SDOCM00104913: Warnings when DRM_PREFIX not set
When DRM_PREFIX isn't set, the Linux test examples issue warnings of
unsafe cross-compilation. This is due to the Makefile setting include
paths based on a non-existent DRM_PREFIX path. The Makefile now
conditionally sets include paths when autotools config variables are
set.
When DRM_PREFIX isn't set, the Linux test examples issue warnings of
unsafe cross-compilation. This is due to the Makefile setting include
paths based on a non-existent DRM_PREFIX path. The Makefile now
conditionally sets include paths when autotools config variables are
set.
Linux: Autoconf script requires updating
The Linux Autoconf script had not been updated after previous changes
to the configure.ac file had been made.
The Linux Autoconf script had not been updated after previous changes
to the configure.ac file had been made.
Use size_t cast instead of fixed-length integer cast
When using fixed-length integer-based cast (unsigned int or unsigned long)
to or from a pointer type, errors/warnings are thrown from one compiler
or another, depending on the integer size vs. the pointer size. The
size_t type adjusts based on pointer size, so it allows a clean compile.
Also had to add a cast after existing cast, creating an ugly double-cast.
Signed-off-by: Robert Tivy <rtivy@ti.com>
When using fixed-length integer-based cast (unsigned int or unsigned long)
to or from a pointer type, errors/warnings are thrown from one compiler
or another, depending on the integer size vs. the pointer size. The
size_t type adjusts based on pointer size, so it allows a clean compile.
Also had to add a cast after existing cast, creating an ugly double-cast.
Signed-off-by: Robert Tivy <rtivy@ti.com>
Remove ducati trace thread in QNX
The ducati trace thread was deemed redundant wrt. the IPC trace daemon,
and was stealing early trace when the -g option is thrown due to a longer
delay between the time when the slave is started and when the thread is
created. Hence it has been removed. This patch also adjusts the trace buffer size
in GateMPApp to match up with its resource entry. This addresses SDOCM00103794.
Signed-off-by: VW <vwan@ti.com>
The ducati trace thread was deemed redundant wrt. the IPC trace daemon,
and was stealing early trace when the -g option is thrown due to a longer
delay between the time when the slave is started and when the thread is
created. Hence it has been removed. This patch also adjusts the trace buffer size
in GateMPApp to match up with its resource entry. This addresses SDOCM00103794.
Signed-off-by: VW <vwan@ti.com>
SDOCM00104689 Remove MultiProc_setBaseIdOfCluster API
This API never worked and the intended behavior is not supported.
This API never worked and the intended behavior is not supported.
Merge 3.10.02.13 into ipc-next
Modify MessageQApp and MessageQBench in QNX to match slave-side changes
The tests would fail due to previous changes in the messageq_single
server in how the Message Id was being assigned and used. Updated the
QNX-side test example to reflect these changes.
Signed-off-by: VW <vwan@ti.com>
The tests would fail due to previous changes in the messageq_single
server in how the Message Id was being assigned and used. Updated the
QNX-side test example to reflect these changes.
Signed-off-by: VW <vwan@ti.com>
Update message protocol between host and server.
The message protocol changed on the server side. Updated host
application to use the new protocol.
The message protocol changed on the server side. Updated host
application to use the new protocol.
Test: Modified MessageQBench due to server changes
The test would fail due to previous changes in the messageq_single
server in how the Message Id was being assigned and used. Updated the
Linux-side test example to reflect these changes. Performed some code
clean-up in the process.
The test would fail due to previous changes in the messageq_single
server in how the Message Id was being assigned and used. Updated the
Linux-side test example to reflect these changes. Performed some code
clean-up in the process.
Prevent MMU registers from becoming unreadable
This commit changes the DSP domain power state transition register
CM_DSP_CLKSTCTRL from forcing a wake-up transition on the domain (SW_WKUP),
to using automatic transitions (HW_AUTO). This prevents the MMU registers from
becoming unreadable after repeatedly pulling the MMU out of reset and putting
it back into reset.
Signed-off-by: VW <vwan@ti.com>
This commit changes the DSP domain power state transition register
CM_DSP_CLKSTCTRL from forcing a wake-up transition on the domain (SW_WKUP),
to using automatic transitions (HW_AUTO). This prevents the MMU registers from
becoming unreadable after repeatedly pulling the MMU out of reset and putting
it back into reset.
Signed-off-by: VW <vwan@ti.com>
QNX IPC: OMAP5 - Enable DSP-MMU Fault Recovery
Add the DSP MMU interrupt to listen for DSP-MMU faults.
Also fix a bug where the shm_phys_addr_dsp was not being
properly reset to 0, resulting in recovery failing to
happen properly.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
Signed-off-by: VW <vwan@ti.com>
Add the DSP MMU interrupt to listen for DSP-MMU faults.
Also fix a bug where the shm_phys_addr_dsp was not being
properly reset to 0, resulting in recovery failing to
happen properly.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
Signed-off-by: VW <vwan@ti.com>
change 'swiPriority' to module config param
Setting 'swiPriority' had no effect because it was an instance parameter
and the swi is created internally by IpcMgr. Change 'swiPriority' to be
a module config parameter. This allows users to change the priority of
the swi thread which created in their configuration script.
Setting 'swiPriority' had no effect because it was an instance parameter
and the swi is created internally by IpcMgr. Change 'swiPriority' to be
a module config parameter. This allows users to change the priority of
the swi thread which created in their configuration script.
SDOCM00104757 Fixed VirtQueue notification race condition
Enable notification before checking message pool. This might
result in an unnecessary notification but this is benign. Disable
notification as before (when new message has been acquired).
Enable notification before checking message pool. This might
result in an unnecessary notification but this is benign. Disable
notification as before (when new message has been acquired).
SDOCM00103636: ping_rpmsg OMAP5 test broken
The OMAP5 and DRA7XX Linux kernels require slave-side code to
use RPMSG_NS_2_0-compatible calls to NameMap_register(). When
moving to the XDC-provided ti.platforms.sdp5430 platform, the
check for which platforms need to throw -DRPMSG_NS_2_0 broke.
