Add low level event recording infrastructure for system and user
defined events. The infrastructure is able to record high frequency
events such as
* SMP lock acquire/release,
* interrupt entry/exit,
* thread switches,
* UMA zone allocate/free, and
* Ethernet packet input/output, etc.
It allows post-mortem analysis in fatal error handlers, e.g. the last
events are in the record buffer, the newest event overwrites the oldest
event. It is possible to detect record buffer overflows for consumers
that expect a continuous stream of events, e.g. to display the system
state in real-time.
The implementation supports high-end SMP machines (more than 1GHz
processor frequency, more than four processors).
Add a new API instead. The implementation uses per-processor data
structures and no atomic read-modify-write operations. It is uses
per-processor ring buffers to record the events.
The CPU counter is used to get the time of events. It is combined with
periodic uptime events to synchronize it with CLOCK_REALTIME.
The existing capture engine tries to solve this problem also, but its
performance is not good enough for high-end production systems. The
main issues are the variable-size buffers and the use of SMP locks for
synchronization. To fix this, the API would change significantly.
Update #3665.
Allocate the per-CPU data for secondary processors directly from the
heap areas before heap initialization and not via
_Workspace_Allocate_aligned(). This avoids dependency on the workspace
allocator. It fixes also a problem on some platforms (e.g. QorIQ) where
at this early point in the system initialization the top of the RAM is
used by low-level startup code on secondary processors (boot pages).
Update #3507.
This configuration option was undocumented and not really helpful. To
debug issues in <rtems/confdefs.h> it is better to save the
preprocessesd file (GCC "-save-temps" option) and use the GCC "-Wp,-dD"
pre-processor option.
Statically allocate the objects information together with the initial
set of objects either via <rtems/confdefs.h>. Provide default object
informations with zero objects via librtemscpu.a. This greatly
simplifies the workspace size estimate. RTEMS applications which do not
use the unlimited objects option are easier to debug since all objects
reside now in statically allocated objects of the right types.
Close#3621.
Use the maximum ID for the ID to object translation. Using the maximum
ID gets rid of an additional load from the object information in
_Objects_Get(). In addition, object lookups fail for every ID in case
the object information is cleared to zero. This makes it a bit more
robust during system startup (see new tests in spconfig02).
The local table no longer needs a NULL pointer entry at array index
zero. Adjust all the object iteration loops accordingly.
Remove Objects_Information::minimum_id since it contains only redundant
information. Add _Objects_Get_minimum_id() to get the minimum ID.
Update #3621.
Rename Objects_Information::allocation_size in
Objects_Information::objects_per_block. Adjust integer types in
_Objects_Shrink_information() and _Objects_Free().
Update #3621.
Rename Objects_Information::size to Objects_Information::object_size.
Change its type from size_t to uint16_t and move it to reduce the size
of Objects_Information.
Update #3621.
The sbintime_t is a non-POSIX type and not visible if strict standard
options are selected.
Move implementation details from <rtems/score/timestamp.h> to
<rtems/score/timestampimpl.h>.
Update #3598.
The following rtems_cache_*_processor_set() cache manager API functions
are exotic, complex, very hard to use correctly, not used in the RTEMS
code base, and apparently unused by applications.
Close#3622.
Tracking references lets us manage when an object file can be
unloaded. If an object file has references to it, it cannot be
unloaded.
Modules that depend on each other cannot be unloaded.
Updates #3605
The use of separate text and data results in uninitialised variables
being placed in the common section. There is no section in ELF for
the common variables so the loader needs to create the section and
allocate the variables in that section. This patch does that.
The patch adds a second pass over the symbols.
The issue can also be seen as a section 65522 error.
Updates #3604
The RTEMS_USE_16_BIT_OBJECT define is not set by an RTEMS port. Remove
support for 16-bit object identifiers. If someone really wants to use
RTEMS on a 16-bit target, then it is better to use self-contained
objects instead of playing around with object identifier optimizations.
Update #3603.
Sebastian Huber
Marçal Comajoan Cara
Chris Johns