In pm_state_set we can't just call k_cpu_idle() because
this will clear out PS.INTLEVEL. Use k_cpu_atomic_idle instead
since Zephyr's expect interruptions to be locked after pm_state_set.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
We are arbitrarily setting a value to PS after power gates and
losing valid information like OWB, CALLINC and INTLEVEL.
We need to properly save/restore them to avoid possible wrong behavior.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Only when CONFIG_MP_MAX_NUM_CPUS > 1, then .bss is put in
uncached region. Otherwise, .bss is in cached region.
So the assertion that g_key_read_holder must be in uncached
region must take into account how many CPUs are enabled on
build.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Commit 3b99fb1b4a ("xtensa: do not imply atomic ops kconfig") removed
ATOMIC_OPERATIONS_ARCH at xtensa arch level. This triggers a bug in
intel_adsp cavs builds with XCC compiler as
CONFIG_ATOMIC_OPERATIONS_BUILTIN is not defined but neither is
CONFIG_ATOMIC_OPERATIONS_ARCH anymore, resulting in failed builds.
Fix the XCC build by defining CONFIG_ATOMIC_OPERATIONS_ARCH at
soc level.
Signed-off-by: Kai Vehmanen <kai.vehmanen@linux.intel.com>
use CONFIG_SOC_INTEL_ACE15_MTPM instead of CONFIG_ACE_VERSION_1_5.
CONFIG_ACE_VERSION_1_5 leaked from SOF.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
This patch enables time stamping controlled by DSP Timers / Time Stamping
logic on ACE1.5 / ACE2.0 platforms.
Signed-off-by: Tomasz Lissowski <tomasz.lissowski@intel.com>
The soc.c interrupt-related definitions are supposed to
provide support for multi-level interrupts. At the moment,
the way the functions work is they only process the LEVEL 1
interrupt from the encoded INTID and treats the provided INTID
as if it were simply a LEVEL 1 interrupt, which is wrong. Another
issue with soc.c is the fact that the definitions from it clash
with the ones provided by the IRQSTEER driver. To fix this, remove
the soc.c file altogether and change the corresponding CMakeLists.txt
to only contain the necessary statements.
Signed-off-by: Laurentiu Mihalcea <laurentiu.mihalcea@nxp.com>
When exiting power gated state, call the CPU start function
passed to arch_start_cpu().
Signed-off-by: Rander Wang <rander.wang@intel.com>
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Use 'switch' to emphasise that we're handling different values of
'state' in pm_state_set().
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
'ret' in pm_state_set() is always set before it's used, no need to
initialise it.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
pm_device_runtime_get() must be called after pd_intel_adsp_init() is
called for each device, because the latter calls
pm_device_runtime_enable(), which sets the device runtime PM use
count to 0. The current wrong calling order causes a DSP panic
because of an unbalanced pm_device_runtime_put(). Fix this by
delaying pm_device_runtime_get() until the POST_KERNEL initialisation
step.
Fixes commit c3a6274bf5 ("intel_adsp: ace: power: Prevent HST
domain power gating")
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
.bss and .data are uncached in Zephyr builds for intel_adsp. No need
to try to manipulate cache of objects in those sections.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
SOC_ESP32_NET is now SOC_ESP32_APPCPU, following espressif's
naming convention in the same manner as ESP32S3 app cpu.
SOC_ESP32_APPCU is now a subset of SOC_SERIES_ESP32.
This commit also changes the necessary files, samples and tests
for bisect purposes.
Signed-off-by: Lucas Tamborrino <lucas.tamborrino@espressif.com>
This patch enhances the power-down sequence for the HOST (HST) domain
within the Intel ADSP ACE 1.5 architecture. It introduces a check to
ensure that a specific condition, represented by a magic key value, is
met before disabling the HST domain. This additional verification step
ensures that the HST domain is only powered down when it is safe to do
so, thereby maintaining the stability and reliability of the system.
Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
This patch introduces power management for the HOST (HST) domain within
the Intel ADSP ACE IP. It adds macros to access the node identifier and
device pointer for the HST power domain and integrates power management
calls into the system initialization and power state transition
functions.
The patch ensures that power gating of the HST domain is prevented when
the primary core of the audio DSP is active. Preventing power gating is
crucial for maintaining the functionality of the HST domain while the
primary DSP core is performing critical tasks.
Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
Adds common thread-local-storage.ld provided
by Zephyr. This also fixes a wrong xtensa_core entry
that should be riscv_core.