As a result, the ping_rpmsg test, specifically for OMAP5, was
building incorrectly.
The OMAP5 and DRA7XX Linux kernels require slave-side code to
use RPMSG_NS_2_0-compatible calls to NameMap_register(). When
moving to the XDC-provided ti.platforms.sdp5430 platform, the
check for which platforms need to throw -DRPMSG_NS_2_0 broke.
As a result, the ping_rpmsg test, specifically for OMAP5, was
building incorrectly.
Examples: Added ping_tasks to talk w/ Linux app
A slave-side ping_tasks example was added that interfaces with the
Linux kernel's rpmsg_client_sample. This example demonstrates how to
leverage the Linux kernel's rpmsg driver to communicate directly to
the slave-side BIOS RPMessage interface.
A slave-side ping_tasks example was added that interfaces with the
Linux kernel's rpmsg_client_sample. This example demonstrates how to
leverage the Linux kernel's rpmsg driver to communicate directly to
the slave-side BIOS RPMessage interface.
Examples: Modify BIOS ping_rpmsg with dynamic Tasks
The ping_rpmsg test example has be modified to create it's Task
dynamically. This is in anticipation of upcoming examples that will
reuse the ping_rpmsg configuration on the BIOS-side(slaves).
The ping_rpmsg test example has be modified to create it's Task
dynamically. This is in anticipation of upcoming examples that will
reuse the ping_rpmsg configuration on the BIOS-side(slaves).
Fix TCI6636 platform typo in comment
Minor comment change to fix TCI6636 platform variable name.
Minor comment change to fix TCI6636 platform variable name.
OMAP5: Fix test build of nameserverapp
When moving scripts from ti/configs/omap5 into ti/ipc/tests,
some scripts were renamed (to avoid conflicts with ti/configs/vayu
scripts of the same name). One of the renames was missed in
nameserverapp.cfg.
Ideally this would have been included in
3cb709f6d6ebbf62b5d2ac25c9fb9f84ab2a8a1a. Oh, well.
When moving scripts from ti/configs/omap5 into ti/ipc/tests,
some scripts were renamed (to avoid conflicts with ti/configs/vayu
scripts of the same name). One of the renames was missed in
nameserverapp.cfg.
Ideally this would have been included in
3cb709f6d6ebbf62b5d2ac25c9fb9f84ab2a8a1a. Oh, well.
Merge tag '3.10.02.12_eng' into ipc-next
Mask off signals before creating thread pool in QNX resource manager
This commit ensures that SIGTERM is caught by the main thread by masking off signal
handling in the other threads of the thread pool. This allows the IPC resource
manager to terminate correctly after MessageQ is used (SDOCM00102268).
Signed-off-by: VW <vwan@ti.com>
This commit ensures that SIGTERM is caught by the main thread by masking off signal
handling in the other threads of the thread pool. This allows the IPC resource
manager to terminate correctly after MessageQ is used (SDOCM00102268).
Signed-off-by: VW <vwan@ti.com>
Modified MessageQApp and messageq_single to support > 65K messages.
The test would fail when sending more then 65K messages because the
message id was only 16 bits wide. Moved message id from header into
message payload and made it 32 bits wide. Cleaned up loop to be one
based (instead of 0-based). Turned off slave side prints.
The test would fail when sending more then 65K messages because the
message id was only 16 bits wide. Moved message id from header into
message payload and made it 32 bits wide. Cleaned up loop to be one
based (instead of 0-based). Turned off slave side prints.
SDOCM00104583 RPMessage crash after 65k messages
The semaphore was posted for every message added to the pool.
Added VRING_USED_F_NO_NOTIFY flag to only post semaphore when
needed. Moved clearing of flag to avoid race condition with
remote processor. Also changed the semaphore from counting
mode to binary mode.
The semaphore was posted for every message added to the pool.
Added VRING_USED_F_NO_NOTIFY flag to only post semaphore when
needed. Moved clearing of flag to avoid race condition with
remote processor. Also changed the semaphore from counting
mode to binary mode.
Examples: Move ti.configs.vayu into tests
ti.configs.vayu was a convenience package for reusable config
snippets for IPC tests. To reinforce this, and discourage outside
use of these snippets, this commit moves the content into the test
package and deletes the ti.configs.vayu package.
ti.configs.vayu was a convenience package for reusable config
snippets for IPC tests. To reinforce this, and discourage outside
use of these snippets, this commit moves the content into the test
package and deletes the ti.configs.vayu package.
Examples: Move ti.configs.omap54xx into tests
ti.configs.omap54xx was a convenience package for reusable config
snippets for IPC tests. To reinforce this, and discourage outside
use of these snippets, this commit moves the content into the test
package and deletes the ti.configs.omap54xx package.
ti.configs.omap54xx was a convenience package for reusable config
snippets for IPC tests. To reinforce this, and discourage outside
use of these snippets, this commit moves the content into the test
package and deletes the ti.configs.omap54xx package.
SDOCM00104194 Missing vayu package dependencies
In close() function, bring in ISA specific module using the
program's target ISA chain. Look for oldest ISA which is
supported by device processor.
In close() function, bring in ISA specific module using the
program's target ISA chain. Look for oldest ISA which is
supported by device processor.
Create a separate BIOS configuration file for NameServerApp
This commit decouples the configuration of NameServerApp from the MessageQ
tests. This is done because NameServerApp does not use MessageQ, and can
use a simpler configuration.
Signed-off-by: VW <vwan@ti.com>
This commit decouples the configuration of NameServerApp from the MessageQ
tests. This is done because NameServerApp does not use MessageQ, and can
use a simpler configuration.
Signed-off-by: VW <vwan@ti.com>
Linux: Update autotools to require KERNEL path
The IPC Linx-side build now requires KERNEL path to be set to
build all libraries and test applications redardless of platform.
This change is needed since rpmsg has been made Linux kernel version
aware.
The IPC Linx-side build now requires KERNEL path to be set to
build all libraries and test applications redardless of platform.
This change is needed since rpmsg has been made Linux kernel version
aware.
