CONFIG_FREERTOS_ISR_STACKSIZE was set to 2100 when ELF core dump was
enabled, which resulted in a non-16-byte-aligned interrupt stack
offset. This triggered "is SP corrupted" check in the backtrace,
terminating the backtrace early.
Fix the default value, and make sure that the stack is always aligned,
regardless of the value of CONFIG_FREERTOS_ISR_STACKSIZE.
This MR uses an intermediary function `start_app` to call after system
initialization instead of `app_main`.
In RTOS builds, freertos provides `start_app` and calls `app_main`.
In non-RTOS builds, user provides `start_app` directly.
Changes the startup flow to the ff:
hardware -> core libraries init -> other libraries init -> os
init (optional) -> app_main
- hardware init resides in the port layer, and is the entry point
- core libraries init executes init functions of core components
- other libraries init executes init functions of other components (weak
references)
- after other lib is init, the app_main function is called, however,
an OS can wrap the real call to app_main to init its own stuff, and
*then* call the real app_main
FreeRTOS scheduler uses additional stack space, as in some functions
variables are placed onto the stack instead of registers.
This issue resulted in occasional stack overflows in dport task, when
compiling at -O0 optimization level.
- Increase the configMINIMAL_STACK_SIZE to 1kB.
- Enable the watchpoint at the end of stack in CI startup test for
this optimization level.
This fixes the issue where XTOS_SET_INTLEVEL would lower INTLEVEL from
4 to 3, when eTaskGetState is invoked during the core dump, triggered
from the interrupt watchdog.
`xQueueGenericCreateStatic` is placed into flash by the linker script to
reduce IRAM usage. This will also cause the `xRingbufferCreate` not
not callable when cache is disabled.
The SPI bus lock on SPI1 introduces two side effects:
1. The device lock for the main flash requires the
`CONFIG_FREERTOS_SUPPORT_STATIC_ALLOCATION` to be selected, however this
option is disabled by default in earlier IDF versions. Some developers
may find their project cannot be built by their old sdkconfig files.
2. Usually we don't need the lock on the SPI1 bus, due to it's
restrictions. However the overhead still exists in this case, the IRAM
cost for static version of semaphore functions, and the time cost when
getting and releasing the lock.
This commit:
1. Add a CONFIG_SPI_FLASH_BYPASS_MAIN_LOCK option, which will forbid the
space cost, as well as the initialization of the main bus lock.
2. When the option is not selected, the bus lock is used, the
`CONFIG_FREERTOS_SUPPORT_STATIC_ALLOCATION` will be selected explicitly.
3. Revert default value of `CONFIG_FREERTOS_SUPPORT_STATIC_ALLOCATION`
to `n`.
introduced in 49a48644e4.
Closes https://github.com/espressif/esp-idf/issues/5046
Configurable option to use IRAM as byte accessible memory (in single core mode) using
load-store (non-word aligned and non-word size IRAM access specific) exception handlers.
This allows to use IRAM for use-cases where certain performance penalty
(upto 170 cpu cycles per load or store operation) is acceptable. Additional configuration
option has been provided to redirect mbedTLS specific in-out content length buffers to
IRAM (in single core mode), allows to save 20KB per TLS connection.
1. Clarify THREADPTR calculation in FreeRTOS code, explaining where
the constant 0x10 offset comes from.
2. On the ESP32-S2, .flash.rodata section had different default
alignment (8 bytes instead of 16), which resulted in different offset
of the TLS sections. Unfortunately I haven’t found a way to query
section alignment from C code, or to use a constant value to define
section alignment in the linker script. The linker scripts are
modified to force a fixed 16 byte alignment for .flash.rodata on the
ESP32 and ESP32-S2beta. Note that the base address of .flash.rodata
was already 16 byte aligned, so this has not changed the actual
memory layout of the application.
Full explanation of the calculation below.
Assume we have the TLS template section base address
(tls_section_vma), the address of a TLS variable in the template
(address), and the final relocation value (offset). The linker
calculates:
offset = address - tls_section_vma + align_up(TCB_SIZE, alignment).
At run time, the TLS section gets copied from _thread_local_start
(in .rodata) to task_thread_local_start. Let’s assume that an address
of a variable in the runtime TLS section is runtime_address.
Access to this address will happen by calculating THREADPTR + offset.
So, by a series of substitutions:
THREADPTR + offset = runtime_address THREADPTR = runtime_address - offset
THREADPTR = runtime_address - (address - tls_section_vma + align_up(TCB_SIZE, alignment)) THREADPTR = (runtime_address - address) + tls_section_vma - align_up(TCB_SIZE, alignment)
The difference between runtime_address and address is same as the
difference between task_thread_local_start and _thread_local_start.
And tls_section_vma is the address of .rodata section, i.e.
_rodata_start. So we arrive to
THREADPTR = task_thread_local_start - _thread_local_start + _rodata_start - align_up(TCB_SIZE, alignment).
The idea with TCB_SIZE being added to the THREADPTR when computing
the relocation was to let the OS save TCB pointer in the TREADPTR
register. The location of the run-time TLS section was assumed to be
immediately after the TCB, aligned to whatever the section alignment
was. However in our case the problem is that the run-time TLS section
is stored not next to the TCB, but at the top of the stack. Plus,
even if it was stored next to the TCB, the size of a FreeRTOS TCB is
not equal to 8 bytes (TCB_SIZE hardcoded in the linker). So we have
to calculate THREADPTR in a slightly obscure way, to compensate for
these differences.
Closes IDF-1239
spin_lock: cleaned-up port files and removed portmux files
components/soc: decoupled compare and set operations from FreeRTOS
soc/spinlock: filled initial implementation of spinlock refactor
It will decouple the spinlocks into separated components with not depencences of freertos
an similar interface was provided focusing the readabillity and maintenance, also
naming to spinlocks were adopted. On FreeRTOS side the legacy portMUX macros
gained a form of wrapper functions that calls the spinlocks component thus
minimizing the impact on RTOS side.
This feature aims to close IDF-967
soc/spinlock: spinlocks passed on unit test, missing test corner cases
components/compare_set: added better function namings plus minor performance optimization on spinlocks
soc/spinlock: code reordering to remove ISC C90 mix error
freertos/portmacro: gor rid of critical sections multiline macros, placed inline functions instead
soc/spinlock: improved spinlock performance from internal RAM
For cases where the spinlock is executed from IRAM, there is no
need to check where the spinlock object is placed on memory,
removing this checks caused a great improvement on performance.
components/freertos: cleaned up multicore option scheduler.
components/freertos: more cleanup and test optimization to present realistic results
components/freertos: remove unused macros of optimized task selection when multicore is used
freertos/Kconfig: fix trailing space on optimized scheduler option
freertos/tests: moved test context variables inside of test task.
