Restore ability to use (and test!) zephyr/docker-build.sh locally
without overwriting work in progress.
Fixes commit b371373f7d ("scripts: docker-build.sh updated with
changes from xtensa-build-zephyr.py")
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
Connecting buffers belongs to pipeline initialisation, it can be done
completely on the master core.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Kconfig options with no description cannot be selected from
defconfig. They can only be selected from other Kconfig options.
This means, that CONFIG_INCOHERENT doesn't get selected on TGL. Move
it to Kconfig to fix this.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
test_keyword_params() must acquire and release the source buffer
appropriately for access.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
When a component or a buffer is freed, it has to be made inaccessible
from everywhere. Make sure to always delete buffers from lists in
such cases and also to use correct direction macros.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Buffers, attached to mixers, can be remaining parts of pipelines,
being freed. Then their component pointers might no longer point to
valid components. Check them before dereferencing.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Freeing components and buffers is performed one by one from IPC
commands, sent by the host. When a component is freed, buffers,
connected to it might still exist and keep pointers to the component,
that is being freed. Then, for example if such a buffer is also
connected to a mixer from a different pipeline, that other pipeline
might still be active. Then the mixer might walk its connected
pipelines and come across the no longer existing component. Remove
those stale pointers to avoid such issues.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Both IPC3 and IPC4 stream termination paths use uncached buffer
addresses, which can contain stale data with core-local buffers. Fix
both to use cached buffer aliases.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Buffer control structures are accessed at run-time via their cached
aliases. But buffer enumeration involves accessing list headers,
embedded in those control structures. To be able to use the standard
list traversal routines that access is performed via uncached
aliases. Therefore the life-cycle of a buffer control object looks as
follows:
- initialisation
1. allocate memory
2. obtain a cached alias
3. use cached access to initialise the audio stream
4. release cached access
5. initialise some fields, using the uncached alias
- linking
6. add to the list
7. obtain a cached alias
8. bind to a component
9. release cached access
- run-time
10. scan buffer list
11. obtain a cached alias
12. use the buffer
13. release cached access
This patch fixes buffer object initialisation and linking to ensure
data consistency between cached and uncached accesses.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Current component object cache synchronisation is wrong. Firstly
component objects are always allocated with uncached addresses, so no
cached access ever takes place. Secondly cache manipution should be
performed via cached addresses. Thirdly if we decide to access
component objects via cached addresses we'll likely use the coferent
API to do this. Remove wrong cache manipulation to fix cases with
componentson different cores within a single pipeline.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Previously we temporarily added two functions
comp_update_buffer_cached_produce() and
comp_update_buffer_cached_consume() to be able to migrate users of
comp_update_buffer_produce() and comp_update_buffer_consume() to them
respectively. Now that all users have been migrated we can remove
original implementations and restore original names.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Zephyr removed support for header inclusion without the zephyr/ path.
Update SOF with correct paths.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Added versioning to scripts/xtensa-build-zephyr.py to get version
information when incompatible changes are done to the script.
Added yml schema version number to west.yml manifest.
Signed-off-by: Andrey Borisovich <andrey.borisovich@intel.com>
Removed old flags from python script call and replaced with
new one. Added one more working directory to zephyr-build container
now SOF is placed in /workdir/sof and west workspace in /workdir .
Signed-off-by: Andrey Borisovich <andrey.borisovich@intel.com>
xtensa-build-zephyr.py had been using some execute_command
calls when building rimage and using rimage keys over
symlink from modules/audio/sof. Changed it so that calls are
executed now over normal sof directory and symlink is not needed
for script execution.
Signed-off-by: Andrey Borisovich <andrey.borisovich@intel.com>
Added new flag for non-interactive mode that should be used
when script is invoked programmatically.
In default (interactive) mode user is asked using a prompt whether to
reinitialize west manifest if it is initialized to manifest other
than SOFs.
Signed-off-by: Andrey Borisovich <andrey.borisovich@intel.com>
Added usage of west manifest in the script to update both
SOF and zephyr dependencies.
Renamed "clone-mode" to "update" flag - the logic where
sof dependencies and zephyr with dependencies is cloned.
Removed "point-mode" from script as it will no longer work
(west manifests require fixed directory structure).
Removed "zephyr revision" or "-z" argument as it is no longer
needed (now revision may be set in west.yml manifest).
Added "-p" pristine flag to rebuild platforms while removing build
directory.
Co-developed-by: Krzysztof Frydryk <krzysztofx.frydryk@intel.com>
Signed-off-by: Krzysztof Frydryk <krzysztofx.frydryk@intel.com>
Signed-off-by: Andrey Borisovich <andrey.borisovich@intel.com>
Added west.yaml file, that manages zephyr repo and its dependencies.
Additionally west manifest may now control sof submodules.
Added submanifests directory with README.txt file so the
submanifests directory exists in version control - otherwise
west update command returns error. This is bug described in
https://github.com/zephyrproject-rtos/west/issues/594 .
Co-developed-by: Andrey Borisovich <andrey.borisovich@intel.com>
Signed-off-by: Andrey Borisovich <andrey.borisovich@intel.com>
Signed-off-by: Krzysztof Frydryk <krzysztofx.frydryk@intel.com>
Don't fail immediately when the driver is not loaded. Use inotify
instead to wait for /sys/kernel/debug/sof/[e]trace to appear.
This makes it possible to start before the driver is loaded which
reduces considerably the chances of missing early logs.
Fixes a small part of https://github.com/thesofproject/linux/issues/3275
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
Open /sys/kernel/debug/sof/fw_version _after_
/sys/kernel/debug/sof/[e]trace because reading the former is optional
and the latter is not.
So when the driver is not loaded, we get the same (missing trace) error
trace message whether we use the -n option or not.