Linux: Update user AF_RPMSG define for 3.9 kernel
IPC's Linux user space support contains a copy of a few
kernel-defined structs and definitions. This is necessary
because the kernel headers are not include-able by user space.
One of these values is 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 kernel's
socket.h.
Prior to Linux 3.9, this was aways 40. In Linux 3.9, a new
address family was introduced (AF_VSOCK), and as a result we
needed to set AF_RPMSG to 41. To enable this, the user space
header reflects on the kernel version and appropriately assigns
the value of AF_RPMSG.
Note that in addition to adding 3.9+ support, this commit also
cleans up some dead code.
IPC's Linux user space support contains a copy of a few
kernel-defined structs and definitions. This is necessary
because the kernel headers are not include-able by user space.
One of these values is 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 kernel's
socket.h.
Prior to Linux 3.9, this was aways 40. In Linux 3.9, a new
address family was introduced (AF_VSOCK), and as a result we
needed to set AF_RPMSG to 41. To enable this, the user space
header reflects on the kernel version and appropriately assigns
the value of AF_RPMSG.
Note that in addition to adding 3.9+ support, this commit also
cleans up some dead code.
Build: Align products.mak variables across OS's
The make variable to indicate a platform is now consistent
between BIOS, QNX and Linux, and only needs to be set once.
Note that this is a compatibility break for consumers (e.g. SDKs)
which automate the consumption/build of IPC. This is consciously
being introduced in a new Major.minor release (3.20) to keep
from breaking 3.00 and 3.10 users.
This resolves SDOCM00103415.
The make variable to indicate a platform is now consistent
between BIOS, QNX and Linux, and only needs to be set once.
Note that this is a compatibility break for consumers (e.g. SDKs)
which automate the consumption/build of IPC. This is consciously
being introduced in a new Major.minor release (3.20) to keep
from breaking 3.00 and 3.10 users.
This resolves SDOCM00103415.
Remove BIOS-side rpmsg trace for NameServerApp and MessageQ tests
Signed-off-by: VW <vwan@ti.com>
Signed-off-by: VW <vwan@ti.com>
Remove static allocation of IPU_MEM_IOBUFS memory in OMAP5 QNX
This commit removes unnecessary static memory allocation for
IPU_MEM_IOBUFS in the IPC resource manager on OMAP5 (SDOCM00104300)
Signed-off-by: VW <vwan@ti.com>
This commit removes unnecessary static memory allocation for
IPU_MEM_IOBUFS in the IPC resource manager on OMAP5 (SDOCM00104300)
Signed-off-by: VW <vwan@ti.com>
GateMP: typo fixes
Resolve a few typos in the GateMP.xdc file. One notable
change is that an Assert message was corrected. While this
should be merely a cosmetic change, end users experiencing the
assert will see a change in the message.
Resolve a few typos in the GateMP.xdc file. One notable
change is that an Assert message was corrected. While this
should be merely a cosmetic change, end users experiencing the
assert will see a change in the message.
OMAP5: Provide default Notify delegate
The default Notify delegate for OMAP5 is NotifySetupNull. This
satisfies the most common use case of the A15 running an HLOS,
and the slave cores not communicating with each other.
This addresses SDOCM00103112.
The default Notify delegate for OMAP5 is NotifySetupNull. This
satisfies the most common use case of the A15 running an HLOS,
and the slave cores not communicating with each other.
This addresses SDOCM00103112.
Removed unnecessary cast in parameter assignment.
QNX IPC: RscTable - Use MAP_NOINIT Flag to Allocate Mem
The QNX IPC dynamically allocates memory for the remote
core sections based on the resource table. These sections
are allocated using an mmap call and can be quite large.
There is a performance improvement if the allocated
memory is not zero-initialized during allocation with
mmap. It may be desireable to trade off zero-initialization
of the memory for the performance improvement, so to
enable the memory to not be zero-initialized, the MAP_NOINIT
flag is added to the flags in the mmap call.
It is important to note that the MAP_NOINIT flag will only
relax the zero-init requirement if the memory being mapped
was previously freed using UNMAP_INIT_OPTIONAL. This is
considered set for any memory that hasn't been allocated before,
so can help with boot-up time. Also, The default behavior of munmap
is to not have this flag set. However, if the "-m~i" option is
used on the procnto command line, the default behavior of
munmap is changed to act like UNMAP_INIT_OPTIONAL is set.
So, this change will only gurantee the zero-init to be relaxed if the
additional change is made to use the "-m~i" option.
To be safe, this patch also explicitly zero-initializes the
vring memory after allocation, so that even if the zero-init
requirement is relaxed, the vring memory will be zero-ed.
Also, a flag named "ZEROINIT_CHUNKS" is added in the file, and
set to 0 by default. If it is desired to restore the behavior of
zero-initialization of the chunks, this can be set to 1, and
the QNX IPC rebuilt.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
The QNX IPC dynamically allocates memory for the remote
core sections based on the resource table. These sections
are allocated using an mmap call and can be quite large.
There is a performance improvement if the allocated
memory is not zero-initialized during allocation with
mmap. It may be desireable to trade off zero-initialization
of the memory for the performance improvement, so to
enable the memory to not be zero-initialized, the MAP_NOINIT
flag is added to the flags in the mmap call.
It is important to note that the MAP_NOINIT flag will only
relax the zero-init requirement if the memory being mapped
was previously freed using UNMAP_INIT_OPTIONAL. This is
considered set for any memory that hasn't been allocated before,
so can help with boot-up time. Also, The default behavior of munmap
is to not have this flag set. However, if the "-m~i" option is
used on the procnto command line, the default behavior of
munmap is changed to act like UNMAP_INIT_OPTIONAL is set.
So, this change will only gurantee the zero-init to be relaxed if the
additional change is made to use the "-m~i" option.
To be safe, this patch also explicitly zero-initializes the
vring memory after allocation, so that even if the zero-init
requirement is relaxed, the vring memory will be zero-ed.
Also, a flag named "ZEROINIT_CHUNKS" is added in the file, and
set to 0 by default. If it is desired to restore the behavior of
zero-initialization of the chunks, this can be set to 1, and
the QNX IPC rebuilt.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
OMAP5: resource table dead code cleanup
Removed dead "#ifdef OMAP5" code chunks from two omap5-specific
files. These files are already OMAP5-specific, they don't support
any other devices.