The public variables used on scheduling time test now were packed into a structure allocated on test case task stack and passed to tasks as arguments saving RAM comsumption.
FreeRTOS have an platform dependent configuration to enable selection task in a optimized way.
Provided the platform dependent functions in order to allow the scheduler to use the optimized algorithms by telling to the port layer where to found bitscan instruction i.e. NSAU.
This closes IDF-1116
components/freertos: added option to disable the optimized scheduler
bugfix/pthread: fix pthread_once() race condiion possibility adding critical section in compare and set function
Closes IDFGH-2448
See merge request espressif/esp-idf!7236
DISABLED_FOR_TARGETS macros are used
Partly revert "ci: disable unavailable tests for esp32s2beta"
This partly reverts commit 76a3a5fb48.
Partly revert "ci: disable UTs for esp32s2beta without runners"
This partly reverts commit eb158e9a22.
Partly revert "fix unit test and examples for s2beta"
This partly reverts commit 9baa7826be.
Partly revert "efuse: Add support for esp32s2beta"
This partly reverts commit db84ba868c.
newlib/assert: replace unlikely with likely to keep original assertion
newlib/assert: fix assert macro that uses likely
freertos/port: add the missing sdkconfig.h back
newlib/assert: assert macro back to a single line
esp_common/esp_compiler: renamed esp_macros file to a more specific one
esp_common/esp_compiler: removed CONTAINER_OF macro, it was a duplicate
components/freertos: placed likely macros around port and critical sections
component/freertos: placed likely macros on lists module
components/freertos: placed unlikely macros inside of assertion points, they likely wont fail
components/freertos: added likely macros on queue modules
FreeRTOS queues are one of most hot code path, because to queues itself tend to
be used a lot by the applications, besides that, queues are the basic primitive
to form both mutexes and semaphores, The focus here is to place likely
macros inside lowest level send and receive routines, since they're common
from all kobjects: semaphores, queues, mutexes and FR internals (like timer queue)
components/lwip: placed likely/unlikey on net-interfaces code
components/fatfs: added unlikely macros on disk drivers code
components/spiffs: added unlikely macros on low level fs driver
components/freertos: added likely/unlikely macros on timers and ticker
freertos/event_group: placed likely/unlikely macros on hot event group code paths
components/sdmmc: placed likely / unlikely macros on lower level path of sdmmc
components/bt: placed unlikely macros around bt HCI functions calling
components/lwip: added likely/unlikely macros on OS port code section
components/freertos: fix code style on tick handler
Since in b0491307, which has introduced the optimized window spill
procedure, _xt_context_save did not work correctly when called from
_xt_syscall_exc. This was because unlike _xt_lowint1, _xt_syscall_exc
does not save PS and EPC1. The new version of _xt_context_save
modified PS (on purpose) and EPC1 (accidentally, due to window
overflow exceptions), which resulted in a crash upon 'rfi' from the
syscall.
This commit adds restoring of PS and EPC1 in _xt_context_save. It also
slightly reduces the number of instructions used to prepare PS for
window spill.
Unit test for setjmp/longjmp (which were broken by this regression)
is added.
Closes https://github.com/espressif/esp-idf/issues/4541
1. add hal and low-level layer for timer group
2. add callback functions to handle interrupt
3. add timer deinit function
4. add timer spinlock take function
Regression introduced in commit 79e74e5d5f
It is possible that some FreeRTOS APIs are invoked prior to
scheduler start condition (e.g. flash initialization in unicore mode).
In that condition these asserts should not trigger (scheduler state being yet to be started),
hence changes per this fix.
This commit fixes thread safety issues with configASSERT() calls
regarding the value of uxSchedulerSuspended. A false negative
occurs if a context switch to the opposite core occurs in between
the getting the core ID and the assesment.
Closes https://github.com/espressif/esp-idf/issues/4230
esp32s2beta: Merge support to master
Closes IDF-513, IDF-756, IDF-758, IDF-999, IDF-753, IDF-749, IDF-754, IDF-840, and IDF-755
See merge request espressif/esp-idf!6100
1. Since BLE full-scan feature for BLE mesh change the controller code cause this problem,
it cause coex semaphore take in "interrupt disable", then it may cause task schedule
and cause crash in freertos
2. Fix newlib lock ISR context and critical section check
3. Fix bt controller ISR context and critical section check
Rename and add multiple kconfig compiler options. New compiler options
COMPILER_OPTIMIZATION_PERF and COMPILER_OPTIMIZATION_NONE have been added.
Optimize "Debug" and "Release" options to "Default" and "Size" respectively.
This commit also does the following:
- The COMPILER_OPTIMIZATION_PERF option introduced multiple bug.
This commit fixes those bugs.
- build.yml also updated to test for the new optimization options.
This macro is used in places which expect it to work even without dual core being on.
Still make "mux" functions in FreeRTOS into no-ops as the mux is not needed.
Do not include bootloader in flash target when secure boot is enabled.
Emit signing warning on all cases where signed apps are enabled (secure
boot and signed images)
Follow convention of capital letters for SECURE_BOOT_SIGNING_KEY
variable, since it is
relevant to other components, not just bootloader.
Pass signing key and verification key via config, not requiring
bootloader to know parent app dir.
Misc. variables name corrections
This commit refactors backtracing within the panic handler so that a common
function esp_backtrace_get_next_frame() is used iteratively to traverse a
callstack.
A esp_backtrace_print() function has also be added that allows the printing
of a backtrace at runtime. The esp_backtrace_print() function allows unity to
print the backtrace of failed test cases and jump back to the main test menu
without the need reset the chip. esp_backtrace_print() can also be used as a
debugging function by users.
- esp_stack_ptr_is_sane() moved to soc_memory_layout.h
- removed uncessary includes of "esp_debug_helpers.h"
!4452 used setting LINK_LIBRARIES and INTERFACE_LINK_LIBRARIES to link
components built under ESP-IDF build system. However, LINK_LIBRARIES does
not produce behavior same as linking PRIVATE. This MR uses the new
signature for target_link_libraries directly instead. This also moves
setting dependencies during component registration rather than after all
components have been processed.
