Hello Mikko! On 06/18/2018 09:41 AM, Mikko Perttunen wrote: > There seems to be an issue where if the receive interrupt is enabled > when the receive queue is emptied, the interrupt status register will > get updated but the actual interrupt line may not update. This can be > worked around by disabling the receive interrupt in IER and then > restoring the desired value to IER. In some downstream code, this is > done in the IRQ handler when we detect a spurious interrupt, but > presumably it could be done elsewhere as well. > > David, can you try if doing that fixes the issue? Actually I don't think this would help here, as it seems to be the *TX* queue being the culprit. But I will give it a try shortly. Also I might have found something helpful last friday. Testing over the weekend shows promising results: no spurious interrupts. It does NOT involve the wrongly muxed LPW2 though. As I wrote earlier, the error occurs in the following sequence, when we want to write something to the serial port: * when the TX queue is empty: * UARTA.IER := 0x15. Enable RHR (received data), RXS (receiver line status) and RX_TIMEOUT interrupts. * we want to write a character -> enable int. for an empty TX queue: * UARTA.IER := 0x17. Enable THR (transmitter holding reg. empty) Int. * we get an interrupt, as the queue has been empty for a while: * UARTA.IIR == 0xC2 * now we write the char (in interrupt context) * UARTA.TX := some char <<< added FIX: still in ISR, before writing 0x15, another write is added that also disables the ELSI interrupt: UARTA.IER := 0x11 this seems to reliably shut down the spurious interrupt. Note that double-writing 0x15 here did not help. So this seems to relate to the RXS receiver line status interrupt. <<< end of FIX. * still in ISR, kernel-side tx queue is empty, so disable: * UARTA.IER := 0x15. * WITHOUT above write, interrupt would be retriggered again, with: * UARTA.IIR == 0xC1 Currently this seems a bit like another silicon bug, where maybe the interrupt line of the RXS interrupt is in a metastable state, and disabling and re-enabling it properly resets to a valid state. When I have sufficiently tested I will provide a proper patch for review and hopefully some further context. If you already have some comment on this, I'd be glad to hear it! David -- To unsubscribe from this list: send the line "unsubscribe linux-tegra" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