Removed dead "#ifdef OMAP5" code chunks from two omap5-specific
files. These files are already OMAP5-specific, they don't support
any other devices.
Android: Added build support for ping_rpmsg app
Add headers and libs to QNX 'make install' (part 2)
Export MessageQ-related headers and libs for QNX. Also clean
up exported files in general to remove headers that are not needed.
This commit adds an additional file to a previous commit that
addressed SDOCM00102141.
Export MessageQ-related headers and libs for QNX. Also clean
up exported files in general to remove headers that are not needed.
This commit adds an additional file to a previous commit that
addressed SDOCM00102141.
Merge recent 3.00 changes into 3.10
SDOCM00104043 MessageQ_open() returns not found
Concurrency issues in NameServerRemoteNotify caused message corruption
which resulted in keys not being found. Fixed by moving state variables
from message to object state and by splitting existing swi into two
separate ones, one for request and one for response. Updated ROV.
Concurrency issues in NameServerRemoteNotify caused message corruption
which resulted in keys not being found. Fixed by moving state variables
from message to object state and by splitting existing swi into two
separate ones, one for request and one for response. Updated ROV.
Fix wrong usage description of resource manager (SDOCM00103947)
This patch updates the description that appears when calling 'use ipc'.
It also removes the command line options -f and -d that are obsolete,
and updates the error messages that comes up when wrong arguments
are passed, to guide people towards correct usage.
Signed-off-by: VW <vwan@ti.com>
This patch updates the description that appears when calling 'use ipc'.
It also removes the command line options -f and -d that are obsolete,
and updates the error messages that comes up when wrong arguments
are passed, to guide people towards correct usage.
Signed-off-by: VW <vwan@ti.com>
Add defaultErrFxn and an err fxn to module state
Update TransportCirc_setErrFxn to set the module errFxn at runtime.
Create a default Error Fxn which does nothing for TransportCirc.
If an error fxn is not specified it will be set to the default.
Update TransportCirc_setErrFxn to set the module errFxn at runtime.
Create a default Error Fxn which does nothing for TransportCirc.
If an error fxn is not specified it will be set to the default.
SDOCM00103043:Resolve Concerto IPC stress test failures
SDOCM00103935:Update to work with different cluster base id
SDOCM00103934:allow Swi priority to be configurable
Android: Added support for MessageQ
Support for MessageQ and related application(LAD) was added for Android.
Android specific build files have been created/modified to build all
user-side libraries (libmmrpc, libtiipc, libtiutils) and applications
from within the Android filesystem. This include test apps (MesssageQ*).
A few changes were also needed by the common source files that
are shared across Linux and Android builds.
Support for MessageQ and related application(LAD) was added for Android.
Android specific build files have been created/modified to build all
user-side libraries (libmmrpc, libtiipc, libtiutils) and applications
from within the Android filesystem. This include test apps (MesssageQ*).
A few changes were also needed by the common source files that
are shared across Linux and Android builds.
Linux: Removed circular dependencies from ipc libs
The circular dependencies between libtiipc and libtiipcutils has
been removed by moving the assignment of the user side application's
MultiProc config structure to the libtiipcutils library via a new
function (_MultiProc_initCfg).
This also forces the creation of a LAD specific libtiipcutils library
(libtiipcutils_lad), which the daemon needs to link against.
The circular dependencies between libtiipc and libtiipcutils has
been removed by moving the assignment of the user side application's
MultiProc config structure to the libtiipcutils library via a new
function (_MultiProc_initCfg).
This also forces the creation of a LAD specific libtiipcutils library
(libtiipcutils_lad), which the daemon needs to link against.
Test: Add autotools-generated Makefile.in
This is the second half of the Makefile.am changes in
47557586d086d61d2c361796388a61598563b413. That commit changed
Makefile.am, this commit is the associated, generated Makefile.am.
This is the second half of the Makefile.am changes in
47557586d086d61d2c361796388a61598563b413. That commit changed
Makefile.am, this commit is the associated, generated Makefile.am.
Implement reference counting in NameServer_create/delete (SDOCM00103730)
Add headers and libs to QNX 'make install'
Export MessageQ-related headers and libs for QNX. Also clean
up exported files in general to remove headers that are not needed.
This addresses SDOCM00102141.
Export MessageQ-related headers and libs for QNX. Also clean
up exported files in general to remove headers that are not needed.
This addresses SDOCM00102141.
New unit test for MessageQ delivery.
This test validates SDOCM00103664 MessageQ_get() can discard
msgs when a msg is available from more than one remote core.
This test validates SDOCM00103664 MessageQ_get() can discard
msgs when a msg is available from more than one remote core.
Android.mk: remove HEAPTRACK library dependency
BUILD_HEAPTRACKED_SHARED_LIBRARY is a local define
in hardware/ti/omap4xxx. using this build macro inside
ipc component, unnecessarily creates build dependency with
omap4xx path. And this heaptrack feature is not required
for ipc. Hence removed the HEAPTRACK option.
Signed-off-by: Sunita Nadampalli <sunitan@ti.com>
BUILD_HEAPTRACKED_SHARED_LIBRARY is a local define
in hardware/ti/omap4xxx. using this build macro inside
ipc component, unnecessarily creates build dependency with
omap4xx path. And this heaptrack feature is not required
for ipc. Hence removed the HEAPTRACK option.
Signed-off-by: Sunita Nadampalli <sunitan@ti.com>
Build: Enable builds without SMP support
Some customers (including the TI RTOS SDK) do not use SMP
BIOS and therefore don't require the libraries. To improve
build times and reduce the disk size for these users, a
new variable, BIOS_SMPENABLED, was created in products.mak.
Its default value is 1 (enabled), which maintains the
behavior of previous releases. Setting it to 0 disables
the build of SMP-enabled libraries.
Note that some platforms require SMP libraries to be built,
otherwise their test executables may fail to build.
This resolves SDOCM00103413.
