SPI4 RXNE never raises on STM32G4

[amo25]

Hello, my project is using all 4 SPI interfaces on the STM32G474. SPI1-3 all work flawlessly, but SPI4 never triggers the RXNE flag. All SPI buses are configured identically, with the only differences being the pinouts and APB1 vs 2 bus. I am polling the transactions directly.

Initialization:

if (hspi_.Instance == SPI1)
  LL_APB2_GRP1_EnableClock(LL_APB2_GRP1_PERIPH_SPI1);
else if (hspi_.Instance == SPI2)
  LL_APB1_GRP1_EnableClock(LL_APB1_GRP1_PERIPH_SPI2);
else if (hspi_.Instance == SPI3)
  LL_APB1_GRP1_EnableClock(LL_APB1_GRP1_PERIPH_SPI3);
else if (hspi_.Instance == SPI4)
  LL_APB2_GRP1_EnableClock(LL_APB2_GRP1_PERIPH_SPI4);

LL_SPI_SetMode(hspi_.Instance, LL_SPI_MODE_MASTER);
LL_SPI_SetTransferDirection(hspi_.Instance, LL_SPI_FULL_DUPLEX); 
LL_SPI_SetDataWidth(hspi_.Instance, LL_SPI_DATAWIDTH_8BIT); 
LL_SPI_SetClockPolarity(hspi_.Instance, LL_SPI_POLARITY_LOW); 
LL_SPI_SetClockPhase(hspi_.Instance, LL_SPI_PHASE_1EDGE); 
LL_SPI_SetBaudRatePrescaler(hspi_.Instance, LL_SPI_BAUDRATEPRESCALER_DIV256);
LL_SPI_SetTransferBitOrder(hspi_.Instance, LL_SPI_MSB_FIRST);
LL_SPI_SetStandard(hspi_.Instance, LL_SPI_PROTOCOL_MOTOROLA);
LL_SPI_SetNSSMode(hspi_.Instance, LL_SPI_NSS_SOFT); 
SET_BIT(hspi_.Instance->CR1, SPI_CR1_SSI); 
LL_SPI_SetRxFIFOThreshold(hspi_.Instance, LL_SPI_RX_FIFO_TH_QUARTER); 
LL_SPI_DisableNSSPulseMgt(hspi_.Instance); 
LL_SPI_DisableCRC(hspi_.Instance);
LL_SPI_Enable(hspi_.Instance);

Transfer:

void write(uint8_t* data, size_t len) {
  for (uint16_t i = 0; i < len; i++)
  {
    while (!LL_SPI_IsActiveFlag_TXE(hspi_.Instance)) {}
    LL_SPI_TransmitData8(hspi_.Instance, data[i]);

    while (!LL_SPI_IsActiveFlag_RXNE(hspi_.Instance)) {}
    (void)LL_SPI_ReceiveData8(hspi_.Instance);
  }
  while (LL_SPI_IsActiveFlag_BSY(hspi_.Instance)) {}
  return;
}

For SPI1-3, write works fine. For SPI4, it hangs on the while RXNE wait forever. 

I verified SPI4 is enabled (check RCC_APB2ENR) and not held in reset (check RCC_APB2RSTR):

# RCC_APB2ENR SPI4_EN
(gdb) p/x *0x40021060 >> 15 & 1
$25 = 0x1

# RCC_APB2RSTR SPI4_RST
(gdb) p/x *0x40021040 >> 15 & 1
$20 = 0x0

I can reproduce this across multiple boards and eval boards. Any idea what is wrong?

[TDK · Accepted solution]

If SCK line isn't toggled, RXNE won't get set. Verify the SCK line is actually active with a scope or logic analyzer.  Barring that, verify the pin can be toggled freely when driven in output mode. Verify pins are set to VERY_HIGH frequency, or whatever is appropriate for the speed you are using.

[amo25]

The pins are AF_PP, NOPULL, and VERY_HIGH speed. I can drive them manually in OUTPUT_PP with no issue. 

SCK is active and toggling, but I notice it is at a way lower voltage level (1.5-2.5V) than expected, though this may be a problem with my board - would this cause issues or is the SCK->RXNE connection internal?  

For what it's worth, I see RXNE fail to respond on the NUCLEO-G474RE eval board as well, which does not have PE2/5/6 exposed. 

[gbm]

Is SSM bit set?

[amo25]

Yes, SSM is set.

(gdb) p hspi_.Instance
$1 = (SPI_TypeDef *) 0x40013c00

(gdb) p/x hspi_.Instance->CR1
$2 = 0x37c
(gdb) p/x hspi_.Instance->CR1 >> 9 & 1
$3 = 0x1

[Andrew Neil]

---

@amo25 wrote:

SCK is active and toggling, but I notice it is at a way lower voltage level (1.5-2.5V) than expected,  

---

If the level is too low, the STM32 will just see it as a constant low - ie, as no clock.

As already noted, if the STM32 sees no clock, nothing will be clocked-in and RXNE won't get set.

[waclawek.jan]

> The pins are AF_PP, NOPULL, and VERY_HIGH speed. I can drive them manually in OUTPUT_PP with no issue.

> SCK is active and toggling, but I notice it is at a way lower voltage level (1.5-2.5V) than expected,

Is SCK PE2? And if you toggle it "manually" as GPIO Output, is its amplitude correct, or lower too?

>  would this cause issues or is the SCK->RXNE connection internal?

As others said, yes, this would cause issues, the Rx even in Master uses the input version of the SCK.

> For what it's worth, I see RXNE fail to respond on the NUCLEO-G474RE eval board as well, which does not have PE2/5/6 exposed.

That's irrelevant, there's no PE port on the 64-pin (and 48-pin) 'G474, thus "no" SPI4 - look at the peripherals list/table in the datasheet, for the 48/64-pin variant, there are only 3 SPIs.

JW

[amo25]

Thank you all for the input. The voltage sag was indeed the problem, depopulating the SPI4 bus brings the line back up to 3.3V and RXNE raises as expected.