ESP32 Core v2.0.1 / 2.0.1 RC1 crashes if using float in ISR

[khoih-prog]

Core ESP32 Core v2.0.1 / 2.0.0 crashes if using float inside ISR. No issue with core v2.0.0-

This creates issue with some ESP32 TimerInterrupt-related libraries, such as mentioned in ESP32_New_TimerInterrupt Important Notes

With the following simple sketch, it's working OK if using

#define USING_FLOAT_IN_ISR        false

but crashes if using

#define USING_FLOAT_IN_ISR        true

---

Minimal, Reproducible Example (MRE) sketch

#define USING_FLOAT_IN_ISR        true  //false

volatile int    interruptCounter;
volatile float  interruptFloatCounter;

int totalInterruptCounter;

hw_timer_t * timer = NULL;
portMUX_TYPE timerMux = portMUX_INITIALIZER_UNLOCKED;

void IRAM_ATTR onTimer()
{
  portENTER_CRITICAL_ISR(&timerMux);

#if USING_FLOAT_IN_ISR
  interruptFloatCounter += 1.0f;
#else
  interruptCounter++;
#endif

  portEXIT_CRITICAL_ISR(&timerMux);
}

void setup()
{
  Serial.begin(115200);
  while (!Serial);

  delay(100);

  Serial.print(F("\nStarting ESP32TimerInterrupt_Float_Bug on ")); Serial.println(ARDUINO_BOARD);

  timer = timerBegin(0, 80, true);
  timerAttachInterrupt(timer, &onTimer, true);
  timerAlarmWrite(timer, 1000000, true);
  timerAlarmEnable(timer);
}

void loop()
{
  if (interruptCounter > 0)
  {
    portENTER_CRITICAL(&timerMux);

#if USING_FLOAT_IN_ISR
    interruptFloatCounter -= 1.0f;
#else
    interruptCounter--;
#endif

    portEXIT_CRITICAL(&timerMux);

    totalInterruptCounter++;

    Serial.print("An interrupt as occurred. Total number: ");
    Serial.println(totalInterruptCounter);
  }
}

---

Debug Output if not using float

Starting ESP32TimerInterrupt_Float_Bug on ESP32_DEV
An interrupt as occurred. Total number: 1
An interrupt as occurred. Total number: 2
An interrupt as occurred. Total number: 3
An interrupt as occurred. Total number: 4
An interrupt as occurred. Total number: 5
An interrupt as occurred. Total number: 6
An interrupt as occurred. Total number: 7
An interrupt as occurred. Total number: 8
An interrupt as occurred. Total number: 9
An interrupt as occurred. Total number: 10
An interrupt as occurred. Total number: 11
An interrupt as occurred. Total number: 12
An interrupt as occurred. Total number: 13
An interrupt as occurred. Total number: 14
An interrupt as occurred. Total number: 15
An interrupt as occurred. Total number: 16
An interrupt as occurred. Total number: 17
An interrupt as occurred. Total number: 18
An interrupt as occurred. Total number: 19

---

Debug Output if using float

Starting ESP32TimerInterrupt_Float_Bug on ESP32_DEV
Guru Meditation Error: Core  1 panic'ed (Coprocessor exception). 

Core  1 register dump:
PC      : 0x4008110d  PS      : 0x00060033  A0      : 0x800d1168  A1      : 0x3ffbedb0  
A2      : 0x3ffbdb68  A3      : 0x00000001  A4      : 0x80089bd1  A5      : 0x00000000  
A6      : 0x3ffc15dc  A7      : 0x3ffc15dc  A8      : 0x3ffc10f8  A9      : 0x3ffbee30  
A10     : 0x00060021  A11     : 0x3ffc15dc  A12     : 0x3ffc15d4  A13     : 0xb33fffff  
A14     : 0x00000001  A15     : 0x00000000  SAR     : 0x00000020  EXCCAUSE: 0x00000004  
EXCVADDR: 0x00000000  LBEG    : 0x40085b68  LEND    : 0x40085b73  LCOUNT  : 0xffffffff  

Backtrace:0x4008110a:0x3ffbedb00x400d1165:0x3ffbee90 0x400840ad:0x3ffbeeb0 0x400899c7:0x3ffb2750 0x40088601:0x3ffb2770 0x400d1f08:0x3ffb27b0 0x400d146d:0x3ffb27e0 0x400d1505:0x3ffb2800 0x400d1baa:0x3ffb2820 

ELF file SHA256: 0000000000000000

Rebooting...
ets Jun  8 2016 00:22:57

rst:0xc (SW_CPU_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
configsip: 0, SPIWP:0xee
clk_drv:0x00,q_drv:0x00,d_drv:0x00,cs0_drv:0x00,hd_drv:0x00,wp_drv:0x00
mode:DIO, clock div:1
load:0x3fff0030,len:1420
ho 0 tail 12 room 4
load:0x40078000,len:13540
load:0x40080400,len:3604
entry 0x400805f0

---

Exception Decoder

PC: 0x4008110d: onTimer() at /home/kh/Arduino/Testing/ESP32TimerInterrupt/ESP32TimerInterrupt_Float_Bug/ESP32TimerInterrupt_Float_Bug.ino line 17
EXCVADDR: 0x00000000

