By removing the driver on ifdown and loading it on ifup we workaround some
problems found in long run suspend stability testing.
https://jira.digi.com/browse/DEL-1084
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
This updates the driver to upstream AR6003-ISC-v3.4p6-b3.4.0.225.
https://jira.digi.com/browse/DEL-1127
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
This adds the calibration file for US regulatory (still missing the
world-regulatory calibration file)
https://jira.digi.com/browse/DEL-1102
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
User-space has been updated in meta-fsl-arm for the general availability
(GA) FSL release, so use the kernel branch with the GA support.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
The 'v3.10/master' master branch has been updated with the latest
general availability (GA) release from FSL, but the user-space is not
ready yet, so using the new kernel the build is broken:
imx-lib: error: 'PXP_IOC_GET_CHAN' undeclared
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Needing to set the branch where a specific SHA1 revision is contained is
a unnecessary constraint that can be bypassed setting 'nobranch=1' in
the SRC_URI.
We do that for all the external repos (GitHub) plus the internal ones
for the im-xbootlets and the atheros kernel module (as those two repos
are stable enough and we are not developing actively on them).
For U-Boot and Linux we keep the SRCBRANCH framework for internal repos
as this is needed for the AUTOREV yocto feature.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Starting with Yocto 1.6 there is a need to configure in which branch a
specific SHA1 revision is included. We need a framework to set such
branches both for internal git repos and for external ones (Github).
If the branch is not specified in the SRC_URI, the do_fetch task may
fail even though the SHA1 revision exists in the repository.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Otherwise building Yocto 1.6 for ccardimx28js fails with:
ERROR: Logfile of failure stored in:
/home/jviguera/projects/yocto/master/gui/ccardimx28js/tmp/work/ccardimx28js-dey-linux-gnueabi/kernel-module-atheros/1.0-r1/temp/log.do_fetch.38442
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Fetch the source from the dey-1.6/maint branch which contains a fix
for a kernel oops introduced by the Yocto GCC 4.8.1 compiler.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
This workaround is not needed anymore as the git fetcher bug is fixed in
poky's master branch.
This reverts commit 982af63902.
https://jira.digi.com/browse/DEL-996
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
PR was based on meta-fsl-arm's 'linux-imx.inc' INC_PR, but this was
removed after Dora, so we need to remove PR from our kernel recipes to
avoid a bitbake parser error:
${INC_PR}.0/git')} exception SyntaxError: EOL while scanning string literal
https://jira.digi.com/browse/DEL-996
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
linux-dey.inc is a copy of meta-fsl-arm's linux-imx recipe file, so
instead of copying the whole recipe just use that (including the file)
and code the differences.
While on it, add RDEPENDS_kernel-base="" to remove the dependence of the
rootfs image recipes on the linux image. We don't need/use the kernel
image in the rootfs as we load it from a different partition. This makes
rootfs images a bit smaller.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
change the directory (cd) and make the symbolic link in a subshell to
preserve the current working directory for any subsequent command.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
bitbake git's fetcher uses 'ls-remote' to get the branch revision to
checkout. If the repository has two branches ending equally the 'git
ls-remote' just picks the first one regardless of that being the correct
one.
In our case we were bitten by this bug because our git repo has:
...sha1-revision... refs/heads/android/v3.10/master
...sha1-revision... refs/heads/v3.10/master
and the fetcher was using the android branch instead of linux's
v3.10/master.
A patch has been sent upstream but while that gets applied to Dora
maintenance branch the workaround here is to set the absolute path in
the kernel's bitbake recipe.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
The do_update_dts function was hardcoded for ccardimx28js. We need to
make it mxs specific so the recipe can be used for other platforms that
do not share the same device tree configuration (for example the
upcoming ccimx6js)
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
To be sure the 'cfg80211' module is settled down before the
'ath6kl_sdio' module tries to use it.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Version v3.4p6 seems to have some problems with WEP encription. Version
v3.4p4 seems to work better with WEP, so revert to that version.
https://jira.digi.com/browse/DEL-843
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
There seems to be a race condition in the load of the 'cfg80211' and
'ath6kl_sdio' modules. If we try to load the atheros module before the
cfg80211 is settled down, the driver crash.
