Age | Commit message (Collapse) | Author |
|
With newer QNX kernels, there are issues unless the MAP_SHARED/MAP_PRIVATE
flag is added to the mmap64 call alongside the MAP_PHYS parameter
in rpmsg-rpc.
This patch adds the MAP_SHARED flag to address the issues.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Add the DEH and Watchdog module to the DSP2 tests for DRA7XX.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Add a new module config to the Watchdog module that
allows to disable the module. The Watchdog
module is automatically included when Deh module is
included in an image. This parameter allows the user
to disable the watchdog while still using the Deh
module and allows this to be done from a config param
at the application level on the BIOS-side application.
The following can be added to the application config
file to disable Watchdog module (if desired) when including
Deh module.
WD = xdc.useModule('ti.deh.Watchdog');
WD.disableWatchdog = true;
There are other ways to disable the watchdog from Linux/Android
that already exist. This method just provides additional
flexibility to the user.
The default value is false if this parameter is not set.
So, if this parameter is not set, the behavior is the same as
before this patch.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
The RPMSG_NS_2.0 feature was introduced to distinguish multiple channels of
the same service type to the host OS. The ping_tasks example however has been
using the same descriptor name for both the services. Use unique names for
each channel so that it conforms to the NS_2.0 design. Also, fixed some minor
incorrect whitespace indentation issues.
Issue found with the recent changes on upstream Linux 4.9 kernel.
Signed-off-by: Suman Anna <s-anna@ti.com>
|
|
IPC's Linux user space support contains a copy of a few
kernel-defined structs and definitions including the socket
address family used for rpmsg-based sockets - AF_RPMSG. This
definition is always 'one more' than the last address family
defined by the mainline kernel's socket.h.
Prior to Linux 3.9, this was aways 40. Couple of new address
families were introduced in 4.6 (AF_KCM) and 4.7 (AF_QIPCRTR),
and as a result, we needed to set AF_RPMSG to 43 post 4.7 kernel.
The user space header has been updated to reflect on the kernel
version and appropriately assign the value of AF_RPMSG.
Do note that a specific value of AF_RPMSG can always be provided
through the products.mak when KERNEL_INSTALL_DIR is not set.
Signed-off-by: Suman Anna <s-anna@ti.com>
|
|
This allows freedom in memory placement of data memory sections.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
The Android service start already takes care of checking
if the service is already running, so it's not required
to check again if the daemon is running or not.
It's recommended to add an Android service in the init.rc
script file for the lad_dra7xx. For instance:
service lad_dra7xx /system/bin/lad_dra7xx -g
class main
user media
group system
oneshot
The service can be set up to automatically launch when
Android boots or manually started later:
$ start lad_dra7xx
Directly running the lad_dra7xx executable in Android is
discouraged as the capability to enforce that only one
instance is running is lost.
Signed-off-by: Misael Lopez Cruz <misael.lopez@ti.com>
|
|
lad_dra7xx is being launched as an Android service. The
service status is tracked based on the pid of the binary
run when the service is launched.
In daemon mode, the pid that Android keeps track of is that
of the parent process that started the daemon. So, the
lad_dra7xx service status is not properly shown.
The lad_dra7xx will be compiled in app mode for Android to
avoid the problem described above. The lad_dra7xx will
continue as a daemon for Linux.
Signed-off-by: Misael Lopez Cruz <misael.lopez@ti.com>
|
|
This change is aimed to avoid parallel build issues, due to interfaces
not built.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Captures change with subdir-objects removed
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
automake has a bug handling subdir-objects.
Removing subdir-objects for now and live with warning to avoid build
issue.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Check in autoreconf generated files after update to Makefile.am files
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Without these fixes, the following warnings were seen and now fixed
configure.ac:54: warning: Unknown LT_INIT option `gcc'
linux/src/api/Makefile.am:61: warning: source file
'$(top_srcdir)/linux/src/heaps/HeapStd.c' is in a subdirectory,
linux/src/api/Makefile.am:61: but option 'subdir-objects' is disabled
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Based on gcc documentation and other discussions online, looks like
"-pthread" is required in compile options.
Needed to add -pthread with CFLAGS
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
On Poky based builds, the ti-ipc compilation results in the below
error:
.libs/libtitransportrpmsg.so: undefined reference to `pthread_join'
~~~~~~~~~~~~~~~~~~~
Signed-off-by: Karthik Ramanan <a0393906@ti.com>
|
|
This helps to relocate and place data sections without any restrictions.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Without this fix the NonInstrumented example builds fail for K2
platforms with IPC configured for Build.LibType_NonInstrumented
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
In the current MessageQ_create implementation, the NameServer
registration is done early, before local creation in the user
library is done. It is therefore possible that a remote core
could open a MessageQ that is still being created and put a
message to that MessageQ. This can result in dropped messages
or in the worst case a crash.
