ESP32S3 esp_partition_write & irqs

[Demirug]

I have an ESP32S3 project that emulates an old arcade game (I,Robot). To make sure anything keeps running all the time the different subsystems are all Interrupt triggered. All the code that runs within these interrupts is IRAM_ATTR and the data DRAM_ATTR flagged.
The original game had an NVRAM chip to store high scores and some statistics. To emulate this I am tracking changes and use esp_partition_write to store these to the flash chip. This all works great in general but after some more code changes to track down an issue I now have a reproducible Core 0 Panic

Guru Meditation Error: Core 0 panic'ed (Cache disabled but cached memory region accessed).

I know this happens if the cpu tries to execute code that is on the flash but based on the callstack I am getting this should not be the case

--- 0x40377a57: IRobotSystem::AudioOut(short*, unsigned long) at C:/projects/ESPIRobot/main/IRobotSystem.cpp:379
0x40377183: AudioCallback(short*, unsigned long) at C:/projects/ESPIRobot/main/main.cpp:169
0x40377e5d: AudioOutIrqFunc(void*) at C:/projects/ESPIRobot/main/AudioOutput.cpp:41
0x40382f0d: _xt_medint2 at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/xtensa/xtensa_vectors.S:1329
0x403841e1: delay_us at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/spi_flash/spi_flash_os_func_app.c:177
0x4038f63e: spi_flash_chip_generic_wait_idle at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/spi_flash/spi_flash_chip_generic.c:409
0x40390237: spi_flash_chip_winbond_page_program at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/spi_flash/spi_flash_chip_winbond.c:86
(inlined by) spi_flash_chip_winbond_page_program at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/spi_flash/spi_flash_chip_winbond.c:73
0x4038f102: spi_flash_chip_generic_write at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/spi_flash/spi_flash_chip_generic.c:295
0x40383e65: esp_flash_write at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/spi_flash/esp_flash_api.c:1075
0x420092f1: esp_partition_write at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/esp_partition/partition_target.c:78
0x420081f0: NVRam::Update() at C:/projects/ESPIRobot/main/NVRam.cpp:111
0x40377e33: IRobotSystem::UpdateNVRAM() at C:/projects/ESPIRobot/main/IRobotSystem.cpp:844
0x42006fde: RunIRobot(void*) at C:/projects/ESPIRobot/main/main.cpp:441
0x420070a9: app_main at C:/projects/ESPIRobot/main/main.cpp:483
0x42020d53: main_task at C:/Users/ralfk/esp/v5.3.1/esp-idf/components/freertos/app_startup.c:208

Line 379 is the third call in this block.

Code: Select all

m_Pokey[0].AudioTick(clkpersample);
m_Pokey[1].AudioTick(clkpersample);
m_Pokey[2].AudioTick(clkpersample);
m_Pokey[3].AudioTick(clkpersample);

The function has the IRAM_ATTR

Code: Select all

void IRAM_ATTR Pokey::AudioTick(int clocks)

Based on the map file it has been placed there:

Code: Select all

 .iram1.8       0x4037b988      0x18e esp-idf/main/libmain.a(Pokey.cpp.obj)
                                0x1a6 (size before relaxing)
                0x4037b988                Pokey::AudioTick(int)

Is the error message here missleading for some reason?

I am a little bit surprised that the Interrupt is handled with the stack of the currently running task. Or is this only the case because it is the main task?

[MicroController]

.
I know this happens if the cpu tries to execute code that is on the flash

... or accesses some constant/literal data in flash, including virtual method tables, or data in PSRAM.

The EXCVADDR should point to the location on which the access was attempted.

[Demirug]

... or accesses some constant/literal data in flash, including virtual method tables, or data in PSRAM.

I have no virtual functions as I know they might cause trouble in this setup. Neither is there any PSRAM. I need to inspect the generated assemblercode about other constant and literals but based on the code I do not expect any on this location.

The EXCVADDR should point to the location on which the access was attempted.

It's null.So I am not sure how usefull is it in this case. The function has called many time before this happens and I am pretty sure that the objects have not deleted as there isn't ven code for this.