I have a strange observation that I can't quite determine the cause for. I have a PCI-based board that communicates with the host CPU (Linux) via a memory mapped area and an interrupt line. I can trigger a board-to-host interrupt from the host via access to memory mapped registers. If I write a loop that generates this interrupt, turn on debugging in my driver to see how many times the IRQ handler gets called, and parse /proc/interrupts for the IRQ count of the board, and the ERR count, I see the following: - either the IRQ count in /proc/interrupts increments by 2 and the IRQ handler gets called twice (the first time through the handler clears the interrupt source, and the second time through it does nothing). or - the IRQ count increments by 1, the IRQ handler gets called once, and the ERR counter in /proc/interrupts gets incremented. This last counter is isr_err_count which is incremented in arch/i386/kernel/i8259.c an is often associated with spurious interrupt messages in dmesg - however in this case the counter just gets incremented. I looked at the interrupt line with a scope. The interrupt is a single low pulse for about 10us. So there is no double interrupt to generate the two observed interrupts. (The machine is a 700MHz P-III). One possibility is that the interrupt source is still valid when the interrupt acknowledge is sent to the interrupt controller. However, my interrupt handler follows what appears to be the recommended format of Linux drivers, so if I experience this problem, then I'm sure others must have too! (However, ERR doesn't seem to be counting up on my other linux boxes!). Also, if I add enough printk statements for debugging in the ISR handler, the IRQ count increments by 1, and ERR does not! So making the handler slower helps! (Or perhaps it is the spin locks within the printk code that disable interrupts ...). I tried registering the handler with SA_INTERRUPT to have the handler run with interrupts disabled, however I still got two interrupts. Any help/comments/tests!? Thanks, Dave Hawkins -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
