On Tue, Sep 11, 2012 at 9:06 PM, Anton Vorontsov anton.vorontsov@linaro.org wrote:
On Tue, Sep 11, 2012 at 08:42:46PM -0700, Colin Cross wrote: [...]
The "problem" is in the last step. If we exit NMI without making UART know that we're done with the interrupt, we will reenter the NMI immediately, even without any new characters from the UART.
The UART irq line should go low when you read the character out of the
Probably some controllers may lower the line by themselves, but not all, and probably most of them need an explicit clear.
Anything 8250-based will clear the interrupt automatically, assuming you read the status registers as well as the character register.
receive buffer, or the polling rx function should clear the interrupt for you.
Yes, that's an option. But that way we add a new semantic for the polling routines, and effecitvely we just merge the two callbacks.
Of course, if Alan is OK with this, I'm more than OK too. :-)
(But the polling routines would need to clear all interrupts, not just rx/tx. For example, if the controller indicated some error, and nobody clears it, then we'll start reentering infinitely.)
For exynos5, the only non-8250 based serial port I've come across, we clear all interrupts in the rx poll function (see https://android.googlesource.com/kernel/exynos/+/ef427aafffb7153dde59745e440...).
If you use a clear_irqs callback, you can drop characters if one arrives between the last character buffer read and calling clear_irqs.
Only if we call clear_irqs() after reading the characters, but we do it before. So if new characters are available, we will reenter NMI, which is OK.
But if used incorrectly, it truly can cause dropping (or staling) of characters, so I'd better add some comments about this.
What does clear_irqs() mean for a status or tx interrupt? The tx interrupt will generally re-assert as long as the tx fifo is empty, which would require disabling it. On 8250 ports, status interrupts will re-assert until the corresponding status register is read.