In case of CONFIG_XENO_PSHARED, the main thread's tcb is allocated on
the shared heap. Normally, threadobjs are released via the destructor
of threadobj_tskey. But this is not called when the main thread
terminates so that we leak the object. Fix that by installing an atexit
handler that only needs to perform the release now.

Reported-by: Mauro S. <mau.salvi@tin.it>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This includes can trigger warnings when -Wsign-conversion is set. All
conversions are fine, so silence them by explicit casts.

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

Extracted from upstream commit e2bd9c76c89fbe25df351fc5902cbbcca6a7d372.
We can apply this unconditionally as the adjfine interface was already
present in 5.10, the oldest supported kernel here.

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

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

The kernel claims this namespace, and we now collide on timer_delete
with 6.2.

At this chance, drop some inappropriate used 'inline' declarations from
large function used more than once.

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

Now that we are running both pipelines on the same server, we can also
make use of native gitlab-ci trigger mechanism. This both obsoletes the
need to configure a token and avoids starting a real job on a real
runner, just for calling curl.

This also improves the visualization of the two related pipelines in the
gitlab UI.

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

Tobias Schaffner authored Dec 13, 2022 and Jan Kiszka committed Dec 14, 2022

Make the rtuo thread switch between primary and secondary mode while in the
RTTST_RTIOC_SWTEST_SWITCH_TO syscall.

Signed-off-by: Tobias Schaffner <tobias.schaffner@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Tobias Schaffner authored Dec 12, 2022 and Jan Kiszka committed Dec 12, 2022

Switches to secondary mode do not only occur when a relax is explicitly
called via the cobalt api but also e.g. when a linux syscall is called in
primary mode.

Also test migrations to secondary mode based on linux syscalls in the rtuo task.

Signed-off-by: Tobias Schaffner <tobias.schaffner@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Tobias Schaffner authored Dec 12, 2022 and Jan Kiszka committed Dec 12, 2022

Split the set_mode method in separate methods for switching to primary and
secondary mode for improved readability.

This also fixes a wrong evaluation of the mode variable that lead to never
checking the FPU registers in secondary mode.

Signed-off-by: Tobias Schaffner <tobias.schaffner@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

That should give the auto-cancel option its desired power.

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

According to our workflow, it was always already tested via 'next'.

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

Avoids having to set PIPELINE_KERNEL for every job while it is the same
for them all by now.

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

This also gives a chance to prepare the sorting rule for the transition
v6.9 -> v6.10. It was likely not even correct when it was once
introduced for ipipe rules.

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

This was forgotten in 5af4f166

.

CC: Gunter Grau <gunter.grau@philips.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This is an RTnet driver, so it has no need to interact with NAPI as the
kernel provides. Cleaning up also fixes building over 6.1.

CC: Gunter Grau <gunter.grau@philips.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Align with mainline regarding DMA mask setting to allow compilation
against recent kernels up to 6.1.

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

prandom_u32 was dropped from 6.1, but get_random_u32 is not oob-safe. So
we only use it for initializing a seed and then rely on our own version
of next_pseudo_random32. The latter is still available via
linux/prandom.h, but given that there are no more in-tree users, it
could be dropped at any point in the future.

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

Zqiang authored Sep 21, 2022 and Jan Kiszka committed Sep 26, 2022

When kernel enable CONFIG_HARDENED_USERCOPY option, since not marking
rtskb_slab_pool cache is usercopy type, will trigger the following
calltrace.

[  594.413602] invalid opcode: 0000 1 PREEMPT SMP PTI IRQ_PIPELINE
[  594.413603] CPU: 2 PID: 492 Comm: rtcfg Tainted:
[  594.413603] IRQ stage: Linux
[  594.413604] RIP: 0010:usercopy_abort+0x77/0x79
[  594.413610] Call Trace:
[  594.413610]  __check_heap_object+0xa6/0xc0
[  594.413611]  __check_object_size+0x136/0x150
[  594.413611]  copy_stage_1_data+0x48/0x70 [rtcfg]
[  594.413612]  rtnet_rtpc_dispatch_call+0x151/0x320 [rtnet]
[  594.413612]  ? cleanup_cmd_del+0x70/0x70 [rtcfg]
[  594.413612]  ? wait_woken+0x70/0x70
[  594.413613]  rtcfg_ioctl+0xa0/0x250 [rtcfg]
[  594.413613]  rtnet_ioctl+0xdf/0x170 [rtnet]
[  594.413613]  __x64_sys_ioctl+0x83/0xb0
[  594.413613]  do_syscall_64+0x39/0x50
[  594.413614]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

This commit make rtskb_slab_pool cache being an usercopy source.

