GitLab

Rebase trace statements in the Cobalt core over the generic tracepoint
API. A few former trace markers which did not help in analysing the
dynamic behavior of the system have been dropped in the process.

This work is originally based on Jan Kiszka's trace events conversion
patch series for Xenomai 2.6.

Make sure to have the destination string null-terminated on character
overflow.

These are redundant with CONFIG_XENO_DEBUG, CONFIG_XENO_DEBUG_FULL
respectively, available from xeno_config.h.

This flag is returned by rt_alarm_inquire() into the RT_ALARM_INFO
structure, reflecting the current state of the queried object. If
non-zero, the alarm is enabled (i.e. started).

Cobalt requires a 3.10 kernel or later.

Fix the valid range of runlevels in the O(1) scheduler queue to
[XNSCHED_RT_MIN_PRIO..XNSCHED_RT_MAX_PRIO], as threads may cross
policies freely during PIP boosts, therefore all possible priority
level must be valid for all scheduler queues, regardless of the
policy.

Defining XENO_CONFIG_OUTPUT in the environment causes the application
program linked against libcobalt to dump the configuration settings
used for building the user-space programs/libraries. The program exits
immediately afterwards.

The builtin --dump-config switch is introduced to dump the
configuration information from any copperplate-based application.

We need a way to retrieve the configuration settings used for building
the user-space stack.

To this end, those settings are snapshot during the build process and
stored in a NULL-terminated array of pointer to strings, so that any
client code linking against libboilerplate can retrieve them.

Since dropping the lock may result in the memory holding the restored
cancel state to become stale, we'd better read it prior to unlocking.

Typically, this case would happen with call sites like
threadobj_unlock(), fetching the cancel state word from the TCB of a
thread which pends on its own lock, prior to finalizing. If that
thread preempts threadobj_unlock() (as a result of the lock drop) then
exits, the cancel state may become stale before threadobj_unlock()
refers to it.

This change fixes a crash when a zero-sized private area is specified
for the driver:
http://www.xenomai.org/pipermail/xenomai/2014-April/030628.html

More fixups are included to sanitize this handler a bit further.

This fixes a long-standing bug which caused random duplicate output
from the systemfs handlers, due to misusing the fuse read() method.