The way these hooks are implemented at the moment, once you register a tick or idle task hook, it is run by both cores, leading to intermittent crashes that are difficult to explain (and debug). One solution is to register the hook with the core that is currently running, which is what this patch attempts to do.
Merges https://github.com/espressif/esp-idf/pull/781
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
The two task watchdog timer bugs are as follows...
1) If only a single task existed on the wdt task list, and esp_task_wdt_feed()
was only called once, the watchdog triggers but fails to print task name
2) If a single task already exists on the task wdt list, and another task calls
esp_task_wdt_feed() once, the watchdog fails to trigger
Problem stemmed from the loop responsible for resetting the watchdog timer
having incorrect loop parameters. The loop failed to traverse the full length
of the task wdt list
In release mode, switches in rtc_clk_bbpll_set would be converted to
jump tables. These tables would be placed into .rodata, which caused
rtc_clk_cpu_freq_set to be unusable while cache is disabled. This
manifested itself in crashes when exiting from light sleep:
https://esp32.com/posting.php?mode=reply&f=13&t=3089#pr14590
1. add adc2 coexistence
2. change NOW() to WDEV_NOW()
3. call ets_timer_done() before calling ets_timer_setfn()
4. fix the bug that api lock is not unlocked when call esp_wifi_stop()
twice
5. more hmac tx buffer limit from 32 to 64.
1. Support built-in ADC for I2S.
2. Modify code of ADC, made no change to the original APIs.
3. Add APIs in I2S:
esp_err_t i2s_set_adc_mode(adc_unit_t adc_unit, adc1_channel_t adc_channel);
4. Add I2S ADC/DAC example code.
5. add old-fashion definition to make it more compatible
6. replase spi_flash_ APIs with esp_partition_ APIs
7. add example of generating audio table from wav
8. change example sound
1. Occupy conn[9] for adv. The max connection number of bluetooth controller decrease to 9. (Previously, it is 10)
2. modify the enum of setting BLE TX power corresponding to connection's.
3. fix libbtdm_app.a cwitch jump table
ESPNOW is a kind of WiFi communication bettween WiFi devices, no
matter they are connected to each other or not. Once two WiFi
devices are paired, they can send data to or receive data from
each other. The data is transmitted in action frame which can
be encrypted with CCMP method. ESPNOW also support mutilcast frame
transmitting.
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
Since timestamps are 64-bit, loosing one bit of range due to sign does
not present an issue, however for applications doing calculations on
timestamps, signed return type is more convenient.
The implementation of esp_timer_get_time used a critical section, which
resulted in a call time of ~1.8us. To make esp_timer_get_time more
useable as a high-resolution time source, this change replaces the lock
with polling. Call time is reduced to ~0.7us.
All peripheral clocks are default enabled after chip is powered on.
When CPU starts, if reset reason is CPU reset, disable those clocks
that are not enabled before reset. Otherwise, disable all those
useless clocks.
These peripheral clocks must be enabled when the peripherals are
initialized and disabled when they are deinitialized.
If two different types of SHA hashes were active in the hardware
concurrently, a race condition meant the SHA unit could be incorrectly
reset leading to all-zero results.
If scheduler switches cores in narrow window during esp_dport_access_stall_other_cpu_start(), could cause the stall
interrupt to occur on the running CPU - halting the CPU until WDT cleans up.
Related to https://github.com/espressif/esp-idf/issues/630
1. add sens_struct.h
2. add definition of RTCCNTL and RTCIO
3. modify touch pad examples
4. update example code.
5. add comments add option in menuconfig
6. fix issue that pad index 8 and 9 are mismatched
7. add touch_pad_read_filtered() api to get value filtered by iir filter
8. modify touch pad isr func
9. Make the items in perihperal.ld in the sequence of address
10. delete Kconfig for touch pad
11. add touchpad filter APIs to adjust the filter
12. add touch_pad into index.rst
13. add touch_pad in Doxyfile
14. add touch_pad.rst
Previously VFS driver for UART could only use simple non-blocking
functions to read from and write to the UART. UART driver provides more
complex blocking and interrupt-driven functions, which can be used
instead.
This commit adds optional support for using UART driver's functions.
Also added is a more flexible mechanism for configuring newline
conversion rules on input and output.
This commit also fixes a bug that all UARTs shared one static variable
used as a character buffer in newline conversion code. This variable is
changed to be per-UART.
1. Hello World application shows no footprint difference before and
after this change
2. examples/ethernet/ethernet application compiles properly (can't
test with my board)
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.
Since only the used interface's start function gets called, it pulls
in only the functions that are required in the current application,
thereby saving footprint.
Restart being a lower-layer system-level function, needn't depend on
the higher level Wi-Fi libraries.
This also enables us to get rid of one more WIFI_ENABLED ifdef check
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.
NVS is used to store PHY calibration data, WiFi configuration, and BT
configuration. Previously BT examples did not call nvs_flash_init,
relying on the fact that it is called during PHY init. However PHY init
did not handle possible NVS initialization errors.
This change moves PHY init procedure into the application, and adds
diagnostic messages to BT config management routines if NVS is not
initialized.
