This patch introduces new types for expressing CPU affinities. Instead
of dealing with physical CPU numbers, affinities are expressed as
rectangles in a grid of virtual CPU nodes. This clears the way to
conveniently assign sets of adjacent CPUs to subsystems, each of them
managing their respective viewport of the coordinate space.
By using 2D Cartesian coordinates, the locality of CPU nodes can be
modeled for different topologies such as SMP (simple Nx1 grid), grids of
NUMA nodes, or ring topologies.
When destroying a thread, which was not bound to a protection domain via kernel
primitives beforehand, it is critical to change the pager and exregs the thread
in the destruction process. Therefore, this commit introduces two thread states:
DEAD and RUNNING. On the basis of the thread state, we can decide whether to
reset the thread before destroying it, or not.
The new core-internal 'Address_space' interface enables cores RM service
to flush mappings of a PD in which a given 'Rm_client' thread resides.
Prior this patch, each platform invented their own way to flush mappings
in the respective 'rm_session_support.cc' implementation. However, those
implementations used to deal poorly with some corner cases. In
particular, if a PD session was destroyed prior a RM session, the RM
session would try to use no longer existing PD session. The new
'Address_space' uses the just added weak-pointer mechanism to deal with
this issue.
Furthermore, the generic 'Rm_session_component::detach' function has
been improved to avoid duplicated unmap operations for platforms that
implement the 'Address_space' interface. Therefore, it is related to
issue #595. Right now, this is OKL4 only, but other platforms will follow.
This patch introduces the functions 'affinity' and 'num_cpus' to the CPU
session interface. The interface extension will allow the assignment of
individual threads to CPUs. At this point, it is just a stub with no
actual platform support.
The Cap_mapping abstraction in core shouldn't use a Cap_index directly, but
use Native_capability instead, as it can break reference-counting, as long as
the same Cap_index gets used in a Cap_mapping and a Native_capability. This
commit finally fixes#208.
This commit introduces a Cap_index class for Fiasco.OC's capabilities.
A Cap_index is a combination of the global capability id, that is used by Genode
to correctly identify a kernel-object, and a corresponding entry in a
protection-domain's (kernel-)capability-space. The cap-indices are non-copyable,
unique objects, that are held in a Cap_map. The Cap_map is used to re-find
capabilities already present in the protection-domain, when a capability is
received via IPC. The retrieval of capabilities effectively fixes issue #112,
meaning the waste of capability-space entries.
Because Cap_index objects are non-copyable (their address indicates the position
in the capability-space of the pd), they are inappropriate to use as
Native_capability. Therefore, Native_capability is implemented as a reference
to Cap_index objects. This design seems to be a good pre-condition to implement
smart-pointers for entries in the capability-space, and thereby closing existing
leaks (please refer to issue #32).
Cap_index, Cap_map, and the allocator for Cap_index objects are designed in a way,
that it should be relatively easy to apply the same concept to NOVA also. By now,
these classes are located in the `base-foc` repository, but they intentionally
contain no Fiasco.OC specific elements.
The previously explained changes had extensive impact on the whole Fiasco.OC
platform implementation, due to various dependencies. The following things had to
be changed:
* The Thread object's startup and destruction routine is re-arranged, to
enable another thread (that calls the Thread destructor) gaining the
capability id of the thread's gate to remove it from the Cap_map, the
thread's UTCB had to be made available to the caller, because there
is the current location of that id. After having the UTCB available
in the Thread object for that reason, the whole thread bootstrapping
could be simplified.
* In the course of changing the Native_capability's semantic, a new Cap_mapping
class was introduced in core, that facilitates the establishment and
destruction of capability mappings between core and it's client's, especially
mappings related to Platform_thread and Platform_task, that are relevant to
task and thread creation and destruction. Thereby, the destruction of
threads had to be reworked, which effectively removed a bug (issue #149)
where some threads weren't destroyed properly.
* In the quick fix for issue #112, something similar to the Cap_map was
introduced available in all processes. Moreover, some kind of a capability
map already existed in core, to handle cap-session request properly. The
introduction of the Cap_map unified both structures, so that the
cap-session component code in core had to be reworked too.
* The platform initialization code had to be changed sligthly due to the
changes in Native_capability
* The vcpu initialization in the L4Linux support library had to be adapted
according to the already mentioned changes in the Thread object's bootstrap
code.
This patch unifies the Native_capability classes for the different kernel
platforms by introducing an appropriate template, and eliminating naming
differences. Please refer issue #145.
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.
Norman Feske
Alexander Boettcher
Stefan Kalkowski
Martin Stein
Genode Labs