The XTOS allocator aligns non-cachec allocations to the platform
cacheline size. Test results indicate some SOF application code
relies on this behaviour, so align Zephyr's rmalloc implementation
to match the behaviour.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Implement the cached SOF heap zones using a single sys_heap
object.
Compared to old implementation which used two separate sys_heaps,
this implementation enables dynamic cached/uncached partitioning
of the heap.
To ensure coherency of the Zephyr heap operation, if cached heap
is enabled, all allocations must be aligned to dcache linesize.
Add a Kconfig option CONFIG_SOF_ZEPHYR_HEAP_CACHED to turn on/off
the cached heap logic, with default set to no for all platforms.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
The heap code was invalidating blocks on allocation, but that's the
wrong side of the pipe. By definition, new heap memory
will/should/must be written before being used (because the memory is
undefined), so any cached contents are irrelevant as they'll be
overwritten.
But when the user is finished with the block and frees it, there may
still be live dirty cache lines in the region on the current CPU.
Those must be invalidated, otherwise they will be evicted from the
cache at some point in the future, on top of the memory region now
being used for different purposes on another CPU.
Remove the invalidate on allocation. Add it back in free. Leverage a
new Zephyr sys_heap_usable_size() routine to get the size so we don't
have to store it in an extra header.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Building SOF with Zephyr for Apollolake with XCC fails because of
insufficient RAM. Reduce the heap size to fix the problem.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
We have change the size of runtime_shared zone to be decided at link
stage, here hardcode HEAP_RUNTIME_SHARED_SIZE to unblock the Zephyr
building.
Signed-off-by: Keyon Jie <yang.jie@linux.intel.com>
sof_heap variable is used by all platform, with or
without ENABLE_CACHED_HEAP.
Without this fix, on i.MX with Zephyr we get:
error: 'sof_heap' undeclared (first use in this function);
Fixes: fed69b6a34 ("zephyr: wrapper: disable cached heap on Intel platforms")
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Tests with dynamic-pipeline topologies increase the complexity
of alloc/free patterns for heap and these tests have uncovered
new issues with heap corruption. Tests are reporting failed
allocs although heap still has free space. Running same test
with cached heap disabled shows no failures.
The cached heap is already disabled for IMX targets, so build on
this code and add a local ENABLE_CACHED_HEAP define to toggle
the cached heap.
BugLink: https://github.com/thesofproject/sof/issues/4494
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Clean up the ifdefs for ipc_pm_context_save(). By implementing
platform_timer_stop() in the Zephyr wrapper, the original context
save sequence can be used also for Zephyr builds.
By doing this, we can fix issues with IPC timeout on CTX_SAVE.
Disabling all interrupts before the IPC reply is needed also on
Zephyr.
BugLink: https://github.com/thesofproject/sof/issues/4507
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
In XTOS SOF, ipc_send_queued_msg() is run by task_main_primary_core().
In Zephyr we need to schedule ipc_send_queued_msg() using a notifier
triggered by the periodic ll_scheduler.
This is similar to commit c194125b83 ("zephyr:
add notifier_register(ipc_send_queued_msg) in task_main_start()")
For i.MX we need to use this temporary fix for SOF_SCHEDULE_LL_DMA, also.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
On i.MX the DMA interrupts are routed via IRQ_STEER.
In order for this to work we need to:
- make any second level interrupts handling go
through interrupt-irqsteer.c;
- use first level interrupt handling from
wrapper.c.
TODO: Implement a driver for the IRQ_STEER in Zephyr,
to replace the legacy code (interrupt-irqsteer.c).
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Commit 34bb9b7282 ("zephyr: wrapper: build SOF with Zephyr for imx")
broke coherent memory allocations on Intel DSP platforms. Restore the
original code.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
When building SOF with Zephyr, for i.MX, we get the following
error: undefined reference to `z_xtensa_cache_inv'.
i.MX has only one DSP core, so CONFIG_KERNEL_COHERENCE is not set.
Therefore include arch/xtensa/cache.h, where z_xtensa_cache_inv
is defined and implemented, in cases when the kernel is not built in
a mode where all shared data is placed in uncached memory.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
When building SOF with Zephyr for i.MX the following
additional adjustments, to the wrapper, are required:
1. include heapmem variable in .heap_mem section,
otherwise the HEAP_SIZE is duplicated in two sections and
the sdram0 region overflows;
2. no need to split heap into shared and unshared since
we only have 1 DSP core; In this case, the kernel will
never be built in a mode where all shared data is placed
in multiprocessor-coherent (generally "uncached") memory.
