**[STM32N6 / CubeMX Bug] ADC Oversampling RightBitShift field not generated

[mraehle]

Hi 
I had some problems to setup the ADC1 and ADC2 on STM32N6 with CubeMX. I use 8 channels of ADC1 and 2 channels of ADC2.

Environment
- MCU: STM32N645
- CubeMX version: 6.17 (latest)
- HAL Driver: STM32CubeN6
- IDE: STM32CubeIDE v2.1.1

---

Problem Description

When configuring ADC1 (or ADC2) with Oversampling enabled in CubeMX and selecting a RightBitShift (e.g. 6-bit shift for Ratio=64), the generated `MX_ADC1_Init()` code is **missing the `RightBitShift` field** in the `Oversampling` struct.

CubeMX correctly generates:

hadc1.Init.OversamplingMode = ENABLE;
hadc1.Init.Oversampling.Ratio = 64;
hadc1.Init.Oversampling.TriggeredMode = ADC_TRIGGEREDMODE_SINGLE_TRIGGER;
hadc1.Init.Oversampling.OversamplingStopReset = ADC_REGOVERSAMPLING_CONTINUED_MODE;

But **does NOT generate**:
```c
hadc1.Init.Oversampling.RightBitShift = ADC_RIGHTBITSHIFT_6;
```

As a result, `RightBitShift` defaults to `0` (no shift), and the DMA buffer receives the raw **accumulated** oversampled value instead of the divided 12-bit result.

---

Symptoms

- With Ratio=64 and RightBitShift=6 configured in CubeMX, the expected DMA value for a 1V input (VDDA=1.8V) would be ~2275 (12-bit).
- Actual DMA value received: ~11'876 — which is the raw accumulated value without the hardware right-shift applied.
- The value in the DMA buffer is always larger than 4095, indicating no shift is being applied.

---

Root Cause

The `ADC_OversamplingTypeDef` struct contains a `RightBitShift` field:

typedef struct {
uint32_t Ratio; // Oversampling ratio
uint32_t RightBitShift; // ← NOT generated by CubeMX
uint32_t TriggeredMode;
uint32_t OversamplingStopReset;
} ADC_OversamplingTypeDef;

Since C/C++ does not zero-initialize struct members that are not explicitly set in this context, `RightBitShift` remains `0` (= `ADC_RIGHTBITSHIFT_NONE`), overriding the CubeMX UI setting at runtime.

---

Workaround

The only reliable fix is to write the `CFGR2` register directly after `MX_ADC1_Init()`. Place this in the `USER CODE BEGIN 0` block (preserved on CubeMX regeneration) using wrapper functions:

main.c enhanced

/* Private user code ---------------------------------------------------------*/
/* USER CODE BEGIN 0 */
void ADC1_Reinit(void) {
    MX_ADC1_Init();
    MODIFY_REG(ADC1->CFGR2, ADC_CFGR2_OVSS, (6UL << ADC_CFGR2_OVSS_Pos));
}
void ADC2_Reinit(void) {
    MX_ADC2_Init();
    MODIFY_REG(ADC2->CFGR2, ADC_CFGR2_OVSS, (6UL << ADC_CFGR2_OVSS_Pos));
}
/* USER CODE END 0 */

Use the Reinit wrapper, where you Calibrate the ADC Channel

extern "C"
{
    void ADC1_Reinit(void);
    void ADC2_Reinit(void);
}


//******************************************************************************
// description:
//   Constructor
//******************************************************************************
AdcHandler::AdcHandler()
{
    m_adc1Hdl_p = new AnalogDigitalConverter(&hadc1, ADC1Ch_Nof, ADC1_Reinit);
    m_adc2Hdl_p = new AnalogDigitalConverter(&hadc2, ADC2Ch_Nof, ADC2_Reinit);
}

AnalogDigitalConverter::AnalogDigitalConverter(ADC_HandleTypeDef* hadc_p, uint8_t nofChannels, ReinitCallback reinitCb)
{
   this->m_adcHdl_p = hadc_p;
	this->m_nofChannels = nofChannels;
    m_reinitCb = reinitCb;
}

void AnalogDigitalConverter::Calibrate()
{
    uint32_t tmpNbrOfConversion = m_adcHdl_p->Init.NbrOfConversion;