The consequence is that internally, components have to use the new
signature form as well. This does not affect linking the components to
external targets, such as with idf_as_lib example. This only affects
linking additional libraries to ESP-IDF libraries outside component processing (after
idf_build_process), which is not even possible for CMake<v3.13 as
target_link_libraries is not valid for targets not created in current
directory. See https://cmake.org/cmake/help/v3.13/policy/CMP0079.html#policy:CMP0079
Using xxx_periph.h in whole IDF instead of xxx_reg.h, xxx_struct.h, xxx_channel.h ... .
Cleaned up header files from unnecessary headers (releated to soc/... headers).
port*_CRITICAL_SAFE API calls port*_CRITICAL or port*_CRITICAL_ISR
depending on the context (Non-ISR or ISR respectively).
FREERTOS_CHECK_PORT_CRITICAL_COMPLIANCE Kconfig option added
Signed-off-by: Sachin Parekh <sachin.parekh@espressif.com>
xTaskIncrementTick have to unwind uxPendedTicks on CPU1 and CPU0.
Use case: If an erase operation was run on the CPU1 then it leads
to starving other tasks which waiting time. Waited tasks just skipped.
Closes: https://github.com/espressif/esp-idf/issues/1952
Closes: IDF-183
This MR removes the common dependency from every IDF components to the SOC component.
Currently, in the ``idf_functions.cmake`` script, we include the header path of SOC component by default for all components.
But for better code organization (or maybe also benifits to the compiling speed), we may remove the dependency to SOC components for most components except the driver and kernel related components.
In CMAKE, we have two kinds of header visibilities (set by include path visibility):
(Assume component A --(depends on)--> B, B is the current component)
1. public (``COMPONENT_ADD_INCLUDEDIRS``): means this path is visible to other depending components (A) (visible to A and B)
2. private (``COMPONENT_PRIV_INCLUDEDIRS``): means this path is only visible to source files inside the component (visible to B only)
and we have two kinds of depending ways:
(Assume component A --(depends on)--> B --(depends on)--> C, B is the current component)
1. public (```COMPONENT_REQUIRES```): means B can access to public include path of C. All other components rely on you (A) will also be available for the public headers. (visible to A, B)
2. private (``COMPONENT_PRIV_REQUIRES``): means B can access to public include path of C, but don't propagate this relation to other components (A). (visible to B)
1. remove the common requirement in ``idf_functions.cmake``, this makes the SOC components invisible to all other components by default.
2. if a component (for example, DRIVER) really needs the dependency to SOC, add a private dependency to SOC for it.
3. some other components that don't really depends on the SOC may still meet some errors saying "can't find header soc/...", this is because it's depended component (DRIVER) incorrectly include the header of SOC in its public headers. Moving all this kind of #include into source files, or private headers
4. Fix the include requirements for some file which miss sufficient #include directives. (Previously they include some headers by the long long long header include link)
This is a breaking change. Previous code may depends on the long include chain.
You may need to include the following headers for some files after this commit:
- soc/soc.h
- soc/soc_memory_layout.h
- driver/gpio.h
- esp_sleep.h
The major broken include chain includes:
1. esp_system.h no longer includes esp_sleep.h. The latter includes driver/gpio.h and driver/touch_pad.h.
2. ets_sys.h no longer includes soc/soc.h
3. freertos/portmacro.h no longer includes soc/soc_memory_layout.h
some peripheral headers no longer includes their hw related headers, e.g. rom/gpio.h no longer includes soc/gpio_pins.h and soc/gpio_reg.h
BREAKING CHANGE
1. separate rom include files and linkscript to esp_rom
2. modefiy "include rom/xxx.h" to "include esp32/rom/xxx.h"
3. Forward compatible
4. update mqtt
When xPortGetCoreID() is called twice within a function,
it might only be called once after compilation. This
commit makes the inline assembly of the function volatile.
Closes#3093
This commit adds the ability for backtracing to trace from the itnerrupt to the
task stack, and across nested interrupts. Test cases have also been added.
New unity component can be used for testing other applications.
Upstream version of Unity is included as a submodule.
Utilities specific to ESP-IDF unit tests (partitions, leak checking
setup/teardown functions, etc) are kept only in unit-test-app.
Kconfig options are added to allow disabling certain Unity features.
Having two different spinlocks is problematic due to possibly
different order in which the locks will be taken. Changing the order
would require significant restructuring of kernel code which is
undesirable.
An additional place where taking xTickCountMutex was needed was in
vApplicationSleep function. Not taking xTickCountMutex resulted in
other CPU sometimes possibly advancing tick count while light sleep
entry/exit was happening. Taking xTickCountMutex in addition to
xTaskQueueMutex has shown a problem that in different code paths,
these two spinlocks could be taken in different order, leading to
(unlikely, but possible) deadlocks.
In earlier change this component was decoupled from freertos and hence
regression was introduced which changed default placement to flash. Some
device drivers make use of ringbuffer while flash cache is being disabled
and hence default placement should instead be internal memory.
Closes: https://github.com/espressif/esp-idf/issues/2517
This commit updates some of the API references in task.h so
that stack sizes are referred to in bytes rather than words as
found in the vanilla version.
Closes#2528
If CONFIG_FREERTOS_LEGACY_HOOKS is kept enabled then defining
idle/tick hooks will be applications responsibility as was the
case earlier.
Signed-off-by: Mahavir Jain <mahavir@espressif.com>
This commit adds an option to enclose all FreeRTOS task functions within a
wrapper function. In the case that a task function returns, the wrapper function
will log an error and abort the application immediately.
Closes#2269Closes#2300
This fixes multiple bugs with ring buffers and re-factors the code. The public
API has not changed, however the underlying implementation have various private
functions have been changed. The following behavioral changes have been made
- Size of ring buffers for No-Split/Allow-Split buffers will not be rounded
up to the nearest 32-bit aligned size. This was done to simplify the
implementation
- Item size for No-Split/Allow-Split buffers will also be rounded up to the
nearest 32-bit aligned size.
The following bugs have been fixed
- In copyItemToRingbufAllowSplit(), when copying an item where the aligned
size is smaller than the remaining length, the function does not consider
the case where the true size of the item is less than 4 bytes.
