adsp_memory.h is pretty much the same for all ace platforms.
Generalize it getting register address from devicetree and
and move it to a common place.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Platforms that support IPIs allow them to be broadcast via the
new arch_sched_broadcast_ipi() routine (replacing arch_sched_ipi()).
Those that also allow IPIs to be directed to specific CPUs may
use arch_sched_directed_ipi() to do so.
As the kernel has the capability to track which CPUs may need an IPI
(see CONFIG_IPI_OPTIMIZE), this commit updates the signalling of
tracked IPIs to use the directed version if supported; otherwise
they continue to use the broadcast version.
Platforms that allow directed IPIs may see a significant reduction
in the number of IPI related ISRs when CONFIG_IPI_OPTIMIZE is
enabled and the number of CPUs increases. These platforms can be
identified by the Kconfig option CONFIG_ARCH_HAS_DIRECTED_IPIS.
Signed-off-by: Peter Mitsis <peter.mitsis@intel.com>
headers for dmic are now part of the SoC and maintained per generation,
so create one header for PTL and build the code for PTL in some of the
drivers (dmic_nhlt).
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
This commit adds definition of ACE 3.0 Panther Lake board.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
power_down() expects a cached pointer. Fix the sparse annotation
to match the implementation (sys_cache_cached_ptr_get() returns a cached
pointer so this is correct).
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
This commit introduces support for an alternate linking method in the
LLEXT subsystem, called "SLID" (short for Symbol Link Identifier),
enabled by the CONFIG_LLEXT_EXPORT_BUILTINS_BY_SLID Kconfig option.
SLID-based linking uses a unique identifier (integer) to identify
exported symbols, instead of using the symbol name as done currently.
This approach provides several benefits:
* linking is faster because the comparison operation to determine
whether we found the correct symbol in the export table is now an
integer compare, instead of a string compare
* binary size is reduced as symbol names can be dropped from the binary
* confidentiality is improved as a side-effect, as symbol names are no
longer present in the binary
Signed-off-by: Mathieu Choplain <mathieu.choplain@st.com>
Create a pseudo-terminal to access Zephyr shell on the audio DSP.
The shell terminal is enabled with "-p" command-line option.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Add a new shell backend implemented over a shared memory window
on the Intel audio DSPs. The implementation uses the Zephyr winstream
to manage the data streaming.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Cosmetic change to align code style when initializing DSP registers. The
code in intel_is_ace() branch was moved as-is from acetool.py when the
two tools were merged to make reviewing easier. Fix the code style to be
coherent in the merged cavstool.py. No functional change.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Use the correct register to read ROM status on intel_adsp_ace20.
Without this this fix, firmware load is successful but
boot takes extra 2 seconds and following warning was emitted:
WARNING:cavs-fw:Load failed? ROM_STATUS = 0x0
The log-only mode (-l) was not working at all and is fixed
by this commit.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
This adds support for the local APIC in one-shot mode as the timeout
event source for those cases where the CPU supports invariant TSC but
no TSC deadline capability. It is presented as another timer choice.
Existing Kconfig symbols were preserved to minimize board config
disturbance.
This hybrid approach was implemented kind of backward in the apic_timer
driver but it is far cleaner to carry this here.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Namespaced the generated headers with `zephyr` to prevent
potential conflict with other headers.
Introduce a temporary Kconfig `LEGACY_GENERATED_INCLUDE_PATH`
that is enabled by default. This allows the developers to
continue the use of the old include paths for the time being
until it is deprecated and eventually removed. The Kconfig will
generate a build-time warning message, similar to the
`CONFIG_TIMER_RANDOM_GENERATOR`.
Updated the includes path of in-tree sources accordingly.
Most of the changes here are scripted, check the PR for more
info.
Signed-off-by: Yong Cong Sin <ycsin@meta.com>
The software protocol to write status value of 0x05 (FW_ENTERED)
into memory window 0 at Zephyr boot, is not needed in the ace1.x
boot flow and does not match the semantics host systems are expecting
at this location in the memory window (e.g. write of 0x05 is not
expected).
Make this logic specific to intel_adsp_cavs platforms and move the code
out from common intel_adsp code.
