Consus | Consus is a geo-replicated transactional key-value store | Architecture library

By enabling interrupts inside interrupt handler you don't enable only "others" interrupts, but all interrupts. Including the timer interrupt. (this is too general, payne points out in comment, on x86 there may be different sources of IRQ signals, and particular PIC/APIC chips can be configured to prioritize one over other ... doesn't change anything about the principle of re-entrant IRQ handling, but you can fine-tune which IRQs have priority, which may mean you don't need re-entry lock at all in the end, if your configuration prevents it)

Your code doesn't look like re-entrant-aware handler, so if that is how your handler code is done, then your timer keeps firing up new timer interrupts as configured, and because you have sti there, they keep re-entering the handler code, filling up slowly stack space until it will overflow (or wrap around, I'm not sure if you are in real or protected mode, and how your stack is set up). Only the last one is of course running, rest is dormant (forever, if the last one doesn't return).

A long-duration interrupt handlers are usually sign of bad code architecture (aim for shortest + fastest possible interrupt handlers, often just filling up some queue with data, which is processed in async way outside of interrupt handler). But if you really need one, and you can't leave other interrupts disabled, you must write the handler as re-entrant one.