    /* EOC temporär auf Single-Conversion für Kalibrierung */
    m_adcHdl_p->Init.EOCSelection = ADC_EOC_SINGLE_CONV;
    m_adcHdl_p->Init.NbrOfConversion = 1;          /* nur 1 Kanal aktiv    */
    m_adcHdl_p->Init.ScanConvMode    = DISABLE;
    m_adcHdl_p->Init.ContinuousConvMode = DISABLE;
    if (HAL_ADC_Init(m_adcHdl_p) != HAL_OK) {
        Error_Handler();
    }

    /* Kalibrierung */
    if (HAL_ADCEx_Calibration_Start(m_adcHdl_p, ADC_SINGLE_ENDED) != HAL_OK)
    {
        Error_Handler();
    }

    LL_ADC_Disable(m_adcHdl_p->Instance);
    /* Warten bis disabled */
    uint32_t tick = HAL_GetTick();
    while (LL_ADC_IsEnabled(m_adcHdl_p->Instance))
    {
        if ((HAL_GetTick() - tick) > 10U) { Error_Handler(); break; }
    }


    /* Vollständige Konfiguration via CubeMX-Wrapper wiederherstellen  */
    /* Setzt korrekt: Oversampling, Channel-Ranks, DMA-Link, SampTimes */
    if (m_reinitCb != nullptr)
    {
        m_reinitCb();
    }
}

---

Additional STM32N6-specific ADC pitfalls not covered by CubeMX

During the investigation, several other issues were discovered:

1. **`HAL_PWREx_EnableVddA()` missing from `HAL_ADC_MspInit()`**
CubeMX does not generate this call. Without it, ADC calibration hangs in timeout (ADSTART never clears). Must be added manually to `HAL_ADC_MspInit()`. See also ST Community thread #815299.

2. **`ADC_CR_ADVREGEN` must be set before `HAL_ADC_Init()`**
The internal ADC voltage regulator needs explicit activation and a stabilization time of at least 10µs before calibration can start:
```c
SET_BIT(hadc->Instance->CR, ADC_CR_ADVREGEN);
HAL_Delay(1);
```

3. **`HAL_ADCEx_Calibration_Start()` hangs when `NbrOfConversion > 1`**
The calibration routine internally calls `LL_ADC_REG_StartConversion()` and polls for EOC after each single conversion. With `EOCSelection = ADC_EOC_SEQ_CONV` and `NbrOfConversion > 1`, the EOC flag is only set after the complete sequence, causing a timeout.
Solution: temporarily set `NbrOfConversion=1`, `ScanConvMode=DISABLE`, `ContinuousConvMode=DISABLE` before calibration, then restore the full configuration via the re-init wrapper.

4. **ADC remains ENABLED after `HAL_ADCEx_Calibration_Start()` on STM32N6**
Unlike other STM32 families, the STM32N6 ADC is not automatically disabled after calibration. `LL_ADC_Disable()` must be called explicitly before reconfiguring registers:

HAL_ADCEx_Calibration_Start(&hadc1, ADC_SINGLE_ENDED);
LL_ADC_Disable(hadc1.Instance);
uint32_t tick = HAL_GetTick();
while (LL_ADC_IsEnabled(hadc1.Instance)) {
if ((HAL_GetTick() - tick) > 10U) { Error_Handler(); break; }
}

---

Expected Fix from ST

- CubeMX should generate the `RightBitShift` field when Oversampling with RightBitShift != 0 is configured in the UI.
- `HAL_ADC_Init()` should reliably update `CFGR2` on repeated calls, or at minimum document that a state reset is required.

---

Hope this helps others hitting the same issue. Please confirm if this is a known CubeMX bug and whether a fix is planned.

Best regards
Markus

[Souhaib MAZHOUD]

Hello @mraehle 

Thank you for reporting this issue; your detailed explanation is very helpful.

I will escalate it to the appropriate team for further investigation and review.

BR, Souhaib