- The copy functions will automatically wrap around the write pointers when
the remaining length of the buffer is not large enough to fit a header, but
does not consider if wrapping around will cause an overlap with the read
pointer. This will make a full buffer be mistaken for an empty buffer
closes#1711
- xRingbufferSend() can get stuck in a infinite loop when the size of the
free memory is larger than the needed_size, but too small to fit in the ring
buffer due to alignment and extra overhead of wrapping around.
closes#1846
- Fixed documentation with ring buffer queue set API
- Adding and removing from queue set does not consider the case where the
read/write semaphores actually hold a value.
The following functions have been deprecated
- xRingbufferIsNextItemWrapped() due to lack of thread safety
- xRingbufferAddToQueueSetWrite() and xRingbufferRemoveFromQueueSetWrite()
as adding the queue sets only work under receive operations.
The following functions have been added
- xRingbufferReceiveSplit() and xRingbufferReceiveSplitFromISR() as a thread
safe way to receive from allow-split buffers
- vRingbufferGetInfo()
Documentation for ring buffers has also been added.
Add function xRingbufferCanRead & xRingbufferCanWrite
to be able use queue sets. Without it is not possible
to check to which ringbuffer returned semaphore belongs.
In situations where idle task runs a lot of idle hooks or cleanup code
(due to pthread local storage, etc) it can use more than 1KB of stack.
(I think the trigger is if a context switch happens at the right point
in the TLS cleanup).
Also removes an sdkconfig.default which accidentally set all config items,
including this stack size.
* Philosophical: "explicit is better than implicit".
* Practical: Allows useful errors if invalid directories given in components as the defaults aren't
always used. Also trims the -I path from a number of components that have no actual include
directory.
* Simplifies knowing which components will be header-only and which won't
Bugfix to prevent a self deleting no affinity task's memory from being freed by the
idle task of the other core before the self deleting no affinity task is able to context
switch out. prvCheckTasksWaitingTermination now checks if the task is still on
pxCurrentTCB before freeing task memory.
This is a wrapper API for creating a Ring Buffer, which ensures that
the ringbuffer can hold the given number of items, each item being of the
same given length.
Signed-off-by: Piyush Shah <piyush@espressif.com>
Useful to check if the next item to receive is wrapped or not.
This is valid only if the ring buffer is initialised with type
RINGBUF_TYPE_ALLOWSPLIT.
This is as per the feature request here:
https://github.com/espressif/esp-idf/issues/806
Signed-off-by: Piyush Shah <piyush@espressif.com>
The earlier available API (xRingbufferGetMaxItemSize())just gives
a static max entry value possible for given ring buffer.
There was a feature request for an API which could provide
a real time available buffer size. See below:
https://github.com/espressif/esp-idf/issues/806
Signed-off-by: Piyush Shah <piyush@espressif.com>
- Use `code` tags instead of a mix of `<pre></pre>` and
`@verbatim .. @endverbatim`
- Remove manually added function prototypes from comment blocks
- Remove of grouping (`\defgroup`) — some extra work is needed
to make groups compatible with the way we auto-generate API
reference from Doxygen XML files. It's pretty easy to add the
grouping directives back if/when we implement support for
Doxygen groups in the later stages of documentation build
process.
- Hide private APIs under `@cond .. @endcond`
- Convert some comments into Doxygen-compatible ones
- Fix various documentation issues: missing documentation for
some parameters, mismatch between parameter names in comment
block and in function prototype.
- Add doxygen comments for functions which didn't have them
(thread local storage).
- Add [out] param tags where necessary
- Redefine `xTaskCreate` and `xTaskCreateStatic` as inline
functions instead of macros.
This commit backports vTaskDelete() behavior from FreeRTOS v9.0.0 which
allows for the immediate freeing of task memory if the task being deleted
is not currently running and not pinned to the other core. This commit also
fixes a bug in prvCheckTasksWaitingTermination which prevented the
Idle Task from cleaning up all tasks awaiting deletion. Each iteration of the Idle
Task should traverse the xTasksWaitingTermination list and clean up all tasks
not pinned to the other core. The previous implementation would cause
prvCheckTasksWaitingTermination to return when encountering a task
pinned to the other core whilst traversing the xTasksWaitingTermination list.
The test case for vTaskDelete() has been updated to test for the bugfix and
backported deletion behavior.
This commit backports the following features from FreeRTOS v9.0.0
- uxSemaphoreGetCount()
- vTimerSetTimerId(), xTimerGetPeriod(), xTimerGetExpiryTime()
- xTimerCreateStatic()
- xEventGroupCreateStatic()
- uxSemaphoreGetCount()
Functions backported previously
- xTaskCreateStatic()
- xQueueCreateStatic()
- xSemaphoreCreateBinaryStatic(), xSemaphoreCreateCountingStatic()
- xSemaphoreCreateMutexStatic(), xSemaphoreCreateRecursiveMutexStatic()
- pcQueueGetName()
- vTaskSetThreadLocalStoragePointer()
- pvTaskGetThreadLocalStoragePointer()
Unit tests were also written for the functions above (except for pcQueueGetName
which is tested in a separate Queue Registry MR). The original tlsp and del cb test case
was deleted and integrated into the test cases of this MR.
When using CPP and C combination this particular file threw error on linking.
Merges https://github.com/espressif/esp-idf/pull/1249
(Amended to add INC_FREERTOS_H guard, comment on #endif)
1. Usage of this module required applications to include additional
files. What files to include is not very intuitive. Instead, it is
better for the header file itself to include other files as required.
2. Even though it may seem logical, it is better to explicitly mention
that an item needs to be "Returned" after a Receive
Signed-off-by: Piyush Shah <piyush@espressif.com>
It was observed that if the ring buffer size provided by application
is not a multiple of 4, some checks were failing (as read_ptr and write_ptr
could shoot beyond the ring buffer boundary), thereby causing asserts.
Simply wrapping around the pointers for such cases fixes the issue.
Moreover, because of the logic for maintaining 4-byte boundary,
it was also possible that a wrap-around occurred for some data,
even when the actual size remaining was sufficient for it.
Eg. Buffer available: 34, data size: 34, 4-byte aligned size: 36
Since the logic compares against 36, it writes 34 bytes and does a
wraparound. But since all 34 bytes are already written, there is
nothing to write after wrapping. It is better to just re-set the
write pointer to the dtart of ring buffer in such cases.
Tested send/receive for various arbitrary sizes of data and for
arbitrary sizes of ring buffer.
Alternative Solutions:
1) Allow only sizes which are multiples of 4, and return error otherwise.