The workaround is just separate the load of both modules in two separate
'modprobe' commands and add a small delay between them.
https://jira.digi.com/browse/DEL-814
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Not well supported yet and cleans a debug line:
ath6kl: ath6kl_cfg80211_connect: sme->mfp = 0
https://jira.digi.com/browse/DEL-815
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
The Linux kernel revision was splitted from main bitbake recipe to its
own linux-dey-rev file so that file could be used for linux-headers
recipe to build the toolchain using custom kernel headers (from our
repository).
That is not the recommended way to build a toolchain:
http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=babbf7a46acaefd9b36031483cafce053f607e66
and we are not using our kernel headers to build the toolchain anymore,
so just melt the linux-dey-rev files into the linux-dey bitbake recipes
for cleanness and easier maintenance.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
The wireless module occasionally fails to load in kernel v3.10.
If that is the case, unload the driver and load it again (up to
five retries).
Notes:
- Only the ath6kl_sdio module needs to be unloaded and reloaded.
- The chip is not issued a power cycle via CHIP_PWD_L because
this demonstrated to cause failure, instead of improving
behaviour.
- The 2-seconds sleep after unloading the driver is a security
measure to avoid other problems seen during test cycle.
Signed-off-by: Hector Palacios <hector.palacios@digi.com>
Reviewed-by: Javier Viguera <javier.viguera@digi.com>
https://jira.digi.com/browse/DEL-820
Committed a patch in linux-dey repo to fix perf build. At the same
time update linux-dey revision to get the perf fix.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
They conflict.
Warn the user about the problem and let they change the device tree
manually.
https://jira.digi.com/browse/DEL-840
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
list.h: In function 'INIT_HLIST_NODE':
list.h:582:12: error: 'NULL' undeclared (first use in this function)
list.h: In function 'hlist_move_list':
list.h:661:15: error: 'NULL' undeclared (first use in this function)
list.h: In function 'hlist_move_list':
list.h:661:15: error: 'NULL' undeclared (first use in this function)
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Those boolean helper variables can be used in different recipes so move
them to a common place.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Also change the default ccardimx28js variant to 'web1' (wireless, second
ethernet, bluetooth and 1-wire)
https://jira.digi.com/browse/DEL-819
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Most SRC_URI defaults to DIGI_LOG_GIT. This patch sets a more generic
name (DIGI_GIT) that defaults to DIGI_LOG_GIT, but is configurable by
someone who might want to tune it (DIGI_MTK_GIT, etc.).
Signed-off-by: Seth Bollinger <sethb@digi.com>
To get the latest changes: logo, mac-address, wireless...
Also synchronize the defconfig with the one in the kernel tree.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
If people are creating derivative machines in their layer, make
COMPATIBLE_MACHINE more flexible. Use SOC_FAMILY instead of specific
machine types. This allows users to more easily share these recipes.
https://jira.digi.com/browse/DEL-676
Signed-off-by: Seth Bollinger <sethb@digi.com>
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Modify the atheros and bluetooth init scripts to fetch the hwid details
from the device tree.
https://jira.digi.com/browse/DEL-572
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
In Linux 2.6.35, the kernel module accepting the parameters on load
was 'ath6kl_sdio' while in 3.10 is 'ath6kl_core'.
So we need to apply the load parameters to either module depending on
the kernel version.
This is solved by configuring the modprobe configuration file
'/etc/modprobe.d/atheros.conf' with the parameters to the correct
module instead of passing them in the command line.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
With the old version of the kernel (2.6.35) we build an external atheros
kernel module, while with the new kernel (3.10) the atheros driver is
included with the kernel sources.
However, the new driver needs the firmware files and the driver load
scripts that are stored with the old driver.
