• Find the right way of triggering IRQ through PCI:
  • when to assert, when to deassert IRQ line;
  • when to start, when to cease interrupting;
• Make PCI device driver that would handle IRQs;

• Modify qemu (easy part): add timer to qemu device, use qemu_irq_pulse inside its callback;
• Make PCI device driver: start with init and exit fns, pci_driver structure and device table;
• Make PCI probe fn; main steps:
  • pci_enable_device (enable device IO and memory);
  • request_irq (still there's no device open fn do it in probe);
  • pci_set_drvdata (to make our device accessible from the struct device level);
• Make PCI release fn, do cleanup of what probe done;
• Add simple irq handler and sysfs output;

Code: ftp://kkv.spb.su/pub/home/dumb/ws/lted/20100104

• when to assert/deassert IRQ line: no matter (at least for emulation);
  • although it is generally level-triggered, qemu treats them as edge-triggered, so that qemu_irq_pulse may be used;
  • anyway (e.g. for proper IRQ sharing) separate 'irq status' register is needed for the device function (not implemented in this demo, but hold on :);
• when to start/stop interrupts: doc contradiction;
  • PCI spec says that IRQ is enabled (interrupt disable == 0 in command register) upon reset, but linux pci documentation says that IRQ will go only when you explicitly enable them;
  • anyway, nothing happens (at least, nothing is visible) until request_irq;
  • even more: right place to request_irq is not probe routine, but rather device open routine;
• qemu 0.12.1 yields maximum of ~3000 PCI interrupts/second;
• using tabs in Linux is stupid ):