Appropriate code and documentation change would be required
2) Allow arbitrary sizes, but internally add upto 3 bytes to make
the total size a multiple of 4
Signed-off-by: Piyush Shah <piyush@espressif.com>
In the queue registry test, start_sem is given twice to let both tasks
start the test. Each task takes start_sem, does some work, gives done_sem,
and goes on to wait for start_sem again.
It may happen that one task can grab start_sem, add queues to the
registry, give done_sem, then grab start_sem again, delete the queues
from the registry, and give done_sem again. At this point, main test
task takes done_sem twice and proceeds to verify that queues have been
added to the registry. But in fact, the first task has already deleted
its queues from the registry, and the second one might not have added
the queues yet. This causes test to fail.
This changes the test to use separate start semaphores for each task,
to avoid the race condition.
This commit makes the configQUEUE_REGISTRY_SIZE and
configGENERATE_RUN_TIME_STATS configurable in menuconfig.
- configQUEUE_REGISTRY_SIZE can now be set in menuconfig.
- The functions vQueueAddToRegistry() and vQueueUnregisterQueue() were made
SMP compatbile
- pcQueueGetName() was backported from FreeRTOS v9.0.0
- Added test case for Queue Registry functions
- configGENERATE_RUN_TIME_STATS can now be enabled in menuconfig. CCOUNT or
esp_timer can be selected as the FreeRTOS run time clock in menuconfig as
well, although CCOUNT will overflow quickly.
- Run time stats collection (in vTaskSwitchContext) and generation (in
uxTaskGetSystemState) have been made SMP compatible. Therefore
vTaskGetRunTimeStats() now displays the run time usage of each task as a
percentage of total runtime of both CPUs
Squash
Test cases were added for the following functions
- xTaskNotify(), xTaskNotifyGive(), xTaskNotifyFromISR(), vTaskNotifyGiveFromISR(),
- xTaskNotifyWait(), ulTaskNotifyTake()
- vTaskDelayUntil()
The following function was made smp compatible and tested as well
- eTaskGetState()
This commit reverts the revert on the new task watchdog API. It also
fixes the following bug which caused the reversion.
- sdkconfig TASK_WDT_TIMEOUT_S has been reverted from the unit of ms back to the
unit of seconds. Fixes bug where projects using the new API without rebuilding sdkconfig
would cause the old default value of 5 to be interpreted in ms.
This commit also adds the following features to the task watchdog
- Updated idle hook registration to be compatible with dual core hooks
- Updated dual core hooks to support deregistration for cpu
- Legacy mode has been removed and esp_task_wdt_feed() is now replaced by
esp_task_wdt_reset(). esp_task_wdt_feed() is deprecated
- Idle hooks to reset are now registered/deregistered when the idle tasks are
added/deleted from the Task Watchdog instead of at Task Watchdog init/deinit
- Updated example
This commit updates various test cases throughout esp-idf such that
the values used for timer divider pass the assertions in the timer component.
Timer divider values must be between 2 to 65536
This commit makes configUSE_TRACE_FACILITY and
configUSE_STATS_FORMATTING_FUNCTIONS configurable in kconfig. Test cases fro the
functions enabled by the two configurations above have also been added.
Test cases for the following functions have been added...
- uxTaskGetSystemState()
- uxTaskGetTaskNumber()
- vTaskSetTaskNumber()
- xEventGroupClearBitsFromISR()
- xEventGroupSetBitsFromISR()
- uxEventGroupGetNumber()
- uxQueueGetQueueNumber()
- vQueueSetQueueNumber()
- ucQueueGetQueueType()
Test cases for the following functions were not required...
- prvListTaskWithinSingleList()
- prvWriteNameToBuffer()
- vTaskList()
Previously if multiple tasks had been added to xPendingReadyList for the CPU, only the first one was resumed.
Includes a test case for resuming multiple (pending) tasks on xTaskResumeAll().
Document the limitation that while scheduler is suspended on one CPU, it can't wake tasks on either CPU.
Legacy API of task watchdog used the same function esp_task_wdt_feed() to add
and feed a task. This caused issues of implicitly adding a task to the wdt list
if the function was used in shared code.
The new API introduces init, adding, feeding, deleting, deinit functions. Tasks
must now be explicitly added to the task watchdog using their handles. Deletion
must also be explicit using task handles. This resolves the issue of implicit
task additions to the task watchdog due to shared code calling
esp_task_wdt_feed().
Task watchdog is now fully configurable at runtime by calling the init and
deinit functions.
Also added functions to get the handles of idle tasks of the other core. This
helps when adding idle tasks to the watchdog at run time.
Configuring the task watchdog using menu config is still available, however
menu config will only result in calling the init and add functions for idle
tasks shortly after the scheduler starts.
Menu config also allows for using legacy behavior, however the legacy behavior
willcall the new API functions but with slight variations to make them legacy
compatible.
Documentation and example have also been updated
gcov_rtio.c headers updated to prevent error of freertos header files being
included in the wrong order.
Resolves issue TW#13265
Added documentation about the ESP-IDF changes to FreeRTOS.
The documentation covers changes to the following FreeRTOS aspects.
- Task Creation
- Affects on scheduling (Task skipping, scheduler suspension, tick synchronicity)
- Critical sections and disabling interrupts
- Thread Local Storage Pointers and deletion callbacks
- Configuring ESP-IDF FreeRTOS
When debugging crashes caused by flash cache access errors, OpenOCD may
request the value of uxTopUsedPriority when cache is disabled. Placing
it into IRAM to avoid an error in such case.
Fix warnings where undefined vars are used.
Make Kconfig emit "FOO=" for unset bool options
To ensure make variables are always defined, even if empty.
When writing auto.conf, include symbols disabled by dependency to make sure all make variables are always defined.
Fixesespressif/esp-idf#137
Cherry-picked from https://github.com/espressif/esp-idf/pull/138
Unlocking a never-locked mutex is an assertion failure in debug mode.
In release mode, this further improves performance:
No-Contention -> 153 cycles
Recursion No-Contention -> 138 cycles
Contention -> 378 cycles
The code in xtensa_vectors.S did not handle XT_DEBUG_BACKTRACE
consistently: #if was used in one case, and #ifdef in another. This makes
the checks consistent, and also introduces a Kconfig option to enable
backtrace.
In some cases, xPortGetCoreID was not inlined, and ended up in flash.
Since this function is used in many situations when cache is disabled,
that caused exceptions. Adding IRAM attribute to fix that.