This patch moves the NameServer registration to the end of
the creation sequence so that the MessageQ cannot be opened
before it is fully created.
A new LAD command called MESSAGEQ_ANNOUNCE is added that is specifically
for registering a previously created MessageQ with the NameServer.
Once all creation is completed and the MessageQ is ready to
receive messages, the MessageQ_create api makes the call to LAD
MESSAGEQ_ANNOUNCE to register with the NameServer.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Fix a trace in NameServer_add which was assuming a 32-bit aligned
pointer, but the pointer may not be aligned. So, print only the
first byte, as the size of the entry is varaible anyway.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
The rpmsg channel publishing requires RPMSG_NS_2_0 to be defined
for all Keystone 2 platforms, otherwise the channels are not
created properly on the Linux kernel side and the ping_tasks and
ping_rpmsg base images do not work with corresponding MPU-side
samples.
Signed-off-by: Suman Anna <s-anna@ti.com>
|
|
With latest Android N compiler, we get following error. Fixing it by
removing the always true check.
linux/src/api/NameServer.c:311:17: error: comparison of array 'rsp.get.buf'
not equal to a null pointer is always true [-Werror,-Wtautological-pointer-compare]
if (rsp.get.buf != NULL) {
~~~~~~~~^~~ ~~~~
linux/src/api/NameServer.c:453:17: error: comparison of array 'rsp.get.buf'
not equal to a null pointer is always true [-Werror,-Wtautological-pointer-compare]
if (rsp.get.buf != NULL) {
~~~~~~~~^~~ ~~~~
Signed-off-by: Vishal Mahaveer <vishalm@ti.com>
|
|
The QNX IPC code has a mix of LOGn and GT traces. This patch
modifies all the LOGn traces to be GT traces for consistency,
and to give better control to trace enabling.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
System_printf in this function floods the IPU trace buffer and drowns out other
logs because this code is executed for every MmRpc_call made from the host.
Log_print provides better control to enable or disable logs on a per module
basis.
Signed-off-by: Harinarayan Bhatta <harinarayan@ti.com>
|
|
State
The current implementation will cease checking of the remaining remote cores in
the cluster if NameServer_getRemote fails for any reason other than
NameServer_E_NOTFOUND or NameServer_E_RESOURCE. If one of the remote cores in
the cluster is in an unresponsive state, then either NameServer_E_TIMEOUT or
NameServer_E_FAIL could be returned. And when that error code is returned, no
more remote cores in the cluster will be checked by NameServer_get. So if one
core goes into a bad state, then communication becomes blocked for all remaining
cores in the cluster.
To improve error handling and allow other cores to continue communication with
the HOST even if one core has crashed, the remaining cores should continue to
be checked, regardless of an error being returned from one of the cores.
This patch modifies NameServer_get so that if NameServer_getRemote fails for
a particular core, it no longer breaks, but continues to try the remaining
cores. In this way a crashed core does not block communication with the other
cores.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
The baseaddr for watchdog timer 13 was being incorrectly set
for DRA7XX DSP. This patch fixes the value to be the correct
address.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
If RPMessage_recv returns due to shutdown, then no valid message
is received by the call, and it should not assert if the message
is invalid.
This patch fixes the code to only do the assert checks if shutdown
is false.
Also fixes a minor issue with a trace referencing a parameter
that no longer exists.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
In the case that IGateProvider_NULL is used for one of the remoteCustom
gates, the code will anyway attempt to map it and throw and error trace.
The code continues as this is the last part of the attach and the error
is essentially ignored as part of the design of GateMP support in order
to allow the LAD daemon to continue even if the setup fails. So the error
trace doesn't cause any functionality failure.
Fix up the code so that mappings are only done when there is a valid
size, so that the valid case of IGateProvider_NULL is properly handled.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Update the sr0 addresses to align with default values for
linux kernel jamr dts file.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Avoid use of find command in windows:
Use of find in windows results in build issues, especially with
xdctools/bin in the path.
Also there is no examples modules under packages directory ( examples
are maintained outside packages directory) and so no need to filter
them.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Without this, build will fail, when building for A15 targets alone
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Error print was misleading when this is expected in case the GateMP not
supported in the system. Replaced with Verbose print
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Starting in Linux kernel 4.1, the INTMEM type is removed
in favor of the internal memory sections needed for loading
being handled entirely in the Linux kernel. As such, the
INTMEM type no longer is defined.
This patch updated the rsc_types header file to match the
Linux defined types, which no longer defines the INTMEM
type.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Update the tests to remove usage of the INTMEM type
which is deprecated starting in Linux kernel 4.1.