Some customers (including the TI RTOS SDK) do not use SMP
BIOS and therefore don't require the libraries. To improve
build times and reduce the disk size for these users, a
new variable, BIOS_SMPENABLED, was created in products.mak.
Its default value is 1 (enabled), which maintains the
behavior of previous releases. Setting it to 0 disables
the build of SMP-enabled libraries.
Note that some platforms require SMP libraries to be built,
otherwise their test executables may fail to build.
This resolves SDOCM00103413.
MessageQ: Fix MessageQ_get() dropping messages
MessageQ_get() had a bug where it would loop over all remote cores
looking for and getting a message and would return the message from
the highest numbered processor (in MultiProc order) that had one
available, tossing messages received from lower numbered processors.
The MessageQ_get() loop now terminates as soon as the first message
is found. Further to this change, on the next call to the function
it starts looking for messages from the processor that is "next" in
the (cyclic) list after the one in which it last found a message.
Fixes SDOCM00103664.
Signed-off-by: Robert Tivy <rtivy@ti.com>
MessageQ_get() had a bug where it would loop over all remote cores
looking for and getting a message and would return the message from
the highest numbered processor (in MultiProc order) that had one
available, tossing messages received from lower numbered processors.
The MessageQ_get() loop now terminates as soon as the first message
is found. Further to this change, on the next call to the function
it starts looking for messages from the processor that is "next" in
the (cyclic) list after the one in which it last found a message.
Fixes SDOCM00103664.
Signed-off-by: Robert Tivy <rtivy@ti.com>
MmRpc: updated to new Linux rpmsg_rpc.h API
Two structures changed a field name from reserved to fd.
Also aligned the QNX copy of this header to match.
Two structures changed a field name from reserved to fd.
Also aligned the QNX copy of this header to match.
NameServer: fix Linux-requested remote ns lookups
BIOS-side NameServerRemoteRpmsg.c was recently taught to
support NameServer lookups larger than 32 bits. Prior to
that education, it assumed all requestors were asking for
only 32 bits of NameServer data.
Linux must now indicate how big its NameServer buffer is
so the BIOS-side knows how to fill it. When this > 32-bit
lookup support was added, QNX was correctly modified; Linux
was overlooked.
For users, this commit fixes Linux NameServer lookups (which
most users see as MessageQ_open() calls) which have been broken
since 82e4969d635d0f6cbee3c6fcc3df3a7be39bc852. Tag-wise,
this Linux-only bug was introduced in the 3.10.00.06_eng build.
BIOS-side NameServerRemoteRpmsg.c was recently taught to
support NameServer lookups larger than 32 bits. Prior to
that education, it assumed all requestors were asking for
only 32 bits of NameServer data.
Linux must now indicate how big its NameServer buffer is
so the BIOS-side knows how to fill it. When this > 32-bit
lookup support was added, QNX was correctly modified; Linux
was overlooked.
For users, this commit fixes Linux NameServer lookups (which
most users see as MessageQ_open() calls) which have been broken
since 82e4969d635d0f6cbee3c6fcc3df3a7be39bc852. Tag-wise,
this Linux-only bug was introduced in the 3.10.00.06_eng build.
Fixed compiler warning when assigning buffer address to descriptor.
Linux: Removed install-hook to include *.la files
The Linux autotools install-hook to delete *la files as part of the
installation was removed.
The Linux autotools install-hook to delete *la files as part of the
installation was removed.
Dev: Add new QNX binaries to .gitignore
Removed unused IPU processor name from NotifySetup.
The IPU interrupt setup is not handled by the NotifySetup module.
In addition, the IPU references were incorrect for non-smp mode.
The IPU interrupt setup is not handled by the NotifySetup module.
In addition, the IPU references were incorrect for non-smp mode.
Build: Cosmetic changes to products.mak
Developers added the many variables to products.mak over time.
This commit tries to address inconsistencies throughout the file,
including:
* indication of optional variables
* descriptions of variables
* white space
It also moves the PLATFORM variable (used by both Linux and
BIOS) out of the Linux section and into a more common area.
Developers added the many variables to products.mak over time.
This commit tries to address inconsistencies throughout the file,
including:
* indication of optional variables
* descriptions of variables
* white space
It also moves the PLATFORM variable (used by both Linux and
BIOS) out of the Linux section and into a more common area.
OMAP5 QNX: Changed ipc.c to avoid calling GateMP since it is not yet supported
Signed-off-by: VW <vwan@ti.com>
Signed-off-by: VW <vwan@ti.com>
Fixed make install for QNX (Vayu) to export GateMPApp
Signed-off-by: VW <vwan@ti.com>
Signed-off-by: VW <vwan@ti.com>
Added MmRpc_use()/MmRpc_release() API functions.
These functions call into the driver to make the buffers
persistent from the remote processor view. This allows the
remote processor to maintain persistent references to the
buffers across multiple calls to MmRcp_call().
These functions call into the driver to make the buffers
persistent from the remote processor view. This allows the
remote processor to maintain persistent references to the
buffers across multiple calls to MmRcp_call().
QNX IPC: rpmsg_rpc - Don't Use Primary Offset in Translations
The primary offset for the params is not needed in the
translations. The translation code was incorrectly subtracting
the primary offset when calculating the location of the pointer
to be translated. The translation offset is the offset from the
start of the params data pointer, and not the offset from the
start of the params base. So, the params data pointer plus the
translation offset is all that is needed in order to get the
location of the translation.
Also, fix the reverse translations so that the correct
value is put back in the buffer.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
The primary offset for the params is not needed in the
translations. The translation code was incorrectly subtracting
the primary offset when calculating the location of the pointer
to be translated. The translation offset is the offset from the
start of the params data pointer, and not the offset from the
start of the params base. So, the params data pointer plus the
translation offset is all that is needed in order to get the
location of the translation.
Also, fix the reverse translations so that the correct
value is put back in the buffer.
Signed-off-by: Angela Stegmaier <angelabaker@ti.com>
Add base field to MmRpc_Xlt structure.
Fixed the offset calculation for embedded pointers by allowing
the caller to specify encoded base value. Enhanced some comments.
Code cleanup.
