When introducing an avl-tree to re-find known capabilities in Fiasco.OC
(issue #112), the memory demand of apllications was slightly increased.
So we've to give the pl11x driver more memory by default. Fixes#140.
Separate spin-lock implementation from lock-implementation and put it into a
non-public header, so it can be re-used by the DDE kit's and Fiasco.OC's
capability-allocator spin lock. Fixes issue #123.
The old variant provided 8K capability slots to all processes on core,
which increased binaries by 180 KB for the static allocator. I reduced it
to 4K capabilities stay under 100 KB overhead for the allocator.
Anyway, pci_drv and pl11x_drv need more RAM quota now: 2M for pl11x_drv
and 1M for pci_drv.
In the cap-session component in core when freeing a capability, the
corresponding kernel object should be unmapped from all processes and core.
Until now, the unmap operation for removing the kernel object didn't worked
because of using the wrong rights-map. This patch fixes it.
The re-use of capabilities introduced by the last patch triggered this
problem because its essential for the capability-registry to detect
invalidated capabilities.
This is an interim fix for issue #112. This patch extends the
'Capability_allocator' class with the ability to register the global
ID of a Genode capability so that the ID gets associated with a
process-local kernel capability. Whenever a Genode capability gets
unmarshalled from an IPC message, the capability-allocator is asked,
with the global ID as key, whether the kernel-cap already exists.
This significantly reduces the waste of kernel-capability slots.
To circumvent problems of having one and the same ID for different kernel
objects, the following problems had to be solved:
* Replace pseudo IDs with unique ones from core's badge allocator
* When freeing a session object, free the global ID _after_ unmapping
the kernel object, otherwise the global ID might get re-used in some
process and the registry will find a valid but wrong capability
for the ID
Because core aggregates all capabilities of all different processes, its
capability registry needs much more memory compared to a regular process.
By parametrizing capability allocators differently for core and non-core
processes, the global memory overhead for capability registries is kept
at a reasonable level.
On Ubuntu 11.10 the host library created for the 'lx_hybrid_ctors' test
doesn't get loaded, because the test object in the library is not being
used by the application. This commit fixes the problem by having the
application call a dummy function in the library.
Fixes#120.
The kernel distinguishes local from global IDs by looking at the lowest
6 bits of the thread ID (i.e., in 'L4_ThreadControl'). If those bits are
zero, the ID is interpreted as a local ID. Because those zero bits
overlap with the version bits of global IDs, this invariant could be
violated once the version of a global ID reaches 64. In this case,
'L4_ThreadControl' will return an error on the attempt to create a new
PD. To prevent this from happening, we always set the lowest bit to 1.
- Remove example application source code files which also exist in contrib
- Outsource commonly used parts from target.mk files
- Store the current Qt version only in one place
- Add run scripts for the example applications
Fixes#127.
The 'noux_bash.run' script has become able to present the user with an
interactive bash shell for executing various coreutils programs. It is
still pretty limited, i.e., the environment is not correctly passed to
child processes and pipes are not supported. But bash and coreutils are
operational.
Replace 'Reg_array' in 'Genode::Mmio' by 'Register_array' and 'Subreg'
in 'Genode::Register', 'Genode::Mmio::Register'and
'Genode::Mmio::Register_array' by 'Bitfield'.
Update and beautify comments in the according headers and test programs.
'Reg_array' contains items whose width can be the width of the register
storage type at a max. Nethertheless they can be smaller and iterate all
subregs that are covered by the item width. The array uses as much
successive instances of its storage type as needed.
The test 'run/util_mmio' also tests these new features heavily.
The run script 'run/util_mmio.run' runs a test over basic
functionalities of 'Mmio::Register' and 'Mmio::Register::Subreg'. The
test covers the functions 'read' and 'bits', 'set', 'clear' and 'get'.
Inline function in 'Mmio::Register::Subreg' whose definition otherwise
looks ugly.
To accommodate CPU registers, which have a structured layout but don't
depend on a region base address, this patch introduces the generic
'Genode::Register' and 'Genode::Subreg' to 'register.h'.
'Mmio::Register' and 'Mmio::Subreg' inherit from them.
The MMIO access framework consists of an abstraction for a contiguous
MMIO area with a base address set dynamically. Within this class 'Mmio'
are declarations for 'Register' and 'Subreg'. These two can be
parameterized statically via template parameters to create arbitrary
MMIO structures.
Whereas 'Register' relies to a POD like subregion of 'Mmio', 'Subreg'
relies to a MMIO region within a specific 'Register' and therefore is
smaller or equal then the storage type of its superior 'Register'.
Furthermore with 'Reg_array' and 'Subreg_array', there exists the
possibility to handle arrays of uniform contiguous registers or subregs
by index. 'Subreg_array' therefore abstracts from the width boundary of
its superior 'Register' and handles a steady distance between its
members in addition. Both also check array size limits.
Related to issue #69.
This patch removes the hardcoded parent cap for the new child by
introducing a facility called 'poke' similar to copy-to-user. This
change makes the fork mechanism kernel-agnostic. The 'noux_fork'
test works on L4/Fiasco, OKL4, NOVA, Fiasco.OC, and L4ka::Pistachio.
Linux is not supported yet.
If using the libc for Noux programs, the default implementation of these
functions cannot be used because they rely on a dedicated
timeout-scheduler thread. Noux programs, however, are expected to
contain only the main thread. By turning the functions into weak
symbols, the noux libc-plugin can provide custom implementations.
The resource virtualization layer has been extended to support sub RM
sessions because this feature is used by the dynamic linker. Currently,
the address space of the forking process is copied eagerly. However,
different dataspace types are treated differently. For example, only RAM
dataspaces are copied but no ROM dataspaces.
At the libc side, a setjmp/longjmp-based trampoline mechanism is used to
implement the continuation of the main thread inside the new process.
This procedure is also responsible for re-establishing the relationship
to the new process' parent as well as its Noux session.
This version is still in flux and has many rough edges. It is tied to
OKL4 for now.
The startup procedure of forked processes differs from Genode's
normal process creation by omitting all steps related to ELF loading
and the start of the main thread. To let the process lib support this
distinction, an invalid ELF-binary capability is handled as valid
argument now.
Because 'Noux::Child' implements several abstract Genode interfaces, the
'Genode::' prefix became almost ubiquitous, making the code harder to
read. By importing the 'Genode' namespace into the 'Noux' namespace,
this patch remedies this inconvenience.
To implement fork semantics, we need to customize the bootstrapping of
the newly created process, in particular the startup of the main thread.
The CPU session interface provides a suitable hook. By virtualizing the
CPU connection of the process to core, we can defer (and parametrize)
the startup of the main thread. Furthermore, this enables us to detect
illegal attempts by the Noux process to create threads in addition to
the main thread.
By letting Noux processes talk to service implementations local to the
Noux server, we can track RAM allocations and RM operations. This is
needed as a prerequisite to implement fork.
The 'Child' framework used to perform the transfer of session quota
using 'env()->ram_session()' as hard-wired reference account. When
locally virtualizing the RAM session supplied to the 'Child', this
policy does not work. When closing a session, core would try to transfer
session quota to the virtualized RAM service, which is of course not
possible. This patch makes the reference RAM session configable via the
'Child_policy' interface.
The new function 'Platform_env::reload_parent_cap' triggers a reload
of the parent capability and its respective resources. It is needed
during the bootstrap of a new process forked from an existing Noux
process.
Stefan Kalkowski
Sebastian Sumpf
Norman Feske
Christian Helmuth
Christian Prochaska
Martin Stein
Alexandre Bique