This commit depends on update to cavstool.py to use correct
ROM status register to observe boot state.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Starting with ace1.x, the boot status is no longer reported by
the boot ROM in the SRAM status window as it was done in older
platforms. The current cavstool.py code works on these newer platforms,
as Zephyr soc bootcode writes to same location, but this is not
the recommended boot flow.
Modify boot flow to use a dedicated register to observe boot
state. This change improves usability of cavstool.py on ace1.x
platforms as:
- it is possible to start cavstool.py (e.g. in log-only or shell mode)
while DSP has been already been booted, but is currently in
low-power mode (and SRAM window is not accessible from host)
- more reliable boot and better error reporting as actual ROM
status is observed
Furthermore, this change allows to remove the memory window
writes from Zephyr intel_adsp boot_complete(). This IPC interface
is application and IPC revision specific and the write should not
be done in generic Zephyr SoC code. However, to keep cavstool.py
working, the tool has to be updated first.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Secondary cores are not allowed to be power gated on
runtime-idle. They have to explicitely set off by host command.
Remove this state from secondary CPUs so power management logic
does not need workarounds to enforce this behavior.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
k_sem_init() is called for every IPC message sent in
intel_adsp_ipc_send_message(). This has not had any side-effects
in upstream configurations, but has been linked to a failing
run of test_obj_tracking_sanity test case in downstream Zephyr
use.
Replace k_sem_init() with k_sem_reset() as this is more appropriate
API to reset the semaphore count, and ensure deterministic
behaviour in case a thread is waiting on the semaphore at time
of reset.
Suggested-by: Peter Mitsis <peter.mitsis@intel.com>
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Corrected virtual memory size to match the range supported by the
Translation Lookup Buffer. The TLB size is 16 MB, however the first 128 KB
is dedicated to LPSRAM and bypasses the TLB. This was taken into account in
KERNEL_VM_BASE, so KERNEL_VM_SIZE was reduced accordingly.
Signed-off-by: Adrian Warecki <adrian.warecki@intel.com>
Merge codebases of cavstool.py and acetool.py as the two have
a lot of duplicated code.
To ease with transition, keep acetool.py around with implementation
imported from cavstool.py. This will help to keep any automated
testing flows working that assume both tools exist.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Introduce z_page_frame_set() and z_page_frame_clear() to manipulate
flags. Obtain the virtual address using the existing
z_page_frame_to_virt(). This will make changes to the page frame
structure easier.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
The ipc driver device data (struct intel_adsp_ipc_data) contains a
semaphore. Upon device init, the device data is zeroed out. This is safe
for other fields, but the semaphore should be properly initialized
before use.
This lack of initialization leads to a system crash when CONFIG_POLL is
enabled (e.g. to enable CONFIG_SHELL), IPC driver handles an interrupt
and executes k_sem_give() on a uninitialized semaphore object. This will
eventually lead to null dereference in z_handle_obj_poll_events().
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
This commit updates the device tree and memory header file
for the Intel MTPM 1.5 platform to define the LSBPM and
HSBPM registers.
Changes include:
- Added node definitions for 'lsbpm' and 'hsbpm' in
intel_adsp_ace15_mtpm.dtsi
- Updated adsp_memory.h
Signed-off-by: Damian Nikodem <damian.nikodem@intel.com>
This commit updates the device tree and memory header file
for the Intel LNL 2.0 platform to define the LSBPM and
HSBPM registers.
Changes include:
- Added node definitions for 'lsbpm' and 'hsbpm' in
intel_adsp_ace20_lnl.dtsi
- Updated adsp_memory.h
Signed-off-by: Damian Nikodem <damian.nikodem@intel.com>
This commit updates the device tree and memory header file
for the Intel cAVS 2.5 platform to define the LSBPM and
HSBPM registers.
Changes include:
- Added node definitions for 'lsbpm' and 'hsbpm' in
intel_adsp_cavs25.dtsi and intel_adsp_cavs25_tgph.dtsi
- Updated adsp_memory.h
Signed-off-by: Damian Nikodem <damian.nikodem@intel.com>
A workaround to avoid icache corruption was added in commit be881d4cf2
("arch: xtensa: add isync to interrupt vector").
This patch implements a different workaround by adding custom logic to
idle entry on affected Intel ADSP platforms. To safely enter "waiti"
when clock gating is enabled, we need to ensure icache is both unlocked
and invalidated upon entry.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
IPC has inverted the usage of the state lock API.
