About the SPI BUG in the HAL library (V1.1.0) of STM32H7

dige

About the SPI BUG in the HAL library (V1.1.0) of STM32H7 [Copy link]

 

I've been playing with stm32h743xi recently, and I feel that the chip is quite powerful. During the SPI test, I found that calling the HAL_SPI_Transmit function to send multiple bytes has not been successful, and only the first byte is sent. After debugging, I found that the hal library did not clear the TXTF flag after sending a byte successfully, resulting in subsequent unsuccessful transmission. Solution: Add /* check TXTF flag */ after line 652 if (__HAL_SPI_GET_FLAG(hspi, SPI_FLAG_TXTF)) { __HAL_SPI_CLEAR_TXTFFLAG(hspi); } I hope it will be helpful to those who debug this later.

yl20084784

Like this

流年u7

There should be a column on the bugs of the HAL library

小Q玉米

Encountered the same problem

dige

Here is another pitfall. In the SPI driver of the H7 hal library, when calling the HAL_SPI_TransmitReceive function, the received data will be wrong. For example, it should receive 01 01 64... but actually receives 01 00 01 64... The reason for the problem is that the TXTF flag is cleared too early. The original cycle of sending and receiving data is to write the sent data to TXDR first, then clear TXTF, and then read the RXDR register data. It should be to write the sent data to TXDR first, then read the RXDR register data, and then clear TXTF. Otherwise, the read data will always be wrong.

huo_hu

dige posted on 2018-1-9 15:32 Let me fill in another hole. In the spi driver of the H7 hal library, calling the HAL_SPI_TransmitReceive function will result in an error in the received data. For example,...

You are wrong. If it is receiving, you must use the RXNE mark. Sending empty does not mean that the reception is completed, because the transmission is buffered.

dige

huo_hu posted on 2018-1-9 15:54 You are wrong. If it is receiving, you must use the RXNE mark. Sending empty does not mean that the reception is completed, because the transmission is buffered.

The test result here is that if the TXTF flag is cleared, the RXNE mark is also cleared, and there is a problem with receiving data.

huo_hu

dige posted on 2018-1-9 16:22 The test result here shows that if the TXTF flag is cleared, the RXNE flag is also cleared, and there is a problem with the received data.

There is a slight time difference, and what you see through the simulator may not be correct.

dige

huo_hu posted on 2018-1-9 21:28 There is a slight time difference here, and what you see through the simulator may not be correct.

May I ask how to modify the reception? The test is to read and write ferroelectric memory through SPI. At present, the order is adjusted, and the ferroelectric reading is normal. If it is not adjusted, the reading will be wrong as mentioned before. I don't know if there is a more reasonable modification.

huo_hu

dige posted on 2018-1-9 22:26 Can you tell me how to modify the reception? The test is to read and write ferroelectric memory through SPI. At present, the order is adjusted, and the ferroelectric reading is normal. If it is not adjusted, the reading...

OK, it can be used. I haven't succeeded in using F405 SPI transceiver interrupt. It worked with f103 before, but not with f4.

小Q玉米

huo_hu posted on 2018-1-10 10:56 OK, as long as it works, I haven't succeeded in using F405 SPI transceiver interrupt. It worked with f103 before, but not with f4.

The SPI architecture of H7 has changed, and it is different from the previous model

star_66666

I suggest you don't use the library, it's rubbish, it's more practical to configure the registers directly, everything is under control

小Q玉米

Have you tested the H7 SPI DMA?

dige

Xiao Q Yumi posted on 2018-1-17 18:59 Have you tested the DMA of H7 SPI?

I have tested it. The SPI receiving DMA has the following problems. First, change the receiving fifo threshold from 08 to SPI_FIFO_THRESHOLD_01DATA, otherwise part of the received data will always be stored in the fifo. Second, shield the following code, otherwise the length of the received data will be incorrect. I hope it helps you. // if ((hspi->RxXferCount & 0x1U) == 0x0U) // { // hspi->RxXferCount = hspi->RxXferCount >> 1U; // } // else // { // hspi->RxXferCount = (hspi->RxXferCount >> 1U) + 1U; // }

ddllxxrr

Should report to ST company: time:

star_66666

The st library is an abyss, a bottomless pit. I hate it. I only use it when time is tight. I usually write it myself and never use the library.

huo_hu

star_66666 posted on 2018-1-26 15:07 The st library is an abyss, a bottomless pit. It is very annoying. I only use it when time is tight. I usually write it myself and never use the library

It will be very tiring.

star_66666

I am a little tired, but I am happy.

dige

star_66666 posted on 2018-1-26 15:07 The library of st is an abyss, a bottomless pit, very annoying, only used when time is tight, usually write it yourself, never use the library

Writing it yourself and using the library each have their own advantages, H7 is just out, the library has more problems, it should be perfected and mature later. I have no choice, I have to rush to release the product, I found a lot of pitfalls, but now everything is smooth.

star_66666

Yes, time is tight.