Re: Weird qemu-nvmm problem

[Andrew Doran <ad%netbsd.org@localhost>]

On Wed, Mar 11, 2020 at 06:45:06PM +0100, Maxime Villard wrote:

> Please CC me for issues related to NVMM, there is a number of lists where
> I'm not subscribed.
> 
> My understanding is that this commit is the cause (CC ad@):
> 
> 	https://mail-index.netbsd.org/source-changes/2019/12/06/msg111617.html
> 
> NVMM reschedules the thread when the SPCF_SHOULDYIELD flag is set. But
> after this change the flag never gets set, so the rescheduling never
> occurs, and NVMM is stuck with running the guest forever unless a signal
> is caught in the emulator thread.
> 
> The test program below shows the difference. On NetBSD-9 you have many
> "resched", as expected. On NetBSD-current you have none.
> 
> Andrew, can you have a look? There is a good dozen of places that use
> SPCF_SHOULDYIELD for reschedulings, and they too may potentially be buggy
> now.

Yes that was incorrect and I have fixed it.  I will see about adding a
wrapper function for this too.

Andrew

> 
> Thanks,
> Maxime
> 
> 
> 
> ---------------------------------------------------------------------------
> 
> /*
>  * # gcc -o test test.c -lnvmm
>  * # ./test
>  */
> 
> #include <stdio.h>
> #include <stdlib.h>
> #include <unistd.h>
> #include <sys/mman.h>
> #include <fcntl.h>
> #include <nvmm.h>
> #include <stdint.h>
> #include <string.h>
> 
> int main()
> {
> 	uint8_t instr[] = { 0xEB, 0xFE };
> 	struct nvmm_machine mach;
> 	struct nvmm_vcpu vcpu;
> 	uintptr_t hva;
> 
> 	nvmm_init();
> 	nvmm_machine_create(&mach);
> 	nvmm_vcpu_create(&mach, 0, &vcpu);
> 
> 	hva = (uintptr_t)mmap(NULL, 4096, PROT_READ|PROT_WRITE,
> 	    MAP_ANON|MAP_PRIVATE, -1, 0);
> 	nvmm_hva_map(&mach, hva, 4096);
> 	nvmm_gpa_map(&mach, hva, 0xFFFF0000, 4096, PROT_READ|PROT_EXEC);
> 
> 	memcpy((void *)hva, instr, sizeof(instr));
> 
> 	nvmm_vcpu_getstate(&mach, &vcpu, NVMM_X64_STATE_GPRS);
> 	vcpu.state->gprs[NVMM_X64_GPR_RIP] = 0;
> 	nvmm_vcpu_setstate(&mach, &vcpu, NVMM_X64_STATE_GPRS);
> 
> 	while (1) {
> 		printf("looping\n");
> 		nvmm_vcpu_run(&mach, &vcpu);
> 
> 		switch (vcpu.exit->reason) {
> 		case NVMM_VCPU_EXIT_NONE:
> 			printf("resched\n");
> 			break;
> 		default:
> 			printf("unknown: %d\n", vcpu.exit->reason);
> 			break;
> 		}
> 	}
> }