This is no longer required since the functions automatically get
pulled in based on the usage. A quick summary of footprint
comparisions before and after these set of patches is shown below:
Hello-World: (simplified for readability)
old Total image size:~ 104902 bytes (.bin may be padded larger)
old Total image size:~ 105254 bytes (.bin may be padded larger)
Per-archive contributions to ELF file:
Archive File DRAM .data & .bss IRAM Flash code & rodata Total
old libesp32.a 1973 177 4445 3939 2267 12801
new libesp32.a 1973 185 4473 3939 2267 12837
new libnvs_flash.a 0 92 0 274 8 374
new libstdc++.a 0 0 0 24 0 24
For some reason, nvs_flash.a (~400bytes) gets pulled in (particularly
the nvs_flash_init() function).
Power-Save: (simplified for readability)
old Total image size:~ 421347 bytes (.bin may be padded larger)
old Total image size:~ 421235 bytes (.bin may be padded larger)
old libtcpip_adapter.a 0 81 0 1947 115 2143
new libtcpip_adapter.a 0 69 0 1897 115 2081
The size actually shrinks a bit, since the AP interface function
doesn't get pulled in.
For config-only components, component.mk should now contain "COMPONENT_CONFIG_ONLY := 1"
Also refactored some of the generation of linker paths, library list. This required cleaning up the way the bootloader
project works, it's now mostly independent from the parent.
Implements support for system level traces compatible with SEGGER
SystemView tool on top of ESP32 application tracing module.
That kind of traces can help to analyse program's behaviour.
SystemView can show timeline of tasks/ISRs execution, context switches,
statistics related to the CPUs' load distribution etc.
Also this commit adds useful feature to ESP32 application tracing module:
- Trace data buffering is implemented to handle temporary peaks of events load
Updated test case to include configASSERT cases (+1 squashed commits)
Squashed commits:
[871ec26f] Freertos:Bugfix uxTaskGetSystemState
Bug (github #12142) with uxTaskGetSystemState where
if called immediately after creating a bunch of tasks,
those tasks would be added twice into the TaskStatusArray.
Bug caused due to use old implementation using vTaskSuspendAll
which did not stop newly created task on other core from accessing the
read/waiting task lists whilst the list were being read by
uxTaskGetSystemState. Fixed bug by replacing vTaskSuspendAll
with taskENTER_CRITICAL and added test case for the bugfix
xPendingReadyList[1] was never initialised correctly, so if a task is added to this list (ie by xSemaphoreGiveFromISR()
or similar) while scheduler is disabled, then it causes a null pointer dereference.
Bug produces stack traces similar to:
0x40086e87: vListInsertEnd at /home/gus/esp/32/idf/components/freertos/./list.c:130
0x40084ba3: xTaskRemoveFromEventList at /home/gus/esp/32/idf/components/freertos/./tasks.c:3439
0x40083c54: xQueueGiveFromISR at /home/gus/esp/32/idf/components/freertos/./queue.c:2034
0x400f50a0: timer_group0_isr at /home/gus/esp/32/idf/components/freertos/test/./test_suspend_scheduler.c:27
0x40081d7d: _xt_lowint1 at xtensa_vectors.o:?
Make internal stack size variables in FreeRTOS 32-bit instead of 16-bit
Stock FreeRTOS uses an uint16 to store stack sizes, making it impossible to allocate a stack >64K. This changes this into an uint32, allowing for larger stacks.
See merge request !677
component/esp32 : fix dualcore bug
1. When dual core cpu run access DPORT register, must do protection.
2. If access DPORT register, must use DPORT_REG_READ/DPORT_REG_WRITE and DPORT_XXX register operation macro.
See merge request !742
1. When dual core cpu run access DPORT register, must do protection.
2. If access DPORT register, must use DPORT_REG_READ/DPORT_REG_WRITE and DPORT_XXX register operation macro.
Assert when a new task is started on a nonexisting CPU
Previously, starting a task on a CPU ID higher than the amount of CPUs FreeRTOS is configured with would not start the task and possibly have unintended side effects due to some out-of-bounds array writes. Assert on this so the app aborts cleanly.
See merge request !701
- Implements application tracing module which allows to send arbitrary
data to host over JTAG. This feature is useful for analyzing
program modules behavior, dumping run-time application data etc.
- Implements printf-like logging functions on top of apptrace module.
This feature is a kind of semihosted printf functionality with lower
overhead and impact on system behaviour as compared to standard printf.
FreeRTOS: Fix cross-core event group sync
As above
Also includes fixes which allowed removing some semi-hacky bits from the event group unit tests - specifically, higher priority tasks will always be started immediately even if they run on the opposite core.
See merge request !535
freertos kconfig: Add max task length name, move task stack overflow watchpoint
* Allow setting task name length via kconfig
* Move task stack overflow watchpoint out of internals debugging submenu, place next to other stack overflow checks.
* Thread local pointer feature should depend on Ethernet or WiFi
See merge request !531
1. add a macro in menuconfig for users to choose whether store phy calibration data into NVS or not.
2. rename some disable phy and rf APIs so that existing code which calls old APIS will fail to compile.
1. Add disable PHY and RF when WiFi and BT are both disabled(including call sniffer disable API).
2. Do not init PHY and RF when cpu start. Init PHY and RF when call Wifi or BT start APIs(including sniffer enable API).
3. Add a temporary lib: librtc_clk.a and will delete it when CPU frequency switching function is done.
4. Add an function to get OS tick rate.
5. Do not put the whole pp.a in iram0, only put lmac.o, ieee80211_misc.o, ets_time.o and wdev.o in iram0.
When functions in queue.c calls listLIST_IS_EMPTY() to check queue event list, the queue list is
protected by queue mutex, on the other hand, when xTaskIncrementTick() modify the queue list, the
queue list is protected by xTaskQueueMutex, this may cause xTaskRemoveFromEventList operate on
the empty queue list and cause problem. This commit is to fix this bug.
may cause
fixes for issues observed when using spi_flash
This MR fixes three unrelated issues:
- Race condition in spi_flash_enable_interrupts_caches_and_other_cpu
when operations on unpinned tasks are performed.
The issue is reported in https://github.com/espressif/esp-idf/pull/258
- esp_intr_noniram_disable doesn’t disable interrupts when compiled in
release mode. This issue manifested itself with an illegal instruction
exception when task WDT ISR was called at the time when flash was
disabled.