Signed-off-by: Sylvio Alves <sylvio.alves@espressif.com>
Currently SOF has disabled CONFIG_PM_DEVICE_RUNTIME_EXCLUSIVE option and
use pm_suspend_devices() to suspend and resume IPC device during D3
power flow. The pm_suspend_devices() function skips suspending devices
that are busy. In very rare cases, the IPC device is busy during the
power state transition, which results in the device not being restored
during reboot. This happens when FW sends a message to the HOST and
waits for ACK, and the HOST simultaneously sends a SET_DX message to the
DSP. This suspend/resume logic in IPC driver does not work well when the
system enters the D3 state because it is not a suspend state, but rather
a power-off. IPC does not require suspending, only reinitialization when
exiting D3. We cannot avoid this one missing ACK and it cannot block the
DSP from turning off.
When FW receives a SET_DX message it checks whether it can enter the D3
state and then returns an error (via IPC) or calls the pm_state_force
function. Success response is sent directly from power_down assembly and
not via ipc driver. This is because after receiving the response, the
HOST will turn off the DSP.
In order for the transition to D3 to take place, only the primary core
can be active, all pipes must be stopped (and therefore all modules in
FW). The only active thread at this time is the Idle thread. Driver on
the host will not send another ipc because is still waiting for
response. FW can try to send only two notification:
- FW exception: from this place there is no return to continue the power
transition,
- log buffer status: skipped, they remain in the queue without being
sent.
I'm moving pm_device_busy_clear(dev) from IRQ handler to
intel_adsp_ipc_send_message function so the pending ACK does not block
power transition.
Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
HDA DMA driver uses an excessive value of 128 bytes as required alignment
for DMA buffer size. This may result in the correct buffer size (e.g.
32-byte aligned, which is DT-compliant) being silently truncated before
writing it into DGBS register. This patch changes the requirement to the
value implied by DGBS register format (effectively reduces to 16 bytes).
Signed-off-by: Tomasz Lissowski <tomasz.lissowski@intel.com>
QEMU MMU tracing showed that there might be something wrong with
its Xtensa MMU implementation, which result in access violation
when running samples/userspace/hello_world_user.
Here is the MMU trace from QEMU from failed runs:
get_pte: autorefill(00109020): PTE va = 20000424, pa = 0010c424
get_physical_addr_mmu: autorefill(00109020): 00109000 -> 00109006
xtensa_cpu_tlb_fill(00109020, 1, 0) -> 00109020, ret = 0
xtensa_cpu_tlb_fill(00109028, 1, 0) -> 00109028, ret = 0
xtensa_cpu_tlb_fill(00109014, 0, 2) -> 00103050, ret = 26
The place where it fails is during reading from 0x109014.
From the trace above, the auto-refill maps 0x109000 correctly
with ring 0 and RW access with WB cache (which should be correct
the first time under kernel mode). The page 0x109000 is the libc
partition which needs to be accessible from user thread.
However, when accessing that page, the returned physical address
became 0x103050 (and resulting in load/store access violation).
We always identity map memory pages so it should never return
a different physical address.
After forcing TLB invalidation during page table swaps, the MMU
trace is:
get_pte: autorefill(00109020): PTE va = 20000424, pa = 0010c424
get_physical_addr_mmu: autorefill(00109020): 00109000 -> 00109006
xtensa_cpu_tlb_fill(00109020, 1, 0) -> 00109020, ret = 0
get_pte: autorefill(00109028): PTE va = 21000424, pa = 0010e424
get_physical_addr_mmu: autorefill(00109028): 00109000 -> 00109022
xtensa_cpu_tlb_fill(00109028, 1, 0) -> 00109028, ret = 0
get_pte: autorefill(00109014): PTE va = 21000424, pa = 0010e424
get_physical_addr_mmu: autorefill(00109014): 00109000 -> 00109022
xtensa_cpu_tlb_fill(00109014, 0, 2) -> 00109014, ret = 0
xtensa_cpu_tlb_fill(00109020, 0, 0) -> 00109020, ret = 0
Here, when the same page is accessed, it got the correct PTE
entry, which is ring 2 with RW access mode (but no cache).
Actually accessing the variable via virtual address returns
the correct physical address: 0x109014.
So workaround that by forcing TLB invalidation during page swap.
Fixes#66029
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
There is no need for this config here and it is messing
with total sys heap calculation.
Signed-off-by: Lucas Tamborrino <lucas.tamborrino@espressif.com>
Kconfig options with a HEAP_MEM_POOL_ADD_SIZE_ prefix should be used to
set the minimum required system heap size. This helps prevent
applications from creating a non-working image by trying to set a too
small value.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The DC233C core has support for both i-cache and d-cache.