Currently the last 128KB of DRAM is reserved for the bootloader & early boot stacks. This means if >192KB of static DRAM
is allocated, the only available heap is this region - which is disabled until the scheduler starts. As a result, you
get either heap corruption on early boot if the static data overlaps startup heap (leading to very weird errors), or
FreeRTOS will fail to start when it can't malloc() anything.
Long term fix is to move the stacks & bootloader data to the very end of RAM, and only reserve that part for early
boot. This is a little fiddly because of also wanting to make sure this memory is not preemptively fragmented when it
gets reintroduced to the heap. This will become more important if/when we have more static allocation options in the
future.
For now, these errors make it clear why the boot has failed.
Ref TW13909
Because of errata related to BOD reset function, brownout is handled as follows:
- attach an ISR to brownout interrupt
- when ISR happens, print a message and do a software restart
- esp_restart_nonos enables RTC watchdog, so if restart fails,
there will be one more attempt to restart (using the RTC
watchdog)
RTC watchdog didn’t have any actions configured for any of the stages.
This change configures it to use SW_SYSTEM_RESET at stage 0 and a
full reset at stage 1. The timeout is now calculated based on
RTC_SLOW_CLK frequency.
Specifying -fexceptions for the compiler is not enough.
- add necessary zero padding after .eh_frame section
- link .gcc_except_table_table in a way flash script does not complain
- call __registrer_frame_info before global constructors
Kudos jcmvbkbc for the necessary help.
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
Bug occurs when core dump destination in menuconfig is set to flash. When
programme crashes, xt_unhandled_exception or panicHandler will both trigger
commonErrorHandler. commonErrorHandler calls esp_core_dump_to_flash which
will attempt to use DPORT functions and hang due to trying to a stall and already
stalled processor (already stalled in xt_unhandled_exception and panicHandler).
Program will eventually be rebooted when wdt expires.
Added esp_dport_access_int_deinit after calls to haltOtherCore() so that DPORT
functions don't try to halt and already halted cpu hence preventing hang.
Fixes TW#12944 https://github.com/espressif/esp-idf/issues/646
Small changes to clock calibration value will cause increasing errors
the longer the device runs. Consider the case of deep sleep, assuming
that RTC counter is used for timekeeping:
- before sleep:
time_before = rtc_counter * calibration_val
- after sleep:
time_after = (rtc_counter + sleep_count) * (calibration_val + epsilon)
where 'epsilon' is a small estimation error of 'calibration_val'.
The apparent sleep duration thus will be:
time_after - time_before = sleep_count * (calibration_val + epsilon)
+ rtc_counter * epsilon
Second term on the right hand side is the error in time difference
estimation, it is proportional to the total system runtime (rtc_counter).
To avoid this issue, this change makes RTC_SLOW_CLK calibration value
persistent across restarts. This allows the calibration value update to
be preformed, while keeping time after update same as before the update.
This fixes a bug introduced by !848, where APP CPU would not be reset
during esp_restart, if esp_restart was called from a task running on APP
CPU, and wouldn’t be reset by PRO CPU on startup.
This change replaces stalling APP CPU with resetting it.
Also adds a non-automated esp_restart tests.
DPORT access protection can not work when the other CPU is stalled.
Writes to DPORT registers in esp_restart caused the program to hang due
to access protection, and the reset happened due to RTC_WDT, not SW_RST.
This change adds esp_dport_access_int_deinit function and calls it from
esp_restart once the other core is stalled.
This seems to clean up some of the more wrong addr2line output results (not
sure why, something to do with optimisations I think - perhaps the return
address may also be a jump target from some earlier line of the code.)
Previously, this resulted in task stack frames turning up incorrectly in the backtrace, ie
Backtrace: 0x400d22a0:0x3ffb0fa0 0x40085a3c:0x3ffb0fc0 0x400f32c4:0x3ffb0fe0 0x40081965:0x3ffb1010
0x400d22a0: esp_vApplicationIdleHook at /home/esp/esp-idf/components/esp32/./freertos_hooks.c:
52
0x40085a3c: prvIdleTask at /home/esp/esp-idf/components/freertos/./tasks.c:4431
0x400f32c4: i2c_isr_handler_default at /home/esp/esp-idf/components/driver/./i2c.c:598
0x40081965: _xt_lowint1 at xtensa_vectors.o:?
Fix is to implement abort() via an unhandled exception rather than a breakpoint, I think
because of relative priority of exception types.
Another approach would be to assign a software-only INUM to abort()ing and defined a
PANIC_RSN_ABORTED, but this is more complex and interrupt numbers are more scarce than RAM!
When ‘reset halt’ command is executed, OpenOCD will take the APP CPU
out of reset and enable the clock. At this point, user can set a
breakpoint on code which will run on APP CPU. Previously, app startup
code would do another reset of APP CPU, thereby removing any breakpoints
which may have been set. This change makes APP CPU reset conditional on
DPORT_APPCPU_CLKGATE_EN bit, which is 0 by default but is set to 1 by
OpenOCD after reset.
To increase the chances that the examples work out of the box, this
change is raising the default deep sleep wakeup delay to 1ms.
If GPIO15 is low at startup, ROM code takes less time to execute
(because no logging is performed), so more time may be needed to allow
flash chip to become ready.
