Re: [RFC PATCH 3/3] kvm: Atomic memslot updates

[Paolo Bonzini]

On 11/4/22 16:14, Emanuele Giuseppe Esposito wrote:
> +    g_assert(qemu_mutex_iothread_locked());

Please add a comment here:

    /* Block further invocations of the ioctls outside the BQL.  */

> +    CPU_FOREACH(cpu) {
> +        qemu_lockcnt_lock(&cpu->in_ioctl_lock);
> +    }
> +    qemu_lockcnt_lock(&kvm_in_ioctl_lock);
>   
> -    kvm_set_phys_mem(kml, section, false);
> -    memory_region_unref(section->mr);
> +    /* Inhibiting happens rarely, we can keep things simple and spin here. */

Not making it spin is pretty easy.  You can add a qemu_event_set to 
kvm_set_in_ioctl() and kvm_cpu_set_in_ioctl(), and here something like:

    if (in_kvm_ioctls()) {
        qemu_event_reset(&kvm_in_ioctl_event);
        if (in_kvm_ioctls()) {
            qemu_event_wait(&kvm_in_ioctl_event);
        }
    }

where in_kvm_ioctls() returns true if any (vCPU or KVM) lockcnt has a 
nonzero count.

Also please create a new header sysemu/accel-blocker.h and 
accel/blocker.c or something like that with all the functions, because 
this code can potentially be used by all KVM-like accelerators.

Paolo