Shouldn't clean vector pending before CPU really handle it

[rasoft]

I see you clean pending status of edge vector in
riscv_clic_next_interrupt(void *opaque, int hartid)

it is not correctly! since before the edge vector is really handled, latter IRQ may come and override it, and the edge vector (such as tspend, systick) will not be handled forever.

I think you could clean them when cpu really handle it, for example in
if (enable) branch of bool riscv_cpu_exec_interrupt(CPUState *cs, int interrupt_request)

thx
alan

[romanheros]

Thanks, alan. I think you are right. I don't know why we didn't cover it.
Your adivce about how to fix it is also good. Another way may be register a bh function or event notifier for CLIC, which will make the work for CLIC de-couple with the target.

Do you have a test case? It will be very helpful.

Thanks again,
Zhiwei