GitLab

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Provide choices to switch clock between MONO and REALTIME for getting
different timestamp since gpio driver already provide choosable service.
Signed-off-by: Hongzhan Chen <hongzhan.chen@intel.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

When test is interrupted, avg latency need to be calcuated once
to avoid large latency showing even though it may not be quite
accurate.
Signed-off-by: Hongzhan Chen <hongzhan.chen@intel.com>
[Jan: tuning of wording and output format]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Ensures build-testing. We need to enable a couple of dependencies of the
driver for that.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This ensures that Xenomai GPIO drivers are built. It's a dependency of
XENO_DRIVERS_GPIO that is not automatically enabled.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Add support for Intel Cherryview pin controller driver
Signed-off-by: Hongzhan Chen <hongzhan.chen@intel.com>
[Jan: renamed 'compat_array' to 'label_array', folded in right removal function]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

To find gpiochip for non-OF platforms like x86
Signed-off-by: Hongzhan Chen <hongzhan.chen@intel.com>
[Jan: adjust parameter name 'compat' to 'label', reindent]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Rename interface name because it is independent of OF platform and
non-OF platform would also call it to remove rtdm gpio chip device.
Signed-off-by: Hongzhan Chen <hongzhan.chen@intel.com>
[Jan: choose conflict-free rtdm_gpiochip_remove_by_type as name]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This is in line with other drivers and should avoid surprises like [1]
in the future.

[1] https://xenomai.org/pipermail/xenomai/2021-June/045661.html

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The API's default will change, so select monotonic explicitly.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This allows to compare timestamps of GPIO pins against other devices and
ensures that the stamps are synchronized cross-systems in case PTP is in
use (dovetail-based kernels). Keep default to monotonic for now.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Especially 4.19 is not 2038 safe and 32 bit kernels do not have
CONFIG_64BIT_TIME set, which ends up in __kernel_timespec being defined
in a non y2038 safe way.

cobalt_get_timespec64 will copy the upper bytes of the sec field into
the nsec field, which is always zero.

Testing y2038 syscalls on a non y2038 safe kernel doesn't make sense,
hence limiting the test runs to kernels with 2038 support.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Make sure that named semaphores as well as mqueues are exported to the
procfs.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
[Jan: delay procfs_exists to account for asynchronous /proc update]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The exit code of the child process was never correctly reported to the
parent process. Hence, the status code check was wrong at all. The
variable previously named subprocess_status was not shared between both
processes.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Exporting named semaphores to /proc/xenomai/registry/posix/sem helps
to identify resource leaks by making registered semaphores visible. The
content of all created virtual files is empty. Removal of such files is
not possible and reports "operation not supported".
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Exporting created mqueues to /proc/xenomai/registry/posix/mqueue helps
to identify resource leaks by making mqueues visible. The content of all
created virtual files is empty. Removal of such files is not possible
and reports "operation not supported".
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Will be used later for exporting registry entries, namely named
semaphores as well as mqueues to
/proc/xenomai/registry/posix/{mqueue,sem}
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The refcnt field was never initialized and triggered a BUG() in
vfile_regular_release (vfile.c) when closing / releasing such a virtual
file.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Allows re-using xnregistry_vfreg_ops. All other file options were
already available.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Neither needed nor correct for named semaphores where xnregistry_remove
will trigger a flush_work when running over the secondary domain.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The values in xnclock_gravity are in ticks, not nanoseconds. Make sure
nanosecond values are converted before storing them.

This was originally broken by bc761d8c

, for all architectures except
arm, arm64 and powerpc. During the dovetail port, this also sneaked into
the arm and arm64 headers.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Adding the following trace events to make registry-usage traceable:
  - registry_enter   (object gets added to the registry)
  - registry_unlink  (object is made anonymous)
  - registry_remove  (object is removed from the registry)

These trace events are helpful if there is a need to track down
resource leaks triggered by an Xenomai based application.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
[Jan: add missing include]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Allow the user to disable demo and testsuite to decrease compilation
time and save disk space.
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
[Jan: slightly more verbose help for demo]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This should help to catch breakages of the compat case.