Note that this is a compatibility break with k3.14
and the tests only support k4.1 and newer kernels
now.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
This patch essentially reverts the following commits:
1. "Add fw_rsc_intmem definition to support new TYPE_INTMEM entries in QNX"
(commit 4a5f6fd0864854a99d430c6e6e7c65fb962f00e4)
2. " Add INTMEM resource type support in QNX"
(commit eb3d6bcfd35b823746c6032a9f811fc414824704)
Support for TYPE_INTMEM is being removed. Instead the internal memory
regions will be available by default from the QNX resource manager, without
the remote core image needing to speicify a special resource table
entry. The changes are being made to align with coming BIOS-side changes
to remove the TYPE_INTMEM.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Previously smp was supported for M4 targets only.
Adding A15 support now.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
After getting the NameServer value for _GateMP_TI_dGate, check
to make sure that the length is equal to the expected value.
This is to prevent using some uninitialized values to configure
the GateMP. Indirectly, this can help check if hostSupport
has been enabled on the remote processor or not, because
currently, only if hostSupport has been enabled are there 4
entries. Otherwise there are only 2 entries.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Additional cache invalidates are needed to ensure all data remains fresh.
Signed-off-by: John Godbey <j-godbey@ti.com>
|
|
Call MessageQ_unbind in the fault test to unbind the application
from remote cores that it doesn't care about. This is useful in
the case of remote core recovery, so that recovery of a remote
core doesn't trigger errors in MessageQ APIs if the application
doesn't care about the state of a particular core. In this way,
the MessageQ for the application is only bound to the core(s)
with which it is communicating.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
In the case of error recovery on one core while more than one
core is currently connected to the LAD daemon, the refcount for
the NameServer daemon will never reach zero unless all connected
applications call Ipc_stop. This will prevent the NameServer
daemon from reconnecting to the recovered remote core until all
applications call Ipc_stop. This is not desireable behavior because
this prevents an application opening a MessageQ on the recovered
core until all applications call Ipc_stop.
Instead, the patch makes changes to internally re-connect with the
recovered core within NameServer daemon itself in the case of
remote core error recovery. The NameServer daemon will continue
trying to reconnect to the recovered core until successful. In
this way, the other applications are not required to exit for a new
MessageQ_open to be successful with the recovered core.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
TransportRpmsg_delete was not checking for a NULL handle
before de-referencing it, resulting in an application crash
if it was called with a NULL handle. A NULL handle was
being passed to TransportRpmsg_delete in the case where
TransportRpmsg_create socket connect call failed. In this
case, TransportRpmsg_create called TransportRpmsg_delete
with a NULL handle.
This patch adds a check for a NULL handle before de-referencing
the pointer in TransportRpmsg_delete, and also takes care
to cleanup the socket in TransportRpmsg_create in the failure
case where it cannot be cleaned by TransportRpmsg_delete.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
The default configuration for GateMP remoteCustom2 is to use
GateMPSupportNull. In this case, the remoteCustom2InUse array
will be NULL. Due to this, the virtToPhys translation will
fail. If Asserts are enabled for the GateMP module, this
will result in a failure.
This patch fixes the issue by first checking if the pointer
is NULL before attempting the translation, since it is a valid
case for the pointer to be NULL.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
In the case where UIO is not present, the GateMP_get_sr0() function
should set the baseaddr to 0. However, the code was not correctly
de-referencing the baseaddr parameter when attempting to set it to
zero, resulting in an un-initialized value for baseaddr when the UIO
driver is not present. This can result in an incorrect address for
the inUse arrays with consequences such as failure to map or other
unexpected behavior.
This patch fixes it to properly set the baseaddr to 0 for the case
where UIO is not present.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
Watchdog implementation configures timers to operate in posted-write
mode, requiring SW to poll on TWPS register prior to subsequent write
access.
Therefore, references to the Watchdog timer instance must be declared
as volatile. Otherwise, the initial value read of the TWPS will be
used as the polling value indefinitely, causing a potential lockup
if the application is completely idle.
Signed-off-by: Stephen Molfetta <sjmolfetta@ti.com>
|
|
- Removed following warning "In the definition of a function-like
macro each instance of a parameter shall be enclosed in parentheses
unless it is used as the operand of # or ##
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
There is an issue when GateMP in the remote processor is not available,
resulting in error. This fix handles the case gracefully by
ignoring remote gateMP availability.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Update the cmem and sr0 addresses to align with default values for
linux kernel.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
|
|
- Also update print string for unknown platform
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
- Correct typo in Exception string for RemoteCustom2
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
When reserving SR0 regions for GateMp instances, in some cases,
(for example the NULL gate) has zero resources. This was
not handled properly and resulted in failure. Fix is to avoid doing
unnecessary operations in case of zero resources.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|
|
Before this fix, the rsc table header file was included directly in C
code. And the table was getting optimized away due to no direct
reference to the resource table. This resulted in failure of test
downloaded through user space loader.
Instead added the configuration to include the resource table (without
vdev and vring) through configuration file, which is more cleaner.
Signed-off-by: Sam Nelson <sam.nelson@ti.com>
|