In this API semantics, the get method disallow the policy of
using the given state, while the put() release this constraint.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Fix the inline documentation to match implementation. As IPFL is used,
the correct matching function is xthal_icache_region_lock().
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Fix the inline documentation to match implementation. As IPFL is used,
the correct matching function is xthal_icache_region_lock().
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Checking if the device state is locked is not correct, device runtime
put / get is no longer checking for it and the only place that was
locking device's state was the pm action callback and if the idea
was to protect some concurrent call, using it is not enough.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
There is no needed to lock/unlock a device state in its pm action
callback. The power management subsystem should take care of
serialize these calls.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Prevent the system from entering runtime idle state during IPC
transactions until the HOST acknowledgment is received.
This patch modifies the IPC mechanism to:
- Lock the runtime idle state immediately after sending an IPC message
to the HOST, preventing the system from entering a low-power state.
- Unlock the runtime idle state once the IPC transaction is acknowledged
by the HOST, allowing the system to enter low-power states if
conditions permit.
The changes ensure that the DSP does not enter a power state that could
interrupt the IPC communication process, maintaining the integrity of
the IPC state machine.
Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
Fixes the following errors when sparse (SCA) is enabled:
soc/intel/intel_adsp/ace/power.c:46:12: warning:
cast removes address space '__cache' of expression
/soc/intel/intel_adsp/ace/power.c:48:9: warning:
incorrect type in argument 1 (different address spaces)
Fixes#70725
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Simple rename to align the kernel naming scheme. This is being
used throughout the tree, especially in the architecture code.
As this is not a private API internal to kernel, prefix it
appropriately with K_.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Issue an upstream read transaction through uncached memory to flush
out all pending transactions before power down the host domain.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
This patch updates the power status register bitfield definitions in the
power management header for the Intel ADSP ACE 2.0 LNL platform.
Modifications include:
- Adjusting the 'ioxpgs' field from 4 bits to 2 bits.
- Adding a 'rsvd11' field with 2 bits to reflect reserved space.
- Changing the 'mlpgs' field from 2 bits to 1 bit.
- Updating the 'rsvd14' field from 1 bit to 2 bits for alignment.
These changes ensure that the power status register bitfields match the
latest hardware specification for the ACE 2.0 LNL SoC, which is crucial
for accurate power domain status monitoring.
Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
When non-primary core is powered down and restart with sequence of:
- PM state set to SOFT_OFF
- once target core is idle, cut power with soc_adsp_halt_cpu()
- power up core again with k_smp_cpu_resume()
The execution will continue from stored DSP core context, but
will hit an assert in z_smp_cpu_mobile() as the PS.INTLEVEL
is zero.
Fix this issue by storing and restoring PS register in this flow.
Link: https://github.com/zephyrproject-rtos/zephyr/issues/70181
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
Transition to a low power DMI L1 state should be allowed only after all
pending DMA channels transfers have started.
Signed-off-by: Serhiy Katsyuba <serhiy.katsyuba@intel.com>
In case the core is not power gated, waiti will restore intlevel. In
this case we lock interruption after it.
In the bug scenario, the host starts streaming and via SOF APIs, keeps a
lock to prevent Zephyr from entering PM_STATE_RUNTIME_IDLE. During the
test case, host removes this block and core0 is allowed to enter IDLE
state.
When core0 enters power gated state, interrrupts are left enabled (so
the core can be woken up when something happens). This leaves a race
where suitably timed interrupt will actually block entry to power gated
state and k_cpu_idle() in power_gate_entry() will return. This is rare,
but happens often enough that the relatively short test plan run on SOF
pull-requests will trigger this case.
Fixes#69807
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
z_mp_entry() and power_gate_exit() are only called from assembly code
in the same file, where they're defined. Make them "static" and add
an attribute to let the compiler know, that they aren't unused.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
This enables the use of CONFIG_XTENSA_MORE_SPIN_RELAX_NOPS to
specify how many NOPs for arch_spin_relax(). This is to keep
the atomic_cas() (via s32c1i) from saturating the internal bus
with contant RCW ops. The numbers should be further tuned
for specific application but these should provide a good start.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