Fixed the offset calculation for embedded pointers by allowing
the caller to specify encoded base value. Enhanced some comments.
Code cleanup.
Changes to support non-NameServer-capable slaves on Linux
Ipc setup code was giving up when any slave failed to connect to Linux
sockets. It now succeeds when at least one slave connects. Previously,
both NameServer_setup() (in LAD) and MessageQ_attach()es called from
Ipc_start() were expected to succeed for all remote cores specified in
MultiProc. Now those functions can tolerate slave connection failures and
will report success if at least one slave can connect.
Ipc setup code was giving up when any slave failed to connect to Linux
sockets. It now succeeds when at least one slave connects. Previously,
both NameServer_setup() (in LAD) and MessageQ_attach()es called from
Ipc_start() were expected to succeed for all remote cores specified in
MultiProc. Now those functions can tolerate slave connection failures and
will report success if at least one slave can connect.
Change NameServer_get to ignore 'remote get' timeouts and use a reduced period.
This is to fix SDOCM00102548 for QNX, so that MessageQ_open succeeds when
the IPU is loaded with an executable that does not 'speak' NameServer.
Signed-off-by: VW <vwan@ti.com>
This is to fix SDOCM00102548 for QNX, so that MessageQ_open succeeds when
the IPU is loaded with an executable that does not 'speak' NameServer.
Signed-off-by: VW <vwan@ti.com>
Added GateMPApp test for GateMP verification
Signed-off-by: VW <vwan@ti.com>
Signed-off-by: VW <vwan@ti.com>
GateMP: Introduce support for GateMP on QNX
To enable GateMP,throw the -g option when launching the IPC resource
manager.
On the BIOS side, DSP1 must be configured as the owner of SR0 and
the new GateMP.hostSupport config param must be set to true like
this:
var GateMP = xdc.useModule('ti.sdo.ipc.GateMP');
GateMP.hostSupport = true;
Note that it's initially only been validated on DRA7XX DSP1.
Signed-off-by: VW <vwan@ti.com>
To enable GateMP,throw the -g option when launching the IPC resource
manager.
On the BIOS side, DSP1 must be configured as the owner of SR0 and
the new GateMP.hostSupport config param must be set to true like
this:
var GateMP = xdc.useModule('ti.sdo.ipc.GateMP');
GateMP.hostSupport = true;
Note that it's initially only been validated on DRA7XX DSP1.
Signed-off-by: VW <vwan@ti.com>
MmRpc: Added support for Android OS build
Introduce an initial Android.mk build script, and slight
code change to build the MmRpc library.
Signed-off-by: Sunita Nadampalli <sunitan@ti.com>
Introduce an initial Android.mk build script, and slight
code change to build the MmRpc library.
Signed-off-by: Sunita Nadampalli <sunitan@ti.com>
MessageQ: Fix for Linux msgs sent to wrong slave
Previously the MultiProc-to-Linux-kernel-rproc-index mapping was
hard-coded in the MessageQ-based socket library to be
rprocIndex = MultiProc_id - 1. This was too simple for the async
loading use cases on systems with > 1 slave.
A recent kernel patch looks up the rpmsg channel based on Linux
kernel rproc index, meaning the socket's 'connect' and 'bind'
calls now need to provide the actual rproc index. Today's Linux
kernels ensure the rproc index will be consistent from run to run
(so long as all slaves are configured enabled), so this rproc-
index-to-MultiProc-id can be known and hard-coded ahead of time.
This commit records that hard-coding in the MultiProcCfg_*.c
source files, and now provides the appropriate rproc index in
the socket 'connect' and 'bind' calls.
This user-space commit requires a matching update to the kernel
space drivers.
Previously the MultiProc-to-Linux-kernel-rproc-index mapping was
hard-coded in the MessageQ-based socket library to be
rprocIndex = MultiProc_id - 1. This was too simple for the async
loading use cases on systems with > 1 slave.
A recent kernel patch looks up the rpmsg channel based on Linux
kernel rproc index, meaning the socket's 'connect' and 'bind'
calls now need to provide the actual rproc index. Today's Linux
kernels ensure the rproc index will be consistent from run to run
(so long as all slaves are configured enabled), so this rproc-
index-to-MultiProc-id can be known and hard-coded ahead of time.
This commit records that hard-coding in the MultiProcCfg_*.c
source files, and now provides the appropriate rproc index in
the socket 'connect' and 'bind' calls.
This user-space commit requires a matching update to the kernel
space drivers.
Linux: Remove KERNEL and DRM requirement
Remove the KERNEL and DRM dependency for OMAP54xx and
DRA7xx builds. We'll build as many libs/tests as possible,
given the provided dependencies.
If the user doesn't supply KERNEL, we won't build the MmRpc
libs; if they don't supply DRM, we won't build tests that
depend on DRM... but we'll still build, for example,
MessageQ libs and tests.
Remove the KERNEL and DRM dependency for OMAP54xx and
DRA7xx builds. We'll build as many libs/tests as possible,
given the provided dependencies.
If the user doesn't supply KERNEL, we won't build the MmRpc
libs; if they don't supply DRM, we won't build tests that
depend on DRM... but we'll still build, for example,
MessageQ libs and tests.
MessageQ: Merge new IPC 1.25.03 MessageQ features
The IPC 1.25.03 patch introduced two MessageQ-related features:
SDOCM00099783: MessageQ_free needs a hook function
SDOCM00095724: MessageQ_create should allow a queueId to be
specified instaned of finding free slot
These were driven by MCU/TI-RTOS programs, but may be useful for
other use cases.
This commit brings those features into IPC 3.x.
The IPC 1.25.03 patch introduced two MessageQ-related features:
SDOCM00099783: MessageQ_free needs a hook function
SDOCM00095724: MessageQ_create should allow a queueId to be
specified instaned of finding free slot
These were driven by MCU/TI-RTOS programs, but may be useful for
other use cases.
This commit brings those features into IPC 3.x.
Tests: mmrpc_test: Fix unitialized status variable in mmrpc_test.c
This was causing the mmrpc_test to report failure, when in fact it succeeded.
Signed-off-by: G Anthony <a0783926@ti.com>
This was causing the mmrpc_test to report failure, when in fact it succeeded.
