other: 0.160
device: 0.156
semantic: 0.147
PID: 0.072
performance: 0.070
graphic: 0.061
debug: 0.059
permissions: 0.049
vnc: 0.047
socket: 0.045
network: 0.038
boot: 0.035
files: 0.034
KVM: 0.026
debug: 0.549
other: 0.075
files: 0.066
device: 0.053
semantic: 0.049
boot: 0.037
performance: 0.035
PID: 0.031
permissions: 0.024
socket: 0.022
vnc: 0.018
network: 0.016
graphic: 0.016
KVM: 0.010

RISC-V unreliable IPIs

I am working on a project with custom inter processor interrupts (IPIs) on the RISC-V virt machine.
After upgrading from version 3.1.0 to 4.1.0 which fixes a related issue (https://github.com/riscv/riscv-qemu/issues/132) I am able to use the CPU hotplug feature.

However, if I try to use IPIs for communication between two cores, the wfi instruction behaves strangely. Either it does not return, or it returns on timer interrupts, even though they are disabled. The code, I use on one core to wait for an interrupt is the following.

	csr_clear(sie, SIE_SEIE | SIE_STIE);
	do {
		wait_for_interrupt();
		sipval = csr_read(sip);
		sieval = csr_read(sie);
		scauseval = csr_read(scause) & 0xFF;
	/* only break if wfi returns for an software interrupt */
	} while ((sipval & sieval) == 0 && scauseval != 1);
	csr_set(sie, SIE_SEIE | SIE_STIE);

Since the resulting sequence does not seem to be deterministic, my guess is, that it has something to do with the communication of qemu's threads for the different cores.

Can you post a whole program that reproduces this?  freedom-e-sdk <https://github.com/sifive/freedom-e-sdk> will run bare-metal code on QEMU if you don't want to post the rest of the surrounding infrastructure.

I created a minimal example from my setup. I'm running a kernel 4.19.57 with a custom firmware based on bbl (https://github.com/riscv/riscv-pk). 
An ioctl device from a kernel module is used to execute the code above in kernel space.
In the example, the userspace application proceeds after a couple of seconds without receiving the custom IPI.

The QEMU project is currently considering to move its bug tracking to
another system. For this we need to know which bugs are still valid
and which could be closed already. Thus we are setting older bugs to
"Incomplete" now.

If you still think this bug report here is valid, then please switch
the state back to "New" within the next 60 days, otherwise this report
will be marked as "Expired". Or please mark it as "Fix Released" if
the problem has been solved with a newer version of QEMU already.

Thank you and sorry for the inconvenience.


[Expired for QEMU because there has been no activity for 60 days.]