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
If operation of disabling secondary core fails at any of the steps, host
will only learn about it after checking the SPA/CPA bits.
This patch adds checking core state after it disabling. If core will be
still active ipc response will report error.
Additionally, it removes an outdated comment.
Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
Add possibility to apply gain to mixed sources to avoid
mixed output saturation. Gain (attenuation) is configurable
per source and comes from host via IPC4 large config set.
Signed-off-by: Serhiy Katsyuba <serhiy.katsyuba@intel.com>
Because of the fact that some platforms use a software host DMA
(e.g: i.MX8QM) pipeline_copy() implicitly calls
pipeline_get_timestamp() when a host copy operation occurs.
The issue with this is that when the host copy operation
occurs, pipeline_for_each() is in the middle of a pipeline
graph traversal. Because of this, pipeline_get_timestamp()
is not able to get to the DAI because it will stop the
pipeline graph traversal at the host buffer (because
of the fact that the 'walking' flag of the buffer is set
to true). This is bad because this way, pipeline_get_timestamp()
is unable to get the updated dai position.
The aim of this fix is to bypass the conditioned pipeline graph
traversal (based on the 'walking' flag) and use an unconditioned
pipeline graph traversal (pipeline_get_dai_comp())
in order to reach the DAI component and get its updated position.
Signed-off-by: Laurentiu Mihalcea <laurentiu.mihalcea@nxp.com>
Originally, pipeline_get_dai_comp would be able to find the
DAI only for the PLAYBACK case because the direction would
be hardcoded and would be PPL_DIR_DOWNSTREAM.
The aim of this change is to make pipeline_get_dai_comp more
general and give it the ability to find the DAI for the
CAPTURE case as well.
Signed-off-by: Laurentiu Mihalcea <laurentiu.mihalcea@nxp.com>
This file provides the flexibility to test any Zephyr commit in Linux
CI without any pollution of `git -C sof describe --dirty`. See (too)
long discussion in #6005 for details.
This new file will have no effect until #6005 is merged. However it can
and SHOULD be merged before #6005 because #6005 is backwards
incompatible and requires a "flag day" and the synchronization of
several moving parts. So merging this commit earlier means having one
less moving part to synchronize and it will simplify "flag day"
rehearsals.
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
Patch a0ff2bb277 added field .gain to configuration data for
src_generate() but the previous generator scripts were not
updated to set it. This patch fixes in them the error about
undefined gain by setting it to previous -1 dB default.
Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
This patch updates the SRC tool to export same guidance for
Doxygen as added by a previous patch for coefficients
header files. The added comments to begin and end are:
/** \cond GENERATED_BY_TOOLS_TUNE_SRC */
/** \endcond */
The impacted headers remain identical so they don't need update.
Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
With Zephyr, SOF uses k_cycle_get_64() API in order to
count clock cycles.
Not all platforms have a 64bit counter (e.g i.MX dsp integration
only has xtensa timer which is limited to 32 bits).
So, instead of using k_cycle_get_64 introduce new API sof_cycle_get_64
which keeps the existing behavior for platforms which define
CONFIG_TIMER_HAS_64BIT_CYCLE_COUNTER, otherwise use k_uptime_ticks().
See comments from @andyross at
https://github.com/zephyrproject-rtos/zephyr/pull/48318
Suggested-by: Andy Ross <andrew.j.ross@intel.com>
Signed-off-by: Daniel Baluta <daniel.baluta@nxp.com>
Prefer to use source components direction, rather than dai direction when
setting pipeline components direction. Accept first direction set in
pipeline. This properly handles
host_copier -> module -> host_copier pipelines.
Signed-off-by: Krzysztof Frydryk <krzysztofx.frydryk@intel.com>
while simultaneous playback and internal mic capture usecases
are running, there is a case where dmic driver was clearing the
playback interrupt.This might cause noise in playback or no audio
in speaker.Done changes to clear only dmic interrupt status only
in dmic irq handler.
Signed-off-by: Balakishorepati <balaKishore.pati@amd.com>
Group together the Cadence main Kconfig option with the options related
to various Cadence based codecs.
This is only a visual fix.
Signed-off-by: Paul Olaru <paul.olaru@nxp.com>
When module adapter is enabled, it is best to keep the new related
options next to it, rather than one option after that.
Signed-off-by: Paul Olaru <paul.olaru@nxp.com>
Deprecate the usage of circular buffer check while doing convert, and
use 2-way and 4-way xtensa instructions first to optimize the converter
functions, and process the left samples one by one to avoid potential
risk of memory access overrun. Here are the results:
Optimize pcm_convert_s16_to_s24 and save about 38% cycles;
Optimize pcm_convert_s24_to_s16 and save about 25% cycles;
Optimize pcm_convert_s16_to_s32 and save about 39% cycles;
Optimize pcm_convert_s32_to_s16 and save about 40% cycles;
Optimize pcm_convert_s32_to_s24 and save about 38% cycles;
Optimize pcm_convert_s24_to_s32 and save about 47% cycles;
Signed-off-by: Andrula Song <xiaoyuan.song@intel.com>
Fill in cfg.rate to be used by zephyr native drivers as rate is not
always present in ipc config.
Signed-off-by: Krzysztof Frydryk <krzysztofx.frydryk@intel.com>
Fix the memory overrun risk of odd samples to be processed.
Since the 2-way xtensa instruction would always process 2
samples and there would be risk of memory overrun at the
boundary of buffer.
Signed-off-by: Andrula Song <xiaoyuan.song@intel.com>
Added glue logic to integrate with module adapter
the loadable modules developed under Intel
IADK Framework.
Signed-off-by: Jaroslaw Stelter <Jaroslaw.Stelter@intel.com>
Kai Vehmanen