fix: modify the maintenance of host sp_el0

[hky1999]

BUG decription

There is a bug in current solution of the maintenance of SP_EL0 belonging to host ArceOS, which is used to store the current task pointer.

That is, we save the host's sp_el0 in a wrong place: after restore_vm_system_regs, where guest's sp_el0 is already write to sp_el0 register.

As a result, host save a polluted sp_el0, which is already modified by restore_vm_system_regs.

It will cause a panic when you try to get current task’s pointer during VM-Exit handling.

[  0.467056 0 axruntime::lang_items:5] panicked at /home/hky/.cargo/git/checkouts/arceos-8d7257a02b016867/d4cf00a/modules/axtask/src/task.rs:385:25:
current task is uninitialized

Solution

use percpu region to maintain the correct SP_EL0 belonging to host ArceOS.

#[percpu::def_percpu]
static HOST_SP_EL0: u64 = 0;

fn save_host_sp_el0() {
    unsafe { HOST_SP_EL0.write_current_raw(SP_EL0.get()) }
}

fn restore_host_sp_el0() {
    SP_EL0.set(unsafe { HOST_SP_EL0.read_current_raw() });
}

[hky1999]

@aarkegz Please check this PR.
@guoweikang I think its more proper to store current SP_EL0 in percpu area.

[hky1999]

ref: how kvm in aarch64 maintain the sp_el0 for host
https://github.com/torvalds/linux/blob/baeb9a7d8b60b021d907127509c44507539c15e5/arch/arm64/kvm/hyp/entry.S#L35