This is a bigger update and brings the SAPI queue, readv/writev
and realtime scheduling to the sysmobts software. This will greatly
improve the stability of the software under load.
The previous change modified the defconfig to not include the legacy
pseudo terminal support. Increase the version for the sysmoBTS to have
a kernel without old terminal files.
There is no need to have the various pty files clobber the /dev
directory when they are never used. Do not force a PR upgrade
right now as we don't have a good upgreade strategy for the v1
and the BSC config.
This way I just need to install sysmocom-openvpn-config and copy
a matching key to the device for the VPN config. The configuration
file is marked as a CONFFILE, this means that modifications to it
will not be lost during upgrade.
We have some images that need the real wget for upgrades, the SSL
root certificate of CA-Cert but that doesn't use task-sysmocom to
build the image. Move these utilities into the tools task.
The rootfs in the UBI is configured to grow to the full size of the
UBI, currently these are 999 LEBs (when there is no kernel volume)
but we had limited the LEBs the filesystem can hold to 400. After
this change the filesystem will grow from the 42MB to more than 120
MB.
I'm not sure if we really want to have a rootfs of that size but we
can at least experiment with it and see if the failure rate is changing
or not.
For the sysmoBTS we want to include a kernel volume inside the UBI as
a fallback in case the mounting of the main kernel is failing. We still
need to modify the boot command of the bootloader to look at this UBI
and to modify the kernel to include the tun/tap modules so the kernel
will be capable of opening a OpenVPN connection.
Currently one needs to write INHERIT += images_types_sysmo into the
conf/local.conf and add ubi-sysmo to the IMAGE_FSTYPES variable.
systemd is using the depedencies of the LSB section and according to
the debian documentation all applications accessing /var should depend
on the localfs. This makes the watchdog start properly on systemd systems.
This sanity check is included in the used bbclass directly and to
support both Edison and master we will need to remove the custom
python code due a change in indention in the python code.
We should use perf for future performance problems but it might
require more work (e.g. not to install the source into the -dbg
package, adding newt, etc.). But oprofile is not very nice to
use right now anyway.
Ask the buildsystem to install the manpages to the mandir. This will avoid
the following warning.
WARNING: QA Issue: lmsensors-apps: Files/directories were installed but not shipped
/usr/man
RROR: There is a confusing multiline, partially commented expression on line 22 of file /home/ich/firmen/gesmokom/source/meta-sysmocom-bsp/recipes-extra/gpsd/gpsd_3.7.bb (SRC_URI[md5sum] = "52d9785eaf1a51298bb8900dbde88f98").
Some of our target applications require ext3/ext4 support for mounting
an MMC card. While the sysmobts-v1 hardware does not have an external
MMC card slot it is nice to have the same image available for this
hardware.
Install the sysmobts firmware headers on top of the Osmocom toolchain.
This way the SDK can be used to re-compile both the OpenBSC and the
osmo-bts software.
Make sure the -dev package of the femtobts-api has not runtime
dependencies to a library that does not exist.
The name will be resolved to the hash during parse time. This will
not work when not having access to the internet. Use the sha1 sum
to make offline builds work.
This sadly still prints the firmware reload messages. But this is
necessary right now. The fpgadl/dspdl drivers do not log fatal
failures so the 'success' is the only information we have. This
is to be fixed in the driver.
If the first partition of a MMC card is mounted and has a logs/
directory syslog will start to write log messages to the MMC card
in the directory /media/mmcblk0p1.
This issue is documented in PRJ#72. The easiest is to to keep the
old software running as everything else is racy and start-stop-daemon
does not appear to implement the --retry command. Let's see how
far we get with this workaround.
We want to only log a given loglevel, backport the start up script
from OE-Core to get the LOGLEVEL variable. This removes the MARKINT
and adds DROPDUPLICATES and LOGLEVEL.
The kernel has been rebased to target 3.2.40 with various bug
fixes in the upstream work. The sysmobts v2 config has been
changed to allocate space for two more UARTs. These are required
for the multi-trx configuration and they make a warning go away.
* Fix the link timeout handling as found in post-review.
* Fix the set bts attr parsing due a funny parsing error.
* Sleep 10 seconds after the execution, reload DSP and FPGA firmware
We could use "fakeroot do_install() {}" but this will still not
make the spool directory be owned by uucp. Use a postinst script
to make this up. The logic is copied from the cronie bitbake file
and we will see how this works at image creation time.