h for base-linux and
base-nova and merely use platform-specific implementations.
Issue #1832
This patch establishes a common organization of header files
internal to the base framework. The internal headers are located at
'<repository>/src/include/base/internal/'. This structure has been
choosen to make the nature of those headers immediately clear when
included:
#include <base/internal/lock_helper.h>
Issue #1832
This patch integrates the functionality of the former CAP session into
the PD session and unifies the approch of supplementing the generic PD
session with kernel-specific functionality. The latter is achieved by
the new 'Native_pd' interface. The kernel-specific interface can be
obtained via the Pd_session::native_pd accessor function. The
kernel-specific interfaces are named Nova_native_pd, Foc_native_pd, and
Linux_native_pd.
The latter change allowed for to deduplication of the
pd_session_component code among the various base platforms.
To retain API compatibility, we keep the 'Cap_session' and
'Cap_connection' around. But those classes have become mere wrappers
around the PD session interface.
Issue #1841
This patch removes the SIGNAL service from core and moves its
functionality to the PD session. Furthermore, it unifies the PD service
implementation and terminology across the various base platforms.
Issue #1841
The wrapper functions (e.g., 'Unwind_*' and friends) now have the same signature
as the original function in 'libgcc', reside in a separate C file which is
archived to cxx.lib.a. In supc++.o we prefix the wrapped functions with '_cxx_'.
This also enables support for riscv.
related to #1880
This prevents a sporadic null-pointer dereference in the nic_loopback
test, which occurred once in 100 runs. I'm not sure if there's still a
race window (we may investigate) with context dissolve.
On Linux the linker can now be loaded at arbitrary addresses, this became
necessary for newer kernel versions. The 'linux_arm' target is not supported.
Issue #1728
'block_for_signal' and 'pending_signal' now set pending flag in signal context
in order to determine pending signal. The context list is also used by the
'Signal_receiver' during destruction.
Fixes#1738
Currently, when a signal arrives in the main thread, the signal dispatcher is
retrieved and called from the main thread, the dispatcher uses a proxy object
that in turn sends an RPC to the entry point. This becomes a problem when the
entry point destroys the dispatcher object, before the dispatch function has
been called by the main thread. Therefore, the main thread should simply send an
RPC to the entry point upon signal arrival and the dispatching should be handled
solely by the entry point.
Issue #1738
Destroying an object within the scope of a lambda/functor executed
in the object pool's apply function leads potentially to memory corruption.
Within the scope the corresponding object is locked and unlocked when
leaving the scope. Therefore, it is illegal to free the object's memory meanwhile.
This commit eliminates several places in core that destroyed wrongly in
the object pool's scope.
Fix#1713
* Move the Synced_interface from os -> base
* Align the naming of "synchronized" helpers to "Synced_*"
* Move Synced_range_allocator to core's private headers
* Remove the raw() and lock() members from Synced_allocator and
Synced_range_allocator, and re-use the Synced_interface for them
* Make core's Mapped_mem_allocator a friend class of Synced_range_allocator
to enable the needed "unsafe" access of its physical and virtual allocators
Fix#1697
The commit consumes the argument of a unsupported printf command.
Without the commit - a subsequent command uses the argument of the preceding
command, which may cause memory corruption or page faults for sequences using
string commands, e.g.
Genode::printf("%#x %s\n", 0x20, "Test");
'#' is not supported by Genode::printf. In this scenario a pagefault at
address 0x20 is caused.
Fixes#1701
Instead of holding SPEC-variable dependent files and directories inline
within the repository structure, move them into 'spec' subdirectories
at the corresponding levels, e.g.:
repos/base/include/spec
repos/base/mk/spec
repos/base/lib/mk/spec
repos/base/src/core/spec
...
Moreover, this commit removes the 'platform' directories. That term was
used in an overloaded sense. All SPEC-relative 'platform' directories are
now named 'spec'. Other files, like for instance those related to the
kernel/architecture specific startup library, where moved from 'platform'
directories to explicit, more meaningful places like e.g.: 'src/lib/startup'.