Signed-off-by: Zqiang <qiang1.zhang@intel.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Florian Bezdeka authored Sep 20, 2022 and Jan Kiszka committed Sep 26, 2022

This helps to identify if the testsuite was run in compat mode and if
y2038 support has been requested. Useful when looking at CI logs and
trying to understand the runtime conditions that caused a failure.

Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Florian Bezdeka authored Sep 20, 2022 and Jan Kiszka committed Sep 26, 2022

The rtnet interface was lacking compat support for the following ioctl
commands:
  - _RTIOC_BIND
  - _RTIOC_CONNECT

This patch is heavily inspired by Konstantin's proposal but migrated
to some rtnet APIs that allow us to stay away from
  #ifdef CONFIG_XENO_ARCH_SYS3264
sections.

Reported-by: Konstantin Smola <ksmola51@gmail.com>
Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Florian Bezdeka authored Sep 20, 2022 and Jan Kiszka committed Sep 26, 2022

Once Xenomai was running as compat application (=32bit applicaiton over
64bit kernel) the can smokey test failed. This patch fixes the test by
implementing the necessary compat bits.

Affected ioctls:
  - _RTIOC_BIND
  - _RTIOC_SETSOCKOPT

Affected syscalls:
  - recvmsg
  - sendmsg

Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Florian Bezdeka authored Sep 20, 2022 and Jan Kiszka committed Sep 26, 2022

Several drivers will have to read and write data structures that have
a different memory layout when handling compat applications. Moving
the decission if the compat path has to be taken into the rtdm layer
helps to keep the drivers simple and clean.

Added helpers
  - for reading and writing struct iovec
    (and its compat representation)
  - for reading and writing struct _rtdm_setsockaddr_args
    (and its compat representation)

Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Hongzhan Chen authored Sep 14, 2022 and Jan Kiszka committed Sep 14, 2022

ported basic network function not including TSN feature.

Signed-off-by: Hongzhan Chen <hongzhan.chen@intel.com>
[Jan: enable driver in CI]
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Florian Bezdeka authored Aug 18, 2022 and Jan Kiszka committed Sep 05, 2022

Running rtnet for compat applications failed because the compat ioctl
interface was not activated.

Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Norbert Lange authored Jul 25, 2022 and Jan Kiszka committed Jul 25, 2022

clang will never support 'variable length array in structure',
and there does not seem to be a reason for this weird construct.

Signed-off-by: Norbert Lange <norbert.lange@andritz.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Mauro Salvini authored Jul 22, 2022 and Jan Kiszka committed Jul 24, 2022

Before this change, on TDMA master stations using the TDMA_CONFIG file
the NETMASK parameter was not used and netmask was assigned automatically
by [rt]ifconfig (deriving from IP class of the address assigned to the station),
both on master and on slaves.
Instead, on TDMA master stations not using the TDMA_CONFIG file, the NETMASK
parameter specified in rtnet.conf was used both for master and slaves.
This commit uniforms the NETMASK usage between configurations with and without
TDMA_CONFIG file.

Signed-off-by: Mauro Salvini <mau.salvi@tin.it>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This tests the latest v5.xx[.y]-dovetail-rebase queue in addition to the
5.10 and 5.19 variants. Should help to find build compatibility issues
earlier.

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

Changed in 5.18.

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

The kernel generalized spi_master to spi_controller and renamed several
services. Follow this to avoid breakage when the legacy wrappers are
eventually removed. Should come without functional changes.

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

The legacy one has been retired, and the new one was already available
with 5.10.

CC: Philippe Gerum <rpm@xenomai.org>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

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

Latest kernel with around gcc-10 starts to complain about the type casts
done for error stubs, syscall handlers and nomac_dev_openclose. Those
casts remain fine, so suppress them by casting to the special
"void (*)(void)" type first.

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

The legacy API has been finally dropped, and the new one was "already"
available with 5.10.

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

The legacy API has been finally dropped, and the new one was "already"
available with 5.10.

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

The legacy API has been finally dropped, and the new one was "already"
available with 5.10.

We keep the GFP_ATOMIC allocation type of the original wrapper, even if
some caller likely do not need them - safer and simpler than checking
them all.

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

The legacy API has been finally dropped, and the new one was "already"
available with 5.10.

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

Latest kernel complain about accessing dev_addr directly and push us
toward the accessor functions. Use them and provide wrappers for
kernels prior to 5.15.

CC: Gunter Grau <gunter.grau@philips.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Those moved from the central header into the upstream driver code. We
have to do the same.

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

Was forgotten in a50b60de

.

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

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