Fixes https://github.com/espressif/esp-idf/issues/263.
- Tick hooks on CPU1 were not called if CPU0 scheduler was disabled for
significant amount of time (which could happen when doing flash erase).
The issue manifested itself as “INT WDT timeout on core 1” error.
Fixes https://github.com/espressif/esp-idf/issues/219.
See merge request !441
The block which dispatches ticks on CPU1 was a copy of the code block
for the normal path (CPU0). It used to check uxPendedTicks, with the
logic that uxPendedTicks can be 0 iff the scheduler is not suspended.
On CPU1 however, uxPendedTicks is not related to the state of the
scheduler (as uxPendedTicks is updated on CPU0). Due to this, if CPU0
scheduler was suspended, and uxPendedTicks happened to be nonzero,
tick hooks on CPU1 didn’t run, even though CPU1 scheduler was working.
This change removes the check for uxPendedTicks in CPU1 code path,
so that the tick hooks on CPU1 always get called (as for the CPU0 code
path).
The following issues mentioned during MR!341 review were fixed:
1) Core dump test application description
2) Usage of CONFIG_ESP32_ENABLE_COREDUMP_TO_FLASH and CONFIG_ESP32_ENABLE_COREDUMP_TO_UART
3) FLASH_GUARD_START macro usage is fixed in flash API
4) Core dump module logging facility
5) cache util functions doc updated
6) interactive delay before print core dump to uart
7) core dump partion support in build system
1) PS is fixed up to allow GDB backtrace to work properly
2) MR!341 discussion: in core dump module: esp_panicPutXXX was replaced by ets_printf.
3) MR!341 discussion: core dump flash magic number was changed.
4) MR!341 discussion: SPI flash access API was redesigned to allow flexible critical section management.
5) test app for core dump feature was added
6) fixed base64 file reading issues on Windows platform
7) now raw bin core file is deleted upon core loader failure by epscoredump.py
Complimentary changes:
1) Partition table definitions files with core dump partition
2) Special sub-type for core dump partition
3) Special version of spi_flash_xxx
4) espcoredump.py is script to get core dump from flash and print useful info
5) FreeRTOS API was extended to get tasks snapshots
1. Update wifi lib which contains ampdu and other optimizations
2. Add throughput code debug code
3. Other misc modification about throughput optimization
This commit fixes:
1. xTaskGetCurrentTaskHandle may return wrong TCB when current task switch to a different core
2. Idle task may have problem when it terminate the task in both core
Minor fixes from various sources
- Fix memory debugging code. Noticed by Tuan.
- intr_enable/disable should be in IRAM. Noticed by rojer
- Still old timer code in examples in doxygen comments. Noticed on the forum by jumjum123
- Timer example was broken. Noticed on the forum by jumjum123
See merge request !325
“Full” printf uses more stack space than “nano” printf, and more space than available in the idle task.
This caused stack overflow in TLS delete test. Replacing printf with ets_printf.
Refs:
http://esp32.com/viewtopic.php?f=13&t=550http://esp32.com/viewtopic.php?f=13&t=551
rmt.c should include stdlib.h for malloc, esp_bignum,c &
https_request_main.c for abort().
FreeRTOSConfig.h is only including stdlib if
CONFIG_FREERTOS_ASSERT_ON_UNTESTED_FUNCTION is set. However, it is
included for abort() so needs to be included whenever
CONFIG_FREERTOS_ASSERT_FAIL_ABORT is set.
This change includes unconditionally in FreeRTOSConfig.h. This is to
avoid this kind of bug where compiler errors are dependent on config. I
suggest we don't change this to be more selective until we have 'make
randomconfig' style tests in CI.
Integrate unit tests into build system
This MR moves unit tests from esp-idf-tests repository into 'test' subdirectories of respective components.
Tests are run using a runner app in tools/unit-test-app.
This needs a bit of cleanup:
1. remove extra newlines added to makefiles,
2. re-format unit tests code which has tabs
3. write a document on using this test app
4. maybe some refactoring in project.mk
I think 1&2&4 need to be done in this MR, while 3 may be done in a follow-up one.
See merge request !221
Includes a tweak to make Static_task_t equal size to TCB_t when using
MPU_WRAPPERS . Matches tskSTATIC_AND_DYNAMIC_ALLOCATION_POSSIBLE macro
in tasks.c. This isn't actually a bug (if static task allocation is off,
there is no use for Static_task_t), but it allows us to make consistent
compile-time checks that Static_task_t == TCB_t.
Without this change, building FreeRTOS with static allocation enabled succeeds, but trying to create a queue from a static buffer causes an assert because the size of static and dynamic queues differ.
rename nvs host test folder, modify .gitlab-ci.yml
remove unit-test-app build
re-format unit test files
remove extra newlines in project.mk
some refactoring for unit test part in project.mk
add build files of unit-test-app in gitignore
add README.md for unit test app
correct headings in README.md
remove files and make minor tweaks in unit test app
update .gitlab-ci.yml to use unit test app
delete unused lines in component_wrapper.mk
delete periph_i2s.h and lcd test
add text floating point in components/esp32/test/Kconfig
correct idf test build paths in .gitlab-ci.yml
Default esp-idf builds now show -Wextra warnings (except for a few:
signed/unsigned comparison, unused parameters, old-style C declarations.)
CI building of examples runs with that level raised to -Werror, to catch
those changes going into the main repo.
Per-CPU interrupt handlers and args
Up to now, the interrupt handlers and args were shared between CPUs, that is, if you set an interrupt handler on CPU0, CPU1 would invoke the same interrupt handler on that interrupt. This code gives every CPU its own space for interrupt handlers.
See merge request !192
New makefile component_wrapper.mk allows some variables to be set
before component.mk is evaluated. This properly fixes problems with
sdkconfig being hard to access in all phases of the build.
Including component_common.mk is no longer necessary and will print a
deprecation warning for components which use it.
Add cross-core int to accelerate task being awoken from another CPU.
This adds a per-CPU interrupt that can be used to poke the CPU to go do something. In this case all that is implemented is a request to yield the current task, used in case a CPU unblocks a task that runs on another CPU. This gets rid of the limitation that inter-CPU communication using queues, muxes etc can take up to a FreeRTOS tick to happen.
Specs!