Fix#1673
Instead of returning pointers to locked objects via a lookup function,
the new object pool implementation restricts object access to
functors resp. lambda expressions that are applied to the objects
within the pool itself.
Fix#884Fix#1658
For most platforms except of NOVA a distinction between pager entrypoint
and pager activation is not needed, and only exists due to historical
reasons. Moreover, the pager thread's execution path is almost identical
between most platforms excluding NOVA, HW, and Fisco.OC. Therefore,
this commit unifies the pager loop for the other platforms, and removes
the pager activation class.
Increase internal chunk size of heap only if an allocation succeeded
beforehand. Otherwise the chunk size increases with every unsuccessful
invocation and a upgrade of the used ram session will be insufficient and of
no use at all.
Fixes#1632
Introduces a class Unmanaged_singleton_constructor that can be declared as
friend to be able to call unmanaged_singleton on classes with a private
constructor. Enables the appliance of the singleton pattern.
Ref #1625
This commit eliminates the mutual interlaced taking of destruction lock,
list lock and weak pointer locks that could lead to a dead-lock situation
when a lock pointer was tried to construct while a weak object is in
destruction progress.
Now, all weak pointers are invalidated and dequeued at the very
beginning of the weak object's destruction. Moreover, before a weak pointer
gets invalidated during destruction of a weak object, it gets dequeued, and
the list lock is freed again to avoid the former dead-lock.
Fix#1607
Up to now it was not possible to trace threads that use a different
Cpu_session rather than env()->cpu_session() (as done by VirtualBox).
This problem is now solved by setting the Cpu_session explicitly when
creating the event logger and attaching the trace control area when
creating the thread.
Fixes#1618.
Add a test where a locked pointer shall be taken during object destruction.
Moreover, extend the run-script so it runs on different platforms with
"real" timers.
While importing trace sources as trace subjects into a TRACE session,
the session quota might become depleted. The TRACE session already keeps
track of the session quota via an allocator guard but the 'subjects' RPC
function missed to handle the out-of-memory condition. This patch
reflects the error condition as an 'Out_of_metadata' exception to the
TRACE client. It also contains an extension of the trace test to
exercise the corner case.
This patch enable clients of core's TRACE service to obtain the
execution times of trace subjects (i.e., threads). The execution time is
delivered as part of the 'Subject_info' structure.
Right now, the feature is available solely on NOVA. On all other base
platforms, the returned execution times are 0.
Issue #813
On ARM, the compiler generates calls to memcpy and memset. Most
dynamically linked programs use the libc, which provides these
functions. However, if a dynamically linked program does not use the
libc (e.g., noux/minimal or the new version of cli_monitor), those
symbols remain unresolved. By adding them to ldso's symbol.map, the
dynamic linker will resolve them with the functions of the cxx
library, which is part of the dynamic linker.
Issue #1561
Currently, libc_noux includes the 'base/src/base/env/platform_env.h' file
to be able to reinitialize the environment using the 'Platform_env'
interface. For base-linux, a special version of this file exists and the
inclusion of the generic version in libc_noux causes GCC 4.9 to make wrong
assumptions about the memory layout of the 'Env' object returned by
'Genode::env()'.
This commit moves the reinitialization functions to the 'Env' interface to
avoid the need to include the 'platform_env.h' file in libc_noux.
Fixes#1510
On seL4, we need to convert untyped memory to page frames before being
able to use it as normal memory. There already exists the hook function
'_export_ds' that is principally suitable for such tasks. It is
currently solely used on Linux where we have to create a file for each
dataspace. To make the hook useful also for seL4, we need to call
_export_ds prior _clear_ds. Otherwise, we would try to clear memory that
is still untyped.
This patch avoids the attempt to extend the cxx-local heap during the
startup phase of an application. Originally, the static part of the cxx
was merely 100 bytes, which did not suffice to run the minimalistic test
roottask on seL4.
The thread library (thread.cc) in base-foc shared 95% of the code with
the generic implementation except myself(). Therefore, its
implementation is now separated from the other generic sources into
myself.cc, which allows base-foc to use a foc-specific primitive to
enable our base libraries in L4Linux.