Signed-off-by: G Anthony <a0783926@ti.com>
Tests: mmrpc_test: Removed omapconf comment and getchar from mmrpc_test.
This was previously inserted for debugging purposes.
Signed-off-by: G Anthony <a0783926@ti.com>
This was previously inserted for debugging purposes.
Signed-off-by: G Anthony <a0783926@ti.com>
Tests: rpc_task: Invalidate the compute structure in Mx_compute fxn.
This to ensure the cache is invalidated before reading the fields of this
structure, which may be in cached memory (Tiler).
Signed-off-by: G Anthony <a0783926@ti.com>
This to ensure the cache is invalidated before reading the fields of this
structure, which may be in cached memory (Tiler).
Signed-off-by: G Anthony <a0783926@ti.com>
Build CMEM library only if platform is specified.
CMEM library is platform specific. You just specify a platform
to build the library and programs that use CMEM.
CMEM library is platform specific. You just specify a platform
to build the library and programs that use CMEM.
Added offset pointer support to MmRcp.
New parameter type to support data pointers into managed memory
allocations such that the data pointer is offset from the base
address.
New parameter type to support data pointers into managed memory
allocations such that the data pointer is offset from the base
address.
Moved drm libraries out of common variable.
Ran autoreconf to generate new makefiles and configure script.
tests: mmrpc_test: Updated Mx_compute() to test rpmsgrpc buff translation.
Requires DRM library, to allocate DRM buffers, allowing access to
shared memory via user virtual and DMA_BUF handles.
Signed-off-by: G Anthony <a0783926@ti.com>
Requires DRM library, to allocate DRM buffers, allowing access to
shared memory via user virtual and DMA_BUF handles.
Signed-off-by: G Anthony <a0783926@ti.com>
tests: mmrpc_test: Add DRM library to Linux side makefiles.
DRM library is needed for the Mx_Compute() buffer translation use case.
Add DRM_PREFIX in ipc-linux.mak and Makefile.am
Signed-off-by: G Anthony <a0783926@ti.com>
DRM library is needed for the Mx_Compute() buffer translation use case.
Add DRM_PREFIX in ipc-linux.mak and Makefile.am
Signed-off-by: G Anthony <a0783926@ti.com>
products.mak: Added DRM_PREFIX setting.
This DRM_PREFIX is used by linux/src/tests/Makefile.am
Signed-off-by: G Anthony <a0783926@ti.com>
This DRM_PREFIX is used by linux/src/tests/Makefile.am
Signed-off-by: G Anthony <a0783926@ti.com>
Linux: Added DRM_PREFIX config variable to the Linux build for OMAP5 and DRA7XX devices
Set status field in fw_rsc_vdev after MessageQCopy_attach and before RscTable_update in QNX
Signed-off-by: VW <vwan@ti.com>
Signed-off-by: VW <vwan@ti.com>
MxServer_compute: Incremented num params from 1 to 2 to match signature.
Previously, MxServer_compute function had only 1 paramter (return val included)
in the signature.
This was not being checked by the QnX rpmsgrpc driver, but is being checked by
Linux rpmsgrpc driver, which caught the inconsistency during an RpcCall.
This has been fixed.
Signed-off-by: G Anthony <a0783926@ti.com>
Previously, MxServer_compute function had only 1 paramter (return val included)
in the signature.
This was not being checked by the QnX rpmsgrpc driver, but is being checked by
Linux rpmsgrpc driver, which caught the inconsistency during an RpcCall.
This has been fixed.
Signed-off-by: G Anthony <a0783926@ti.com>
VirtQueue: OMAP5: Poll for VDEV status in VirtQueue_startup()
This resolves a race condition where the first slave RPMSG send would block
in a non-task context on a Semaphore_pend(), when the slave loads faster than
than the host can prime the send buffers in the vring to the host.
This patch implements the solution for OMAP5 separately.
Signed-off-by: G Anthony <a0783926@ti.com>
This resolves a race condition where the first slave RPMSG send would block
in a non-task context on a Semaphore_pend(), when the slave loads faster than
than the host can prime the send buffers in the vring to the host.
This patch implements the solution for OMAP5 separately.
Signed-off-by: G Anthony <a0783926@ti.com>
VirtQueue: Poll for VDEV status in VirtQueue_startup() to sync rpmsg on boot.
Previously, various methods (or none) have been implemented to ensure that
the slave has buffers available in the vring to the host before sending its
first message.
VirtQueue_startup() now polls the status bit in the VDEV struct of the Resource
table for a completion status, written by the host after the VDEV is created
successfully.
This resolves a race condition where the first slave RPMSG send would block
in a non-task context on a Semaphore_pend(), when the slave loads faster than
than the host can prime the send buffers in the vring to the host.
This patch implements the solution for Vayu separately.
Signed-off-by: G Anthony <a0783926@ti.com>
Previously, various methods (or none) have been implemented to ensure that
the slave has buffers available in the vring to the host before sending its
first message.
VirtQueue_startup() now polls the status bit in the VDEV struct of the Resource
table for a completion status, written by the host after the VDEV is created
successfully.
This resolves a race condition where the first slave RPMSG send would block
in a non-task context on a Semaphore_pend(), when the slave loads faster than
than the host can prime the send buffers in the vring to the host.
This patch implements the solution for Vayu separately.
Signed-off-by: G Anthony <a0783926@ti.com>
Resource: Add accessor to read the status field from a VDEV struct.
The Linux host will write a VIRTIO_CONFIG_* status into the remote processor
resource table upon completion of VDEV creation.
As this corresponds to priming of vring buffers for RPMSG type VDEVs, this is
a useful status for the slave to poll before it sends it's first RPMSG
to the host.
Other non-Linux HLOS must implement a similar status update mechanism.
Signed-off-by: G Anthony <a0783926@ti.com>
The Linux host will write a VIRTIO_CONFIG_* status into the remote processor
resource table upon completion of VDEV creation.
As this corresponds to priming of vring buffers for RPMSG type VDEVs, this is
a useful status for the slave to poll before it sends it's first RPMSG
to the host.