So put a dependence on kernel-module-atheros regardless of the kernel
version and then rework the kernel-module-atheros recipe to build the
whole driver for the old kernel or only the firmware for the new kernel.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
The UBOOT_LOADADDRESS and UBOOT_ENTRYPOINT variables are only used in
the kernel recipe, so remove from the machine configuration files.
Also, use the defconfig from the kernel tree which compiles all the
support by default and expects the device tree to be configured with
the desired support. Hence, remove all the empty configuration fragments
files.
https://jira.digi.com/browse/DEL-577
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
ERROR: Fetcher failure: Unable to find file file://atheros anywhere
The FILESPATH variable (used as the search path for files referred to
in SRC_URI) is now (dylan) more restrictive, so move the files where
they can be found.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
This is to work as in previous kernels. For now the fragments are empty
and they will have to be individually included and tested.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Copy including folders. This was causing problems in kernel version 3.10
as the directory included folders that were omitted in the copy and caused
the do_install to fail.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Remove support that is to be added as a config fragment:
- FB support.
- LCD backlight.
- Sound.
Add configuration that was also enabled in previous kernels.
- Serial, ethernet and storage USB gadgets.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
We need separate linux-libc-recipes if we want to support different
kernels.
On the 1.4 meta-fsl-arm layer this recipe will disappear, so review then.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Placing all of the revision info in a common include we avoid the common
issue of forgetting to update the linux-libc-headers append recipe when
the kernel is updated.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
The external repositories have been cleaned up and commits have been
squashed so the sha1s need to be updated.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
The public branches of those repositories have been cleaned up so
the external sha1s have changed.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
The final product will be called "Digi Embedded for Yocto" or DEY for
short.
This commit renames all relevant files.
https://jira.digi.com/browse/DEL-474
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
https://jira.digi.com/browse/DEL-472
While on it, fix minor bug: uImage-<platform> symbolic link was not
being created on deployment directory.
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Another option would be to split it up in machine folders, but it is
similar enough in all platforms not to need this for the time being.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Fix the kernel wireless configuration fragment, and add a couple of kernel
patches to set up the mmc0 interface to be used with the wireless and
remove an incorrect pin configuration.
There are similar patched to this on the gateways/master branch related to
the vehicle bus adapter product.
meta-digi-arm/conf/machine/cpx2.conf
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
It is now based on the ConnectCard defconfig instead of the gateways
product one. This makes more sense in this layer, and upper layers
can overwrite it if needed.
https://jira.digi.com/browse/DEL-462
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
This manually reverts commit ecfdf6ae23.
Once mdev was removed as a dependency from del-image-graphical, we need
udev to load the firmware files for the wireless drivers. This job was
being done by mdev.
The reason for the commit in the first place was that both mdev and udev
were attempting to load the firmware files, so we changed the udev
rule to ignore the atheros driver. This was only needed when mdev and
udev were both running.
https://jira.digi.com/browse/DEL-450https://jira.digi.com/browse/DEL-450
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
I have split the machine configuration in two:
-The module machine configuration include, for example ccimx53.inc
-The machine configuration (for example cimx53js.conf) which includes
the variant configuration.
This will allow for customers to create their own machine keeping the
module include file.
The variant can be configured in the local.conf file with a
MACHINE_VARIANT variable.
The variant names are made up of the interface initials as follows:
w: wifi
e: externel ethernel
b: bluetooth
1: 1-wire
a: accelerometer
g: GPU
v: VPU
128: 128MB RAM
4k: 1024MB NAND (only for MX51 JTAG flash)
For details regarding the supported variants, each machine configuration
file contains a comment with a variants table.
https://jira.digi.com/browse/DEL-330
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
* don't use a login shell to run the script. It overrides the
environment that comes from the caller script.