Issue #1491
Physical CPU quota was previously given to a thread on construction only
by directly specifying a percentage of the quota of the according CPU
session. Now, a new thread is given a weighting that can be any value.
The physical counter-value of such a weighting depends on the weightings
of the other threads at the CPU session. Thus, the physical quota of all
threads of a CPU session must be updated when a weighting is added or
removed. This is each time the session creates or destroys a thread.
This commit also adapts the "cpu_quota" test in base-hw accordingly.
Ref #1464
This patch adds const qualifiers to the functions Allocator::consumed,
Allocator::overhead, Allocator::avail, and Range_allocator::valid_addr.
Fixes#1481
* Instead of using local capabilities within core's context area implementation
for stack allocation/attachment, simply do both operations while stack gets
attached, thereby getting rid of the local capabilities in generic code
* In base-hw the UTCB of core's main thread gets mapped directly instead of
constructing a dataspace component out of it and hand over its local
capability
* Remove local capability implementation from all platforms except Linux
Ref #1443
The global capability ID counter is not used by NOVA and Fiasco.OC
and in the future not needed by base-hw too. Thereby, remove the static
counter variable from the generic code base and add it where appropriated.
Ref #1443
Enable platform specific allocations and ram quota accounting for
protection domains. Needed to allocate object identity references
in the base-hw kernel when delegating capabilities via IPC.
Moreover, it can be used to account translation table entries in the
future.
Ref #1443
There are lots of places where a numeric argument of an argument string
gets extraced as signed long value and then assigned to an unsigned long
variable. If the value in the string was negative, it would not be
detected as invalid (and replaced by the default value), but become a
positive bogus value.
With this patch, numeric values which are supposed to be unsigned get
extracted with the 'ulong_value()' function, which returns the default
value for negative numbers.
Fixes#1472
The emergency dataspace is used to accommodate the corner case where
a signal context capability is created while issuing the first
resource request. Normally, the attempt to upgrade the signal-session
quota under such a constrained situation would fail. By freeing the
emergency dataspace in this situation, we regain enough quota to
upgrade the signal session.
This is a follow up commit for "base: Raise RAM quota of signal session
to 16K" and fixes the resource_request test on 64-bit platforms.
* Introduce hw-specific crt0 for core that calls e.g.: init_main_thread
* re-map core's main thread UTCB to fit the right context area location
* switch core's main thread's stack to fit the right context area location
Fix#1440
The linker scripts use to fill alignment gaps within the text section
with the magic value 0x90909090, which correponds to the opcodes of four
nop instructions on x86. This patch removes this value because it
apparently solves no problem. If, for some reason (e.g., due to a dangling
pointer) a thread executes instructions within alignment paddings, NOP
instructions are not any better than any other instruction. The program
will eventually execute the instructions after the padding, which is
most likely fatal. It would be more reasonable to fill the padding with
the opcode of an illegal instruction so that such an error can be
immediately detected. That said, I cannot remember a single instance,
where the fill value has helped us during debugging.
Even if the mechanism served a purpose on x86, it is still better to
remove it because it does not equally work on the other architectures
where the linker scripts are used. I.e., on ARM, the opcode 0x90909090
is not a NOP instruction.
If newlines are in the string send to the core log service, they don't get
the label properly appended before each output. The messages then look like
they are coming from core.
Fixes#1368
This has to be used during shared object creation and destruction because global
lists are manipulated. We cannot use the 'Elf_object::lock' here because there
may be jump-slot relocations during object initialization.
Fixes#1350
This wasn't necessary before because we built an l4 library for
Pistachio and linked it against each application. With the new linker,
we compile the required files from within Genode and create a syscall
library that is only linked to ldso. If a program uses system calls
directly, for example, DDE kit's spinlock implementation, the required
symbols must be made globally accessible.
Fixes#1306
In the init configuration one can configure the donation of CPU time via
'resource' tags that have the attribute 'name' set to "CPU" and the
attribute 'quantum' set to the percentage of CPU quota that init shall
donate. The pattern is the same as when donating RAM quota.