So mark it as such so we can test caching of Xtensa in QEMU.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Definition of ADSP_FORCE_DECOUPLED_HDMA_L1_EXIT_BIT,
which is used in the intel_adsp_force_dmi_l0_state function,
is missing.
Signed-off-by: Fabiola Kwasowiec <fabiola.kwasowiec@intel.com>
Separating two new functions force and allow l1
to have the current state with separated functions
in the ipc file so that SOF can call these
functions via IPC DMI_FORCE_L1_EXIT. Change related
to the addition of a new parameter to force
DMI L1 exit on IPC request.
Signed-off-by: Fabiola Kwasowiec <fabiola.kwasowiec@intel.com>
This gets rid of the z_ prefix.
Note that z_xt_*() are being used by the HAL so they cannot be
renamed.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This follows the idea to remove any z_ prefix. Since MMU has
a large number of these, separate out these changes into one
commit to ease review effort.
Since these are no longer have z_, these need proper doxygen
doc. So add them too.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Header files under arch/xtensa/include are considered internal
to architecture. There is really no need for two places to
house architecture internal header files.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Simply to provide some consistencies on file naming under
arch/xtensa.
These are all internally used files and are not public.
So there is no need to provide a deprecation path for
them.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This patch enables DSP clock gating for ACE platforms. By default, clock
gating is blocked by the firmware in the hardware configuration. If
CONFIG_ADSP_IDLE_CLOCK_GATING is enabled, this prevent is not active and
clock can be gated when core is in idle state. WIth this option disabled
clock gating will only be enabled in hardware during power gating.
Signed-off-by: Tomasz Leman <tomasz.m.leman@intel.com>
Reset cause reason was not initalized properly, making
hwinfo feature not to work as expected.
Fixes#65634
Signed-off-by: Sylvio Alves <sylvio.alves@espressif.com>
Secondary dsp is idle and waiting for interrupt before it is totally
halted. The other active cores can trigger idc interrupt to this core,
this can wake it up and result to fw panic. Mask idc interrupt as timer
interrupt to prevent this case.
Signed-off-by: Rander Wang <rander.wang@intel.com>
Andy Ross re-implementation of MMU layer with some subtle changes,
like re-using existent macros, fix page table cache property when
direct mapping it in TLB.
From Andy's original commit message:
This is a reworked MMU layer, sitting cleanly below the page table
handling in the OS. Notable differences from the original work:
+ Significantly smaller code and simpler API (just three functions to
be called from the OS/userspace/ptable layer).
+ Big README-MMU document containing my learnings over the process, so
hopefully fewer people need to go through this in the future.
+ No TLB flushing needed. Clean separation of ASIDs, just requires
that the upper levels match the ASID to the L1 page table page
consistently.
+ Vector mapping is done with a 4k page and not a 4M page, leading to
much more flexibility with hardware memory layout. The original
scheme required that the 4M region containing vecbase be mapped
virtually to a location other than the hardware address, which makes
confusing linkage with call0 and difficult initialization
constraints where the exception vectors run at different addresses
before and after MMU setup (effectively forcing them to be PIC
code).
+ More provably correct initialization, all MMU changes happen in a
single asm block with no memory accesses which would generate a
refill.
Signed-off-by: Andy Ross <andyross@google.com>
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Some workarounds were introduced for intel cavs2.5 platform bring up.
It is not general so move them to platform code.
Signed-off-by: Rander Wang <rander.wang@intel.com>
Cavs platforms starts from Apllolake to Raptorlake. Some of them need some
workaround for arch_cpu_idle so create a bespoken one. Each workaround is
configured by kconfig setting.
Signed-off-by: Rander Wang <rander.wang@intel.com>
The xtensa/nxp_adsp_imx8m linker script is missing
the necessary include statements for linker snippets.
So we need to add them.
This fixes compile warnings like: orphan section `.unstable_id'
from `modules/chre/lib..__modules__lib__chre__platform__zephyr.a
(version.cc.obj)' being placed in section `.unstable_id'.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Add _heap_sentry value to fix build errors for
newlib, like: "undefined reference to `_heap_sentry'"
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
When linking, in crtbegin.o for C++ exception support, we pull in
the .tm_clone_table section.
Update the linker scripts to handle this, otherwise we get a
"warning: orphan section `.tm_clone_table'".
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Remove platform.h since is no longer used for SOF.
Move memory.h to include folder and modify the linker
to reflect this.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