Sending an in in a queue of length 1 (essentially a semaphore) as quickly as possible (just a small delay in the sender, to make sure the receiver task gets swapped out) for 10 seconds. Number indicates the amount of ints transferred
Old code:
CPU0->CPU0: 42986
CPU0->CPU1,: 2999
New code:
CPU0->CPU0: 42868
CPU0->CPU1: 62073
See merge request !155
Feature/wdts
This adds two watchdogs to esp-idf:
- An interrupt watchdog. Kicks in if the FreeRTOS timer interupt on either the PRO_CPU or (when configured) the APP CPU isn't called for a configurable time. Panics, displaying which CPU caused the problem and the registers that may lead to the offending code.
- A task watchdog. A task has to feed it every once in a while. If not, it will print the name of the offending tasks, as well as the tasks currently running on both CPUs, and optionally panic.
Also adds a panic reason to the panic call, as well as fixes the panic code a bit.
See merge request !148
Feature/trax
Add Trax support to esp-idf. OpenOCD already has trax support, this allows an esp-idf program to also trigger this when needed. Also included: some more logic to correctly reserve memory blocks for Trax.
See merge request !73
Some small fixes
- Kill unused uxReturn in task.c, https://github.com/espressif/esp-idf/issues/48
- Line end conversion in gpio.c
- Move heap_alloc_caps.h so components can also use it
See merge request !135
* master: (117 commits)
build system: Add -fno-rtti when compiling C++ code
FreeRTOS KConfig: Limit tick rate to 1000Hz
bootloader: Fix accidental tabs introduced in !78
build system: Print a WARNING if any submodule is out of date
Fix stack overflow message format
'make flash' targets: Print serial port when flashing
lwip/esp32: support iperf
Add data memory for RMT peripheral
syscall write: Should return number of bytes written
Also push relevant tags over
esp32: add libsmartconfig.a to link libs
esp32: not link wps
esp32/lib: update wifi lib to a1e5f8b9
esp32: remove esp_wps.h
add smartconfig header files(merge this after updating libsmartconfig.a version v2.6.2)
esp32/lib: update wifi lib to 3853d7ae
Add Comments
Modify spinlock error in periph_ctrl.c
Define xcoreid offset, add warning in tcb struct wrt the need to also change that define when struct changes
components/tcpip_adapter: add some comments
...
# Conflicts:
# components/freertos/queue.c
# components/freertos/tasks.c
>1000Hz breaks portTICK_PERIOD_MS (see gitlab 4)
A working >1000Hz tick rate is possible with some changes, but beyond a
certain point it's dimishing returns to preempt tasks this often.
esp32: Bootloader wake deep sleep stub
App can contain a stub program resident in RTC fast memory. Bootloader
will load the stub on initial boot. If the device wakes from deep sleep,
the stub is run immediately (before any other data is loaded, etc.)
To implement a custom wake stub, implement a function in your program:
```
void RTC_IRAM_ATTR esp_wake_deep_sleep(void)
{
esp_default_wake_deep_sleep();
// other wake logic
}
```
... and it will replace the default implementation.
See merge request !78
Workaround: Automatically pin no-cpu-affinity task to a core when FPU is used
FPU status at the moment does not migrate cleanly between cores, so tasks without affinity that happen to migrate across FPUs will run into problems. As a workaround, this modification will automatically pin the task to the current CPU when FPU activity is detected. If anything, it's better than getting all kinds of weird and wonderful FPU corruption issues...
See merge request !124
Startup flow refactoring
This set of commits changes the startup code in a way that lets the application choose if/when to initialize WiFi/BT.
Application entry point is now a more familiar `main()` function. This function is executed in its own task. Application may choose to do some initialization from main function, create some tasks and then return from `main`. Simple applications may choose to do all their work from `main`.
Additionally this MR splits event handling code into two parts.
- One part is a set of standard handlers for WiFi and DHCP events. Most applications will use this set of handlers, and it is made available via new `esp_event_process_default` function.
- Another part is the default implementation of event handling loop. Some applications may choose to use default event loop through `esp_event_loop_` set of APIs, which start an event handling task and call user-provided event callback from this task. Other applications may create an event queue and implement event loop themselves. In this case application has to provide `esp_event_send` function. In this case the implementation provided by `esp_event_loop_` module is unused.
esp-idf-template has been updated to match this set of changes: https://github.com/espressif/esp-idf-template/tree/feature/init_refactoring
BT example has also been updated.
We need to provide examples of both event handling approaches. This will be done in a separate follow-up MR.
See merge request !112
This feature allows to use static buffers (or from a pool of memory which is not
controlled by FreeRTOS).
In order to reduce the impact of the changes, the static feature has only been added
to the queus (and in consequence to the semaphores and the mutexes) and the tasks.
The Timer task is always dynamically allocated and also the idle task(s), which in the
case of the ESP-IDF is ok, since we always need to have dynamic allocation enabled.
* master: (57 commits)
components/lwip: fix grammar
components/lwip: make SO_REUSE configurable via menuconfig
bootloader: remove trailing newlines from log messages
components/freertos: override per-task __cleanup handler to close stdin, stdout, stderr
components/esp32: move peripheral symbols to separate ld script
components/log: regroup macros for better readability
gitlab-ci: allow running tests for branches, triggered via API
components/log: fix timestamp calculation
components/log: set default runtime log level to ESP_LOG_VERBOSE
components/log: fix error when using ESP_LOGx from C++ code
components/log: fix bugs, add options to override log level for files, components, and bootloader
fix ledc and spi typo
remove prefix and postfix
Enable SO_REUSEADDR in LWIP
freertos: fix memory debug macro issue Define configENABLE_MEMORY_DEBUG according to CONFIG_ENABLE_MEMORY_DEBUG
peripheral structure headers: move volatile keyword from members to typedef
Adding -fstrict-volatile-bitfields to the CFLAGS/CXXFLAGS. Without this, gcc tries to access bitfields using the smallest possible methods (eg l8i to grab an 8-bit field from a 32-bit). Our hardware does not like that. This flag tells gcc that if a bitfield is volatile, it should always use the type the field is defined at (uint32_t in our case) to size its access to the field. This fixes accessing the hardware through the xxx_struct.h headers.
add peripheral module struct headers
build system docs: Add note about no spaces in component names
Docs: Add note about unusual submodule messages when cloning on Windows
...
# Conflicts:
# components/esp32/cpu_start.c
# components/esp32/include/soc/cpu.h
Note that with WiFi stack enabled, system_init will reset frequency to 240MHz.
To make this setting useful, esp32-wifi-libs submodule needs to be updated.
