HAL_I2C_SlaveRxCpltCallback is not executed in I2C DMA (BUG?)

I am trying to use I2C DMA on 2 STM32H7B3I-DKs. After the message is sent to the slave device using HAL_I2C_Master_Transmit_DMA(), the slave seems to receive the message as HAL_I2C_Slave_Receive_DMA() returns HAL_OK but HAL_I2C_SlaveRxCpltCallback() fails to execute afterwards.

Master config:

static void MX_I2C4_Init(void)
{
 hi2c4.Instance = I2C4;
 hi2c4.Init.Timing = 0x00303D5B;
 hi2c4.Init.OwnAddress1 = 0;
 hi2c4.Init.AddressingMode = I2C_ADDRESSINGMODE_7BIT;
 hi2c4.Init.DualAddressMode = I2C_DUALADDRESS_DISABLE;
 hi2c4.Init.OwnAddress2 = 0;
 hi2c4.Init.OwnAddress2Masks = I2C_OA2_NOMASK;
 hi2c4.Init.GeneralCallMode = I2C_GENERALCALL_DISABLE;
 hi2c4.Init.NoStretchMode = I2C_NOSTRETCH_DISABLE;
 if (HAL_I2C_Init(&hi2c4) != HAL_OK)
 {
 Error_Handler();
 }
 
 /** Configure Analogue filter
 */
 if (HAL_I2CEx_ConfigAnalogFilter(&hi2c4, I2C_ANALOGFILTER_ENABLE) != HAL_OK)
 {
 Error_Handler();
 }
 
 /** Configure Digital filter
 */
 if (HAL_I2CEx_ConfigDigitalFilter(&hi2c4, 0) != HAL_OK)
 {
 Error_Handler();
 }
 
 
}
static void MX_BDMA2_Init(void)
{
 
 /* DMA controller clock enable */
 __HAL_RCC_BDMA2_CLK_ENABLE();
 
 /* DMA interrupt init */
 /* BDMA2_Channel0_IRQn interrupt configuration */
 HAL_NVIC_SetPriority(BDMA2_Channel0_IRQn, 0, 0);
 HAL_NVIC_EnableIRQ(BDMA2_Channel0_IRQn);
 /* BDMA2_Channel1_IRQn interrupt configuration */
 HAL_NVIC_SetPriority(BDMA2_Channel1_IRQn, 0, 0);
 HAL_NVIC_EnableIRQ(BDMA2_Channel1_IRQn);
 
}

Master code:

void main()
{
 HAL_Init();
 SystemClock_Config();
 MX_GPIO_Init();
 MX_BDMA2_Init();
 MX_I2C4_Init();
 char Buffer[16]="hi";
 while(HAL_I2C_Master_Transmit_DMA(&hi2c4, 0x11,(uint8_t*)Buffer, sizeof(Buffer))!= HAL_OK)
 {
 if (HAL_I2C_GetError(&hi2c4) != HAL_I2C_ERROR_AF)
 {
 Error_Handler();
 }
 }
 while (HAL_I2C_GetState(&hi2c4) != HAL_I2C_STATE_READY)
 {
 }
}

slave config:

static void MX_I2C4_Init(void)
{
 hi2c4.Instance = I2C4;
 hi2c4.Init.Timing = 0x00303D5B;
 hi2c4.Init.OwnAddress1 = 0x11;
 hi2c4.Init.AddressingMode = I2C_ADDRESSINGMODE_7BIT;
 hi2c4.Init.DualAddressMode = I2C_DUALADDRESS_DISABLE;
 hi2c4.Init.OwnAddress2 = 0;
 hi2c4.Init.OwnAddress2Masks = I2C_OA2_NOMASK;
 hi2c4.Init.GeneralCallMode = I2C_GENERALCALL_DISABLE;
 hi2c4.Init.NoStretchMode = I2C_NOSTRETCH_DISABLE;
 if (HAL_I2C_Init(&hi2c4) != HAL_OK)
 {
 Error_Handler();
 }
 
 /** Configure Analogue filter
 */
 if (HAL_I2CEx_ConfigAnalogFilter(&hi2c4, I2C_ANALOGFILTER_ENABLE) != HAL_OK)
 {
 Error_Handler();
 }
 
 /** Configure Digital filter
 */
 if (HAL_I2CEx_ConfigDigitalFilter(&hi2c4, 0) != HAL_OK)
 {
 Error_Handler();
 }
 
}

slave code:

char Buffer[100];
int main()
{ 
 HAL_Init();
 SystemClock_Config();
 MX_GPIO_Init();
 MX_BDMA2_Init();
 MX_I2C4_Init();
 MX_USART1_UART_Init();
 if(HAL_I2C_Slave_Receive_DMA(&hi2c4, (uint8_t *)Buffer, sizeof(Buffer)) != HAL_OK)
 {
 Error_Handler();
 }
 while (HAL_I2C_GetState(&hi2c4) != HAL_I2C_STATE_READY)
 {
 }
}
 
void HAL_I2C_SlaveRxCpltCallback (I2C_HandleTypeDef * hi2c)
{
	if (hi2c->Instance==hi2c4.Instance)
	{
		HAL_UART_Transmit(&huart1,(uint8_t *)Buffer,strlen(Buffer),1000);
	}
 