We only need to build userland as the kernel is already covered by the
x86 job.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

According to POSIX message queues have kernel persistence, which means
they are not cleaned up until the next reboot.

The test has a mq_unlink() as "precondition", so it will clean up
failed test runs silently but /proc/xenomai/registry/usage will still
report a "leak" once the test has completed.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Based on the tests provided by Song with the following adjustments:
  - #includes cleaned up
  - making sure all resources used during tests have are cleaned up
  - Switched to CLOCK_MONOTONIC for time measurement
    __cobalt_event_wait sets the time mode to XN_ABSOLUTE, so we have to
    use CLOCK_MONOTONIC for comparisons
  - Adding a testcase for calling event_wait64 with a zero timeout,
    which should come back immediately with EWOULDBLOCK/EAGAIN

While at it: Adding a missing cobalt_monitor_destroy() in
             test_sc_cobalt_monitor_wait64
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

If sizeof time_t bigger than 4, which means glibc supports
64bit timespec, go to event_wait64.

otherwise, go to original event_wait.
Signed-off-by: Song Chen <chensong_2000@189.cn>
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Add a syscall specific for event_wait with 64bit time_t.
Signed-off-by: Song Chen <chensong_2000@189.cn>
[Florian: Tracing, reformatting/rebasing]
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The time measurement is a bit different to the other y2038 related
measurements we do. Using a monitor with CLOCK_REALTIME requires an
absolute timeout.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

If sizeof time_t bigger than 4, which means glibc supports
64bit timespec, go to monitor_wait64.

otherwise, go to original monitor_wait.
Signed-off-by: Song Chen <chensong_2000@189.cn>
[Florian: Fix warnings reported by checkpatch.pl, coding style]
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Adding monitor_wait64. Patch is based on v1 sent out by Song and was
re-worked to apply on top of current next/master.

__cobalt_monitor_wait() is now validating the struct timespec64 it
received from userland. That was not the case up to now.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Heavily inspired by the tests sent from Song but with the following
modifications:
  - return -EAGAIN instead of -ETIMEOUT (posix spec) in case of a
    timeout
  - Fix time calculations for timeout detection
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

If sizeof time_t bigger than 4, which means glibc supports
64bit timespec, go to sigtimedwait64.

otherwise, go to original sigtimedwait.
Signed-off-by: Song Chen <chensong_2000@189.cn>
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Add a syscall specific for sigtimedwait with 64bit
time_t.
Signed-off-by: Song Chen <chensong_2000@189.cn>
[Florian: Tracing, fix return value EAGAIN vs. ETIMEOUT]
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Extending the test suite for mq_timedsend64() and mq_timedrecv64()
tests.
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

If libc reports time64_t support mq_timedreceive is dispatched to the
time64_t based syscall.
Signed-off-by: Song Chen <chensong_2000@189.cn>
[Florian: Reformat commit message]
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Add a syscall specific for mq_timedreceive64 with 64bit time_t.
Signed-off-by: Song Chen <chensong_2000@189.cn>
[Florian:
  - Reformat commit msg
  - relocate code
  - tracing
  - fix syscall declarations
]
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Dispatch mq_timesend to the 64bit aware syscall if libc reports
time64_t support.
Signed-off-by: Song Chen <chensong_2000@189.cn>
[Florian: Rephrase commit message]
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Add a syscall specific for mq_timedsend with 64bit time_t.
Signed-off-by: Song Chen <chensong_2000@189.cn>
[Florian: Reformat commit msg, relocate code, tracing, fix compat decl]
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This patch was based on the patch sent out by Song and reworked.
Especially
  - switched from CLOCK_MONOTONIC to CLOCK_REALTIME
    According to POSIX this service is based on CLOCK_REALTIME

  - Fixed some mutex leaks / missing pthread_mutex_destroy()

  - Removed some magic numbers used for making sure the syscall does
    not return too early

  - Fixed several mutex deadlocks. Once mutex_timedlock() was
    successful following calls will deadlock (as long as no special
    DEADLOCK mutex is being used)
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
[Jan: resolve compiler warning about uninitialized t1, fix join result evaluation]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>