3. use simple interrupt_get_irq() to get the Linux interrupt
and later pass it to irq_steer;
4. use Xtensa timer, as we do now with SOF and XTOS.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Fixes#4356
In XTOS SOF, ipc_send_queued_msg() is run by task_main_primary_core(),
the initialization of which has a hard and circular dependency with the
initialization of the EDF scheduler. This EDF scheduler is not part of
Zephyr. Because of the circular dependency it does not seem
possible (nor desirable?) to refactor this code and make it compatible
with Zephyr. So schedule ipc_send_queued_msg() differently in Zephyr:
using a notifier triggered by the periodic Low Latency scheduler.
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
To increase performance of audio application code, the SOF heap should
be mapped to cached address region on platforms with aliased
cached/uncached memory layout.
This splits the heap into two and maps the SOF_MEM_ZONE_BUFFER to
the cached heap and all rest to uncached heap.
The raw heap memory area is still mapped to uncached memory, even for
the cached heap. The allocated blocks are converted to cached addresses
and passed to SOF users of rmalloc() and rballoc_align(). To ensure
application buffers do not share cachelines with heap metadata chunks,
allocation sizes are rounded up to integer multiples of platform
cacheline size.
BugLink: https://github.com/thesofproject/sof/issues/4306
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
The key-phrase-buffer (kpb) component can be included in
Zephyr-based builds, but the component is not initialized
properly in the builds.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Notifier data has to be per-core and its allocation is
performed in init_system_notify(), no need to also do it in
architecture code.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
sof_main() doesn't run on secondary cores. Instead we provide
a new function, that initialises a dummy thread, runs SOF
initialisation by calling secondary_core_init() and switches
over to the dummy thread.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Move logging code and static data to platform/library and
declare a LOG module properly.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
The use of Zephyr LOG_*() API in SOF is breaking compilation.
Replace it with with the native trace logging until a proper
fix is available.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
SOF registers IRQ handlers dynamically, therefore disabling
support for dynamic interrupts for SOF builds doesn't make
much sense. However, this can be used to at least increase
compile-time testing coverage and potentially enable
applications, not using dynamic interrupt registration.
BugLink: https://github.com/thesofproject/sof/issues/4044
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
The Zephyr timer implementation reads high and low 32 bits
in a non-atomic way. Add a loop to make sure the read is
consistent.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Recently added platform_timer_get_atomic() has to be
implemented for Zephyr too.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Add some files from src/platform/library/ to generic arch
builds with Zephyr.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
When building SOF with Zephyr for unsupported architectures
the following additional adjustments to the wrapper are required:
1. define HEAP_RUNTIME_SIZE
2. implement stubs for interrupt handlich code
3. remove arch_timer_get_system() as it's provided by a header
4. hardcode "return 0;" for platform_timer_get()
5. disable sys_module_init() which anyway isn't yet functional
under Zephyr
6. add dummy platform_init() and platform_boot_complete()
7. add log_const_sof
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Broadwell support with Zephyr had only been dry-coded, it has never
been tested, and it contains an infinite recursion bug: functions
platform_timer_get() and arch_timer_get_system() call each other
infinitely. Break the recursion formally, until a proper fix is
available.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
This kconfig variable no longer exists in upstream Zephyr (the feature
it used to control is now zero-overhead and always enabled).
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Audio component samples have been moved to a new location, fix Zephyr
builds to account for it.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
This patch finalises the wrapper to allow LL scheduling of
timer domain audio, EDF sccheduling of IPC, trace re-direction
and kconfig build support.
There are still many items that are WIP and have been marked with
TODO: comments.
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
Add a wrapper to convert SOF API calls to Zephyr equivalents. This allows
development to continue in parallel with xtos/HAL and Zephyr RTOSes until
parity is reached.
The initlial Zephyr APIs to be used are HAL, boot, IRQs, scheduling and
memory. The wrapper reflects this and will probably grow until feature
parity is reached when it can be deleted (as Zephyr APIs can be used
directly).
Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>
Kai Vehmanen