Other non-Linux HLOS must implement a similar status update mechanism.
Signed-off-by: G Anthony <a0783926@ti.com>
VirtQueue: Added capability to see Log_print from Vayu/OMAP5 VirtQueue.
The VirtQueue modules for Vayu and OMAP5 are not XDC modules.
For non-XDC files, a module name needs to be registered so that Log_print
and Diags will work.
This introduces a private _VirtQueue_init() function, called in
VirtQueue_create(), to register VirtQueue with the Diags/Log system.
Signed-off-by: G Anthony <a0783926@ti.com>
The VirtQueue modules for Vayu and OMAP5 are not XDC modules.
For non-XDC files, a module name needs to be registered so that Log_print
and Diags will work.
This introduces a private _VirtQueue_init() function, called in
VirtQueue_create(), to register VirtQueue with the Diags/Log system.
Signed-off-by: G Anthony <a0783926@ti.com>
Merge tag '3.00.02.26' into ipc-next
3.00.02.26
3.00.02.26
.gitignore: Ignore patch files
Add *.patch to .gitignore so that patches can remain in the
tree when a regular git clean is performed.
Signed-off-by: Suman Anna <s-anna@ti.com>
Add *.patch to .gitignore so that patches can remain in the
tree when a regular git clean is performed.
Signed-off-by: Suman Anna <s-anna@ti.com>
Configs: DRA7xx: Change IPU1 BIOS tick timer to use GPT11
Both the IPU1 and IPU2 are currently configured to use the
same GPTimer as the source for BIOS tick timer. Switch the
IPU1 timer to use GPT11 instead so that both the processors
can be run simultaneously.
Signed-off-by: Suman Anna <s-anna@ti.com>
Both the IPU1 and IPU2 are currently configured to use the
same GPTimer as the source for BIOS tick timer. Switch the
IPU1 timer to use GPT11 instead so that both the processors
can be run simultaneously.
Signed-off-by: Suman Anna <s-anna@ti.com>
Configs: DRA7xx: Update the dmtimer addresses
DRA7xx has 16 timers and the MMU equivalent addresses are defined
only for a subset of these dmtimers. Define the addresses for all
the dmTimer addresses as per the resource table configuration for
a HLOS usecase.
Signed-off-by: Suman Anna <s-anna@ti.com>
DRA7xx has 16 timers and the MMU equivalent addresses are defined
only for a subset of these dmtimers. Define the addresses for all
the dmTimer addresses as per the resource table configuration for
a HLOS usecase.
Signed-off-by: Suman Anna <s-anna@ti.com>
Remoteproc: DRA7xx: Remove MPU mapping from resource table
The L4_PER mapping entry is split into three individual L4_PERx
entries. This is done to remove any mappings for the MPU-relevant
address spaces into the slave processor MMUs.
Signed-off-by: Suman Anna <s-anna@ti.com>
The L4_PER mapping entry is split into three individual L4_PERx
entries. This is done to remove any mappings for the MPU-relevant
address spaces into the slave processor MMUs.
Signed-off-by: Suman Anna <s-anna@ti.com>
ping_rpmsg: Updated test to take numLoops and CoreId arguments.
Test updated to allow testing rpmsg_proto socket over multiple cores, and
in multiple processes via command shell.
Signed-off-by: G Anthony <a0783926@ti.com>
Test updated to allow testing rpmsg_proto socket over multiple cores, and
in multiple processes via command shell.
Signed-off-by: G Anthony <a0783926@ti.com>
Vayu: Added DRA7XX platform to configure.ac and Makefile.am for Linux builds.
Also, added DRA7XX MultiProc configuration table to build MessageQ tests for
Vayu.
Signed-off-by: G Anthony <a0783926@ti.com>
Also, added DRA7XX MultiProc configuration table to build MessageQ tests for
Vayu.
Signed-off-by: G Anthony <a0783926@ti.com>
SrvMgr: Publish OMX server for DRA7xx IPU1 & DSP2 processors
Add logic to publish separate rpmsg-omx channel devices to the
host for IPU1 and DSP2 processors for DRA7xx. These are kept
different so that they can be run simultaneously with IPU2 and
DSP1 processors without causing any name conflicts.
Signed-off-by: Suman Anna <s-anna@ti.com>
Add logic to publish separate rpmsg-omx channel devices to the
host for IPU1 and DSP2 processors for DRA7xx. These are kept
different so that they can be run simultaneously with IPU2 and
DSP1 processors without causing any name conflicts.
Signed-off-by: Suman Anna <s-anna@ti.com>
Vayu: Tests: Add support for IPU1 and DSP2
- Resource Table: Vring physical addresses are made unique depending on VAYU
core, and need to match CMA base addresses; given both DSP and IPU
have IOMMU's, the device addresses otherwise remain the same.
- Adjust test image memory maps to fit in 6MB footprint for initial testing.
- Add support to build tests for IPU1 and DSP2 platform instances.
Signed-off-by: G Anthony <a0783926@ti.com>
[s-anna@ti.com: add dsp2 and ipu1 build targets]
Signed-off-by: Suman Anna <s-anna@ti.com>
Vayu: Tests: Adjust memory maps for IPU1 and DSP2
Retain the memory map for IPU2 while creating a newer
memory map for IPU1. Also fix some of the other sizes.
Signed-off-by: Suman Anna <s-anna@ti.com>
- Resource Table: Vring physical addresses are made unique depending on VAYU
core, and need to match CMA base addresses; given both DSP and IPU
have IOMMU's, the device addresses otherwise remain the same.
- Adjust test image memory maps to fit in 6MB footprint for initial testing.
- Add support to build tests for IPU1 and DSP2 platform instances.
Signed-off-by: G Anthony <a0783926@ti.com>
[s-anna@ti.com: add dsp2 and ipu1 build targets]
Signed-off-by: Suman Anna <s-anna@ti.com>
Vayu: Tests: Adjust memory maps for IPU1 and DSP2
Retain the memory map for IPU2 while creating a newer
memory map for IPU1. Also fix some of the other sizes.
Signed-off-by: Suman Anna <s-anna@ti.com>