SPI in async mode will leave IRQ enabled

[mean00]

When using the SPI IRQ in async mode, the SPI IRQ will be left enabled The dma_detach_function will not disable it, with comment / Don't use nvic_irq_disable()! Think about DMA2 channels 4 and 5. / The problem is when alternating between blocking SPI send and async SPI Send, the irq is left enabled and non managed interrupt will happen. Additionnaly if that happens the dma_irq_handler will loop as it does not clear the interrupt if no handler is present.

Crude hack is to do that in the SPI itself , i.e. in SPIClass::onTransmit & SPIClass::EventCallback i.e. explicitely call nvic_irq_disable.

It works, but is there a better way ?

(The same problem probably exists with transfer, but i'm only dealing with transmit)

[stevstrong]

Which device? F1 or F4? Do you have a simple example sketch which shows the problem?

[mean00]

Hi I'm using the F1 (f103 actually) i can try creating a simple use case, it might take a bit of time as i'm using a very different environment than the arduino one, with a lot of dependencies.

But the gist of it is :

• Do a spi.dmaSendAsync()
• Do a spi.dmaSend()

You'll run into the 1st problem, i.e. it loops in dma_irq_handler when the spi.dmaSend() is over because the nvic irq is still enabled. The asymetry in dma_detach_interrupt / dma_attach_interrupt is the root cause i believe

[mean00]

Hi Please find attached a tar ball with 2 files :

• The demo script. It will loop inside dma_irq_handler.
• My current workaround, sort of of fixes the above + trigger a debugger exception if there is a spurious interrupt

Thank you

spiTest.tar.gz

[mean00]

ping ?

[stevstrong]

I just committed a fix for this, I have tested and it works correctly now for me. Please check it and reopen the issue if the problem still persists.

[mean00]

seems to work fine Thanks a lot!