Decoding stack results
0x4008110a: onTimer() at /home/kh/Arduino/Testing/ESP32TimerInterrupt/ESP32TimerInterrupt_Float_Bug/ESP32TimerInterrupt_Float_Bug.ino line 17
0x400d1165: __timerISR at /home/kh/.arduino15/packages/esp32/hardware/esp32/2.0.1/cores/esp32/esp32-hal-timer.c line 110
0x400899c7: xTaskGetSchedulerState at /Users/ficeto/Desktop/ESP32/ESP32S2/esp-idf-public/components/hal/esp32/include/hal/cpu_ll.h line 39
0x40088601: xQueueSemaphoreTake at /Users/ficeto/Desktop/ESP32/ESP32S2/esp-idf-public/components/freertos/queue.c line 1554
0x400d1f08: uartAvailable at /home/kh/.arduino15/packages/esp32/hardware/esp32/2.0.1/cores/esp32/esp32-hal-uart.c line 192
0x400d146d: HardwareSerial::available() at /home/kh/.arduino15/packages/esp32/hardware/esp32/2.0.1/cores/esp32/HardwareSerial.cpp line 193
0x400d1505: serialEventRun() at /home/kh/.arduino15/packages/esp32/hardware/esp32/2.0.1/cores/esp32/HardwareSerial.cpp line 98
0x400d1baa: loopTask(void*) at /home/kh/.arduino15/packages/esp32/hardware/esp32/2.0.1/cores/esp32/main.cpp line 47

[chegewara]

There is few topics on forum about this and it is not arduino related. Here is just few:
https://esp32.com/viewtopic.php?t=1292
https://esp32.com/viewtopic.php?t=1292&start=10
https://www.esp32.com/viewtopic.php?t=831

[khoih-prog]

Thanks @chegewara. I'll use double instead.
But why it's OK with core v1.0.6- ?

[chegewara]

A bug? optimization?
It has something to do with esp-idf, but as you can see the first problems with float in ISR are dated at 2016, so you just got lucky it worked in 1.0.6.

[khoih-prog]

so you just got lucky it worked in 1.0.6.

I don't believe in luck at all. But we have to accept the reality that some mods in esp-idf (and/or core v2.0.0+) disrupted the float usage in ISR. Good to know.

Thanks anyway,

[me-no-dev]

there was something we had to do to get it working. I'll ask around to see if anyone has a clue what was the fix.

[khoih-prog]

Thanks,

There is a possible link here

[me-no-dev]

@khoih-prog I will turn on that option and soon have it in master branch for the next release :)

[atanisoft]

@khoih-prog Do you have a C3/S2 that you can test your minimal sketch with to confirm if they crash similarly?

@me-no-dev that flag appears to be ESP32 specific so it may introduce discrepancy between platform support.

[khoih-prog]

@atanisoft

The ESP32 C3/S2 are single core, so they don't crash as this happens only to dual-core ESP32

1. Test for ESP32-C3 is OK with float using the same sketch

Starting ESP32TimerInterrupt_Float_Bug on ESP32C3_DEV
An interrupt as occurred. Total number: 1
An interrupt as occurred. Total number: 2
An interrupt as occurred. Total number: 3
An interrupt as occurred. Total number: 4
An interrupt as occurred. Total number: 5
An interrupt as occurred. Total number: 6
An interrupt as occurred. Total number: 7
An interrupt as occurred. Total number: 8
An interrupt as occurred. Total number: 9
An interrupt as occurred. Total number: 10
An interrupt as occurred. Total number: 11
An interrupt as occurred. Total number: 12
An interrupt as occurred. Total number: 13
An interrupt as occurred. Total number: 14
An interrupt as occurred. Total number: 15
An interrupt as occurred. Total number: 16

---

2. Test for ESP32-S2 is OK with float using this RPM_Measure, after changing RPM_Measure.ino#L76-L77 to float.

float RPM       = 0;
float avgRPM    = 0;
//uint32_t RPM       = 0;
//uint32_t avgRPM    = 0;

This mod will crash ESP32 immediately while OK with ESP32_S2. Somehow, ESP32_S2 can't use the minimal sketch code

Starting RPM_Measure on ESP32S2_DEV
ESP32_New_TimerInterrupt v1.0.1
CPU Frequency = 240 MHz
Starting  ITimer0 OK, millis() = 643
RPM  = 740.74, avgRPM  = 529.53
RPM  = 740.74, avgRPM  = 625.56
RPM  = 740.74, avgRPM  = 694.70
RPM  = 0.00, avgRPM  = 173.68
RPM  = 0.00, avgRPM  = 43.42
RPM  = 0.00, avgRPM  = 10.85
RPM  = 0.00, avgRPM  = 2.71
RPM  = 0.00, avgRPM  = 0.68
RPM  = 0.00, avgRPM  = 0.17
RPM  = 0.00, avgRPM  = 0.04
RPM  = 0.00, avgRPM  = 0.01

[igrr]

Note that using floats in ISRs shouldn't be a problem on S2 or C3 because neither have a hardware floating point unit. However this might be an issue on the ESP32-S3 (same way as it is an issue on the ESP32).

[me-no-dev]

Merged in master :)