! <start name="test">
! <resource name="CPU" quantum="75"/>
! </start>
This would cause init to try donating 75% of its CPU quota to the child
"test". Init and core do not preserve CPU quota for their own
requirements by default as it is done with RAM quota.
The CPU quota that a process owns can be applied through the thread
constructor. The constructor has been enhanced by an argument that
indicates the percentage of the programs CPU quota that shall be granted
to the new thread. So 'Thread(33, "test")' would cause the backing CPU
session to try to grant 33% of the programs CPU quota to the thread
"test". By now, the CPU quota of a thread can't be altered after
construction. Constructing a thread with CPU quota 0 doesn't mean the
thread gets never scheduled but that the thread has no guaranty to receive
CPU time. Such threads have to live with excess CPU time.
Threads that already existed in the official repositories of Genode were
adapted in the way that they receive a quota of 0.
This commit also provides a run test 'cpu_quota' in base-hw (the only
kernel that applies the CPU-quota scheme currently). The test basically
runs three threads with different physical CPU quota. The threads simply
count for 30 seconds each and the test then checks wether the counter
values relate to the CPU-quota distribution.
fix#1275
Do not support the global construction from of objects from within a global
constructor of another object. This can happen if, for example, dlopen is called
from a global constructor. The construction will be post-boned until the current
constructor has finished.
The memory barrier prevents the compiler from changing the program order
of memory accesses in such a way that accesses to the guarded resource
get outside the guarded stage. As cmpxchg() defines the start of the
guarded stage it also represents an effective memory barrier.
On x86, the architecture ensures to not reorder writes with older reads,
writes to memory with other writes (except in cases that are not
relevant for our locks), or read/write instructions with I/O
instructions, locked instructions, and serializing instructions.
However on ARM, the architectural memory model allows not only that
memory accesses take local effect in another order as their program
order but also that different observers (components that can access
memory like data-busses, TLBs and branch predictors) observe these
effects each in another order. Thus, a correct program order isn't
sufficient for a correct observation order. An additional architectural
preservation of the memory barrier is needed to achieve this.
Fixes#692
The backend allocator for the slab is a sliced heap, which hands out
allocations with page-size granularity (4096 bytes). Therefore, the
slab-block size should also be about a multiple of the page size minus
some bytes of overhead.
Additional adjustments:
- The slab-block size and the default quota-upgrade amount for SIGNAL
sessions depends on the platform bit width now.
- The signal test also stresses the case of many managed context in one
session including creation and destruction of the used signal receiver
in repeated rounds.
So far, the lifetime-management utilities 'Weak_ptr' and 'Locked_ptr'
had been preserved for core-internal use only. However, the utilities
are handy for many use cases outside of core where object lifetimes
must be managed. So we promote them to the public API.
When a page fault cannot be resolved, the GDB monitor can get a hint about
which thread faulted by evaluating the thread state object returned by
'Cpu_session::state()'. Unfortunately, with the current implementation,
the signal which informs GDB monitor about the page fault is sent before
the thread state object of the faulted thread has been updated, so it
can happen that the faulted thread cannot be determined immediately
after receiving the signal.
With this commit, the thread state gets updated before the signal is sent.
At least on base-nova it can also happen that the thread state is not
accessible yet after receiving the page fault notification. For this
reason, GDB monitor needs to retry its query until the state is
accessible.
Fixes#1206.
On ARM it's relevant to not only distinguish between ordinary cached memory
and write-combined one, but also having non-cached memory too. To insert the
appropriated page table entries e.g.: in the base-hw kernel, we need to preserve
the information about the kind of memory from allocation until the pager
resolves a page fault. Therefore, this commit introduces a new Cache_attribute
type, and replaces the write_combined boolean with the new type where necessary.
On ARM, when machine instructions get written into the data cache
(for example by a JIT compiler), one needs to make sure that the
instructions get written out to memory and read from memory into
the instruction cache before they get executed. This functionality
is usually provided by a kernel syscall and this patch adds a generic
interface for Genode applications to use it.
Fixes#1153.
This patch changes the top-level directory layout as a preparatory
step for improving the tools for managing 3rd-party source codes.
The rationale is described in the issue referenced below.
Issue #1082