}

I read from a few other posts that there seems to be a bug in HAL_I2C_MspInit() where the order of how the peripherals are initialised is wrong. But I'm not sure if this is the case. Below shows the intialisation code for I2C.

void HAL_I2C_MspInit(I2C_HandleTypeDef* hi2c)
{
 GPIO_InitTypeDef GPIO_InitStruct = {0};
 RCC_PeriphCLKInitTypeDef PeriphClkInitStruct = {0};
 if(hi2c->Instance==I2C4)
 {
 /* USER CODE BEGIN I2C4_MspInit 0 */
 
 /* USER CODE END I2C4_MspInit 0 */
 
 /** Initializes the peripherals clock
 */
 PeriphClkInitStruct.PeriphClockSelection = RCC_PERIPHCLK_I2C4;
 PeriphClkInitStruct.I2c4ClockSelection = RCC_I2C4CLKSOURCE_D3PCLK1;
 if (HAL_RCCEx_PeriphCLKConfig(&PeriphClkInitStruct) != HAL_OK)
 {
 Error_Handler();
 }
 
 __HAL_RCC_GPIOD_CLK_ENABLE();
 /**I2C4 GPIO Configuration
 PD13 ------> I2C4_SDA
 PD12 ------> I2C4_SCL
 */
 GPIO_InitStruct.Pin = I2C4_SDA_Pin|I2C4_SCL_Pin;
 GPIO_InitStruct.Mode = GPIO_MODE_AF_OD;
 GPIO_InitStruct.Pull = GPIO_NOPULL;
 GPIO_InitStruct.Speed = GPIO_SPEED_FREQ_VERY_HIGH;
 GPIO_InitStruct.Alternate = GPIO_AF4_I2C4;
 HAL_GPIO_Init(GPIOD, &GPIO_InitStruct);
 
 /* Peripheral clock enable */
 __HAL_RCC_I2C4_CLK_ENABLE();
 
 /* I2C4 DMA Init */
 /* I2C4_RX Init */
 hdma_i2c4_rx.Instance = BDMA2_Channel0;
 hdma_i2c4_rx.Init.Request = BDMA_REQUEST_I2C4_RX;
 hdma_i2c4_rx.Init.Direction = DMA_PERIPH_TO_MEMORY;
 hdma_i2c4_rx.Init.PeriphInc = DMA_PINC_DISABLE;
 hdma_i2c4_rx.Init.MemInc = DMA_MINC_ENABLE;
 hdma_i2c4_rx.Init.PeriphDataAlignment = DMA_PDATAALIGN_BYTE;
 hdma_i2c4_rx.Init.MemDataAlignment = DMA_MDATAALIGN_BYTE;
 hdma_i2c4_rx.Init.Mode = DMA_NORMAL;
 hdma_i2c4_rx.Init.Priority = DMA_PRIORITY_LOW;
 if (HAL_DMA_Init(&hdma_i2c4_rx) != HAL_OK)
 {
 Error_Handler();
 }
 
 __HAL_LINKDMA(hi2c,hdmarx,hdma_i2c4_rx);
 
 /* I2C4_TX Init */
 hdma_i2c4_tx.Instance = BDMA2_Channel1;
 hdma_i2c4_tx.Init.Request = BDMA_REQUEST_I2C4_TX;
 hdma_i2c4_tx.Init.Direction = DMA_MEMORY_TO_PERIPH;
 hdma_i2c4_tx.Init.PeriphInc = DMA_PINC_DISABLE;
 hdma_i2c4_tx.Init.MemInc = DMA_MINC_ENABLE;
 hdma_i2c4_tx.Init.PeriphDataAlignment = DMA_PDATAALIGN_BYTE;
 hdma_i2c4_tx.Init.MemDataAlignment = DMA_MDATAALIGN_BYTE;
 hdma_i2c4_tx.Init.Mode = DMA_NORMAL;
 hdma_i2c4_tx.Init.Priority = DMA_PRIORITY_LOW;
 if (HAL_DMA_Init(&hdma_i2c4_tx) != HAL_OK)
 {
 Error_Handler();
 }
 
 __HAL_LINKDMA(hi2c,hdmatx,hdma_i2c4_tx);
 
 /* I2C4 interrupt Init */
 HAL_NVIC_SetPriority(I2C4_EV_IRQn, 0, 0);
 HAL_NVIC_EnableIRQ(I2C4_EV_IRQn);
 HAL_NVIC_SetPriority(I2C4_ER_IRQn, 0, 0);
 HAL_NVIC_EnableIRQ(I2C4_ER_IRQn);
 /* USER CODE BEGIN I2C4_MspInit 1 */
 
 /* USER CODE END I2C4_MspInit 1 */
 }
 
}

Does anyone know how to solve this issue? Or is anyone else experiencing this problem too?

Thank you for your time!

Update 1:

After stepping through the code, I have noticed that the state of I2C4 is always HAL_I2C_STATE_BUSY_RX. However, I have already enabled the event & error interrupts and BDMA interrupts.