* optimize MAC address parsing
* enable 'P2P' and softmac on driver's load
https://jira.digi.com/browse/DEL-412
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
This version of the driver can be redistributed, so remove the
build-from-objects part of the recipes.
https://jira.digi.com/browse/DEL-412
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
They all have a copyright and the all do:
PR = "${DISTRO}.${INC_PR}.0"
Then recipes which include them include:
PR = "${DISTRO}.${INC_PR}.0"
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
All bbappend files will start as follows:
# Copyright (C) 2013 Digi International.
PRINC := "${@int(PRINC) + 1}"
PR_append = "+${DISTRO}"
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
On a release, the kernel will not be built from a git repo but
from a source tarball.
Make the default to build from a tarball.
https://jira.digi.com/browse/DEL-362
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Not yet finished as not all the cpx2 variants support wireless.
This has to be enabled per-variant.
https://jira.digi.com/browse/DEL-134
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
This was causing module loading issues on del-image-graphical.
Fixes https://jira.digi.com/browse/DEL-338
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Following the guidelines in kernel documentation:
linux-tree/Documentation/kbuild/modules.txt
https://jira.digi.com/browse/DEL-134
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Create a symlink on the deploy folder that matches the standard uboot
name for the kernel.
Fixes partially https://jira.digi.com/browse/DEL-283
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
udev will attempt to load the atheros kernel modules because it defines
a modalias that matches an sdio bus event.
In a race condition with the script in if-pre-up, the module will crash.
Recent versions of udev have lost the ability to add an ignore_device
option to the rules, and mdev in del-image-minimal has no module
autoloading ability.
The commit adds a modprobe install rule that does nothing, and allows
the if-pre-up script to load the module as it used to.
Fixes https://jira.digi.com/browse/DEL-277
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
The del-examples-sahara now depends on the kernel modules it needs to
work and these are autoloaded at boot.
Fixes DEL-274.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
For this the CCIMX51 and CCIMX53 kernels are configured to build the
mma7455l module and it is included as a dependency of the del-examples-
accelerometer application.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
This conflicts with user led 2 used in the GPIO example add and with
other interfaces needed in the del-image-graphical image.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Wakeup interrupts were not happening upon USB stick plug
events during normal work. This had the effect of the
interrupt happenign when going to suspend and usually when
the driver is not able to handle it, causing a nobody cared
message from the kernel.
https://jira.digi.com/browse/CCIMX28-326
Signed-off-by: Hector Palacios <hector.palacios@digi.com>
Just a cosmetic name change for the merged wired and wireless platforms.
This and related commits fixes https://jira.digi.com/browse/DEL-188.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Also enable CONFIG_BRIDGE in kernel configs (which is required to use
brctl)
https://jira.digi.com/browse/DEL-156 #resolve
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
Enable RFKILL. Otherwise on boot we get an error message:
rfkill: Cannot open RFKILL control device
Signed-off-by: Javier Viguera <javier.viguera@digi.com>
We need the kernel configuration file in /boot for some of the postint
scripts that aim to do smart things like autoconfigure the interfaces file.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
The bootloader is passing the mac address and wdmode in the kernel command
line.
The driver startup script reads and and sets them.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
The bootloader is passing the mac address in the kernel command line.
The driver startup script reads it and sets it.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
These new defconfigs are needed after merging the wireless and non wireless
products in the linux kernel source.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
We need to split the EXTRA_OEMAKE flags as the do_make_scripts stage
fails if we pass a "M=" argument.
We override module_do_compile and module_do_install to add the extra
make flags as I have not found a working way of overriding EXTRA_OEMAKE.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
This included the recipe where the driver is cloned from git and compiled
from source.
The recipe also copies the firmware files and the if-pre-up scrip.
Pending is the recipe to link from precompiled objects in a .tar.gz file.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
This is the same as the meta-fsl-arm layer and allows for more recipes
for different kernel versions to be considered.
Signed-off-by: Alex Gonzalez <alex.gonzalez@digi.com>
Javier Viguera
Alex Gonzalez
Hector Palacios
Seth Bollinger
Francis Sreeves